ec.europa.eu · Project City4Age Grant Agreement #689731 D2.11 City4Age frailty and MCI risk model 1/183 Elderly-friendly city services for active and healthy ageing City4Age frailty

Project City4Age Grant Agreement #689731

D2.11 City4Age frailty and MCI risk model 1/183

Elderly-friendly city services for active and healthy ageing

City4Age frailty and MCI risk model

Deliverable ID D2.11

Version 1.0

Contractual delivery date 30/11/2018

This version delivery date 28/11/2018

Status1 Final

Dissemination level2 PU

Leading partner MMED

Contributors UNIPV

This project has received funding from the European Union’s Horizon 2020 research and innovation

programme under grant agreement No 689731

1 ToC (v# = 0.0), Draft (v# < 1.0), Final (v# = 1.0), Improvement (v# > 1.0)

2 PU: Public, CO: Confidential, only for members of the consortium (including the Commission Services)



History of changes

Version Date of issue Author(s) Description

0.1 04/11/2018 MMED Draft for peer review

1.0 28/11/2018 MMED Delivered version



Table of contents

1 Abstract ..................................................................................................................... 4

2 Executive summary .................................................................................................. 5

3 Introduction .............................................................................................................. 6

3.1 The case for City4Age unobtrusive behaviour monitoring ....................... 6

3.2 Measuring health......................................................................................... 9

4 City4Age risk modelling ........................................................................................ 14

4.1 Current methods to assess MCI and Frailty risk ...................................... 14

4.2 Assessing MCI and Frailty risk through unobtrusive technology ........... 18

5 Statistical assessment ............................................................................................. 30

5.1 Introduction ............................................................................................... 30

5.2 Correlation analysis .................................................................................. 30

5.3 Machine learning experiment ................................................................... 31

6 Lessons learned and recommendations.................................................................. 41

6.1 Lessons learned ......................................................................................... 41

6.2 Analysis of coverage from Pilots ............................................................. 44

6.3 Technical improvements........................................................................... 46

6.4 Towards a Machine Learning based approach......................................... 48

6.5 Towards market deployment .................................................................... 53

7 Conclusions ............................................................................................................ 55

8 Annex: Current Geriatric Instruments ................................................................... 56

8.1 MCI Instruments ....................................................................................... 56

8.2 Frailty Instruments .................................................................................... 73

9 Annex: Analysis of Items ....................................................................................... 87

10 Annex: Coverage of Geriatric Factors .......................................................... 128

11 Annex: Measures at City4Age Pilots ............................................................ 132

12 Annex: Correlation matrices for Pilots measures ......................................... 152

12.1 Athens Pilot ............................................................................................. 152

12.2 Birmingham Pilot .................................................................................... 154

12.3 Lecce Pilot............................................................................................... 155

12.4 Madrid Pilot ............................................................................................ 165

12.5 Montpellier Pilot ..................................................................................... 167

12.6 Singapore Pilot ........................................................................................ 169

13 Annex: FAI data from Pilots ......................................................................... 170

13.1 Athens Pilot ............................................................................................. 170

13.2 Birmingham Pilot .................................................................................... 171

13.3 Madrid Pilot ............................................................................................ 172

14 Annex: Data pre-processing SQL for the Athens Pilot experiment ............. 173

15 Annex: MultiSchemeAUC modifications ...................................................... 182

16 Annex: Computing *_time_per_visits features ............................................ 183



1 Abstract

The model represents a formalization of current geriatric knowledge which is relevant to MCI and

frailty risk detection, in the City4Age framework.

The main traits of the City4Age geriatric risk modelling are based upon the notions of “GEF” and

“GES” (Geriatric Factors and Sub-factors, i.e. conceptualizations of established geriatric determinants

of frailty/MCI) that can be synthesized from “Measures”, i.e. numbers generated by the various

technologies and methods for collecting data. Measures are analysed in various ways, generating

statistical indicators (called “NUI” in the project) that can be used to actually estimate GEFs and GESs.

Model’s NUIs can also be the basis (features) for building classifiers able to automatically screen

robustness in the elderly population. A proof-of-concept has been produced by training one such

classifier, based on real-life experimental data from the City4Age Athens Pilot site.

The deliverable includes proposals for further improvement, in terms of both (i) future data analytics

investigations that can reveal more about the inner structure of the risk model, and (ii) actions that

City4Age Partners interested in the risk model’s market deployment can put in place to achieve

industry-grade accuracy.



2 Executive summary

A large number of instruments and methods for measuring the risk of MCI and frailty onset have been

proposed in the medical literature, and many of them have been clinically validated and are routinely

used in the geriatric practice.

However, such instruments are based on data collection methods (administration of questionnaires,

meter-based measurement, direct observation) that present several drawbacks in terms of high

deployment costs and reduced timeliness of risk detection.

City4Age aims at harnessing several important technology developments, which appeared on the scene

in the last decade, that offer new ways to help achieve systematic health monitoring and early health

risk detection, targeting elderly populations in smart-cities.

To achieve this goal, a first necessary step has been to accurately survey current geriatric knowledge,

with the aim of identifying health indicators potentially usable within the City4Age paradigm. Such

indicators must feature the following characteristics:

Be routinely applied in clinical settings to diagnose and/or predict the onset of MCI and

frailty conditions in individuals

Be linkable to behavioural traits that can be measured by sensors and other datasets readily

available in the smart-city environment.

Overall, 8 instruments for MCI and 11 instruments for frailty have been identified and analysed.

The analysis revealed that geriatric practice has largely agreed on the identification of the most

important functional domains that are implicated in the course of MCI and frailty onset – examples are:

Mobility, Physical activity, Basic and Instrumental Activities of Daily Living, etc.

Based on this knowledge, a comprehensive, “first-in-the-art” City4Age geriatric risk model has been

outlined.

The model consists of a number of Geriatric Factors (GEFs), possibly subdivided into Geriatric Sub-

factors (GESs), that parallel functional domains with relevant detection/predictive power. Factors can

be gauged through Measures, derived from the application of unobtrusive sensing technology or other

smart-city datasets – examples are WALK_DISTANCE, OUTDOOR_NUM, PUBLICTRANSPORT_TIME, etc.

Two complementary approaches are enabled by such model:

Build Individual Monitoring Dashboards (WP5) that exploit the GEF/GES/Measure

hierarchy to provide clinicians with a comprehensive, visually organized, multi-factorial

geriatric assessment, building on domains that medical professionals are already familiar

with and already know how to interpret

Use Measures to extract relevant features (called “numerical indicators” or NUIs, in the

Project) that can be combined to form “behavioural markers” for frailty/MCI, able to

automatically discriminate between robust and non-robust subjects

Based on the above, forms of effective semi-automated screening of the elderly population in smart

cities can be envisaged.

Proposals for future improvement are discussed, addressing in particular:

Further data analytics investigation avenues, able to capture the full richness and structure of

the model through a Bayesian Network approach (e.g. better identification of conditional

dependence relationships among GEFs, GESs and Measures)

Practical steps that Partners can conduct to achieve industry-grade model accuracy, by

extending model coverage with new unobtrusive data collection technologies and by

optimizing existing ones

With respect to its previous version (i.e. deliverable D2.06), this deliverable presents the following

differences:

Sections 1, 2, 6, and 7 have been fully revised

Section 5 and Annexes 12 to 16 have been added anew

Section 3 and 4 and Annexes 8 to 11 are mostly unchanged (with the exception of subsection

4.2.4.6, which has been partially revised)



3 Introduction

The overarching objective of the City4Age project is to demonstrate that data collection in smart cities

can significantly improve the management of health risks in the elderly populations (in particular, MCI

and frailty) by harnessing the potential of new technologies for unobtrusive behavioural sensing, which

are more and more pervading urban life.

In order to achieve this objective, one of the important areas to be tackled relates to the City4Age early

risk detection subsystem.

In particular, as a starting point, it is important to:

At a first level of detail – state the overall case for applying unobtrusive behaviour

monitoring to address MCI and frailty risk detection; compare it with the methods currently

used in geriatrics for managing this task; and derive an overall City4Age risk detection

concept

At a second level of detail – examine how currently established geriatrics instruments for

measuring health risks are structured, and propose new City4Age approaches that can

effectively take advantage of new technologies for unobtrusive behaviour measurement

The City4Age risk modelling effort – conducted in task T2.1 Modelling risks and resilience profiles

and reported in the rest of this deliverable – aims at structuring the consequences from the above

mentioned thinking into a conceptual framework, to be used as a guidance in the technical design and

implementation of the risk detection subsystem. Together with behavioural modelling – elaborated in

task T2.2 Modelling MCI/frailty related behaviours and reported in deliverable D2.2 – it constitutes the

scientific base for the work planned in the technical work-packages, from WP3 to WP6.

This section briefly introduces the rationale behind these elements and paves the way for the following

Sections, that report on the modelling approach chosen for City4Age.

3.1 The case for City4Age unobtrusive behaviour monitoring

It is well known, especially in geriatrics, that the early detection of risks relating to a specific health

condition improves the chances of enacting appropriate interventions that can halt or at least delay the

condition itself, with beneficial effects on both patients’ quality of life and cost of treatment3.

For instance, in the case of MCI, Petersen et al.4 emphasize the need for the clinician to detect the

earliest signs of cognitive impairment and highlight the importance of this quest. In fact, as MCI is

recognized as a possible precursor of AD, its early detection and subsequent intervention can help to

delay the onset of the disease. Considering, for instance, that an estimated 5.3 million Americans had

Alzheimer's disease in 2015 and the overall cost of caring for these people is around $226 billion5 (i.e.

more than $40,000 per patient per year) detecting and treating MCI such that the progression to AD

can be delayed by even a single year, would imply very important savings. In addition, this analysis

only considers the economic facet of the issue, which is easier to measure, but even more important is

the impact on the quality of life of elderlies and of their carers.

However, under current conditions, MCI detection is a challenge.

Earliest signs of MCI may consist, for instance, in some degree of forgetfulness, beyond what is

justified by normal aging. The forgetfulness may be apparent to those closest to the person but not to

the casual observer6. In fact, relying, as current practice, on self-reported detection of MCI signs and

symptoms (by the patient or by those around her/him) has several drawbacks:

3 Onder et al., Measures of physical performance and risk for progressive and catastrophic disability: Results

from the women’s health and aging study, Journals of Gerontology: Medical Sciences, 2005

4 Petersen et al., Practice parameter: early detection of dementia: mild cognitive impairment (an evidence based

review), Neurology, 2001

5 http://www.alz.org/facts/

6 Petersen, Mild Cognitive Impairment, The New England Journal of Medicine, 2011

http://www.neurology.org/content/56/9/1133.long

http://www.alz.org/facts/



It may not be easy to discriminate signs and symptoms of MCI insurgence from normal

effects of aging

Patients and their carers may not be knowledgeable about exactly which signs or symptoms

to look for

Some patients, especially older ones, may be subject to forms of denial against worsening

conditions (e.g. minimizing symptoms) to avoid care7.

Similar considerations apply to frailty8. Frailty has been shown to be directly correlated with high risk

for adverse health outcomes, such as falls or hospitalization, that increase care costs and decrease

quality of life. On the other hand, when detected early, possibly at the pre-frail stage, frailty can

potentially be prevented or treated. For this reason, the need to recognize frailty in a timely manner is

noted as one of the top priorities in both gerontology, general practice, and public health.

Several important technology developments, which appeared on the scene in the last decade, may offer

new ways to address this type of issues, and help to achieve a systematic health monitoring and early

health risk detection approach.

A paradigmatic example of these technologies is the modern smartphone which – packed with sensors

that can capture many features, such as e.g. orientation, acceleration, location or voice, and equipped

with local computational power and an always-on network connection – can seamlessly interact with

other computer systems in order to conduct complex analysis on the generated data streams and infer

interesting aspects of human behaviour. The smartphone proliferation among users and its continuous

presence and usage, make it the ideal unobtrusive human activity data collection platform9.

In addition, smart cities environments offer extra opportunities, as ever more urban infrastructure is

deployed on the basis of technologies such as RFID cards (e.g. for check-in services), proximity

devices (e.g. BLE beacons for in-shop proximity detection), or intelligent meters of various kind (e.g.

for energy or water consumption)10

.

Prospects are also offered by “software sensors”, such as social network logging applications.

The endless possibilities arising from the combination of these technologies allow to measure, monitor

and analyse human behaviour in unprecedented ways.

With reference to the objectives of City4Age, and among many similar works that can be found in the

literature, Rantz et al. report on a simple yet paradigmatic case that shows what can be practically

achieved11

.

The authors installed an integrated sensors network in apartments of volunteer residents, hosted in an

“aging in Place” retirement community that allows residents to remain in their apartments even if their

health deteriorates.

Among others, the sensors network included several passive infrared (PIR) motion detectors installed

in various locations, to detect presence in different rooms and to consequently infer specific activities.

In particular, the objective of the sensors network is to detect changes in activities that could be linked

to respective changes in health status, and to consequently offer relevant clinical interventions to help

residents age in place.

Alerts are generated and sent to clinicians whenever sensor activity (within definite daytime frames)

deviates from normal (in the particular example, the cut-off point was established at 4 standard

deviations from the mean of the previous 14 days, and it was chosen conservatively, to err on the side

of generating too many alerts rather than possibly missing a crucial one).

7 Trull et al., Ambulatory assessment, Annual Review of Clinical Psychology. 2013

8 Lucas et al., Frailty in the older adult: will you recognize the signs?, Nurse practitioner, 2014

9 Lathia et al., Smartphones for Large-Scale Behavior Change Interventions, IEEE Pervasive Computing, 2013

10 Hancke et al., The Role of Advanced Sensing in Smart Cities, Sensors, 2013

11 Rantz et al. Using Sensor Networks to Detect Urinary Tract Infections in Older Adults, IEEE 13th International

Conference on e-Health Networking, Applications and Services, 2011

http://www.mdpi.com/1424-8220/13/1/393



It is important to note that alerts are not diagnosis. In fact, whenever an alert is generated, the resident

is evaluated by a registered nurse, in order to check if the situation warrants deeper medical

investigation.

In this context, the authors report about several case studies related to the early detection of urinary

tract infections (UTI), a serious health threat for older women that, when treated too late, may lead to

kidney damage, system-wide infections or even death.

Since a common indicator of UTI observed by clinicians is urgency and frequent urination, particularly

at night, alerts based on increased nightly activity of the bathroom PIR motion detector have been used

to test residents for UTIs.

The authors show how, out of three case studies, two of them conducted to an effective early diagnosis

of UTI, which allowed a correspondingly early treatment and full recovery of the patient (the third one

being a false positive). If the diagnosis would have been delayed until the elderly person would have

self-reported relevant symptoms to her GP or caregivers, the situation would have been much worse,

with a less certain and more costly outcome.

Although relatively simple and limited to the indoor environment, this example is paradigmatic of the

state-of-the-art in the field, and clarifies how unobtrusive and continuous behavioural sensing, paired

with an appropriate data interpretation and alerting mechanism, can concretely contribute to the

improvement of health management for elderly people and to the reduction of related care costs.

On this basis, it is possible to highlight some essential characteristics that should also be replicated in

the City4Age risk detection approach (refer to Figure 1 below):

Suitable data streams, coming from sensors and other datasets available in the smart city

environment, shall be linked to valid and reliable health status indicators, established in the

medical practice and associated with specific behavioural patterns, and used to address

relevant health risks through (semi-)automated behaviour recognition and analysis

The ultimate objective of the risk detection subsystem is to interpret data and consequently

generate alerts, against which a conventional medical investigation will possibly lead to

diagnosis and intervention decisions (note that it is out of scope for the system to

automatically generate diagnosis). Alerts are accompanied with information, collected from

the data streams, which is presented to medical personnel through appropriate data

dashboards, in order to support clinical assessment and decision making

Figure 1. Risk detection subsystem in City4Age

A central element to consider in the design of the detection, alerting and data interpretation

mechanism is the achievement of the traditional maximization/balance among higher

City4Age Risk detection subsystem

Alerting mechanism Data dashboards

Visit (diagnosis, interventions)

Behaviour

Datasets

Health indicators



specificity (reduce false positives) and higher sensitivity (reduce false negatives). It is

important to note that this aspect has a clear and direct link with ROI estimation, as (i) true

positives represents actual savings and quality of life improvements made possible by the

enactment of early medical interventions; (ii) false positives represent costs sustained for an

additional, unnecessary medical investigation of a healthy person and (iii) false negatives

represent lost opportunities to provide better care and quality of life to subjects that would

have needed it (in City4Age, this analysis is to be carried out as part of Task T8.2

Exploitation plan and sustainability).

3.2 Measuring health

In order to implement the scheme illustrated in the previous section, a crucial element is the availability

of suitable health measurement indicators, that are:

valid and reliable

able to predict the onset of critical health conditions in individuals (MCI and frailty, in the

case of City4Age)

linked to behavioural traits that can be feasibly measured by sensors and other datasets

available in the city environment.

The effort to derive high quality health measurement indicators is a longstanding one, that will

probably always continue. In the last decades, many indicators of the health or well-being of

individuals have been developed, in order to address three ultimate aims12

:

Diagnosis of illness

Predict the need for care

Evaluate the outcomes of treatment

This is in agreement with the distinction made by Bombardier and Tugwell between the three purposes

of health measurement indicators: diagnostic, prognostic, and evaluative13

.

In particular, the second item – relating to the prognostic/predictive value of indicators – is the one of

interest to City4ge.

In this respect, it is also worth to note that several indicators have been validated for both concurrent

validity and predictive validity, depending on whether the focus is on current or future health status.

For instance, a questionnaire on hearing difficulties may be compared with the results coming from an

audiometric test to assess its concurrent validity, or the same questionnaire may be compared with

future patient’s health outcomes, to assess its predictive validity. It is worth to note that, in the context

of early detection envisaged by City4Age, concurrent validity is also important.

Given the above frame it is important to survey and analyse existing indicators, in order to identify how

they could meet City4Age requirements, and be used to constitute the foundation of the project’s risk

detection model. As a first step, we discuss in details the structure of existing indicators, in order to

base relevant City4Age proposals later on.

3.2.1 Structure of instruments

Defining health measurement indicators normally implies assembling a selection of elementary items,

that are intended to represent the specific clinical aspects that are of interest.

Underlying each indicator is a model, that takes item values as input and – according to relevant

algorithms derived through various techniques, such as domain expertise application, statistics

modelling, knowledge engineering, machine learning, etc. – produces a final outcome that is the

measure that was sought for.

12

McDowell, Measuring Health: A Guide to Rating Scales and Questionnaires, Oxford University Press, 2006

13 Bombardier et al., Methodological framework to develop and select indices for clinical trials: statistical and

judgmental approaches, Journal of Rheumatology, 1982



Examples of elementary items may be a measure on a specimen analysed in a laboratory, the flexion of

a limb observed by a physiotherapist, an estimate of working capacity assessed by a clinician, or a self-

reported behaviour obtained through a questionnaire response.

Items may in turn be grouped into specific categories or domains. For example, in the Mini-Mental

State Examination indicator, the domain “Orientation to time” includes five questions, from broadest to

most narrow, such “What year is this?”, “What season is this?”, “What month is this?”, “What is

today’s date?”, “What day of the week is this?”.

Sometimes, domains correspond to a single item, such as in the Lawton Instrumental Activities of

Daily Living (IADL) Scale, where domains such as “shopping”, “ability to use the telephone”, “mode

of transportation”, directly correspond to eponymous items to be measured.

Ultimately, each item is associated with (a set of) values, among which the item measure, or score, is

drawn. Item scores then contribute – through application of the underlying model, as above mentioned

– to the generation of the instrument outcome score.

In current medical literature there is no uniform terminology to describe these notions, and several

different terms – e.g. instruments, measures, measurement methods, indicators, scales, etc. – are used

more or less interchangeably, sometimes even indicating different concepts.

On the other hand, in City4Age modelling it is important to agree on some kind of formal definition of

the above ideas, upon which to base the subsequent design and implementation work.

For this reason, the specific schema illustrated in Figure 2 below is proposed.

Figure 2. Health measurement instruments strcture

In the figure, the following definitions apply:

Instrument: the overall measuring method, aimed at providing a final, comprehensive score

on a certain health entity

Category: a group of Items (see below) that contribute to the appraisal of a certain facet of

the measured health entity

Items: the actual, basic elements that will be measured to produce scores (see below).

Values: set of values associated to an Item. The Item’s measurement produces a specific

score, drawn from this set.

Model: the mathematical entity (algorithms) that allows to compute the Instrument outcome

score on the basis of the scores measured for single Items.

Examples of the application of these definitions to two established Instruments are illustrated in the

following table.

Category

Item Values

Instrument

1..*

1..* 1

Model 1

1..*



Instrument The Nottingham Extended Activities

of Daily Living

Fried Frailty Index

Example of

Category

Transportation Motility

Example of

Item

Question “Did you travel on public

transport?”

Test “time to walk 15 feet”

Example of

Score

“Not at all”, “With help”, “On your

own with help”, “On your own”

Number of seconds

Example of

(part of) Model

Count 1 if answer is “On your own

with help” or “On your own”, 0

otherwise

Count 1 if threshold is exceeded (Note: threshold depends on sex and

height – not reported here to save space)

Table 1. Example of Instrument definitions

The above terminology has been devised with the ultimate aim of facilitating communication across

the interdisciplinary knowledge areas represented in the project, including geriatricians, behavioural

scientists, information technology engineers, researchers, and data scientists. For this reason, some

trade-offs have been made, e.g. by using the term “Category” instead of the term “Domain”, often used

in geriatrics to denote groups of functionally related items, because the term “Domain” has a strong but

unrelated connotation in the information technology field.

An important observation regards the meaning of the term “score”, as used in the above definitions.

While in principle the outcome of measuring an Item (or the outcome of the overall Instrument) can

always be reduced to a numerical score, the following different types apply:

Nominal or categorical score: numbers are used just as labels for categories (e.g. 0 for false

and 1 for true, for a classifier outcome)

Ordinal score: in addition to the previous type, order is meaningful and represents growing

quantities (e.g. 1 = mild, 2 = moderate, 3 = strong)

Interval score: in addition to the previous type, difference among numbers is meaningful, i.e.

there exists the concept of unit of measure (e.g. temperature in oC)

Ratio score: in addition to the previous type, there is a meaningful zero point, so ratios

among scores are also meaningful (e.g. number of seconds to complete an action)

3.2.2 Instruments for City4Age

While in the 1970s the medical field has been characterized by an intense proliferation of new

Instruments and related models, that were often created relatively hastily and sometimes with

insufficient substantiation, the more recent trend is to focus on a less wide number of high quality

Instruments, making validation and reliability assessment the priority.

Following this trend, the best strategy for City4Age is to also concentrate on a well-chosen set of

existing Instruments, selected on the basis of the following characteristics, as suggested in the opening

to this subsection:

Predictive value for the conditions addressed by the project (MCI and frailty) and/or ability

of timely detect the onset of these condition

Measurability in behavioural terms (i.e. the model underlying the Instrument is linked to the

subject’s behaviour)

Feasibility of measurement through sensors/datasets readily available in a smart city, in the

timeframe expected for the project exploitation (as determined in WP3 and in Task T8.2

Exploitation plan and sustainability)

These points deserve additional discussion.

Concerning measurability in behavioural terms, three approaches are possible (refer to

Figure 3):



Use existing indicators, that rely on models directly linked to behaviours (part a in the

Figure), such as for instance the Lawton Instrumental Activities of Daily Living Scale,

which includes Item values such as “Takes care of all shopping needs independently”. In this

case, the City4Age behaviour reconstruction function shall detect the current behaviour (e.g.

the above “Takes care of all shopping needs independently”) while the risk detection

function should calculate a relevant score and possibly generate an alert if, according to the

model rules, a change is detected (e.g. from “Takes care of all shopping needs

independently” to “Shops independently for small purchases”).

Use existing indicators, that rely on models that, although not directly linked to behaviours,

can be feasibly correlated to them (part b in the Figure). For instance, the Fried Frailty Index

includes the test “time to walk 15 feet”. Although this is not directly a behaviour

(measurement is taken manually, by a physician with a chronometer), the relevant input to

the underlying model (i.e. walking speed) can be correlated to behaviour, for example by

monitoring daily walking speed through smartphone sensors.

Devise new geriatrics Instruments, based on new models – for example, based on machine

learning techniques – that directly connect the sensors and datasets information, gathered by

the City4Age data collection subsystem, to a relevant outcome score (part c in the Figure).

At a first level, the first and second approaches will be the preferred ones in City4Age.

In fact, by relying on existing Instruments for the detection of MCI and frailty onset, which are already

established and accepted by geriatrics researchers and clinicians, the City4Age risk detection

subsystem can count on their proven validity and reliability characteristics, as derived from decades

of geriatrics research and clinical practice, as well as on a higher level of trust from clinicians.

Figure 3. Determining City4Age model

On the other hand, the third approach would be a relatively innovative one. In fact, it is worth to note

that established Instruments have been conceived, validated and published at a time when the very

mechanisms for personal data collection envisaged by City4Age were not generally available to the

medical community. Today, such availability brings in the possibility that correspondingly new

indicators, that directly relate sensor readings to relevant outcome scores, could be developed and

Reconstruct behaviour

Apply existing

Instrument

Sensors and

datasets

Score

Reconstruct behaviour

Apply existing

Instrument

Sensors and

datasets

Score

Apply new Instrument

Sensors and

datasets

Score

Infer input for existing Instrument

(a)

(b)

(c)



tested. In fact, research in using unobtrusive sensing for direct health measurement is in its infancy, and

few studies in this direction have started to appear in the recent literature14

.

Although this is a very stimulating approach, it is also more challenging for two reasons:

Models that directly link data features to MCI or frailty outcomes would be opaque with

respect to the underlying behaviours, and might not sufficiently clarify the link between

specific changes in behaviour and consequences on MCI and frailty; this makes it

problematic for geriatricians to accept such outcomes (this is a general and well-known issue

in the application of automatically, machine-learned models to clinical practice, which is not

specific to City4Age alone)

Relevant machine learning algorithms currently proposed in the literature are generally at the

very early stages of development and, to the best of the City4Age team’s knowledge,

although related research has been conducted in labs or smart-home settings, at the time of

writing no significant effort is yet available for measuring frailty and MCI on the basis of

unobtrusive technologies and datasets deployed in city-wide contexts, for monitoring city-

wide behaviours.

These challenges notwithstanding, the project has still made an effort to preliminary investigate this

third approach in order to at least generate educated recommendations for researchers who are willing

to undertake the full endeavour in the near future. Moreover, several technical achievements in

City4Age (e.g. data collection subsystem, the human activity reconstruction subsystem, the data

analytics platform for risk detection, etc.) represent important enablers for furthering this line of

investigation. In fact, in its last phases the Project has even gone someway further, by proposing a

kind of “hybrid” approach, in which a reasonably performant machine learning component, rather than

being used as “stand-alone”, is applied as a first step in a multi-level screening scheme, that could be

actually implemented in practice. Details are discussed in Sections 5 and 6.

A final note regards technical feasibility. It has been hinted that, when selecting Items to consider for

the City4Age risk detection subsystem, it is also necessary to take into account how the inputs to the

underlying models can be reliably obtained from sensors and other smart-city datasets. For instance

(with reference to previous examples), it may be relatively easy to link an indicator like the number of

nightly visits to the bathroom to a dataset coming from a PIR motion detector, installed in the

bathroom itself (like in the UTI early detection example), but it is quite a different story to understand

in a reliable way if a person is taking care of all her/his shopping needs or if she/he started to shop only

for small items (as in the case of the Lawton IADL scale application). Although seeking technical

feasibility is a crucial task of City4Age work-packages WP3-WP5, relevant issues in this area will still

be consistently considered in this deliverable, in order to ensure that the modelling approach is

conducive to reasonable technical implementations.

14

Ben-Zeev et al., Next-Generation Psychiatric Assessment: Using Smartphone Sensors to Monitor Behavior and

Mental Health, Psychiatric Rehabilitation Journal, 2015



4 City4Age risk modelling

To concretely apply the concepts introduced in the previous Section to the management of health risks

in the elderly populations, it is important to establish how such health risks are to be precisely

characterized.

The project has chosen to concentrate on MCI and frailty, two typical conditions of aging that

influence multiple determinants of health, including both physical and cognitive ones.

Accordingly, in this Section we present how City4Age models MCI and frailty risks in a way that can

be used to support the development of an effective, technology-based early detection system.

In particular, the Section will first illustrate how MCI and frailty are measured today in clinical

practice. On that basis, the challenge of achieving the same measurement through unobtrusive sensing

technology is then addressed and a model to underpin such technology-based assessment – i.e. the

City4Age Geriatric Risk Model – is outlined and discussed.

4.1 Current methods to assess MCI and Frailty risk

Although there is no agreed gold standard to measure MCI or frailty, geriatric medicine has

nevertheless conceived, proposed and clinically validated several Instruments that are routinely

used since many decades in clinical practice for detecting (and in some cases predicting) the onset of

these syndromes.

As such Instruments represent well the currently available medical knowledge on the issue, they are the

natural starting point to formulate a computational risk model to be used with the City4Age approach.

For this reason, in this subsection, after providing a brief introduction to both MCI and frailty, we will

illustrate a survey and analysis of currently established Instruments that has been conducted in the

frame of Task T2.1 (Modelling risks and resilience profiles) and present relevant results, on which the

subsequent discussion is based.

4.1.1 Brief introduction to MCI

Mild cognitive impairment (MCI) represents an intermediate state of cognitive function between the

changes seen in aging and those fulfilling the criteria for dementia and often Alzheimer’s disease.

MCI is classified into two categories:

amnestic MCI, which is characterized by memory impairments that still do not meet the

criteria for dementia, while other areas (e.g. attention, language, executive function,

visuospatial skills) are not affected

non-amnestic (anamnestic) MCI, which impacts non-memory related areas

Amnestic MCI is more common than anamnestic MCI and it has been related to the onset of AD.

Non-amnestic MCI has been related to other kinds of dementia, not linked to AD, such as

frontotemporal lobar degeneration or dementia with Lewy bodies.

Prevalence of mild cognitive impairment ranges from 10 to 20% in persons older than 65 years of age.

For example, the Mayo Clinic Study of Aging, a prospective, population-based study of persons

without dementia who were between 70 and 89 years of age at enrolment, found a prevalence of

amnestic mild cognitive impairment of 11.1% and of non-amnestic mild cognitive impairment of 4.9%.

Given its role as a possible precursor of more severe forms dementia, MCI has recently received a lot

of attention in clinical practice as well as in research settings.

In fact, on one side, its detection can help to enact effective disease-delaying lifestyle interventions,

and, on the other side, being the earliest manifestation of cognitive disorders, it may also be important

in formulating research hypotheses.

Discriminating between MCI and normal aging can be a challenge. Slight forgetfulness, e.g.

misplacing objects and having difficulty recalling words, can be a normal effect of aging.



Amnestic MCI is characterized by a more prominent memory impairment. For example, patients may

forget important information that they used to remember easily, such as appointments, telephone

conversations, or recent events of interest to them (e.g., sporting events), while all other aspects of

function are normally preserved.

The American Academy of Neurology recommends the following criteria for an MCI diagnosis:

self-reported memory problems, preferably confirmed by another person

greater-than-normal memory impairment, measured with standard tests

normal thinking and reasoning skills

no impairments in ADLs

Currently, there is no approved pharmacological intervention for MCI.

There is evidence of benefits from cognitive rehabilitation interventions (e.g. use of mnemonics,

association strategies, computer assisted training programs) for amnestic MCI patients15

.

Other recommended interventions include aerobic exercise, involvement in intellectually stimulating

activities, and participation in social activities, given that these – even if not fully confirmed – might be

beneficial and pose little risk.

4.1.2 Brief introduction to frailty

Frailty is a geriatric syndrome of decreased reserve and resistance to stressors, resulting from

cumulative declines across multiple physiologic systems, causing increased vulnerability to adverse

outcomes16

.

People living with frailty are more at risk of dramatic health outcomes after apparent minor events, like

for instance an infection or a new medication. These outcomes include falls, disability,

institutionalization, hospitalization or even mortality.

On the other hand, frail people may be relatively low users of health care services, and be little known

to their GP, until they undergo a major health decline, possibly as a result of a marginal episode.

There is evidence that older people may not recognize themselves as living with frailty and do not

accept to be considered as ‘frail’, a term that is heavily associated with vulnerability and dependence17

.

There are two broad models of frailty:

the Phenotype model, based on a group of patient characteristics (unintentional weight loss,

reduced muscle strength, reduced gait speed, self-reported exhaustion and low energy

expenditure) which, if present, can predict poorer outcomes

the Cumulative Deficit model, that assumes an accumulation of deficits (ranging from

symptoms e.g. loss of hearing or low mood, through signs such as tremor, through to various

diseases such as dementia) which can occur with ageing and which combine to increase the

‘frailty index’ which in turn will increase the risk of an adverse outcome

Usually, three ordered frailty levels are identified: frail, pre-frail and robust (non-frail).

In City4Age, the ‘pre-frail’ level is an important one, as it has been referred to in the literature as a state

of less ‘inevitability’ that may be more amenable to interventions than the frail state18

. In fact, frailty

may be preventable and early detection and interventions can minimize transitions from the pre-frail

(and pre-disabled) state to the frail state, reduce the chance of adverse outcomes, and reduce healthcare

costs19

.

15

Jean et al., Cognitive intervention programs for individuals with mild cognitive impairment: systematic review

of the literature, The American journal of geriatric psychiatry, 2010

16 Fried et al., Frailty in Older Adults: Evidence for a Phenotype, Journal of Gerontology, 2001

17 British Geriatrics Society, Fit for frailty (Report), 2014

18 Gill et al., Transitions between frailty states among community living older persons, Archives of internal

medicine, 2006

19 Morley et al., Frailty consensus: a call to action, Journal of the American Medical Directors Association, 2013



The prevalence of frailty in community-dwelling older Europeans (65 years and older) varies between

5.8% and 27.3%. In addition, between 34.6% and 50.9% are classified as ‘pre-frail’20

.

Current recommendations to detect frailty include:

periodic social service assessment

review after referral for community intervention

primary care review when interacting with older people

assessment from home carers

assessment by ambulance crew, when called e.g. for a fall or other urgent matter

In terms of prevention, unhealthy behaviours that are implicated include:

insufficient physical activity, particularly resistance and aerobic exercise, which is beneficial

in preventing and treating the physical performance component of frailty

poor diet, particularly in terms of suboptimal protein/total calorie intake and vitamin D

insufficiency

In relation to this, the Action Group A3 of the European Innovation Partnership on Active and Healthy

Ageing has specific areas related to Food and Nutrition and to Physical Activity that investigate how

these aspects can be addressed in order to promote frailty prevention and to enact related multi-modal

interventions.

4.1.3 Analysis of Instruments to measure MCI and Frailty risk

To conduct the analysis of currently used methods for detecting and predicting MCI and frailty, a total

of 19 Instruments have been selected and surveyed: 8 for MCI and 11 for frailty.

The selection has been conducted on the basis of the requirements expressed in Section 3, including in

particular:

capability of the Instrument to detect or predict the onset of MCI or frailty, respectively, as

documented in the literature

possibility to (at least partially) implement the Instrument through behavioural monitoring

prospected technical feasibility of unobtrusive data collection and/or activity recognition

It is important to note that, with reference to the last two points, the selection presented here is rather

liberal, preferring to include Instruments that may eventually reveal as unmeasurable through

behavioural monitoring or technically unfeasible, rather than incurring the risk of leaving out good

ones.

The full results of the analysis work are reported in Annex, in Section 8.

For each Instrument listed in Section 8, the following elements have been compiled:

overall description, including the Instrument’s motivation, fundamental characteristics, and

why it is worth considering in City4Age

reference to literature, pointing to full information about the Instrument as well as to studies

that have addressed its diagnostic or predictive validity, as relevant

a synoptic view of the structure of the Instrument, along the definitions given in Section 3.

In particular, for each Instrument, this view includes:

o the list of Items

o for Items whose values correspond to potentially trackable behaviours, a full list of

those values

o the Category (i.e. geriatric domain) to which each Item belongs

o Moreover, in order to allow reference and tracking, each Item is given a unique ID,

which also identifies the Category and the Instrument to which it belongs.

20

Santos-Eggiman et al., Prevalence of frailty in middle-aged and older community-dwelling Europeans living in

10 countries, The journals of gerontology. Series A, Biological sciences and medical sciences, 2009



It is important to observe that the underlying geriatric domains probed by the surveyed Instruments (as

denoted by Items’ Categories) are generally shared from a relatively well recognizable and common

pool of few dozens of them.

This observation is a meaningful hint that – broadly speaking – geriatric practice has roughly agreed

on the identification of the most important domains implicated in MCI and frailty onset.

Code Category Remarks

Mt Motility21

Slowness, ambulation, balance

Ac Activity Physical activity level, energy expenditure

Ad ADLs Basic activities of daily living in general

Ia IADLs Instrumental activities of daily living in general

Fo Food Meal and/or food preparation

Ho Housekeeping

Ln Laundry

Co Communication Telephone and other types of communication

Sh Shopping

Tr Transportation Transportation beyond walking distance

Fi Finances Ability to manage own finances

Me Medication Ability to manage own medication

So Socialization

Cu Culture Cultural or entertainment activities

En Environment

Dp Dependence Dependence/disabilities and social support

He Health General health status, illnesses, sensory issues

We Weight

Wk Weakness

Ex Exhaustion Exhaustion, fatigue, energy reserve level

Ab Abstraction

At Attention Attention and calculation

Mr Memory Recall (short-term verbal memory)

Mo Mood Mood or depression issues

Ti Time Orientation to time

Sp Space Orientation to space

Vi Visuospatial Visuospatial/constructional praxis

La Language

De Demographics

Table 2. Categories for diagnosis/prediction of MCI/frailty in current Instruments

Given the crucial importance of this fact for the project endeavour, an effort has been made to extract

and list such domains, resulting in the array presented in Table 2 above.

21

with hindsight, “mobility” would be the correct term for this domain; however, to maintain consistency with

other project’s deliverables, we stick to “motility” asking the reader to keep this note in mind



It has to be noted that the above classification has not been always easy, and in some cases Items may

justifiably address more than one Category. In these cases, an effort has been made to identify the main

one and assign the Item to that Category.

This information – reported in full details in Section 8, to which the reader is referred – provides the

project’s common understanding about how geriatric risk is currently measured and assessed in clinical

practice, and it is used in the following as the foundation for City4Age risk modelling.

4.2 Assessing MCI and Frailty risk through unobtrusive technology

The next logical step to undertake, is to harness existing geriatric knowledge – as derivable from the

Instruments discussed in the previous sub-section – and use it to build a model able to organize data

collected through unobtrusive technology and present it to geriatricians in a form that matches such

existing knowledge, so that they can formulate relevant interpretations, according to the current norms

and practices they have been trained to use.

The result of this effort is presented in this sub-section.

Work has been based on a comprehensive, multifarious, joint effort including:

The survey of existing Instruments, presented above (as completed at M4, i.e. March 2016)

Feedback received from work-package WP3 (Unobtrusive acquisition of personal data), that

performed a revision of the proposed Instruments and Items to be used in City4Age and

provided an initial assessment of which technologies could be feasibly deployed to detect

which measures (as made available in several WP3 deliverables at M12, i.e. November

2016)

Feedback from the six City4Age Pilot testbeds (in Athens, Birmingham, Lecce, Madrid,

Montpellier, Singapore), that revised Instruments and Items and provided an assessment of

which measures they were ready to address through the deployment of which technologies

(as reported in several deliverables and subsequent mapping work conducted in the frame of

work-package WP7, from M5 to M18, i.e. from April 2016 to June 2017)

4.2.1 Two case studies

As mentioned in sub-section 3.2.2 (Figure 3, part a), one possibility for City4Age modelling would be

to re-use unmodified, existing instruments and directly measure their Items through unobtrusive

technology, substituting questions addressed to care recipients or informants, direct observation or

obtrusive measurement, with automatic data collection.

However, this straightforward solution proves to be very difficult to achieve in practice, with reduced

chances of success.

In order to better illustrate this point and understand what are the underlying challenges, an analysis of

two case studies – for the landmark instruments Fried Frailty Index (frailty) and Lawton scale of

Instrumental Activities of Daily Living (MCI) – is proposed below.

Fried Frailty Index

The Fried Frailty Index (refer for details in sub-section 8.2.1) is composed of six Items, covering five

Categories, associated to corresponding geriatric domains:

Item We.Fi.01 measures Category “Weight” through the detection of unintended weight loss

of at least 10 pounds in last year. The detection of this Item is performed by directly asking

the person

Item Ex.Fi.02 measures Category “Exhaustion” by asking the person how frequently she/he

felt that everything she/he did was an effort in last week, and classifying the response along

a 4-point Likert scale, with cut-offs 1, 3 and 5 days

Item Ex.Fi.03 also measures Category “Exhaustion” by asking the person how frequently

she/he felt that she/he could not get going in last week, and classifying a response along a

4-point Likert scale, with cut-offs 1, 3 and 5 days



Item Mt.Fi.04 measures Category “Motility” by gauging walking speed along a 15-feet

stretch with a stopwatch and assessing if it is greater than specific cut-off thresholds,

depending on gender and height

Item Ac.Fi.05 measures Category “Activity” by assessing if the number of kCal expenditure

per week is above 270. The assessment is performed by asking the person which of 18

moderately strenuous household chores she/he performed in the last two weeks and applying

a formula that includes activity-specific MET22

, body weight, number of sessions, session

duration, number of months per year the activity was done. Input parameters are obtained by

directly asking the person

Item Wk.Fi.06 measures Category “Weakness” by comparing the person’s grip strength,

gauged through a Jamar hand dynamometer, to specific cut-off thresholds depending on

gender and BMI23

.

The Fried Frailty Index is a classifier that categorizes people into three different stages: not frail, pre-

frail, and frail. The classification is performed as follows: a total score is computed by adding up 1 for

every Category that results as positive after measurement (the two Items for Exhaustion are composed

with or disjunction); then, a score of 0 corresponds to the not frail stage, a score of 1 or 2 corresponds

to the pre-frail stage, and a score of 3, 4 or 5 corresponds to the frail stage.

As it can be seen, all Items are measured obtrusively: four of them by posing direct questions to the

person under examination (Items We.Fi.01, Ex.Fi.02, Ex.Fi.03 and Ac.Fi.05) and two of them through

the application of a relevant measurement meter (Item Mt.Fi.04 with a stopwatch, and Item Wk.Fi.06

with a Jamar dynamometer).

The fundamental question here is if and how these Items can be measured through unobtrusive

technologies instead.

For two of them the answer is positive:

For Item Mt.Fi.04, a measure of walking speed can be derived by combining GPS and

accelerometer readings in a smartphone or wristband worn by the person. Some issues have

to be solved – such as for instance how to identify the right time windows when speed is

better measured – but literature shows that this can be feasibly done (see for example papers

mentioned in 8.1.8.2)

For Item Ac.Fi.05, accelerometer readings can estimate the kCal expenditure of a person by

recognizing several physical activities, including some mentioned in Fried which can be

detected by devices appearing on the market (see for instance the Withings [formerly Nokia

Health] smart-watches, that are trained to recognize more than 10 such activities24

)

For one Item the answer is a qualified yes:

Item We.Fi.01 can be measured semi-unobtrusively by fitting the home of a person under

examination with a smart scale, able to record periodic weight measures performed by the

person herself. While an explicit action form the person is required (she/he has to decide to

use the scale), this can be considered a normal, daily activity, that in most cases she would

have done anyway, while the rest of the burden (keeping track of historical measures) is

taken care of by the system

The remaining two Items are clearly infeasible, given the current state-of-the-art:

Items Ex.Fi.02 and Ex.Fi.03 require obtaining an assessment of feelings from the person

(although the area of emotional computing is studying how such feelings could be possibly

derived from available measurements, this is still an open and very challenging research

question that falls outside the scope of City4Age)

Item Wk.Fi.06 requires the application of a dynamometer, which is a necessarily obtrusive

action. To the knowledge of the project’s WP3 team, that surveyed existing technologies for

22

https://en.wikipedia.org/wiki/Metabolic_equivalent

23 https://en.wikipedia.org/wiki/Body_mass_index

24 https://www.withings.com/eu/en/watches

https://en.wikipedia.org/wiki/Metabolic_equivalent

https://en.wikipedia.org/wiki/Body_mass_index

https://www.withings.com/eu/en/watches



personal data collection, no feasible technology exist to derive muscular strength

unobtrusively.

In conclusion, the analysis above shows that less than half of the Fried Frailty Index Items can actually

be measured fully unobtrusively, making the idea of devising a direct “City4Age-enhanced” version of

such Instrument difficult to achieve in practice.

Lawton scale of Instrumental Activities of Daily Living

The Lawton scale of Instrumental Activities of Daily Living, described in details in sub-section 8.1.1,

is composed by eight Items, covering eight Categories, associated to corresponding geriatric domains:

Item Co.Li.01 measures Category “Communication” by assessing the person’s capability to

correctly use a telephone. The assessment is obtained by directly asking the person or an

informant how well the telephone is used (discriminating among behaviours such as:

“operates telephone on own initiative, looks up and dials numbers”; “dials a few well-known

numbers”; “answers telephone, but does not dial”; “does not use telephone at all”)

Item Sh.Li.02 measures Category “Shopping” by assessing the person’s capability to

independently shop for all her/his needs. The assessment is obtained by directly asking the

person or an informant how well shopping is performed (discriminating among behaviours

such as: “takes care of all shopping needs independently”; “shops independently for small

purchases”; “needs to be accompanied on any shopping trip”; “completely unable to shop”)

Item Fo.Li.03 measures Category “Food” by assessing the person’s capability to

independently prepare her/his own meals. The assessment is obtained by directly asking the

person or an informant how well meals are prepared (discriminating among behaviours such

as: “plans, prepares, and serves adequate meals independently”; “prepares adequate meals if

supplied with ingredients”; “heats and serves prepared meals or prepares meals but does not

maintain adequate diet”; “needs to have meals prepared and served”)

Item Ho.Li.04 measures Category “Housekeeping” by assessing the person’s capability to

keep her/his house clean and in order. The assessment is obtained by directly asking the

person or an informant how well housekeeping is performed (discriminating among

behaviours such as: “maintains house alone with occasional assistance (heavy work)”;

“performs light daily tasks such as dishwashing, bed making”; “performs light daily tasks,

but cannot maintain acceptable level of cleanliness”; “needs help with all home maintenance

tasks”; “does not participate in any housekeeping tasks”)

Item Ln.Li.05 measures Category “Laundry” by assessing the person’s capability to

independently take care of her/his laundry. The assessment is obtained by directly asking the

person or an informant how well laundry is performed (discriminating among behaviours

such as: “does personal laundry completely”; “launders small items, rinses socks, stockings,

etc.”; “all laundry must be done by others”)

Item Tr.Li.06 measures Category “Transportation” by assessing the person’s capability to

independently use transportation means, in order to travel to places beyond walking distance.

The assessment is obtained by directly asking the person or an informant how well

transportation means are used (discriminating among behaviours such as: “travels

independently on public transportation or drives own car”; “arranges own travel via taxi, but

does not otherwise use public transportation”; “travels on public transportation when

assisted or accompanied by another”; “travel limited to taxi or automobile with assistance of

another”; “does not travel at all”)

Item Me.Li.07 measures Category “Medication” by assessing the person’s capability to

independently take care of her/his own medications. The assessment is obtained by directly

asking the person or an informant how well medication responsibility is addressed

(discriminating among behaviours such as: “is responsible for taking medication in correct

dosages at correct time”; “takes responsibility if medication is prepared in advance in

separate dosages”; “is not capable of dispensing own medication”)

Item Fi.Li.08 measures Category “Finances” by assessing the person’s capability to

independently manage her/his own finances. The assessment is obtained by directly asking



the person or an informant how well finances are managed (discriminating among

behaviours such as: “manages financial matters independently – budgets, writes checks, pays

rent and bills, goes to bank”; “collects and keeps track of income”; “manages day-to-day

purchases, but needs help with banking, major purchases, etc.”; “incapable of handling

money”)

The Lawton scale assigns a score of 0 or 1 to every Item, depending on the behaviour that most closely

resembles the person’s highest functional level. For example, for Item Tr.Li.06, a score of 1 is assigned

to the first three behaviours (“travels independently on public transportation or drives own car”;

“arranges own travel via taxi, but does not otherwise use public transportation”; “travels on public

transportation when assisted or accompanied by another”) and a score of 0 to the remaining two

(“travel limited to taxi or automobile with assistance of another”; “does not travel at all”). A summary

score is then computed by summing up the individual Items’ scores, obtaining a result varying from 0

(low function) to 8 (high function).

As in the case of the Fried Frailty Index, all Items are measured obtrusively, by posing direct questions

to the person under examination or to an informant.

The challenge presented by the Lawton scale is somewhat different with respect to the Fried index. The

Items in the Lawton scale are based on complex activities that that are supposed to require a mix of

both physical and cognitive abilities, and are effective to detect the initial stages of MCI or frailty,

when deterioration is still mild and only affects more involved behaviours. Consequently, a possible

way to address automatic detection of this type of Items is through activity recognition.

In some cases the activities can be recognized directly or through observation of relevant proxies, that

can be easily measured with specific unobtrusive sensing technology. For example:

Item Co.Li.01 can be measured by directly observing telephone usage patterns in the

person’s smartphone

Item Sh.Li.02 can be measured by detecting shop entrances, as provided (for example) by

shop keepers that have installed appropriate technologies such as BLE beacons (e.g. to

support their business intelligence tasks – several City4Age Pilots are deploying such

solution)

Item Tr.Li.06 can be measured by detecting bus entrances from appropriate datasets

provided by the bus company (one City4Age Pilot is deploying such solution)

Other cases require more advanced human activity recognition methods, as those researched in the

project’s work-packageWP5, where activities are reconstructed from elementary actions, detected by

sensors. For example:

Item Fo.Li.03 can be measured by reconstructing cooking and meal preparation activities

from elementary actions such as opening/closing furniture elements in the kitchen, switching

on/off the oven, the cookers or other kitchen appliances, etc.

Item Ho.Li.04 can be measured by reconstructing the activity from elementary actions such

as switching on/off housekeeping appliances, tracking movement across different rooms in

the house (if local indoor positioning technologies are available, e.g. as in an appropriately

equipped smart home), etc.

Item Ln.Li.05 can be measured by tracking usage of the washing machine and, possibly,

movement across rooms

The above Items present different degrees of implementation difficulty, depending on how “dense” the

sensing environment is. For examples, activities that requires the usage of many electric appliances

(like Fo.Li.03 and, to a certain degree, Ho.Li.04) may be easier to recognize, as on/off switching can be

detected by deploying cheap, market available technologies, like for instance smart plugs. Others, like

Ln.Li.05 may be more difficult.

At least two Items – namely, Item Me.Li.07 and Fi.Li.08 – seem to be outside the reach of current

technologies, as they require detecting inner thought processes of the person that result in many

elementary actions that cannot be easily correlated with one another (neither in space nor in time) and

thus fall outside the capabilities of the City4Age Human Activity Recognition System described in

deliverable D5.3 The activity recognition system for smart cities.



4.2.2 Deriving a computational model for geriatric risk assessment

The challenges illustrated above with reference to the cases of the Fried Frailty Index and the Lawton

scale of Instrumental Activity of Daily Living, are representative of difficulties that are found with

respect to any of the current geriatric Instruments studied in sub-section 4.1.3.

The fundamental conclusion to be derived is that, in general, only a limited number of an Instrument’s

Items can be unobtrusively measured through technology and, consequently, the creation of

technology-enhanced “unobtrusive doubles” of existing Instruments is difficult to achieve.

A strategy to get around this limitation can be based on the following three observations:

Although individual Instruments rely on few Categories (corresponding to relevant geriatric

domains), collectively the geriatric practice has identified a wider set of relevant

Categories, used across different Instruments. For example, the Fried Frailty Index and the

Lawton scale of Instrumental Activity of Daily Living include 5 and 8 Categories

respectively, while the survey presented in sub-section 4.1.3 identified a total of 29

Categories. It is worth to note that geriatricians already implicitly recognize the need to

cover more Categories than those addressed by a single Instrument, when they propose

assessment protocols based on test batteries that include multiple Instruments, collectively

covering a more comprehensive set of domains25

. A first investigation line thus consists of

aggregating (at least potentially) all Categories identified by clinicians as having relevant

diagnostic and/or predictive power, in order to increase the dimensions along which

available, technology-derived information can be harnessed

Although some individual Instruments rely on few Items per Category, collectively, the

geriatric practice has identified and experimented with a variety of different Items for

each Category, in different Instruments. For example, using again the Fried Frailty Index

and the Lawton scale of Instrumental Activity of Daily Living examples, the former

Instrument has a maximum of two Items per Category (in particular, for Exhaustion) and a

latter has only one Item per Category. However the survey presented in sub-section 4.1.3

identified 226 Items for 29 Categories, that is an average of almost 8 Items per Category. A

second investigation line thus consists in considering all Items that have been associated to a

certain Category (in one or more established Instruments) and assessing the feasibility of

implementing such Items through technology.

Finally, it is also possible to consider several additional proxy measures for each Item, that

can be more easily collected through sensing technology, with respect to corresponding

manually collected versions. In other words, when a certain Item, as formulated in a certain

current Instrument, cannot be directly collected through unobtrusive technology, perhaps a

strongly correlated proxy can be identified and validly used in its place. This is a line that

has been successfully applied in the literature, including to frailty measurement26

. To this

respect, it is also important to note that even several established Items are actually proxy

themselves of a deeper determinant that clinicians would like to measure. For example, the

hand grip strength measurement in Item Wk.Fi.06, in Fried Frailty Index, is a proxy of

sarcopenia, the actual determinant to be assessed.

All three lines ultimately rely on the basic observation that the current limited set of Categories and

Items addressed by established Instruments is not dictated by considerations of diminishing returns

when collecting more data, but rather by the increasing costs associated to the collection process (both in terms of time and money spent). In fact, the major advantage of an automated, technology

based approach to data collection – as advocated by City4Age – is that it does not have such limits, and

thus collection and usage of more data about more Categories and more Items – underlying

multiple and varied geriatric determinants – becomes both possible and logical.

25

See for example the effort to define a Comprehensive Geriatric Assessment diagnostic process for the frail

elderly population: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4164377/

26 Palmer K. et al., Frailty, prefrailty and employment outcomes in Health and Employment After Fifty (HEAF)

Study, Occupational and Environmental Medicine journal, 2017

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4164377/



For this reason, the strategy chosen for the City4Age geriatric risk modelling is to identify and

integrate in the model the largest possible number of Categories and Items, including proxies (as hinted

in subsection 3.2.2, Figure 3, part b), that can be extracted from established Instruments and feasibly

measured through the project data collection platform (feasibility is assessed not only according to

findings by the project’s WP3 work, but also by considering the actual technology proposed by Pilot

Partners for the testbed experiments).

To proceed in this direction a thorough and accurate reclassification work, addressing all of the Items,

from all Instruments surveyed in sub-section 4.1.3 above, has been conducted. In particular, the

following activities have been performed:

Identifying and grouping together all Items that refer to the same Category, even if

belonging to different Instruments

Make explicit the way each Item is measured in current practice

Label the diagnostic and/or predictive power of the Instrument to which each Item belongs

List the way(s) each Item is proposed for measurement through unobtrusive data collection

technologies at City4Age Pilots, in the frame of respective testbed experiments

The result of this work is presented in Table 9, in Section 9.

Such Table, to which the reader is referred, represents the fundamental basis for the derivation of the

City4Age model.

Before proceeding, some supplementary considerations are worth making.

Regarding the current data collection methods (fourth column in Table 9, heading Current

measurement), the Table confirms that such methods can be subdivided into three classes, illustrated in

Table 3 below.

Method Description Example

Report A report (in the form of a response

to a question) from a person – either

the person under examination or an

informant, such as a caregiver or a

family member

Mr.Ti.09

Question “Do you have problems

with your memory?”

Possible responses “Yes /

Sometimes / No”

Observation Direct observation of the conduction

of an assigned task by the health

professional conducting the

assessment

Fi.Da.06

Task: “The person is invited to

identify four different coins and

three notes”

Observation: “Correct / Incorrect

performance for each object”

Meter Gauging a physical quantity through

a relevant meter

Mt.Fi.04

Meter: “Stopwatch”

Measure: “Time to walk 15 feet

(4.57 meters)”

Table 3. Data collection methods in current Instruments

Table 3 clarifies, almost “at a glance”, how costly it is to measure Items with current methods and

confirms the rationale of the City4Age endeavour, that aims at substitute them with unobtrusive

technology.

Regarding the predictive vs. diagnostic value (fifth column in Table 9, heading Geriatric validity), it is

worth to note the good coverage of the former, that ensures that early detection of conditions’ onset is

possible.

Regarding Items measurement in City4Age, we need to recall that work-package WP3 (Unobtrusive

acquisition of personal data) identified the concept of Measure, that is a periodic number (mostly

daily, but also weekly or monthly) obtained from the City4Age data collection platform that, according

to previous discussion, represents either:



a direct quantification of the respective Item or

a proxy quantification of the respective Item

The identifiers reported in the last column of Table 9 (with heading City4Age measurement at Pilots)

represents the names of such Measures, as they are collected at City4Age Pilots, in the experiment

deployment. The full list of Measures that Pilot Partners proposed to collect in the experiment trials, is

presented in the Annex reported in Section 11.

Based on the above work, the following subsection illustrates how the City4Age risk modelling has

been formulated.

4.2.3 A model of Geriatric Factors

The work presented in the previous subsections conducted to the formulation of the City4Age

geriatric risk model summarized in Table 4 below.

The model consists of a number of Geriatric Factors (GEFs), some of which subdivided into Geriatric

Sub-factors (GESs), that parallel Categories – i.e. the diagnostic/predictive geriatric functional domains

discussed in Section 3.2 and identified in Section 4.1.3 above.

With respect to Categories, GEFs and GESs have been partially restructured, in order to address

several requirements aimed at obtaining the best possible match between what has been shown by

medical research to correlate well with the risk of MCI and frailty and what can be feasibly obtained

through unobtrusive sensing technologies. In particular:

GEFs with large sets of Items have been subdivided into composing GESs, able to better

capture the specifics of different groupings of Items, that share extra functional

cohesiveness. This has been done for the following GEFs: Motility, Basic ADLs (split along

the standard ADL list, identified in the literature), Instrumental ADLs (split along the

standard IADL list, identified in the literature), Environment, Health – Physical

GEFs targeted by a large number of Measures, derivable from unobtrusive sensing

technology, have been subdivided into composing GESs, able to isolate several specific

elements that such Measures allow to address. This has been done for the following GEFs:

Socialization, Cultural Engagement

Addition of GESs that, although not directly addressed in current Instruments, are

considered by many in the geriatrics practice to be a useful supplement in the overall

assessment of MCI/frailty risk and for which feasible detection technologies are available.

This includes in particular the addition of GESs Still/Moving, Moving across rooms (both of

them addressing the mobility domain) and New media communication (addressing the

Instrumental Activity of Daily Living domain)

Addition of the Gait balance GES, to represent a group of balance related Items, which are

assessed in several current Instruments (e.g. enter/exit car, enter/exit bed, TUG test, etc.),

with the expectation that such GES could be measured through smartphone/wristband

accelerometer readings (see also sub-section 6.2)

Explicit addition of the GES Going out to the Basic Activity of Daily Living GEF, to

complete the assessment of Basic ADLs with an activity frequently reported in literature for

this GEF and relatively easy to get from unobtrusive sensing

Four GESs have been added to the Health – Physical GEF, in order to complete physical

health assessment with data that geriatricians normally collect, beyond those mandated by

risk assessment Instruments (i.e. Falls, Appetite, Pain, Quality of sleep)

The Health – Cognitive GEF has been added to group the related Abstraction, Attention,

Memory and Mood functional domains and to parallel the Health – Physical GEF

The model also makes a distinction among two different types of GEFs, that is important to note with

reference to the City4Age approach:

Behavioural factors: i.e. the factors that depend on the actions and activities performed by

the person under monitoring; they are essentially the factors that should undergo a

continuous unobtrusive measurement, as advocated by the City4Age paradigm



Context and status factors: these are factors that are related to the person’s health profile

(both physical and cognitive) and her/his socio-economic status (environment quality and

dependence). Most of them are typically intended to be measured once, at the beginning of

the monitoring, and then updated with a periodicity much lower than for the behavioural

factors (e.g. once a year)

A number of Categories related to cognitive health, which are used in several existing Indicators (i.e.

Time, Space, Visuospatial, Language) have been dropped because they have been deemed redundant

with respect to those that have been included in the model (i.e. Abstraction, Attention, Memory, Mood).

The Demographic category has not been included, as these data are part of the Health Record that

geriatricians store for each monitored person and have always available at their hand, when performing

the assessment.

Geariatric factors Corresponding

Category

Remarks

Behavioural factors

Motility Mt Subdivided into sub-factors, to capture

sub-domains

Walking Mt Specific Items from Mt

Climbing stairs Mt Specific Items from Mt

Still/Moving n/a Proposed to enrich the assessment of the

mobility functional domain

Moving across rooms n/a Proposed to enrich the assessment of

mobility functional domain

Gait balance Mt Specific Items from Mt

Physical activity Ac

Basic Activities of Daily Living Ad Subdivided into sub-factors, to capture

sub-domains

Bathing and showering Ad Specific Items from Ad

Dressing Ad Specific Items from Ad

Self-feeding Ad Specific Items from Ad

Personal hygiene and

grooming Ad Specific Items from Ad

Toilet hygiene Ad Specific Items from Ad

Going out Ad Specific Items from Ad

Instrumental Activities of Daily

Living

Ia Subdivided into sub-factors, to capture

sub-domains

Ability to cook food Fo

Housekeeping Ho

Laundry Ln

Phone usage Co Specific Items relating to phone usage

New media

communication n/a Proposed to enrich the assessment of the

communication functional domain

Shopping Sh

Transportation Tr

Finance Management Fi




Category

Remarks

Medication Me

Socialization So Subdivided into sub-factors, to test

potential sub-domains

Visits So

Attending senior centers So

Attending other social

places So

Restaurant So

Cultural engagement Cu Subdivided into sub-factors, to test

potential sub-domains

Visit entertainment /

culture places Cu

Watching TV Cu

Reading newspapers Cu

Reading books Cu

Context and status factors

Environment En Subdivided into sub-factors, to capture

sub-domains

Quality of housing En Specific items from En

Quality of neighbourhood En Specific items from En

Dependence Dp

Health – Physical He Subdivided into sub-factors, to capture

sub-domains

Falls n/a Proposed to complete physical health

assessment

Weight We

Weakness Wk

Exhaustion Ex

Pain n/a Proposed to complete physical health

assessment

Appetite n/a Proposed to complete physical health

assessment

Quality of sleep n/a Proposed to complete physical health

assessment

Visit to doctors He Introduced as a proxy of general health

(Items He.Fx.17, He.Ed.03, He.Ti.01,

He.Gr.05)

Visit to health related

places He Introduced as a proxy of general health

(Items He.Ed.06, He.Gr.09, He.Sb.02)

Health – Cognitive n/a Introduced to group 4 factors relating to

cognitive health

Abstraction Ab




Category

Remarks

Attention At

Memory Mr

Mood Mo

Table 4. City4Age Geriatric Risk Model

Each GES or GEF with no GES, represented in Table 4, is in then gauged by one or more Measures

(as defined in subsection 4.2.2 above), that represent particular embodiments of geriatric Items

associated to the relevant GEF/GES. The mapping among Measures and GEFs/GESs is illustrated in

Annex, in Section 10, while the definition of Measures collected at City4Age Pilot sites is reported in

Section 11.

It has to be noted that the model does not mandate which specific set of Measures, among those

available to assess a certain GEF/GES, should be used in practice: this choice has been left to Pilots, on

the basis of considerations regarding technical feasibility, cost of deployment, local needs, etc.

The ultimate objective of the above described effort is to obtain a reliable model that balances two

(somehow conflicting – see subsection 6.1) requirements:

Provide geriatricians and health professional with a tool capable of presenting them with a

comprehensive geriatric assessment along categories of factors (GEFs and GESs) that they

are already knowledgeable with and already know how to interpret, because they are

rooted in established, validated Instruments. Such requirement allows, for instance, the

immediate usage of the City4Age geriatric risk model in the Individual Monitoring

Dashboards (IMD), specified in Deliverable D2.13 Requirements, user scenarios and data

visualization mock-ups for apps/dashboards

Enact a truly “data driven geriatrics” where all feasibly and unobtrusively collectable

datasets – embodied in Measures associated to GEFs and GESs – are potentially

harnessed and made available to form the geriatric assessment, compared to a

conventional “Instrument driven geriatrics”, where only a restricted amount of data is used,

as limited by relevant cost and time constraints

As a final note, we highlight that the GEFs/GESs hierarchy presented in Table 4 also represents a first

tentative causal structure for a probabilistic model that can be used to support the application of an

alternative Machine Learning approach to risk assessment, anticipated in subsection 3.2.2 and further

discussed in subsection 6.4 below.

4.2.4 How to compute Geriatric Factors from Measures

Although the implementation of appropriate risk detection algorithms that, using the model illustrated

in sub-section 4.2.3 above, actually computes a relevant quantitative assessment of risk is a task to be

addressed in work-package WP5 (Real time data analytics and health risks detection), we propose in

this sub-section a first exploratory approach aimed at completing the previous discussion and showing

the model “in action”. Part of this approach will also be used to conduct a quantitative assessment of

the effectiveness of the geriatric risk model, as reported later in Section 5.

4.2.4.1 Measures

As previously mentioned, data is provided by the City4Age data collection platform in the form of

Measures, i.e. numbers produced periodically (in this discussion we assume that the period is once per

day, although other periods are possible), that correspond to unobtrusively measurable Items,

associated to the geriatric model’s GEFs and GESs.

Measures can be used to assign a synthetic value to a GES (or to a GEF with no GESs) through the

process illustrated below.



As an example, suppose that a Pilot installation has chosen to measure the GES Phone usage with daily

measures PHONECALLS_PLACED, PHONECALLS_RECEIVED and PHONECALLS_MISSED

(refer to Section 11 for the definition of these Measures).

4.2.4.2 Compute Numerical Indicators (NUIs)

NUIs represent specific features extracted from Measures. We assume that NUIs are assessed monthly.

Given a certain Measure, like for example PHONECALLS_PLACED, examples of NUIs are:

AVG(PHONECALLS_PLACED): the average number of daily phone calls placed, across

the month

CV(PHONECALLS_PLACED): coefficient of variation of daily phone calls placed, across

the month

BEST(PHONECALLS_PLACED): best quartile divided by average of daily phone calls

placed, across the month

DELTA(PHONECALLS_PLACED): difference among best quartile and average divided by

average of daily phone calls placed, across the month

4.2.4.3 Compute monthly values for a Measure

This value is computed for any Measure, every month, on the basis of that Measure’s NUIs.

It is a real number between 1 (worst) and 5 (best).

It expresses the variation of the current month Measure with respect to comparison with a reference

month called M0.

It can be computed in the following way.

Let’s call all NUIs for a given Measure at the current month as Ni (e.g. N1 = AVG, N2 = CV, etc.).

Let’s call all NUIs for a given Measure at M0 as Ni M0.

For each NUI a Diff function is computed as follows: Diff(Ni) = (Ni – Ni M0) / Ni M0.

(Note: some NUI may be negative, i.e. best geriatric values correspond to lower numbers; in that case

the formula to use is Diff(Ni) = – (Ni – Ni M0) / Ni M0; for example, this may be true for the CV NUI, if

geriatricians consider that higher variability for the specific Measure is a bad geriatric sign)

An individual value for the NUI, T(Ni), is computed according to the following rule:

1, if Diff(Ni) < –0,25

2, if –0,25 < Diff(Ni) < –0,10

3, if –0,10 < Diff(Ni) < +0,10

4, if +0,10 < Diff(Ni) < +0,25

5, if Diff(Ni) > +0,25

Eventually, the monthly value for the Measure is computed as a linear combination of T(Ni):

V(Measure) = i wvi*T(Ni)

The weights wvi are normalized (i.e. i wvi = 1)

4.2.4.4 Compute a monthly score for a simple Factor (GES or GEF with no GESs)

This score is computed for the factor, every month, on the basis of that factor’s Measures monthly

values.

It is a real number between 1 (worst) and 5 (best).

It expresses the variation of the current month factor with respect to a comparison with a reference

month called M0.

It can be computed in the following way.

Let’s call all Measures’ monthly values for a given factor at the current month as Vi (e.g. V1 = value

for PHONECALLS_PLACED, V2 = value for PHONECALLS_RECEIVED and V3 = value for

PHONECALLS_MISSED).



Let’s call all Measures’ monthly values for a given factor at M0 as Vi M0.

For each Measure monthly value a Diff function is computed as follows: Diff(Vi) = (Vi – Vi M0) / ViM0.

(Note: some Measures are negative, i.e. best geriatric values correspond to lower numbers; in that case

the formula to use is Diff(Vi) = – (Vi – Vi M0) / Vi M0; for example, this is true for the Measure

PHONECALLS_MISSED)

An individual score for the Measure, T(Vi), is computed according to the following rule:

1, if Diff(Vi) < –0,25

2, if –0,25 < Diff(Vi) < –0,10

3, if –0,10 < Diff(Vi) < +0,10

4, if +0,10 < Diff(Vi) < +0,25

5, if Diff(Vi) > +0,25

Eventually, the monthly score for the factor is computed as a linear combination of T(Vi):

S(Factor) = i wsi*T(Vi)

The weights wsi are normalized (i.e. i wsi = 1)

4.2.4.5 Monthly value for a composite Factor (i.e. GEF with composing GESs)

It is computed as a linear combination of values of composing GES:

S(GEF) = i wfi*S(GES)

The weights wfi are normalized (i.e. i wfi = 1)

4.2.4.6 Conclusions

It is important to reiterate that the above formulation has been provided as a first reasonable hypothesis

to support work on WP5, particularly for the implementations of IMDs, as specified in Deliverable

D2.13 Requirements, user scenarios and data visualization mock-ups for apps/dashboards.

To progress on it, additional work has been completed in WP5 and WP7, for example to

experimentally assess the values of the different weights in the proposed linear combinations,

according to information provided by Pilot testbeds and accompanying geriatricians judgments.

However, the above notwithstanding, by adopting the proposed vision we are finally able to derive a

fully applied instance of the City4Age Geriatric Risk Model, that builds on four different

abstraction levels:

GEFs, which represent higher level geriatric domains, and are evaluated on the basis of27

:

GESs, which represent lower level geriatric domains, are evaluated on the basis of:

NUIs, which represent features periodically extracted from:

Measures, which represent datasets provided by the City4Age unobtrusive data collection

platform (WP3)

Such a model flexibly supports different functions, such as:

Mapping measures of health status to GEFs/GESs, synthesizing them in a way that makes it

easier for clinicians to understand and to trust them

Using NUI features to train, validate and test machine-leaned models for fully automated

detection of MCI/frailty onset

Using the GEFs/GESs/Measures structure to infer causal networks among different

determinants of MCI/frailty

The first line has been implemented in WP5, embodied in City4Age IMDs. The second line has been

implemented in an experiment that proves the model’s validity, presented in Section 5 below. The third

line has been proposed as a future research agenda, based on Bayesian Network analysis, in subsection

6.4 below.

27

As previously illustrated, some GEFs are evaluated directly on the basis of NUIs, skipping the GES level



5 Statistical assessment

5.1 Introduction

Previous Sections have illustrated how the City4Age Geriatric Risk Model has been formulated, based

on a comprehensive investigation of current clinical knowledge, accompanied by in-depth discussions

with the Project WP3 team and the Pilot testbed teams.

In this section we attempt to provide a statistical assessment of the Model, examining on how well it

can perform in practice, based on experimental data collected at Pilot sites.

The endeavour is approached as a classification problem: i.e. to verify if it is possible to build a

classifier that (i) receives as input Pilot Measures that have been collected and organized consistently

with the City4Age risk model (see 4.2.3 and 4.2.4), and (ii) produce as output a binary classification of

elderly participants into robust and non-robust subjects, which is statistically well correlated with

ground truth evidence obtained through a clinically validated indicator of functional robustness.

In the subsections that follow, we will first examine the datasets collected at each Pilot site, in order to

assess their quality in relation to the above mentioned task. Then, we select a Pilot on which to conduct

the experiment, illustrate in details how the classification problem was setup and run, present the

quantitative results that were obtained, and finally discuss their meaning.

5.2 Correlation analysis

As a preliminary statistical assessment, we look at the correlation matrices that illustrate the

dependencies among the City4Age risk model’s Measures, collected at each Pilot site.

Data for this analysis has been gathered in June 2018, based on the Measures collected by Pilot sites

and uploaded to the City4Age Shared Repository up to that date, which amounted to several months-

worth of information for each of them (see Table 5 below).

Pilot Start date Assessment date Months

covered

Subjects

covered

Athens April 2017 June 2018 15 40

Birmingham January 2017 June 2018 18 35

Lecce January 2018 June 2018 6 24

Madrid September 2017 June 2018 10 15

Montpellier November 2016 June 2018 20 18

Singapore September 2016 June 2018 22 15

Table 5. Pilot data from City4Age Shared Repository used for correlation analysis

All correlation matrices are included in Annex, in Section 12.

As it could be expected, several Measures are relatively highly correlated with one another, across

several Pilots, as for example walk_distance and walk_steps (they both represents the same distance) or

seniorcenter_visits and seniorcenter_time (more visits are associated to more time), etc.

However, more interesting patterns can be observed through careful analysis:

Data coming from fitness wristbands are in general among the most strongly correlated. This

can be seen more clearly in the matrix from the Birmingham Pilot (subsection 12.2) the

Measures of which are based on the commercially available Withings wristband

(https://www.withings.com/, formerly Nokia Health). As hinted, walk_distance and

walk_steps are almost deterministically linked, with a coefficient of 0,96. In addition, these

Measures are in turn correlated with other three Measures related to physical activity:

physicalactivity_soft_time, physicalactivity_moderate_time and physicalactivity_calories,

with coefficients ranging from 0,44 to 0,82. These latter Measures are also somewhat

https://www.withings.com/



mutually correlated, with coefficients greater than 0,3. Finally, the sleep quality Measures

sleep_awake_time and sleep_wakeup_num are strongly mutually correlated with a

coefficient of 0,76. A general conclusion to be derived is that market available fitness

trackers – although very convenient for their low cost and wider availability – provide an

informative content that is more limited than initially apparent from the number of collected

features that they advertise. In a sense, this limit has to be expected, if we consider that all of

the mentioned features are basically derived from just one type sensor, i.e. the three-axis

accelerometer integrated in the wristband.

A similar conclusion can be reached about smart-homes systems based on presence detectors

in rooms, as visible in the Singapore Pilot matrix, which mostly uses this type of sensing

equipment. Apart from the obvious correlations among visits to rooms and time spent in the

same rooms, data shows that strong correlations also exist among Measures relating to

different rooms: for instance kitchen_visits correlates strongly with livingroom_time

(coefficient 0,48), livingroom_visits (coefficient 0,67) and restroom_visits (coefficient 0,65).

In turn restroom_visits is strongly linked to livingroom_time (coeffient 0,52) and

livingroom_visits (coefficient 0,64). Again, this can be explained by the fact that what

presence detectors actually measure is the pattern of movements across the home

In comparison with the above, the correlation matrices of Pilots that have addressed a larger

number of “city-scale” activities – such as Athens, Lecce, Madrid and Montpellier – are

relatively more “white”, signalling that such Measures actually bring in greater variety,

associated to more diverse and informative content that can be used to distinguish

comparatively weaker signals of incipient health decay.

In conclusion, the correlation analysis supports the City4Age rationale, i.e. that behaviour detection in

cities is likely to increase the chances of implementing better systems for the early detection of aging

conditions, as compared to existing, market available solutions.

5.3 Machine learning experiment

In order to further assess if the model’s Measures discussed in previous subsections have sufficient

statistical power to actually detect health decays, we relied on a supervised machine learning

experiment.

The underlying idea is to demonstrate that the Measures of the City4Age risk model, when collected

and fed to an appropriately trained classifier, are capable to discriminate robust from non-robust

people, as determined by a clinically validated scale, with an accuracy sufficient to prove the concept.

To proceed along this direction, the following tasks had to be completed:

Select an appropriate machine learning scheme to be used

Define and collect ground truth labels

Chose a Pilot experiment to obtain training, validation and test datasets

Conduct feature extraction (i.e. synthesis and selection) and produce the base dataset

Run the experiment and collect results

Analyse and discuss results

These aspects are respectively presented in the next subsections.

5.3.1 Choice of a machine learning scheme

The choice of an appropriate learning scheme to be used in the experiment is dictated by the following

set of requirements:

The learner should be robust against missing data, as missing data is a relevant issue,

frequently found in real-life, experimental datasets, including City4Age Pilots’ datasets (e.g.

when participants forget to switch their smartphone on, or inadvertently deactivate the

City4Age app).

Given the number of available data-points at City4Age Pilot sites (several tens of

participants) the learner should not fractionate the dataset (as typical of some popular



schemes like decision trees or rule learners) as fractions would contain an insufficient

number of data-points for the scheme to work properly

The learner should be robust against irrelevant features as – given the investigation

purpose of this work – it cannot be established in advance what is the best combination of

relevant features. In fact, such combination should be discovered by trial and error, based on

the generation of many candidate features and the support of dimensionality reduction

algorithms, as common in many data mining efforts28

.

In consideration of the above, the Naïve Bayes scheme was selected as the algorithm of choice, as it

satisfies all of the three mentioned requirements.

The price that has to be paid in exchange for the above benefits is that Naïve Bayes assumes that

features are conditionally independent of one another, which is not true in many datasets. However,

two considerations suggest that this downside can be manageable:

the Naïve Bayes algorithm has shown to perform surprisingly well in practice, even in cases

in which the conditional independence assumption among features is clearly violated29

most “city-scale” Measures collected at City4Age Pilots are in fact mutually uncorrelated, as

confirmed by the analysis presented in the previous subsection

5.3.2 Definition and collection of ground truth

To apply a supervised machine learning approach it is necessary to select and enact a method for

measuring ground truth labels. The method should be based on a relevant geriatric instrument for

measuring robustness in elderly people, addressing the following three requirements:

The instrument must be able to detect incipient functional decline in elderly citizens, in

community settings

Its administration must not be burdensome for elderly participants at Pilot sites, and it should

be efficient for Pilot Partners to administer

It must have been clinically validated

These requirements are well addressed by the Functional Ability Index (FAI) 30

, an instrument defined

in the frame of the longitudinal cohort study LUCAS [Note: the City4Age Consortium came into

contact with LUCAS researchers during the face-to-face meeting of the EIP on AHA’s Action Group

D4 on Age Friendly Environments, held in Hamburg on 16-17 November 2017, confirming the

invaluable networking potential of the EIP on AHA31

. Usage of FAI in City4Age was subsequently

discussed in an email exchange with instrument’s authors, in the second half of November 2017].

The index consists of a questionnaire with 11 questions, that can be easily administered (or even self-

administered) without the application of complex equipment that require appropriate management (e.g.

Jamar hand dynamometers, chronometers, etc.).

The FAI instrument excludes disabilities and targets items that are related to the early signs of decay,

including – similarly to the City4Age risk model – outdoor mobility and social interaction.

Differently from other indicators, FAI considers two different types of factors: (i) functional risk

factors, addressed by 6 questionnaire items, and (ii) resources, also addressed by 6 items (one of the 11

questions has double weight, being considered for both types of factors, with inverted scores).

Combinations of fitness/non-fitness classes along these two dimensions produce 4 different

possibilities:

28

Witten et al. Data Mining: Practical Machine Learning Tools and Techniques, Morgan Kaufman, 2017

29 Domingos et al. On the optimality of the simple Bayesian classifier under zero-one loss, Machine Learning,

1997

30 Dapp et al. Long-term prediction of changes in health status, frailty, nursing care and mortality in community-

dwelling senior citizens – results from the longitudinal urban cohort ageing study (LUCAS), BMC Geriatrics,

2014

31 https://ec.europa.eu/eip/ageing/home_en

https://ec.europa.eu/eip/ageing/home_en



People high on resources and low on health risks (Robust)

People high on resources and high on health risk (postRobust)

People low on resources and low on health risk (preFrail)

People low on resources and high on health risk (Frail)

According to conventional approaches (e.g. Fried Frailty Index, see subsection 8.2.1), this introduces a

new class, postRobust, that, according to authors, is intended to distinguishing between fully robust

persons and those not fully robust, but better off than pre-frail, i.e. very early on functional decline.

The predictive power of FAI has been validated in the frame of the LUCAS longitudinal study, with

data spanning over a period of 93 months.

Before usage in City4Age, the FAI questionnaire text has been slightly modified, as administration

rounds where closer in City4Age (3 months, see below) with respect to LUCAS (several years).

Consequently, reference periods of 6 months and 12 months in the original FAI formulation had to be

respectively shortened.

Three City4Age Pilots were available to collect FAI data for their elderly participants. In particular:

The Athens Pilot collected FAI data for 40 participants, in two rounds: in January/February

2018 and April 2018, generating 80 ground truth labels

The Birmingham Pilot collected data for 15 participants in one round, in March 2018,

generating 15 ground truth labels

The Madrid Pilot collected FAI data for 15 participants in January 2018 (1 participant), in

February 2018 (11 participants, including a reassessment of the participant measured in

January), in March 2018 (3 participants) and in June 2018 (1 participant), generating 15

ground truth labels (the measurement obtained in January 2018 was discarded, as it was too

close to the reassessment of the same participant in February 2018, as confirmed by no

change in the classification)

[Note: what above reported is the status at the time of writing: more rounds of FAI measurement are

foreseen at the mentioned Pilot sites, during the last month of the Project work-plan].

A summary of collected ground truth information is presented in Annex, in Section 13.

5.3.3 Selection of a Pilot experiment to obtain training and test datasets

In order to ensure a higher level of homogeneity in collected data and in the procedures for ground

truth measurement, it was decided to conduct the analysis on a Pilot per Pilot basis.

Looking at available ground truth data for the three Pilots mentioned above (see Section 13) it becomes

evident that the best Pilot site to conduct the machine learning assessment is the Athens one. In fact:

The Athens Pilot offers the highest potential number of data-points, having collected ground

truth labels for 40 participants in two rounds (totalling 80 labels). This guarantees a more

solid base for the machine learning algorithm to pick up signals in the data.

The Athens Pilot offers a more balanced number of labels from different robustness classes,

while participants in Birmingham and Madrid are predominantly Robust people, with only

one in fifteen being postRobust, in both cases. This class imbalance would make the

application of a supervised machine learning approach more difficult.

In addition, looking at the correlation matrix for the Athens Pilot, reported in subsection 12.1

and discussed in subsection 5.2 above, it can be seen that this testbed has the further

advantage of including a larger set of mutually independent Measures, that is advantageous

when applying the Naïve Bayes scheme, as postulated in subsection 5.3.1.

5.3.4 Feature extraction and dataset preparation

Having selected the Pilot experiment to work on, the next step is to decide about the features that will

be provided as input to the machine learning algorithm.

Features need to correspond to the City4Age risk model Measures (as collected at the Athens Pilot, in

this case) since the purpose of the experiment is the statistical assessment of the model’s detecting

power.



The application of a supervised learning approach imposes the following additional considerations:

Feature synthesis: the Measures collected at the Athens Pilot belong to the City4Age

Measures dictionary, reported in Section 11. As illustrated in that Section, these Measures

are collected periodically, with different time granularities, down to daily frequency. A

method should be devised in order to synthesize the Athens Measures into relevant features,

that need to be associated with ground truth labels which, as mentioned in subsection 5.3.2,

have been gathered in two rounds, distanced approximately three months from one another

Feature selection: not all Measures are worth including in the experiment’s dataset. In

particular, having chosen to rely on a Naïve Bayes learner, it is beneficial to exclude all

those features that are believed to be obviously correlated with other ones, based on pre-

existing domain knowledge

Data cleaning: differently from other City4Age data analytics components developed on top

of the geriatric risk model, such as for instance the Individual Monitoring Dashboard, where

data interpretation is filtered by a domain expert that can apply her/his knowledge and

judgment to get around unclean data (missing data, outliers, etc.), in a data mining approach

an appropriate data cleaning process needs be run in advance, before feeding the data to the

learning algorithm

To illustrate in detail how these considerations have been applied in the present work, it is convenient

to examine the SQL query that has been used to extract the Athens Pilot’s dataset, in the first step of

the learning workflow. The text of the query is reported in Annex (Section 14), while the following

paragraphs illustrate the pre-processing operations that the query implements.

5.3.4.1 Features synthesis

Having held two rounds of ground truth collection based on FAI measurements with a distance of

approximately three months from one another (see subsection 5.3.2), it has been decided to associate

each FAI label with features that synthesise the City4Age risk model’s Measures collected at the Pilot

site in the trimester preceding the FAI assessment [Note: FAI scores in Athens were collected on

slightly different dates for different subjects: they have been considered as related to a given trimester

if measured within a temporal distance of less than 15days from the trimester’s end]. Such set of

synthesised features with the associated FAI labels, constitute the individual data-points to be used for

training and testing the machine learning algorithms.

Consistently with the approach suggested in subsection 4.2.4, the synthesis has been obtained by

computing the following values for each model Measure:

Average of Measures’ values collected across the trimester: this parallels the “AVG” NUI in

subsection 4.2.4

Standard deviation of Measures’ values collected across the trimester: this parallels the

“CV” NUI in subsection 4.2.4

Best quartile of Measures’ values collected across the trimester: this parallels the “BEST”

NUI in subsection 4.2.4

The “DELTA” NUI, which is also considered in subsection 4.2.4, has not been included here because,

being deterministically dependent on two of the above listed NUIs, it would have risked to confuse the

Naïve Bayes learner that, as mentioned, is based on the conditional independence assumption.

The above computations produce three features for each Measure, as discernible in the SQL query

reported in Section 14. For example, in the case of the cinema_time Measure, the query extracts the

three features, cinema_time_avg, cinema_time_sd, and cinema_time_best in the following way:

avg(cinema_time) as cinema_time_avg,

stddev(cinema_time) as cinema_time_sd,

percentile_cont(0.75) within group ( order by cinema_time ) as cinema_time_best

[Note: for most Measures, the best quartile is the 75th percentile; however, for home_time_best,

pharmacy_time_best and pharmacy_visits_best the best quartile is the 25th percentile, as these

Measures are negatively correlated with geriatric outcomes in the City4Age risk model].



The grouping by trimesters can be discerned in the query in Section 14, in particular (a) in the selection

of the time span covered by the query:

date(time_interval.interval_start) >= '2017-11-01' and

date(time_interval.interval_start) <= '2018-04-30'

(b) in the assignment of the temporary field trimester:

case when time_interval.interval_start <= '2018-01-31' then 1 else 2 end as trimester

and (c) in the grouping over trimester:

group by ct.cr_m, ct.trimester

5.3.4.2 Feature selection

As anticipated, some Measures have been excluded from the dataset because of their deterministic

correlation with other ones. In particular, among the 24 Measures collected at the Athens Pilot (listed in

subsection 12.1) the following 6 have been discarded:

cinema_visits_month and pharmacy_visits_month have been discarded because they report

the same information as, respectively, cinema_visits and pharmacy_visits (just differently

grouped along time)

supermarket_time and supermarket_visits have been discarded because they contain the

same values as, respectively, shops_time and shops_visits

walk_distance_outdoor_fast_perc has been discarded because it reports information which is

already conveyed by the best quartile feature of walk_distance_outdoor

walk_distance_outdoor_slow_perc has been excluded because it is deterministically

computed from walk_distance_outdoor_fast_perc (namely, 100% less

walk_distance_outdoor_fast_perc)

As a result of the selection, a total of 18 Measures remained, which generated a total of 54 features,

which are visible in the main select clause of the query in Section 14.

5.3.4.3 Data cleaning

As common in data mining workflows, especially those based on real-life experimental data, a data

cleaning step must be applied to remove values that are evidently the result of measurement errors.

In particular, the following data errors are targeted in the query in Section 14:

Missing data: this is a classical issue, present in all non-trivial data mining efforts. In the

case of City4Age experiments, the main cause of missing data is participants forgetting to

switch on or bring with them the smartphone or deactivating the City4Age georeferenced

app that enables the unobtrusive collection of model’s Measures. In the current experiment,

point-wise missing data were not considered problematic, as the Naïve Bayes scheme is

quite robust against this problem. However, care recipients that did not activate the

smartphone and/or the Ciy4Age georeferenced app for the whole extension of one trimester

where excluded from that trimester. This is done by checking the following condition over

the trimester group, assuming that being apparently at home for the whole trimester duration

would imply, with probability close to 1, that the City4Age app was simply never used:

having avg(home_time) != 86400. As a consequence, out of the 80 potential data-points

labelled in Athens, 34 had to be discarded, leaving a dataset with 46 valid data-points

Duplicated Measures: this issue is specific to the Athens Pilot and regards the Measure

restaurants_visits_week, which is reported twice with different values for few days. Analysis

revealed that the doubly reported values actually represent the cumulated sum of previous

weeks’ values, reported monthly. The problem was easily solved by grouping double

Measures (clause group by time_interval.interval_start, care_recipient.id,

variable.detection_variable_name in the query) and selecting the minimum value (field

min(measure.measure_value) in the query, the larger value being the incorrectly reported

monthly sum)



Outliers, another typical issue in data mining, are also likely to be present in the City4Age datasets,

including in the Athens Pilot dataset. However, in the current experiment they were not corrected as:

They generally represent a relatively low share of the total number of samples (few

percentage points, on a rough estimation)

Although few values are clearly outside reasonable ranges, it is still not easy to determine a

clear-cut threshold, aimed to discard some values while keeping others (see considerations in

subsection 6.3)

It is expected that the probabilistic Naïve Bayes learner would help in mitigating outliers’

effects on the overall results.

5.3.5 Running the machine learning workflow

The machine learning experiment was run within the Weka data mining environment32

.

An additional pre-processing step was conducted with the Weka MergeManyValues filter, in order to

combine the postRobust, preFrail and Frail labels into a comprehensive nonRobust class, to be

separated by the classifier from the Robust class. This has been done for two reasons:

The limited size of the dataset makes it difficult to train a classifier with several class values

The rationale of the City4Age project is to discriminate robust persons from those that are

affected from early health decay, which is consistent with a binary classification problem

After execution of this pre-processing step, the final Athens Pilot dataset was produced, including 46

data-points, with 54 features and a binary class attribute each.

A first glance at the distribution of class values across individual features seems to confirm the

presence of informative content. For example, the following Figure 4 illustrates how the

seniorcenter_visits_avg feature can reasonably differentiate between Robust elderly citizens (who visit

the Athens Friendship Clubs, on average, up to every other day) and nonRobust elderly citizens (who

visit the Friendship Clubs every 5 days on average, at most). Similarly encouraging patterns can also

be observed in many of the other features.

Figure 4. Class distribution for the seniorcenter_visits_avg feature

32

https://www.cs.waikato.ac.nz/ml/weka/

https://www.cs.waikato.ac.nz/ml/weka/



On that basis, a learning experiment was thus set up and run with the following characteristics:

Execution of a preliminary dimensionality reduction step, through the Weka

PrincipalComponents filter, that operates Principal Components Analysis (PCA) before the

features are fed to the learner. This step is useful to further reduce redundancy among

features, and make them more suited to Naïve Bayes learning. Different values for the

varianceCovered parameter (the algorithm selects the smaller possible set of components

that covers at least this variance) have been tried in order to determine the most performant

setting. In particular, the following values have been tried: 0.75, 0.8, 0.85, 0.9, 0.95 plus no

PCA at all.

Running a Naïve Bayes scheme instance, for each different value of variance covered, as

above illustrated, and selection of the one with the best performance. The result of this step

is illustrated in Table 6. Performance has been measured in terms of AUC33

. Performance

assessment in each run was conducted through 10-fold cross-validation, to prevent

overfitting. Significance is assessed by using the corrected resampled paired t-tester34

, to

mitigate the issue of cross-validation folds insisting on the same underlying dataset. The

table shows that PCA is a useful step that improves performance, and that the best value for

the varianceCovered parameter is 0.8, corresponding to the selection of 7 components (see

second row in the table, shaded)

Running an outer, 10-times, 10-fold cross-validation testing experiment with the same

model, to obtain an unbiased quantitative measure of the performance for the above

described workflow, using different datasets for validation and testing to avoid “peeking at

data”. In particular, the experiment was conducted by using a modified version of the Weka

MultiScheme meta-learner, that uses AUC rather than percent correct to compare schemes

(renamed for this reason as MultiSchemeAUC, see Section 15).

The resulting unbiased performance estimate, obtained from the last step listed above, yields an

AUC = 0.710 different from the null hypothesis (random classification) with significance < 0.05,

which represents a convincing proof-of-concept for the City4Age Geriatric Risk Model

effectiveness.

Variance covered by

PCA step

(at least)

Number of

components selected

by PCA step

AUC Significantly different

from classification by

chance?

(significance level = .05)

0.75 6 0.71 yes

0.80 7 0.72 yes

0.85 7 0.71 yes

0.90 9 0.70 no

0.95 12 0.66 no

No PCA n/a 0.69 no

Table 6. Performance versus different values of variance covered

Starting from this persuasive result, seven additional experiments have been conducted, in order to

further investigate the validity of the City4Age risk modelling assumptions, by tweaking the feature

extraction process and observing resulting changes in performance.

In particular, the following modifications have been tried:

33

https://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_the_curve

34 Nadeau et al. Inference for the generalization error, Machine Learning, 2003

https://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_the_curve



1. Remove standard deviation features: this correspond to removing the “CV” NUI from the

model

2. Remove best quartile features: this correspond to removing the “BEST” NUI from the model

3. Substitute average features with corresponding median features (i.e. 50th percentile): this is

to try a different way to measure central tendency (“AVG” NUI in the model)

4. Remove all “city-bound” features, i.e. only home-bound (home_time) and fitness tracker

provided (walk_*) features where left in: this is to check the relative importance of market

available unobtrusive data collection with respect to the full City4Age risk model

5. Only leave “city-bound” features (the complement of the previous selection): this is to check

the relative importance of smart-city-only unobtrusive data collection with respect to the full

City4Age risk model

6. Merge together all socializing Measures (other_social_*, restaurant_*, seniorcenter_*) in a

single Measure: this is to check if more dense but coarser features are preferable to the

sparser (i.e. with “many zeros”) but more granular ones currently included in the City4Age

risk model

7. Combine paired *_time and *_visits Measures in order to produce 6 additional

*_time_per_visit, features: this is to assess, with an example, the opportunity to further tune

model’s Measures, in order to improve machine learning algorithms performance in

extracting meaning from them (the query modification needed to add these new features is

reported in Annex, in Section 16)

The results from these additional experiments are summarized in Table 7 below. Row 0 in the Table

refers to the original experiment, for sake of comparison. We remind that all experiments presented in

the table have been conducted according to the previously described workflow, summarized at the

bottom of the table.

# Experiment description AUC Significance

0 Original model, with all 54 features

included

0.710 0.05

1 Removing standard deviation

features

0.674 0.1

2 Removing best quartile features 0.689 0.1

3 Substitute average features with

median (50th percentile) features

0.671 0.1

4 Remove city-bound features 0.668 0.1

5 Leave only city-bound features 0.648 0.2

6 Merge socialization features 0.616 0.3

7 Add *_time_per_visit features 0.715 0.05

All experiments are based on:

PCA with best coveredVariance parameter, selected in the range {0.75, 0.80, 0.85, .90,

0.95, no PCA}

Naïve Bayes learning with:

o inner 10-fold cross-validation for parameter selection, based on best AUC

o outer 10-times, 10-fold cross-validation for unbiased performance assessment

Significance calculated with corrected resampled paired t-tester

Table 7. Experiments to assess the City4Age Geriatric Risk Model

In the following subsection, the above reported results are discussed and general conclusions on the

validity of the City4Age Geriatric Risk Model are offered.



5.3.6 Discussion

The first conclusion to be derived from the quantitative results presented in the previous subsection is

the confirmation of the validity of the City4Age Geriatric Risk Model.

Row 0 in Table 7 clearly shows that Measures established by the City4Age risk model are able to

significantly discriminate among Robust and nonRobust subjects, and thus support the enactment of

the early detection paradigm that is central to the Project’s rationale.

The following Figure 5, that illustrates a Receiver Operating Characteristic (ROC) curve from one

experiment run, visually shows how the City4Age risk model behaves differently from the null

hypothesis (i.e. random classification – red line in the Figure, corresponding to AUC = 0.5).

Obviously, the resulting AUC performance, although very good, does not yet achieve the levels

required for a market deployable system. However, it clearly represents a convincing proof-of-concept

that confirms the success of the City4Age modelling endeavour and lay the basis for its industrial

scale-up (see next Section).

Figure 5. Example of ROC curve from one experiment run

Other notable conclusions can be derived from the results of the additional experiments, reported in

rows 1-7 of Table 7:

Removing standard deviation or best quartile features or substituting average features with

modal ones (rows 1-3 in Table 7) decreases both AUC performance and statistical

significance, confirming that the NUIs proposed in subsection 4.2.4.2 are a sound base

hypothesis

Removing either city-bound or non-city-bound features (rows 4-5 in Table 7) decreases both

AUC performance and statistical significance, confirming the assumption of the City4Age

model that both types of features should be used in order to accurately detect health decay

Merging socialization features (that contain several daily zero values) in a single

consolidated feature to reduce sparseness (row 6 in Table 7) decreases both AUC

performance and (dramatically) statistical significance. This effect vindicates the City4Age

risk model assumption that it is better to collect more granular Measures, even at the cost of

more sparseness, as differences in their expression can be meaningful for the detection of

decaying conditions

Although the signal is relatively weak (a 0.05 increase in AUC with no increase in

significance), row 7 in Table 7 also deserves attention, as it highlights that the City4Age risk



model may still be augmented with additional interesting Measures, which have escaped the

first drafting phase of the model and can contribute to improve its accuracy

A last interesting consideration that can be derived from experimental data, regards putting into

perspective the number of Measures that should be included in a model for screening frailty/MCI,

depending on (i) the applied data collection method (obtrusive vs. unobtrusive) and (ii) the size of the

datasets used to train it. The landmark, non-technology-based Fried Frailty Index includes 6 Items,

collected through self-reported questions and meter-based measurements covering 5 domains (see

subsection 8.2.1.3). The Index was derived from a cohort of more than 5,000 subjects (see subsection

8.2.1.2). On the other hand, for example, the study discussed in subsection 8.1.8 shows the possibility

to predict MCI on the basis of a single Measure (gait speed) with an SVM learner that achieves an

AUC = 0.81 when trained on 984 data-points. However, although such measure was collected through

automatic sensing equipment, it still required a specific “lab-like” installation in the subject’s home,

that is more obtrusive than allowed by City4Age requirements. Finally, the work discussed in this

Section yields AUC = 0.71 with a Naïve Bayes learner trained on 46 data-points based on a set of 18

Measures, which cover 9 geriatric factors. All Measures are collected in a fully automated,

unobtrusive way and all of them are necessary to achieve good performance, as shown by experiments

#4 and #5 in the previous subsection.

Thus, it seems that when committing to an automated, unobtrusive data collection approach, the

number of Measures to be included in the model needs to be relatively increased. This is

reasonable, as unobtrusive measurement is likely to provide less information content “per Measure”,

with respect to manual or technically more burdensome, but more precise methods; and this

disadvantage must be counterbalanced by a corresponding increase in the number of Measures used.

From a cost point of view, this is not an issue as City4Age demonstrates that modern technologies do

allow to collect larger numbers of Measures at much lower cost (for instance, the 18 Measures

collected at the Athens Pilot were all obtained from just a single, georeferenced smartphone app).



6 Lessons learned and recommendations

Critical analysis of the model presented in Section 4 above and, in particular, its application to the

Pilots’ testbed cases, allowed to point out several areas that may be addressed to further advance and

improve the City4Age risk modelling strategy.

Results of such analysis are presented in this Section.

In particular, subsection 6.1 provides an overview of the lessons learned during the development of the

City4Age Geriatric Risk Model, that provide the basis for the recommendations presented in the rest of

the section.

Such recommendations are subsequently articulated as follows:

subsection 6.2 presents a detailed study of the coverage of the geriatric risk model in relation

to the data collection technologies deployed at the Pilot testbeds and discusses additional

research that can be implemented to improve the situation

subsection 6.3 presents possible technical improvements, relating to the way current

measures are collected, that can have positive impact on the model’s accuracy

as anticipated in subsection 3.2.2 and building upon the encouraging results presented in

Section 5, subsection 6.4 proposes the investigation of a new approach to geriatric risk

modelling, based on machine learning methods

subsection 6.5 proposes a sequence of steps that Partners interested in the commercial

exploitation of the City4Age risk model may want to apply in order to achieve industry-

grade accuracy, and move beyond the proof-of-concept milestone that was the main

objective of the Research and Innovation Action conducted within the City4Age Project

6.1 Lessons learned

In this section, we review the whole process that has been followed to address health risk detection,

from the initial conception of the geriatric risk model, to its further review based on additional insights

obtained from the implementation of Pilot sites’ data collection platforms, to the final assessment based

on experimental data generated by Pilots. This review allows to identify difficulties that have been

encountered and ways to overcome them.

During the course of the Project, work has proceeded along the following three phases:

In a first phase, as illustrated in Section 4, the effort has been focussed in identifying the

largest possible number of domains to be included in the model, by looking at established

geriatric knowledge, crystallized in existing instruments for detecting MCI and Frailty,

widely used in clinical practice. This decision has been driven by two observations: (i) that

“automating” the measurement of individual, established geriatric instruments is difficult,

since they are normally characterized by few domains, of which even fewer are addressable

with unobtrusive data collection technologies, and (ii) that the parsimoniousness of existing

instruments is not due to diminishing returns when collecting more data, but to the need to

contain the relatively high costs associated with conventional, “manual” data collection. The

approach proposed for City4Age has a double advantage:

o as it still relies on valid geriatric knowledge expressed in several established

instruments, it is clinically strong and minimizes the risks of moving along

medically unsound directions

o it follows a consistent strategy, common to many data analytics efforts, based on

initially widening the number hypothesised features, so as to maximize the

probability that good ones are not left out of consideration

In particular, the result of the first phase led to the identification of 45 geriatric factors, that

were then reorganized around 10 high level factors, 8 of which subdivided in 43 sub-factors (2

higher level factors do not have sub-factors). Each sub-factor (or factor without sub-factors) is

in turn linkable to several ways to measure it through unobtrusive technology. It is worth to

note that such reorganization also represents a first attempt at eliciting a causal model of



robustness in elderly citizens, conducive to more sophisticated data analysis strategies, e.g.

based on Bayesian Networks (sub-section 6.4).

In a second phase, the risk model obtained as above described was discussed with Pilots

Partners, in particular to verify the possibility to implement it on top of the specific data

collection platforms planned at the 6 different City4Age testbed sites. This work led to

several important results:

o The City4Age risk model was completed with the addition of a dictionary of 107

candidate measures – linked to factors and/or sub-factors, as previously mentioned –

to be potentially implemented at testbeds

o It turned out that Pilots were able or willing to collect relatively small subsets of the

above measures, targeting a correspondently restricted subset of geriatric factors

o It also turned out that geriatricians, although recognizing the vast potential of

counting on bigger datasets, had difficulty in considering a large number of

individual factors and measures, having been trained to mostly assess synthesized

scores, like those produced by currently used instruments (e.g. those reported in

Section 8)

These issues led to the elaboration of Pilot specific configurations of the risk model, to be used

in Individual Monitoring Dashboards (IMDs – developed within WP5 and discussed in that

work-package’s deliverables) that could be flexibly adapted to both (i) the factors and

measures actually collected at each Pilot site and (ii) the specific needs expressed by local

clinical and/or geriatric staff. An example of such configuration can be seen in Table 8 below,

for the Athens Pilot case discussed in Section 5. The underlying assumption is that a well-

crafted IMD configuration would make it possible to synthesise the right information needed

by medical professionals at each specific site, for efficiently assessing the robustness level of a

subject and to decide about appropriate interventions.

Two main difficulties had to be dealt with, while applying this approach:

o devising the right IMD configuration is a challenging task, since it implies deciding

on a large number of parameters (weights relating measures with sub-factors, and

sub-factors with factors, as illustrated in subsection 4.2.4) which represent a

relatively new and unfamiliar terrain for geriatricians

o the fact that – although IMD assessment represents a clear step forward in the

direction of a much earlier and efficient detection for frailty or MCI, when compared

with the existing situation – the careful analysis of the IMD for a certain subject still

represents a complex action to be manually carried out by a medical professional

(e.g. the GP), which is relatively hard to scale up without more automated support

In a third phase, when experimental Pilot data have become available for analysis, a

reassessment of the above described scenario has been possible, that led to the following

conclusions regarding the application of the City4Age risk modelling in practice:

o As discussed in Section 5, it has been proved that even a relatively limited set of

measures and factors (18 measures covering 9 sub-factors, in the case of the Athens

Pilot) are in principle capable to act as a kind of “behavioural marker” that, while

not sufficient for a full diagnosis of frailty or MCI, can still be validly used to

“screen out” people who are robust enough to exclude, with reasonable confidence,

that they need further medical attention. If this approach is properly scaled up (e.g.

by applying the recommendations suggested in the next subsections) a new, data

driven approach to screening for frailty could be developed, that overcomes one of

the main obstacles that have so far led geriatric societies to recommend against such

screening – i.e. the expensiveness vs low specificity of currently available

instruments35

.

35

It is worth to note that geriatric societies invite to reconsider the recommendation against screening for frailty if

a “suitably validated electronic frailty index” could be devised, i.e. the very concept proved in Section 5; see for

example BGS conclusion in https://www.bgs.org.uk/resources/introduction-to-frailty

https://www.bgs.org.uk/resources/introduction-to-frailty



o At a second level, only those subjects for which the above mentioned “screen out”

procedure yielded a positive result, would undergo a more in-depth analysis,

conducted through IMDs at the GP practice, as above illustrated. Such IMD-based

examination would still be a preliminary assessment (with respect e.g. to a full

CGA), but it can be conducted quickly and relatively inexpensively, as (i) data has

already been collected and synthesised (respectively) by the City4Age data

collection platform and the City4Age IMD, and (ii) only the time needed for the GP

to review IMD information regarding the few “screened in” cases has to be

accounted for. In particular, in this scenario, the purpose of the IMD would be to (a)

provide a potentially larger “quantity” of information to support patient

examination, as collected in smart environments; while at the same time (b) provide

clinicians with an organized synthesis of such information, based on the

GEF/GES/Measure hierarchy and a specific “local” configuration, that make it

easier for them to navigate such more complex knowledge asset

o Only the small number of subjects that are “red-flagged” by the IMD assessment

would then be referred to face-to-face visits at the geriatrician’s or neurologist’s

practice, for further assessment and, possibly, for a CGA.

This two-level screening procedure can contribute to increase overall specificity while still

significantly containing costs, since false positives that pass through the automated “screen

out” procedure will most likely be filtered out by subsequent IMD analysis.

Geriatric factors Measures NUIs

Motility

AVG

SD

BEST_QUARTILE

for each Measure

Walking WALK_DISTANCE_OUTDOOR

WALK_TIME_OUTDOOR

WALK_SPEED_OUTDOOR

Basic Activities of Daily Living

Going out –HOME_TIME36

Instrumental Activities of Daily Living

Shopping SHOPS_VISITS

SHOPS_TIME

SHOPS_OUTDOOR_TIME_PERC

Transportation TRANSPORT_TIME

Socialization

Attending senior centers SENIORCENTER_VISITS

SENIORCENTER_VISITS_WEEK

Attending other social places OTHERSOCIAL_VISITS

OTHERSOCIAL _TIME

Restaurant RESTAURANTS_VISITS_WEEK

RESTAURANTS_TIME

Cultural engagement

Visit entertainment / culture

places CINEMA_VISITS

CINEMA_TIME

Health – Physical

Visit to health related places –PHARMACY_VISITS

–PHARMACY _TIME

Table 8. City4Age Geriatric Risk Model configuration for the Athens Pilot case

36

A minus sign indicates that the Measure is negatively correlated with geriatric outcomes



The claim resulting from the above described work is that the City4Age Geriatric Risk Modelling is a

flexible tool, with proven validity, that can be used in different ways, adapted to specific application

cases and implementing multi-level approaches to screening for robustness in elderly populations, that

can optimize healthcare management.

We also claim that the City4Age Geriatric Risk Modelling is an innovative, first-in-the-art

contribution to the field, as:

(1) to the best of our knowledge, at the time of writing no other effort exists that targets the early

detection of age related health decay through city-scale behaviour analysis (e.g. visiting shops,

using transportation means, socializing, engaging in cultural activities, etc.). In fact, current

efforts are mostly focussed on “home-scale” behaviours (e.g. ADLs) detected through smart

home kits, or “body-scale” behaviours (e.g. walking, physical activity, sleep quality, etc.)

detected through fitness tracking apps

(2) our experiments have shown that city-scale behaviour analysis can achieve superior

classification performance, with respect to “home-scale” and “body-scale” only approaches.

On the other side, it is acknowledged that the above results have been achieved in the form of a “proof-

of-concept” scheme, which is satisfactory for a low-TRL, Research and Innovation effort like

City4Age, but that needs to be improved in several directions to achieve the higher TRLs that are

needed for large scale market deployment. Many opportunities to proceed along this path have been

discovered during the Project research work, and are discussed in details in the following subsections.

6.2 Analysis of coverage from Pilots

Table 10 in Section 10, lists the geriatric factors included in the City4Age risk model and, for each of

them, reports which Measures – obtained from the data collection platforms deployed at Pilots – can be

used to assess them (e.g. according to the guidelines provided in sub-section 4.2.4 above).

For each factor, the union of all the Measures collected at each of the six project Pilots is considered,

arguing that if one Measure could be feasibly deployed at one site, it could be similarly deployed at

another one, and thus the union of all of them realistically represents the potential offered by the

City4Age project to any new City wishing to install the system – for example, during the exploitation

phase, after the end of the Project.

By looking at Table 10, we can infer a good general coverage for the model, through both direct and

proxy Measures, and, in the case of some factors, even an overabundant one.

On the other hand, there are still areas where coverage is absent or poor, relying only on very few or on

only tentative Measures.

In the following paragraphs, we review the most important of these areas and propose interventions

that, if applied, would correct the situation.

Gait balance

GES Gait balance is not addressed by any Pilot, although this is an important factor, that summarizes a

considerable number of Items, as discussed in sub-section 4.2.3 above.

The current ways to assess such factor are varied: through reports from the monitored persons (e.g.

Items Mt.Ne.03, Mt.Ne.04 or Mt.Ne.05), through direct observation (e.g. Mt.So.02) or through a

meter-based measurement (e.g. Mt.Ed.11).

However, since the advent of affordable accelerometers (such as those integrated in smartphones or

fitness wristbands), several authors have researched the usage of such unobtrusive gauges in order to

assess gait balance, for instance to address the needs of Parkinsonian patients’ treatment.

A relevant survey in this direction, conducted by Hubble and co-authors, has been published on PLoS

One in 2015 (Wearable Sensor Use for Assessing Standing Balance and Walking Stability in People

with Parkinson’s Disease: A Systematic Review37

). Authors have found that, out of 26 papers surveyed,

addressing gait balance measurement through wearable sensors, 31% were of low methodological

37





quality, 58% were of moderate methodological quality and 11% were of high methodological quality.

81% of the studies reported differences in sensor-based measures of standing balance or walking

stability between different groups of Parkinson’s disease patients and/or healthy controls.

Yang and co-authors have studied the application of the principle of accelerometry measurement to a

wide variety of balance and postural problems, as those addressed by the factors included in the

City4Age risk model (not limited to Gait balance) including posture and movement classification,

estimation of energy expenditure, fall detection and balance control evaluation. Their paper, published

in Sensors in 2015 (A Review of Accelerometry-Based Wearable Motion Detectors for Physical

Activity Monitoring38

) also mentions the possibility to automatically obtain TUG test timings.

Clearly, the above mentioned papers report about research endeavours that certainly need more work to

be transferred into practice; however, they still point to stimulating directions that could be worth

investigating in further research efforts on geriatric risk modelling.

Basic ADLs

Basic ADLs are still poorly covered: out of six Basic ADLs identified in the geriatric practice only one

(Going out) is reliably measured in City4Age testbeds. Some Pilots have proposed to address three

more (Bathing and showering, Personal hygiene and grooming and Toilet hygiene) with tentative

proxies; however their reliability is limited. Two remaining ADLs are not addressed by any Pilot

(Dressing and Self-feeding).

The general problem with Basic ADLs is that they are composed by sets of many granular actions that

are difficult to reliably detect through sensors (as an example, consider the description for Item

Ad.Da.13: “The person is taken to a bathroom and asked to take the cap off a tube of toothpaste, put

toothpaste on a toothbrush, turn on the tap, brush teeth, dampen washcloth, put soap on washcloth,

wash the face and turn off the tap”).

In fact, this level of behaviour requires a “dense sensing environment” in order to be effectively

captured, i.e. an environment where sensors are attached to almost all objects that are related to the

activities under monitoring.

City4Age’s component HARS (Human Activity Recognition System), developed within work-package

WP5, has been shown to be very successful at detecting activities in a dense sensing environment,

including recognizing many geriatrically meaningful ones, such as “Use toilet” (addressing GES Toilet

Hygiene), “Take shower” (addressing GES Bathing and showering), “Get dressed” (addressing GES

Dressing) “Brush teeth” (addressing GES “Personal hygiene and grooming”)39

.

However, few City4Age Pilots are currently equipped with the dense sensing infrastructure needed to

recognize such activities, as this is typically found in highly advanced AAL dwellings, such as some

nursing homes, while it is more difficult to implement in conventional homes or in City

neighbourhoods.

Given the above difficulties, a possible innovative way to be investigated in order to recognize Basic

ADLs without requiring dense sensing, is to appropriately filter accelerometer data, detected (as

previously mentioned) by wearables such as fitness wristband or smartwatches.

For example, the already mentioned Withings smartwatch trackers have been trained to recognize more

than 10 sports activities in this way (e.g. run, swim, ping-pong, volleyball, dancing, etc.40

). A similar

direction could be investigated for recognizing Basic ADLs.

38


39 Relevant details are reported in Deliverable D5.3 The activity recognition system for smart cities, v3

40 It is worth to note that one of the City4Age Pilots is using this capability of Withings trackers in order to

measure GEF Physical Activity




Instrumental ADLs

Only four Instrumental ADLs are convincingly covered, while one is addressed through a tentative

proxy (Laundry) and four are not addressed by any Pilot (Housekeeping, New media communication,

Financial management, Medication).

Some of the Instrumental IADLs can be analysed with the HARS system. For example, one particular

good candidate could be GES Housekeeping, given the involvement of several electrical appliances,

the usage of which can be detected for instance through easy-to-deploy smart-plugs and the

involvement of patterns of rooms changes, which are detected at several City4Age Pilots. Such

elements might make available the right kind of “densely sensed” elementary actions that are needed to

effectively drive HARS.

GES New media communication could be relatively easily implemented, given the fact that this activity

actually “happens online”.

Context and status factors

GEFs in the Context and status factors class (Environment, Dependence, Health-Physical and Health-

Cognitive) are also less covered, with many GESs not addressed by any Pilot and several of them

addressed through obtrusive measurement. However this issue is less essential since – as mentioned in

sub-section 4.2.3 – these factors are mostly intended to be measured with very low frequency (typically

yearly) and thus are not a central part of the continuous unobtrusive monitoring paradigm of City4Age.

6.3 Technical improvements

In addition to improving model coverage, benefits can be extracted also by making factors detection

more accurate through technical improvement in the derivation of underlying measures.

Here we mainly refer to some possible developments that are suggested by the experiment conducted

on the Athens Pilot site, illustrated in Section 5, but the approach can be generalised to all Pilot sites, in

order to benefit potential new instances of the City4Age platform.

As already mentioned, outliers present in experimental data can reveal areas where sensor

improvement is possible.

For instance, Figure 6 below shows the distribution of measure restaurant_visits_week across all care

receivers in the Athens Pilot41

. As it can be seen, one particular care receiver seems to have visited a

restaurant 42 times in a single week (third from left).

This value may be an outlier (it means 6 restaurant visits per day) that may signal a sensor

malfunctioning (e.g. the person is living close to a restaurant and the system misfires

POI_ENTER(location_type: restaurant) LEAs when the persons passes by). However, things are not so

clear as it may seem at a first glance: it may be that the person’s family owns a restaurant and the

system is misclassifying visits to family as visits to restaurants; in this case the sensing equipment

would be OK, but the geriatric value of the measure becomes questionable. Other care receivers in the

Figure are difficult to assess, too. For example, the second more frequent visitor to restaurants accounts

for a week with 13 visits: is this an outlier, signalling sensing issues? Or is she simply a person that has

the habit to have daily meals outside home? Would this be still meaningful from a geriatric point of

view?

Answering these questions would require a deeper analysis of the outliers cases, with a joint technical

and geriatric assessment, to make sound decisions. Such decisions could also lead to modifications in

the measures’ dictionary reported in Section 11 (for example, restaurant_visits_week may be deemed a

geriatrically valid measure up to 14 visits, while after that threshold it should be considered as

missing).

Other issues are more straightforward and may just require better filtering. For instance, Figure 7

shows that the distribution of the walk_speed_outdoor measure (again, in the Athens Pilot) has two

modes, one towards lower speed values, which is likely to be the actual walking speed, and another

41

City4AgeID pseudonyms on the x-axis have been removed for anonymization purposes



around 11 km/h that is probably biking speed, rather than walking speed. Moreover, there are outlier

values up to 130 km/h which certainly represent car speeds.

Figure 6. Value of measure restaurant_visits_week across care receivers

This type of issues can be addressed technically, by better filtering algorithms (in this case it should not

be difficult to establish reasonable cut-off values) or by jointly consider other measurements that allow

to better discriminate the situation (e.g. looking at accelerometer signal patterns, to discriminate

walking from biking from car riding).

Figure 7. Distribution of measure walk_speed_outdoor values (in m/s)

An interesting observation is that the walk_speed_outdoor measure was in fact not filtered in the

experiment illustrated in Section 5: it was included as it is, as represented in Figure 7. Nevertheless, the

experiment gave good results. This is far from being unreasonable: as it is clear from the discussion

in previous sections of this deliverable, biking is a physical activity which has relevant positive impact

on frailty status assessment (see e.g. the Fried Frailty Index in subsection 8.2.1, which explicitly

mention biking). Even outliers that show usage of a motorized transportation vehicle like a car – a

IADL factor – can be meaningful. Thus, after all, the current, unfiltered formulation of

walk_speed_outdoor can be totally acceptable in the model. On the other side, as experiment #6 in

subsection 5.3.5 shows, it seems to be better to keep different features separated, rather than merged in

a single “macro-feature” that summarizes them; thus, even if the current formulation of the



walk_speed_outdoor measure is a good indicator, separating it into different walking, biking and car

usage measures may lead to better model accuracy.

This multi-faceted issue has been mentioned as a paradigmatic example, to underline that there are no

simple cases and each issue should be analysed individually and in depth, if valid indications are to be

derived.

A last opportunity for technical improvement regards the very choice of measures, to be included in the

model. The current dictionary reported in Section 11 has been drafted based on extensive discussion

with Project Pilot representatives, on the backdrop of the first version of the geriatric factors’

modelling.

However, as showed by experiment #7 in subsection 5.3.5, new, more informative measures could still

be devised. In the case of experiment #7, for instance, measures regarding Point of Interest (POIs)

visits and time spent at POIs have been combined to derive a new measure, reporting the average time

spent on a visit for a give type of POI. The addition of these new features seems to have slightly

improved the accuracy of the model.

Other opportunities may arise if more attributes are collected for existing features. E.g., if walking

speed measurements are associated with a time stamp that defines when a certain sample was collected,

a, e.g., walk_speed_outdoor_morning measure can be added that, according to some authors, could be

more meaningful from a geriatric point of view (see relevant paper mentioned in 8.1.8.2).

6.4 Towards a Machine Learning based approach

In the previous sections of this deliverable, and in particular in Section 4, we have derived the

City4Age risk model by basing it as much as possible on current geriatric knowledge, as represented by

established Instruments and Items that are routinely used in the clinical practice today.

The big advantage of this approach is that geriatricians and medical professionals are presented with

information that they already know how to interpret.

However, we have also seen that, when making the effort to substitute current measurement methods

(based on reports, observations or meter application, see Table 3 in sub-section 4.2.2) with more

continuous and unobtrusive methods, based on technology, several challenges have to be tackled.

It seems that these challenges stem from the following basic fact: given the underlying factors that are

affected by MCI and frailty conditions, as identified in geriatric research and practice (which are

represented by GEFs and GESs in the City4Age model) the external variables influenced by such

factors that can be observed by current methods (on one side), and those observable with

unobtrusive sensing technologies (on the other side), are generally different.

Take for instance the GES Walking.

Established Instruments assess this factor with Items like the following:

Question “Did you walk around outside without help?” (e.g. Mt.Ne.01, observed through

self-report)

Question “Are you able to walk 100m?” (e.g. Mt.Sh.04, observed through self-report)

Walking speed (e.g. Mt.Fi.04, observed through meter-based measurement)

Of these, the second and last Items can also be observed by unobtrusive sensing (Measures

WALK_DISTANCE and WALK_SPEED_OUTDOOR) while the first one cannot, because sensing

technologies cannot easily detect the presence or absence of help.

On the other side, sensing technologies may provide additional Measures that are clearly related to

GES Walking, but cannot be readily measured with conventional methods. Examples are:

Walk steps

Walk time

More sophisticated features, such as walk distance (or speed, steps, time) in specific periods

of the day – e.g. in the morning, afternoon, evening or night

This is partly an effect of the fact that Instruments used in today’s geriatrics practice have been

developed and clinically validated several decades ago, when new technologies were not yet

extensively available and thus could not be considered.



Given this context, an alternative strategy for risk assessment is to relax the goal of addressing as many

existing Items as possible, and – while still being guided by the underlying causal structure of the MCI

and frailty phenotypes, as reflected in the City4Age model – aim instead at capturing the full

richness of datasets made available by unobtrusive sensing, that are expected to correlate well

with the GEFs and GESs in the model.

In other words, in addition to populating Individual Monitoring Dashboards to support interpretation

by geriatricians, the model can also be used to train appropriate Machine Learning algorithms and

to derive new MCI/frailty classifiers, able to detect and predict the conditions, similarly to (or

possibly better than) individual current Instruments (see sub-section 3.2.2, Figure 3, part c).

This opportunity is further reinforced by the encouraging results obtained from the experiment

described in Section 5 that, although initially conducted for the sake of model’s statistical assessment,

definitely represents a first step in this direction.

In this sub-section, we clarify the link among the City4Age risk modelling and the possible application

of a general Machine Learning-based approach to modelling, in order to support possible future

research work.

6.4.1 Conceptual framework

As already mentioned, the City4Age geriatric risk model outlines the structure of the causal

connections leading from the MCI and Frailty status of the person to the evidence represented by

relevant behavioural datasets, obtained through available unobtrusive sensing technologies.

In this subsection, we will delineate more precisely this assertion.

For the sake of explanation, we will assume that the risk of MCI/Frailty is modelled by the following

probabilistic variable:

MCIFrailtyRisk = <care recipient will become MCI or frail within 1 year>

This definition can certainly be changed in many ways, as for instance:

Consider classification along more states (instead of Boolean one); for instance, the Fried

Frailty Index classifies persons along three states: not frail, pre-frail and frail; the Lawton

scale of Instrumental ADLs classifies persons on an integer scale from 0 to 8

Creating two different classifiers, one for MCI and one for frailty

Establishing a different prospective period for the onset of the condition

Etc.

However for the purpose of this section and for sake of simplicity, the above mentioned Boolean

classifier is sufficient.

On the other side, we have a number of Measures, i.e. probabilistic variables that represent an

elaboration over the readings of the various sensors deployed at a certain City4Age instance. Some

examples of these variables are:

WALK_SPEED_OUTDOOR = <average outdoor walking speed in meters/seconds in the day>

SHOPS_VISITS = <number of visits to monitored shops in the day>

PUBLICTRANSPORT_RIDES_MONTH = <number of times the user gets on the bus in a month>

(for brevity, we will refer to the above variables as WSO, SV, PRM, in the subsequent text).

Each of the variables has an associated periodicity (e.g. daily or monthly, in the case of the examples

above).

In order to keep the illustration simple, in the rest of this section we assume that each of these variables

is represented through a uniquely established monthly feature (e.g. the Measure’s value itself for

monthly Measures and the average over the month for daily Measures – details on other feature

extraction approaches are discussed in subsections 4.2.4 and 5.3.4 above, and subsection 6.4.4 below).

Framed in this way, the fundamental problem of risk detection in City4Age can be stated as follows:

Each month, we would like to infer the distribution of the query variable MCIFrailtyRisk

given the evidence variables WSO, SV, PRM, etc.



In other words, we want to compute:

P(MCIFrailtyRisk | WSO’, SV’, PRM’, …)

where WSO’, SV’, PRM’, … are specific values for variables WSO, SV, PRM, … respectively, that

have been measured for a given care receiver in a given month.

This means in turn that we need to know the full joint probability distribution table (FJD table) for the

query variable and all the evidence variables:

P(MCIFrailtyRisk, WSO, SV, PRM, …) (1)

Representing such FJD table is clearly infeasible, since evidence variables are too many. At present,

107 of them have already been defined (as illustrated in Section 11), and more of them could be added

in future. This means that, even in the very simplified case of transforming those variables into

discrete, Boolean form (see below) the FJD table in (1) would still have 2108

entries.

We need a different, more tractable way to represent the FJD table in (1) that can be used to reason

about the model. In the following sub-sections we propose a Bayesian Network representation for

achieving this objective, exploiting conditional independence assumptions among GEF/GES variables.

6.4.2 Bayesian network description of the geriatric modelling

The geriatric model illustrated in Section 4 naturally lends itself to a representation as a Bayesian

Network.

In fact, the model has been built based on the established, current geriatric knowledge (see subsection

4.1 and Annex in Section 8), including in particular:

1. The definition of relevant geriatric functional domains (Categories) involved in MCI and

Frailty, that have been hypothesized to be implicated in the conditions’ development

2. The definition of related evidence variables (Items) that have been hypothesized to correlate

well with the above domains

3. A number of clinical trials, that have consistently validated the hypotheses mentioned in 1)

and 2) above, making it possible to build Instruments for detecting and/or predicting

MCI/Frailty onset, which are routinely used in the clinical practice

This medically sound basis allow us to drive out the causal relationships that are needed to build an

initial Bayesian Network model.

To streamline the discussion, before proceeding we will make one more simplifying step: we will

convert each evidence variable into a discrete, Boolean form, by establishing a relevant cut-off

threshold, and assigning true or false labels to all values laying, respectively, on one or the other side of

the threshold.

This approach is in fact common in many widely used scales. For instance, the Fried Frailty Index (see

8.2.1) defines the (Boolean) evidence variable WALK_SPEED for men taller than 173cm as follows:

WALK_SPEED = if “time to walk 15 feet < 7 secs” then true else false

In a similar way, we can also consider GEFs and GESs illustrated in sub-section 4.2.3 as Boolean

variables, ending up with the conditions needed to draw a discrete Boolean Bayesian Network, an

excerpt of which is represented in Figure 8 below.

For space reasons, the Figure only represents the MCIFrailtyRisk variable, the first level of GEFs and

the expansion of (just) the Motility GEF into its composing GESs. Of these GESs, only the Walking

GES is furtherly expanded in its composing Measures.

It should be appreciated that, when fully expanded, the network is much bigger than what appears in

the Figure: in fact, it may include (depending on Pilot installations) up to 10 GEFs, 43 GESs and 107

Measures, for a total of 161 nodes.

In general, in the network presented in Figure 8, MCIFrailtyRisk represents the query variable,

Measures represent evidence variables and GEFs and GESs represents hidden variables.



Approximations notwithstanding, the network of Figure 8 conveys a solid, basilar semantics for the

City4Age geriatric risk model, that can be used to ground an appropriate probabilistic risk detection

system.

In particular, the network causally models:

how the person’s MCI/Frailty status, represented by the MCIFrailtyRisk node, influences a

number of functional “macro-domains”, represented by the GEFs nodes. For instance: the

capability of the person to perform coordinated movement (variable Motility), her ability to

carry out basic ADLs (variable Basic ADLs), or more demanding Instrumental ADLs

(variable IADLs), or her interest in cultural activities (variable Cultural engagement) or

socialization activities (variable Socialization), etc.

Figure 8. Excerpt of a Bayesian Network representing the City4Age geriatric risk model

how macro-domains may in turn influence a number of further geriatric functional domains

represented by GESs. This is the level normally addressed in established geriatric

Instruments, and it includes for example, ability to walk (variable Walking), shopping habits

(variable Shopping), ability to prepare own meals (variable Ability to cook food), etc.

how geriatric domains in turn influence the evidence variables, i.e. the Measures which

relate to the datasets on care receivers’ behaviour, as actually collected by the City4Age

sensing technologies developed in work-package WP3. These include, for example, walking

speed (variable WALK_SPEED_OUTDOOR), number of visits to shops (variable

SHOP_VISITS), number of cooking sessions carried out (variable MEALS_NUM), etc.

Fixing our attention on the Bayesian Network so derived, we are able to highlight an initial set of

important challenges to be tackled in exploring this research avenue, as discussed in the following

subsections.



6.4.3 Validity of conditional independence assumptions

The initial Bayesian Network depicted in the previous subsection makes some conditional

independence assumptions that may require further checking.

For example, according to the network, GEFs and GESs do not interact with each other, and the same

is assumed for Measures.

However, a more focussed scrutiny may suggest additional potential conditional dependencies among

such variables.

For instance, the Motility GEF would probably have partial influence on GEFs Socialization and

Cultural engagement, or on some GESs, like for instance Shopping, as a declining movement

capability can in turn reduce the possibility to conduct the activities measured by such variables, other

conditions remaining unchanged.

On the other hand, such extra conditional dependencies are expected to be reasonably limited.

Moreover, with the availability of larger datasets it would be possible to learn better network structures

directly from data and to define and describe the most meaningful of them. Dependencies that reveal to

be rather tenuous would be discarded, while others may be added to the network.

6.4.4 Feature extraction

Measures are available at the City4Age testbed experiments as discussed in sub-section 4.2.2 and listed

in Section 11.

Extracting features from such Measures is not an easy task, as in the literature there are not many

endeavours that, like City4Age, aim at predicting MCI/frailty on the basis of a large set of attributes,

addressing simultaneously many different functional domains. This leaves us with limited cases that

can be used as a starting point.

Among the works that address a single factor, an interesting example that parallels the City4AAge

attempt in many aspects, is Akl et al., Autonomous Unobtrusive Detection of Mild Cognitive

Impairment in Older Adults, published in the IEEE Transactions on Biomedical Engineering in 201542

and mentioned in subsection 8.1.8.2.

In particular, the paper addresses the GES Walking, through the measurement of (indoor) walking

speed.

In addition to statistical pre-processing similar to what has been illustrated in sub-section 4.2.4, the

authors have investigated different feature types, as follows:

Average of measures, where features were extracted by taking the average of the individual

measures across a temporal window of appropriate size

Probability densities of measures, where features were extracted by estimating the

probability densities of the individual measures in the temporal window

Trajectories of measures, where features were extracted by concatenating the trajectories of

the individual Measures across the temporal window into one vector

The first of these strategies has been applied in City4Age too, in the experiment illustrated in Section 5

above, with good results.

However, with the availability of more and larger datasets, it could be worth to investigate the other

two approaches, particularly in view of the fact that the authors eventually found that third one

produced the best accuracy.

It is also important to underline that the feature extraction process need not be limited to the measures

listed in Section 11. In fact, all City4Age Pilots also store in the Project’s repository Low-level

Elementary Actions (LEAs), i.e. basic events denoting actions performed by the monitored person,

that can be used to compute – through appropriate pre-processing steps – new Measures that may lead

to more interesting features.

42

https://www.ncbi.nlm.nih.gov/pubmed/25585407




It is out of the scope of this deliverable to discuss LEAs (they are the subject of work-package WP3’s

deliverables), however we provide here a brief example to show what is possible:

In many City4Age Pilots the SHOPS_VISIT Measure (i.e. number of shops visited in the

day) is computed on the basis of a couple of LEAs – for example detected by BLE Beacon

technology – called POI_ENTER(location_type: Shop, location_id: shop_ID) and

POI_EXIT(location_type: Shop, location_id: shop_ID), that detect the entrance to and the

exit from a given shop

Such LEAs carry with them additional attributes, such as a timestamp and, in some cases, a

geographical position

Exploiting the above attributes, several extra Measures, more sophisticated than

SHOPS_VISIT, can be derived, such as for instance SHOPS_VISITS_MORNING,

SHOPS_VISITS_FAR, SHOPS_VISITS_LONG, etc.

As hinted in subsection 5.3.6, when discussing experiment #7, the addition of some of these measures

can contribute to accuracy improvement.

6.4.5 Further developments

In the above sub-sections we have proposed to formalize the City4Age geriatric model as a Bayesian

Network structure, and have hinted at some of the issues that need to be solved in order to derive a

classifier based on such network, able to predict the posterior probability of MCI/Frailty given the

evidence represented by Measures obtained by the City4Age data collection platform.

The purpose of the preceding discussion is to clarify several preliminary aspects that may impact the

design and learning of such classifier. However, to proceed in this direction, fundamental aspects have

to be additionally investigated, such as:

Selection of the type of Machine Learning algorithms to be used. For example, the work

from Akl et al. mentioned in the previous sub-section has investigated support vector

machines and random forests. Other algorithms may be more appropriate, given the structure

of the City4Age classification problem and the fact that, as mentioned, it simultaneously

considers many geriatric determinants

Opportunity to relax the discretization hypothesis, possibly considering more general models

that involve both discrete and continuous variables

An additional issue is to consider if it makes sense to use temporal models, such as for

instance Dynamic Bayesian Networks or Long-Short Term Memory networks as hinted in

the project’s DoA. The current interpretation of the City4Age geriatric risk model is that it is

used monthly to periodically assess the MCI/Frailty risks of a care recipient on the basis of

behaviour changes detected by Measures during that same month. In principle this would not

require temporal modelling, given the low temporal variability of robustness conditions,

such as MCI or frailty. On the other hand, to make a well grounded decision it would be

necessary to more precisely assess if the additional dependencies among the GEFs/GESs

variables in a given month and those in previous months, introduced by temporal modelling,

would be actually reflected in an increased predictive accuracy, that it is worth the additional

effort.

6.5 Towards market deployment

Counting on the soundness of the results presented in this deliverable, City4Age Partners, and in

particularly business Partners, may want to apply the lessons learned and the recommendations that

have been presented in this Section to enact a practical and feasible plan that leads to improvement of

the City4Age risk model effectiveness, and achieves industry-grade accuracy. Building on the results

from the Athens Pilot study, reported in Section 5, such plan may be articulated in the following steps:

First, consider larger sets of candidate model’s Measures, covering a wider range of geriatric

factors and sub-factors. In fact, while the Athens Pilot collects 24 measures that covers 9

factors on a total of 45 offered by the City4Age risk model, other Pilots show that many



more can be feasibly addressed by unobtrusive technologies. By just by merging all

technologies experimented in-the-field in all 6 City4Age Pilots, a total of up to 18 different

factors can be covered, addressing up to 70 different measures, more than doubling the

“informative stock” experimented at the Athens Pilot and correspondingly expanding the

opportunities to look for optimal subsets of features, that can provide superior accuracy

Second, as previously mentioned, current implementation of existing measures can be

improved. For example, we discussed the presence of outliers and missing data, that we have

in part left within the dataset used for the statistical assessment conducted on the Athens

Pilot model. It is a relatively straightforward engineering task – and a common one for

commercial partners – to go through each of the measures and try to mitigate or outright

eliminate such problems, boosting in the process the model performance.

Third, the market is continuously producing dramatic innovations in personal data collection

equipment, that can be easily integrated in the City4Age risk model (see recommendations in

subsection 6.2). In particular, since the start of the Ciy4Age Project, new, affordable, and

reliable technologies have appeared that can provide additional, interesting measures to

address several of the City4Age model’s geriatric factors, or even new factors (e.g.

wristband that can measure HR/HRV, recognise a range of sporting or other physical

activities, open APIs e.g. from Google, that offer travelling patterns and activity recognition

on Android platforms, etc.)

Fourth, by conducting large scale experiments (i.e. with thousands of subjects – rather than

10s or 100s as typical of Research and Innovation Actions like City4Age) more

sophisticated, highly parametrized learning schemes (such as Bayesian Networks, as

discussed in subsection 6.4, but also SVMs or even DNNs) can be used in place of the

simple Naïve Bayes model used for the assessment reported in Section 5 – which, although

convenient for investigating smaller datasets, it is a parametric, fixed-size model, that is

known to saturate performance when increasing data sizes, well before achieving the AUC

values that are needed for commercial deployment43

By proceeding in a coordinated way along this path, it should be feasible for City4Age Partners to

achieve industry-grade performance levels – e.g. by identifying more efficient “behavioural markers”

for automatically screening robustness in elderly citizens, and/or by better tuning IMD configurations

to support geriatric assessment by health care professionals – that would ultimately allow to bring the

City4Age early risk modelling and detection strategy to the market.

A final remark, linked to market exploitation, regards the extensibility of the model to other types of

conditions and diseases, that may also benefit from the City4Age approach to unobtrusive, smart-city

based data collection, beyond health in aging populations.

For instance, these may include “behaviour-linked” conditions, such as for instance:

Obesity (monitor physical activity behaviours)

Mental wellbeing at the workplace (monitor cognitive behaviours)

Quality of life after aggressive medical treatments, like for instance in oncology (monitor

ADLs and IADLs behaviours)

Etc.

Most of the City4Age features would still be valid in cases like these, while new, domain-specific

measures would need to be added, by following the same risk modelling processes that have been

illustrated in details in this deliverable.

43

Domingos, A Few Useful Things to Know About Machine Learning, Communications of the ACM, 2012



7 Conclusions

This deliverable has illustrated the work performed to develop and assess a consistent geriatric risk

detection model, appropriate for City4Age’s needs.

A thorough review of established geriatric Instruments for detecting and predicting the onset of MCI

and frailty, allowed to identify the functional domains implicated in these phenotypes and to derive a

set of Factors (Geriatric Factors – GEFs, and Geriatric Sub-factors – GESs) and Measures (collected by

unobtrusive technology) that constitute the basis for deriving the City4Age Geriatric Risk Model.

A tentative proposal to compute a quantitative assessment of GEFs and GESs on the basis statistical

features (or Numerical Indicators – NUIs) extracted from Measures is presented. These computational

rules enable the immediate application of the model for driving the Individual Monitoring Dashboards,

that are specified in deliverable D2.13 Requirements, user scenarios and data visualization mock-ups

for apps/dashboards, v3. In particular, IMDs based on the City4Age model present geriatricians with a

synthesized view of multiple determinants involved in MCI and frailty detection and prediction,

that they wish to see and are trained to interpret.

Measures and NUIs can also be used to train machine learning classifiers, based on the City4Age risk

model. A statistical assessment of the potential accuracy of the City4Age model has been conducted by

training one such classifier on real-life experimental data collected at the Athens Pilot site. A Naïve

Bayes model running on top of a PCA with covered variance of about 0.8 achieved an AUC = 0.710,

statistically different from random classification with significance < 0.05. This represents a

convincing proof-of-concept, that vindicates the original City4Age rationale: that unobtrusive

data collection in smart-city contexts can lead to effective early detection of health decay in

elderly citizens.

The above results envisage the application of the City4Age Geriatric Risk model at two levels:

To build “behavioural markers” based on machine learning techniques, that can be used to

“screen out” robust elderly people, which do not need further medical attention

To provide domain experts (e.g. geriatricians, neurologists) with Individual Monitoring

Dashboards to quickly and efficiently assess the health conditions of those subjects that are

not screened out by the previous step, in order to more precisely decide about the subset that

actually warrant a further comprehensive geriatric assessment

This two-levels protocol underpins a virtual, validated “electronic frailty index” that can be used to

(semi-automatically) screen the city-dwelling elderly population for robustness, achieving at the

same time sufficient specificity (less false positives) and reasonable cost, as advocated by major

geriatric societies.

An account of invaluable lessons that have been learned during the conduction of the City4Age

modelling work and several directions for further research activities and for the Model’s industrial

scale-up have also been presented and discussed.



8 Annex: Current Geriatric Instruments

This Annex lists the Instruments (respectively, for MCI and frailty) that have been selected and

surveyed to derive the City4Age risk modelling, as discussed in sub-section 4.1.3 above.

8.1 MCI Instruments

8.1.1 The Lawton Instrumental Activities of Daily Living (IADL) Scale

8.1.1.1 Overview

This Instrument has been devised in order to provide an assessment of the functional status in older

adults and deliver objective data to assist with targeting individualized care needs.

It guides the clinician to focus on the person’s baseline capabilities, facilitating early recognition of

changes, that may represent decline and/or need for additional medical check.

It is based on the assessment of a set of independent living skills, which are considered more complex

than the basic activities of daily living as measured by e.g. the Katz Index of ADLs.

For this reason, it can detect milder forms of cognitive impairment, with respect to basic ADLs. In

particular, studies have addressed its relationship with MCI assessment and its value in predicting

cognitive decline.

It is based on 8 domains of function. Persons are scored according to their highest level of functioning

in each Category. A summary score ranges from 0 (low function, dependent) to 8 (high function,

independent).

The questionnaire has to be administered by a trained interviewer and collects self-reported

information.

The Instrument validity and reliability have been assessed in some studies, also by determining its

correlation with four other Instruments: Physical Classification (6-point rating of physical health),

Mental Status Questionnaire (10-point test of orientation and memory), Behaviour and Adjustment

rating scales (4-6-point measure of intellectual, person, behavioural and social adjustment), and the

PSMS (6-item ADLs).

8.1.1.2 References

A complete description of the Instrument can be found at the ConsultGeri website of the

Hartford Institute for Geriatric Nursing, URL: https://consultgeri.org/try-this/general-

assessment/issue-23

A study addressing the usage of Instrumental Activities of Daily Living for assessing MCI

and for predicting future cognitive decline is described in the paper referred by the following

Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/22053873

8.1.1.3 Structure

ID Category Item Values

Co.Li.01 Communication Ability to Use Telephone Operates telephone on own

initiative; looks up and dials

numbers /

Dials a few well-known

numbers /

Answers telephone, but does not

dial /

Does not use telephone at all

https://consultgeri.org/try-this/general-assessment/issue-23

https://consultgeri.org/try-this/general-assessment/issue-23

http://www.ncbi.nlm.nih.gov/pubmed/22053873




Sh.Li.02 Shopping Shopping Takes care of all shopping needs

independently /

Shops independently for small

purchases /

Needs to be accompanied on any

shopping trip /

Completely unable to shop

Fo.Li.03 Food Food Preparation Plans, prepares, and serves

adequate meals independently /

Prepares adequate meals if

supplied with ingredients. /

Heats and serves prepared meals

or prepares meals but does not

maintain adequate diet /

Needs to have meals prepared

and served

Ho.Li.04 Housekeeping Housekeeping Maintains house alone with

occasion assistance (heavy

work) /

Performs light daily tasks such as

dishwashing, bed making /

Performs light daily tasks, but

cannot maintain acceptable level

of cleanliness /

Needs help with all home

maintenance tasks /

Does not participate in any

housekeeping tasks.

Ln.Li.05 Laundry Laundry Does personal laundry

completely /

Launders small items, rinses

socks, stockings, etc. /

All laundry must be done by

others

Tr.Li.06 Transportation Mode of Transportation Travels independently on public

transportation or drives own car /

Arranges own travel via taxi, but

does not otherwise use public

transportation /

Travels on public transportation

when assisted or accompanied by

another /

Travel limited to taxi or

automobile with assistance of

another /

Does not travel at all




Me.Li.07 Medication Responsibility for Own

Medications

Is responsible for taking

medication in correct dosages at

correct time /

Takes responsibility if

medication is prepared in

advance in separate dosages /

Is not capable of dispensing own

medication

Fi.Li.08 Finances Ability to Handle Finances Manages financial matters

independently (budgets, writes

checks, pays rent and bills, goes

to bank); collects and keeps track

of income /

Manages day-to-day purchases,

but needs help with banking,

major purchases, etc. /

Incapable of handling money

8.1.2 The OARS Multidimensional Functional Assessment Questionnaire

8.1.2.1 Overview

The OARS Multidimensional Functional Assessment Questionnaire (OMFAQ) has been designed at

the Duke Center for the Study of Aging and Human Development, as a tool for assessing individual

functional status of older persons (including ability to carry out Activities of Daily Living), in order to

group together people of comparable functional status.

The questionnaire has to be administered by a trained interviewer and collects self-reported

information.

In this subsection we consider the part regarding the Instrumental Activities of Daily Living, which

includes questions from 56 to 62. This part has shown good validity and reliability, suggesting it is

superior to many other Instruments. (Even if the recent trend is to do so, it is to be noted that the OARS

team counsels against using the ADL and IADL parts separately from the other questionnaire

components).

The Instrument has also been used as a screening tool and has been studied in assessing the correlation

among MCI and IADLs, also in view of improving prediction of the outcomes of MCI.

8.1.2.2 References

The main reference about the Instrument can be found on the site of the Duke Center for the

Study of Aging and Human Development, at the following URL:

https://sites.duke.edu/centerforaging/services/older-americans-resources-and-services/

A study on the association of the Instrument with MCI is referred by the following Pubmed

URL: http://www.ncbi.nlm.nih.gov/pubmed/22337146

https://sites.duke.edu/centerforaging/services/older-americans-resources-and-services/




8.1.2.3 Structure


Co.Oa.01 Communication Can you use the telephone? Without help, including looking

up numbers and dialling /

With some help (can answer

phone or dial operator in an

emergency, but need a special

phone or help in getting the

number or dialling) /

Completely unable to use the

telephone

Tr.Oa.02 Transportation Can you get to places out of

walking distance?

Without help (drive your own

car, or travel alone on buses, or

taxis) /

With some help (need someone

to help you or go with you when

traveling) /

Unable to travel unless

emergency arrangements are

made for a specialized vehicle

like an ambulance

Sh.Oa.03 Shopping Can you go shopping for

groceries or clothes?

Without help (taking care of all

shopping needs yourself,

assuming you had

transportation) /

With some help (need someone

to go with you on all shopping

trips) /

Completely unable to do any

shopping

Fo.Oa.04 Food Can you prepare your own

meals?

Without help (plan and cook full

meals yourself) /

With some help (can prepare

some things but unable to cook

full meals yourself) /

Completely unable to prepare

any meals

Ho.Oa.05 Housekeeping Can you do your housework? Without help (can clean floors,

etc.) /

With some help (can do light

housework but need help with

heavy work) /


housework




Me.Oa.06 Medication Can you take your own

medicine?

Without help (in the right doses

at the right time) /

With some help (able to take

medicine if someone prepares it

for you and/or reminds you to

take it) /

Completely unable to take your

medicines

Fi.Oa.07 Finances Can you handle your own

money?

Without help (write checks, pay

bills, etc.) /

With some help (manage day-to-

day buying but need help with

managing your check-book and

paying your bills) /

Completely unable to handle

money

8.1.3 The Nottingham Extended Activities of Daily Living

8.1.3.1 Overview

The Nottingham Extended Activities of Daily Living (NEADL) is a ranked assessment of daily living

scale that has been developed to assess activities which may be important to stroke patients who have

been discharged home.

However, its role has also been investigated in the transition from no cognitive impairment to mild

cognitive impairment (MCI) and dementia, in comparison to neuropsychological tests.

The questionnaire incorporates 22 ADL Items in four sections. Some of them regards basic ADLs, of

less use to City4Age, while others regard IADLs, that are more interesting. Subsection 8.1.3.3 below

lists all the Items, including ADLs.

Answers should reflect what has actually been done in the last few weeks. The questionnaire should be

intended as a record of activity rather than capability.

The questionnaire can be administered by an interviewer (including via telephone) or it can be self-

administered (e.g. via mail, as it has been initially designed as a postal questionnaire).

8.1.3.2 References

The paper that describes the Instrument can be found at the following URL:

http://cre.sagepub.com/content/1/4/301.short

A study finding a powerful predictive value for the Instrument, comparable to that of

neuropsychological tests such as MMSE, MoCA, Addenbrooke cognitive assessment and

frontal assessment battery, can be found at the URL:

http://www.neurores.org/index.php/neurores/article/view/316

http://cre.sagepub.com/content/1/4/301.short

http://www.neurores.org/index.php/neurores/article/view/316



8.1.3.3 Structure


Mt.Ne.01 Motility Did you walk around outside? Not at all /

With help /

On your own with help /

On your own

Mt.Ne.02 Motility Did you climb stairs? Not at all /

With help /


On your own

Mt.Ne.03 Motility Did you get in and out of a car? Not at all /

With help /


On your own

Mt.Ne.04 Motility Did you walk over uneven

ground?

Not at all /

With help /


On your own

Mt.Ne.05 Motility Did you cross roads? Not at all /

With help /


On your own

Tr.Ne.06 Transportation Did you travel on public

transport?

Not at all /

With help /


On your own

Ad.Ne.07 Food Did you manage to feed

yourself?

Not at all /

With help /


On your own

Fo.Ne.08 Food Did you manage to make

yourself a hot drink?

Not at all /

With help /


On your own

Fo.Ne.09 Food Did you take hot drinks from one

room to another?

Not at all /

With help /


On your own

Ho.Ne.10 Housekeeping Did you do the washing up? Not at all /

With help /


On your own




Fo.Ne.11 Food Did you make yourself a hot

snack?

Not at all /

With help /


On your own

Fi.Ne.12 Finances Did you manage your own

money when out?

Not at all /

With help /


On your own

Ln.Ne.13 Laundry Did you wash small items of

clothing?

Not at all /

With help /


On your own

Ho.Ne.14 Housekeeping Did you do your own

housework?

Not at all /

With help /


On your own

Sh.Ne.15 Shopping Did you do your own shopping? Not at all /

With help /


On your own

Ln.Ne.16 Laundry Did you do a full clothes wash? Not at all /

With help /


On your own

Cu.Ne.17 Culture Did you read newspapers or

books?

Not at all /

With help /


On your own

Co.Ne.18 Communication Did you use the telephone? Not at all /

With help /


On your own

Cu.Ne.19 Culture Did you write letters? Not at all /

With help /


On your own

So.Ne.20 Socialization Did you go out socially? Not at all /

With help /


On your own

Ac.Ne.21 Activity Did you manage your own

garden?

Not at all /

With help /


On your own




Tr.Ne.22 Transportation Did you drive a car? Not at all /

With help /


On your own

8.1.4 The Direct Assessment of Functional Status

8.1.4.1 Overview

The Direct Assessment of Functional Status is a performance-based measure for evaluating a broad

spectrum of behaviours related to Instrumental Activities of Daily Living.

Although it has been designed and validated as tool for the assessment of the functional competencies

of patients with dementia, rather than predicting cognitive degradation, the Instrument is included here

as a paradigmatic example of a performance-based method (i.e. method relying on direct observation

of subject’s behaviour in controlled settings) a practice that bear some resemblance to the City4Age

unobtrusive monitoring approach.

Items’ Categories are similar to other IADL related Instruments.

The Instrument needs to be administered by an expert team in appropriate settings.

It is worth to note that performance-based measures in general, and DAFS in particular, have been

found to be better in detecting cognitive deficits, than self- or collateral report questionnaires, although

their broad application is limited by higher costs, as they are more time consuming, require space,

specialized equipment and expert examiners.

8.1.4.2 References

A paper describing the Instrument is referred by the following Pubmed URL:


A paper that shows DAFS higher sensitivity in detecting MCI, compared with self- or

collateral report questionnaire, is referred by the following Pubmed URL:

http://jgp.sagepub.com/content/27/4/253.short

A paper addressing the validity of the Instrument and also mentioning the current economic

and organizational limitations to performance-based methods (that can possibly be overcome

by the City4Age paradigm) is referred by the following Pubmed URL:


8.1.4.3 Structure


Ti.Da.01 Time The person is shown four

different times (0300 h, 0800 h,

1030 h and 1215 h) using a large

model of a clock and is asked to

tell the time

Correct / Incorrect answer for

each item

Ti.Da.02 Time The person is asked to state the

date, the day, the month and the

year.


each item


http://jgp.sagepub.com/content/27/4/253.short





Co.Da.03 Communication The person is invited to dial the

operator, dial from a list of

telephone numbers, dial from

oral presentation and dial from

written presentation.

Correct / Incorrect performance

for each item

Co.Da.04 Communication The person is observed on

picking up the receiver, dialling,

hanging up and operating the

telephone in the correct

sequence.


for each item

Co.Da.05 Communication The person is invited to fold a

letter in half, put it in an

envelope, seal the envelope, put

on a stamp, address it (from a

presented stimulus card) and add

a return address (the person’s

own current address, without a

postal code).


for each item

Fi.Da.06 Finance The person is invited to identify

four different coins and three

notes


for each item

Fi.Da.07 Finance The person is invited to count

four amounts of money Correct / Incorrect performance

for each item

Sh.Da.08 Shopping Before the preparation of the

letter (Co.Da.05), the examiner

instructs the person that in 10

min she/he will be going to a

grocery store to select six items:

orange juice, spaghetti, cherry

jam, tuna fish, rice and

tomatoes. After 1 min to recall

as many grocery items as

possible

Number of recalled items

Sh.Da.09 Shopping The person is taken to a

simulated grocery store to pick

out the items from a total of 25

Number of recalled items

Sh.Da.10 Shopping The examiner then gives the

person a written grocery list

(milk, crackers, eggs and

laundry detergent) and asks

them to select the four items and

to hand them to the examiner

Correctly / Incorrectly picked

items

Fi.Da.11 Finance Given a note to pay for the

items, the person is invited to

make change

Change correctly / incorrectly

identified

Tr.Da.12 Transportation The examiner asks the person to

identify a driver's correct

response to 13 road signals

Correctly / Incorrectly identified

signals




Ad.Da.13 ADLs The person is taken to a

bathroom and asked to take the

cap off a tube of toothpaste, put

toothpaste on a toothbrush, turn

on the tap, brush teeth, dampen

washcloth, put soap on

washcloth, wash the face and

turn off the tap


for each item

Ad.Da.14 ADLs The person is invited to use a

hairbrush, put on a coat, button a

coat (three buttons), fasten a zip

and tie shoelaces


for each item

Ad.Da.15 ADLs The person, sitting at a table,

shows how she/he would cut a

steak, take a bite of it, eat soup

and pour water into a glass and

drink it


for each item

8.1.5 The Mini–Mental state Examination

8.1.5.1 Overview

The mini–mental state examination (MMSE) is a 30-point questionnaire that is used extensively in

clinical and research settings to measure cognitive impairment. It was designed to give a practical

clinical assessment of change in cognitive status in geriatric inpatients.

It is also commonly used to screen for dementia, to estimate the severity and progression of cognitive

impairment and to follow the course of cognitive changes in an individual over time.

The Instrument has to be administered by a trained examiner.

Although one of the most frequently noted disadvantages of the MMSE relates to its lack of sensitivity

to MCI, we report it here because of its wider adoption and acceptance. Another disadvantage is that it

is affected by demographic factors, age and education being the most important.

Performance of specific MMSE domains as predictors of subsequent overall cognitive decline has been

object of study.

8.1.5.2 References

MMSE has been copyrighted by Psychological Assessment Resources (PAR). However,

their site (www.parinc.com) was down at the time of writing. A version of the MMSE

questionnaire can be found on the British Columbia Ministry of Health website at the URL:

http://www2.gov.bc.ca/gov/content/health/practitioner-professional-resources/bc-

guidelines/cognitive-impairment#resources

A review of information regarding the psychometric properties and utility of the Mini-

Mental State Examination can be found in the paper referred by the following Pubmed URL:


Studies on the value of single MMSE domains in predicting decline over time can be found

at the following Pubmed URLs: http://www.ncbi.nlm.nih.gov/pubmed/19382130 and


http://www.parinc.com/

http://www2.gov.bc.ca/gov/content/health/practitioner-professional-resources/bc-guidelines/cognitive-impairment#resources

http://www2.gov.bc.ca/gov/content/health/practitioner-professional-resources/bc-guidelines/cognitive-impairment#resources






8.1.5.3 Structure


Ti.Mm.01 Time The person is asked to state the

year, season, month, date and

day of the week.


each element

Sp.Mm.02 Space The person is asked to state the

country, province, city/town,

address (home: street address, in-

facility: building name) and

location (home: room name, in-

facility: floor)


each element

Mr.Mm.03 Memory The examiner says “I am going

to name three objects. When I am

finished, I want you to repeat

them. Remember what they are

because I am going to ask you to

name them again in a few

minutes.” The examiner says the

following words slowly at 1‐second intervals ‐ ball/ car/ man

Correct / Incorrect repetition for

each word

At.Mm.04 Attention The examiner asks the person to

spell the word WORLD. Now

spell it backwards. (in

alternative, the interviewer

counting down from one hundred

by sevens)

Correct / Incorrect answer

(position of the first error)

Mr.Mm.05 Memory The examiner asks the person

what were the three objects

she/he asked to remember?

Correctly / Incorrectly recalled

words

La.Mm.06 Language The examiner show a wristwatch

and asks the person “What is this

called?”. Repeats with a pencil

Correctly / Incorrectly identified

items

La.Mm.07 Language The examiner asks the person to

repeat this phrase “No ifs, ands

or buts.”

Correct / Incorrect repetition

La.Mm.08 Language The examiner asks the person to

read the words on a page and

then do what it says. Then hand

the person the page with

“CLOSE YOUR EYES” on it. If

the subject reads and does not

close their eyes, repeat up to

three times. Score only if subject

closes eyes

Closing / Not closing eyes

La.Mm.09 Language The examiner hands the person a

pencil and paper, then says

“Write any complete sentence on

that piece of paper”

Writing a sentence that make /

does not make sense (ignore

spelling errors)




Vi.Mm.10 Visuospatial The examiner places a drawing

(interlocking pentagons, i.e. two

five-sided figures intersecting to

make a four-sided figure), eraser

and pencil in front of the person

and asks “Copy this drawing

please.” (Allow multiple tries;

wait until person is finished and

hands it back.)

Drawing correctly / incorrectly

copied

La.Mm.11 Language The examiner asks the person if

she/he is right or left‐handed,

takes a piece of paper and hold it

up in front of the person. Then

says “Take this paper in your

right/left hand [note: whichever

is non‐dominant], fold the paper

in half once with both hands and

put the paper down on the floor”.

Correct / incorrect performance

for each item

8.1.6 The Short Test of Mental Status

8.1.6.1 Overview

The Short Test of Mental Status (STMS) is a screening measure of cognition specifically developed for

use in dementia assessment and was intended to be more sensitive to problems of learning and mental

agility that may be seen in mild cognitive impairment (MCI).

In particular, STMS has been found to be better able than MMSE for predicting MCI.

The Instrument is to be administered by a trained examiner.

8.1.6.2 References

The Instrument is described in the paper referred by the following Pubmed URL:


A study that compares the STMS and MMSE in detecting or predicting MCI is referred by

the following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/14676056

A study that identified STMS as the strongest predictor of MCI risk in NC subjects, among

several other considered demographic and clinical variables, is referred by the following

Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/25788555

8.1.6.3 Structure


Sp.Sm.01 Space The person is asked to state

name, address, current location

(building), city state.


each item

Ti.Sm.02 Time The person is asked to state date

(day), month, year. Correct / Incorrect answer for

each item







At.Sm.03 Attention The following digit spans of

increasing length are presented in

sequence to the person, asking

for repetition: 2-9-6-8-3, 5-7-1-9-

4-6, 2-1-5-9-3-6-2.

Length of correctly repeated

spans

Mr.Sm.04 Memory The examiner tell four unrelated

words: “apple”, “Mr. Johnson”,

“charity”, “tunnel”. The person is

requested to repeat all words.

Number of trial to repeat all four

words

At.Sm.05 Attention The examiner proposes 4

arithmetic operations to be

computed: 5x13, 65-7, 58/2,

29+11

Correctly / Incorrectly computed

operations

Ab.Sm.06 Abstraction The examiner proposes three pair

of words and asks the person to

state the abstract interpretation:

orange/banana, dog/horse,

table/bookcase (e.g. dog/horse =

animal)

Correctly / Incorrectly abstracted

interpretations

Vi.Sm.07 Visuospatial The examiner asks the person to

draw a clock face showing 11:15.

Clock face correctly / incorrectly

drawn

Vi.Sm.08 Visuospatial The examiner shows the person

the drawing of a cube and asks to

copy it


copied

La.Sm.09 Language The examiner asks the person to

provide three information

element: “first president”,

“define an island”, “number of

weeks per year”

Correct/incorrect information

provided for each item

Mr.Sm.10 Memory The examiner asks the person to

recall the words from Mr.Sm.04

Correctly/incorrectly recalled

words

8.1.7 The Montreal Cognitive Assessment

8.1.7.1 Overview

The Montreal Cognitive Assessment (MoCA) is a 10-minute cognitive screening tool to assist first-line

physicians in detection of MCI as a clinical state that often progresses to dementia.

The MoCA has been assessed as a predictor of worsening cognitive conditions, e.g. in conversion from

MCI to Alzheimer’s disease.

The Instrument has to be administered by a trained examiner.

8.1.7.2 References

The Instrument is described in the paper found referred by the following Pubmed URL:


A study that assesses the usefulness of the Instrument in predicting conversion to AD is

referred by the following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/24635004





8.1.7.3 Structure


Vi.Mo.01 Visuospatial The examiner instructs the

subject: "Please draw a line,

going from a number to a letter

in ascending order. Begin here

[point to (1)] and draw a line

from 1 then to A then to 2 and

so on. End here [point to (E)].".


each item

Vi.Mo.02 Visuospatial The examiner gives the

following instructions, pointing

to the cube: “Copy this drawing

as accurately as you can, in the

space below”.


copied

Vi.Mo.03 Visuospatial The examiner give the following

instructions: “Draw a clock. Put

in all the numbers and set the

time to 10 past 11”.

Clock face correctly / incorrectly

drawn

La.Mo.04 Language The examiner points to three

animal figures (lion, rhino,

camel) in turn and says: “Tell

me the name of this animal”.

Names correctly / incorrectly

repeated.

Mr.Mo.05 Memory The examiner reads a list of 5

words (face, velvet, church,

daisy, red) at a rate of one per

second, giving the following

instructions: “This is a memory

test. I am going to read a list of

words that you will have to

remember now and later on.

Listen carefully. When I am

through, tell me as many words

as you can remember. It doesn’t

matter in what order you say

them”. When the subject

indicates that she/he has finished

(has recalled all words), or can

recall no more words, the

examiner reads the list a second

time with the following

instructions: “I am going to read

the same list for a second time.

Try to remember and tell me as

many words as you can,

including words you said the

first time.”. At the end of the

second trial, the examiner

informs the person that she/he

will be asked to recall these

words again”

Correct / Incorrect repetition for

each word




At.Mo.06 Attention The examiner gives the

following instruction: “I am

going to say some numbers and

when I am through, repeat them

to me exactly as I said them”.

Then she/he reads the five

number sequence at a rate of one

digit per second: 2-1-8-5-4.

Correctly / Incorrectly repeated

sequence


following instruction: “Now I

am going to say some more

numbers, but when I am through

you must repeat them to me in

the backwards order.” Read the

three number sequence at a rate

of one digit per second: 7-4-2

Correctly / Incorrectly repeated

sequence

At.Mo.08 Attention The examiner reads a list of

letters (F B A C M N A A J K L

B A F A K D E A A A J A M O

F A A B) at a rate of one per

second, after giving the


going to read a sequence of

letters. Every time I say the

letter A, tap your hand once. If I

say a different letter, do not tap

your hand”.

Correctly / Incorrectly tapped As


following instruction: “Now, I

will ask you to count by

subtracting seven from 100, and

then, keep subtracting seven

from your answer until I tell you

to stop.”

Correct / Incorrect subtractions

La.Mo.10 Language The examiner gives the

following instructions: “I am

going to read you a sentence.

Repeat it after me, exactly as I

say it [pause]: I only know that

John is the one to help today.”

Following the response, say:

“Now I am going to read you

another sentence. Repeat it after

me, exactly as I say it [pause]:

The cat always hid under the

couch when dogs were in the

room.”





La.Mo.11 Language The examiner gives the

following instruction: “Tell me

as many words as you can think

of that begin with a certain letter

of the alphabet that I will tell

you in a moment. You can say

any kind of word you want,

except for proper nouns (like

Bob or Boston), numbers, or

words that begin with the same

sound but have a different

suffix, for example, love, lover,

loving. I will tell you to stop

after one minute. Are you

ready? [Pause] Now, tell me as

many words as you can think of

that begin with the letter F.

[time for 60 sec]. Stop.”

Words generated

Ab.Mo.12 Abstraction The examiner asks the subject to

explain what each pair of words

has in common, starting with the

example: “Tell me how an

orange and a banana are alike”.

If the subject answers in a

concrete manner, then say only

one additional time: “Tell me

another way in which those

items are alike”. If the subject

does not give the appropriate

response (fruit), say, “Yes, and

they are also both fruit.” Do not

give any additional instructions

or clarification. After the

practice trial, say: “Now, tell me

how a train and a bicycle are

alike”. Following the response,

administer the second trial,

saying: “Now tell me how a

ruler and a watch are alike” Do

not give any additional

instructions or prompts.

Correctly / Incorrectly abstracted

interpretations

Mr.Mo.13 Memory The examiner gives the

following instruction: “I read

some words to you earlier,

which I asked you to remember.

Tell me as many of those words

as you can remember.”

Correctly / Incorrectly recalled

words




Ti.Mo.14 Time The examiner gives the

following instructions: “Tell me

the date today”. If the subject

does not give a complete

answer, then prompt accordingly

by saying: “Tell me the [year,

month, exact date, and day of

the week].”


each element

Sp.Mo.15 Space Then the examiner says: “Now,

tell me the name of this place,

and which city it is in.”


each element

8.1.8 Predicting MCI onset from gait speed analysis

8.1.8.1 Overview

Although not actually an established Instrument used in research or clinical settings, we would like to

present here a significant result that is representative of new directions that can be potentially pursued.

We refer to the study of Buracchio et al. (see References subsection below) that assesses how gait

speed can be used as predictor of MCI onset.

Subjects for the study were 204 healthy seniors (58% women) from the Oregon Brain Aging Study,

who have been evaluated for up to 20 years with annual neurologic, neuropsychological and motor

examinations.

The authors found that rates of change with aging of gait speed were significantly different between

MCI converters and non-converters (p<0.001).

Change points occurred approximately 14 years prior to MCI onset in men and approximately 6 years

for women.

As gait speed can be measured in a relatively easy and unobtrusive way through a smartphone, this

result is significant for City4Age. In fact, research work aimed at devising unobtrusive classifier

systems based on this model have recently appeared in the literature (see References subsection below).

Authors have also analysed tapping speed and found it significantly different between MCI converters

and non-converters both in the dominant hand (p<0.003) and non-dominant hand (p<0.001). However,

in this case, change points occurred after the onset of MCI, and thus the feature could not be used for

prediction.

8.1.8.2 References

The relevant paper from Buracchio er al. is referred by the following Pubmed URL:


A study that applies the Buracchio et al.s’ model to build an appropriate classifier, based on

datastreams coming from motion sensors placed in a smart home, is referred by the

following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/25585407





8.1.8.3 Structure


Mt.Gm.01 Motility Time in seconds to walk from a

starting point to a marker 15 feet

away, turn, and back at a normal

casual gait for a total of 30 feet

(9.14 meters).

Decrease of 0.013m/s/yr in all

subjects /

Decrease of 0.023m/s/yr in men

who are MCI converters, at 14.2

yr prior to MCI diagnosis /

Decrease of 0.025m/s/yr in

women who are MCI converters,

at 6.0 yr prior to MCI diagnosis

8.2 Frailty Instruments

8.2.1 Fried Frailty Index

8.2.1.1 Overview

The Fried Frailty Index (FFI) is a landmark Instrument that has been conceived as a phenotype

framework for defining frailty and it is one of the most popular and widespread.

It encompasses the assessment of five dimensions that are hypothesized to reflect systems the impaired

regulation of which underlies the syndrome: unintentional weight loss, exhaustion, muscle weakness,

slowness while walking, and low levels of activity.

Corresponding to these dimensions are five specific Items indicating adverse functioning, which are

implemented using a combination of self-reported and performance-based measures.

The instrument classifies people as

‘frail’, when three or more Items are found positive

‘pre-frail’, when one or two Items are found positive

‘non-frail’, when no Items are found positive

8.2.1.2 References

The seminal study of Fried et al. that defined FFI is referred by the following Pubmed URL:

http://www.ncbi.nlm.nih.gov/pubmed/11253156; the same study also offers concurrent and

predictive validity for the Instrument, based on data from the Cardiovascular Health Study

A paper that confirmed FFI validity with data from the Women's Health and Aging Studies

is referred by the following Pubmed URL: https://www.ncbi.nlm.nih.gov/pubmed/16567375

8.2.1.3 Structure


We.Fi.01 Weight The person lost >10 pounds

unintentionally last year True / False

Ex.Fi.02 Exhaustion The person felt that everything

she/he did was an effort in last

week

Rarely or none of the time (<1

day) /

Some or little of the time (1 to 2

days) /

Moderate amount of the time (3

to 4 days) /

Most of the time






Ex.Fi.03 Exhaustion The person felt that she/he could

not get going in last week

Rarely or none of the time (<1

day) /

Some or little of the time (1 to 2

days) /

Moderate amount of the time (3

to 4 days) /

Most of the time

Mt.Fi.04 Motility Time to walk 15 feet (4.57

meters) Number of seconds beyond/not

beyond thresholds:

For men with height <= 173 cm:

7 secs /

For men with height > 173 cm: 6

secs /

For men with height <= 173cm:

7 secs /

For men with height > 173cm: 6

secs

Ac.Fi.05 Activity Physical expenditure on activity

scale per week on 18 items

(Walking for exercise,

moderately strenuous household

chores, mowing or raking the

lawn, gardening, hiking,

jogging, biking, exercise cycle,

dancing, aerobics, bowling, golf,

singles or doubles tennis,

racquetball, calisthenics,

swimming. To compute kcals

expended per week, use the

formula: kcal/week = [activity-

specific MET (kcal/kg × hour) ]

× [duration per session (min) /

60 min] × [body weight (kg)] ×

[number of sessions in the last 2

wk / 2] × [number of months per

year activity was done])

kCal beyond / not beyond

threshold of 270 kCal

Wk.Fi.06 Weakness Grip strength (average of 3

trials, dominant hand, measured

with Jamar hand dynamometer)

Kg beyond/not beyond

thresholds:

For men, BMI <= 24: 29 kg /

For men, BMI 24.1-26: 30 kg /

For men, BMI 26.1-28: 30 kg /

For men, BMI > 28: 32 kg /

For women, BMI <= 23: 17kg /

For women, BMI 23.1-26: 17.3

kg /

For women, BMI 26.1-29: 18 kg

/

For women, BMI > 29: 21 kg



8.2.2 Study of Osteoporotic Fractures index

8.2.2.1 Overview

The Study of Osteoporotic Fractures index (SOF index) is an attempt to propose a simpler alternative

to the FFI Instrument, easier to use in the clinical practice.

It has been developed in the frame of the Study of Osteoporotic Fractures, within a cohort of 6701

women 69 years or older, and its predictive validity has been compared with that of the FFI Instrument.

It uses 3 components: weight loss, the subject's inability to rise from a chair 5 times without using her

arms, and reduced energy level.

Other studies have subsequently extended validation to additional cohorts, including men.

SOF index can classify people as ‘robust’ (no Item is found positive), ‘pre-frail’ (one Item is found

positive), or ‘frail’ (two or more Items are found positive)

8.2.2.2 References

The paper that describes the development of the SOF index Instrument and the comparison

of its predictive validity with that of the FFI Instrument, is referred by the following Pubmed

URL: http://www.ncbi.nlm.nih.gov/pubmed/18299493

A paper that further validates the SOF index and compares it with FFI in a diverse elderly

community-dwelling sample of men and women is referred by the following Pubmed URL:


A study aiming at determining the ability of the SOF index criteria to predict adverse health

outcomes at a one-year follow-up in a sample of older outpatients in Italy is referred by the


8.2.2.3 Structure


We.So.01 Weight Weight loss (irrespective of

intent to lose weight) of 5% or

more in a 2 years period

True / False

Mt.So.02 Motility Subject's inability to rise from a

chair 5 times without using her

arms

True / False

Ex.So.03 Exhaustion Do you feel full of energy? Yes / No

8.2.3 SHARE-FI

8.2.3.1 Overview

SHARE-FI (SHARE Frailty Instrument) is an Instrument based in the on the Survey of Health, Ageing

and Retirement in Europe (SHARE).

SHARE-FI represents an attempt to operationalise the FFI Instrument in a very large European

population-based sample, offering an alternative to FFI in the European context.

To develop the Instrument, SHARE-FI authors selected the five SHARE variables that, in their view,

were the closest to FFI’s Items. On the other hand, their selection was not without significant

departures from FFI and thus SHARE-FI has been validated on its own.

The ultimate goal of the Instrument is to provide European community practitioners with a simple and

valid tool that addresses people over the age of 50.






The SHARE-FI Instrument is publicly available as an HTML/Javascript downloadable calculator44

.

8.2.3.2 References

The main study that defines SHARE-FI can be found at the following Pubmed URL:

http://www.ncbi.nlm.nih.gov/pubmed/20731877. The paper also assesses the validity of the

Instrument in predicting mortality

A paper that provides further prospective validation of SHARE-FI, with a focus on

disability, is referred by the following Pubmed URL:


A study that shows that SHARE-FI predicts mortality similarly to a more complex frailty

Instruments based on Comprehensive Geriatrics Assessment (CGA) is referred by the


8.2.3.3 Structure


Ex.Sh.01 Exhaustion In the last month, have you had

too little energy to do the things

you wanted to do?

Yes / No

We.Sh.02 Weight What has your appetite been

like? Diminution in desire for food

and/or eating less than usual /

No change in desire for food

and/or eating the same as usual /

Increase in desire for food and/or

eating more than usual.

Wk.Sh.03 Weakness Grip strength (highest among

four measures, two for each

hand, taken with a

dynamometer)

Kg (continuous measure)

Mt.Sh.04 Motility Because of a health or physical

problem, do you have any

difficulty walking 100 metres?

(Exclude any difficulties that

you expect to last less than three

months)

Yes / No

Mt.Sh.05 Motility Because of a health or physical

problem, do you have any

difficulty climbing one flight of

stairs without resting? (Exclude

any difficulties that you expect

to last less than three months)

Yes / No

Ac.Sh.06 Activity How often do you engage in

activities that require a low or

moderate level of energy such as

gardening, cleaning the car, or

doing a walk?

Hardly ever, or never /

One to three times a month /

Once a week /

More than once a week.

44

http://bmcgeriatr.biomedcentral.com/articles/10.1186/1471-2318-10-57




http://bmcgeriatr.biomedcentral.com/articles/10.1186/1471-2318-10-57



8.2.4 FRAIL scale

This 5-Items Instrument has been demonstrated as an excellent screening test for clinicians that need to

identify frail persons at risk of developing disability, declining in health functioning and mortality.

The FRAIL scale has been developed with the intent of providing an Instrument that does not requires

face-to-face examination, and could thus result in more efficient identification of the syndrome, that

could be accomplished by telephone or through self-administered forms. These features are aimed at

earlier recognition and treatment by practitioners.

8.2.4.1 References

The International Academy of Nutrition, Health, and Aging proposed the FRAIL scale in the

following papers: http://www.ncbi.nlm.nih.gov/pubmed/18165842 and


Another study that shows that the Instrument can predict future disability before the person

becomes disabled, is referred by the following Pubmed URL:


8.2.4.2 Structure


Ex.Fr.01 Exhaustion How much of the time during

the past 4 weeks did you feel

tired?

Rarely or none of the time /

Some or little of the time /

Moderate amount of the time /

Most of the time

Mt.Fr.02 Motility By yourself and not using aids,

do you have any difficulty

walking up 10 steps without

resting?

Yes / No

Mt.Fr.03 Motility By yourself and not using aids,

do you have any difficulty

walking several hundred yards?

Yes / No

He.Fr..04 Health Did a doctor ever tell you that

you have [illness]? [where

illness is: hypertension,

diabetes, cancer (other than a

minor skin cancer), chronic lung

disease, heart attack, congestive

heart failure, angina, asthma,

arthritis, stroke, and kidney

disease

Number < 5

Number >= 5

We.Fr.05 Weight How much do you weigh with

your clothes on but without

shoes? One year ago in (MO,

YR), how much did you weigh

without your shoes and with

your clothes on?

Decrease > 5% /

Decrease < 5%

8.2.5 PRISMA-7

The PRISMA-7 is a seven-Item, self-completion questionnaire Instrument.

It is intended to be used as a postal questionnaire, or for people who are too unwell to undertake the 4

metre walking speed test.






One point is scored for each of its seven questions and a score of 3 points or more is considered to

identify frailty.

As other similar endeavours, it has been conceived as a cost-effective tool to assess large number of

people, with good sensitivity and specificity. It aims to preview moderate to severe disabilities and may

be less efficient in detecting milder ones. In fact, it is used more as a case finding Instrument (current

state: prevalent cases) rather than a screening one (predicting incident cases).

8.2.5.1 References

A paper that illustrates how PRISMA-7 was conceived and validated is referred by the


A paper that compares PRISMA-7 properties with those of four other Instruments is referred

by the following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/23108163

8.2.5.2 Structure


De.Pr.01 Demographics Are you more than 85 years? Yes / No

De.Pr.02 Demographics Are you male? Yes / No

He.Pr.03 Health In general, do you have any

health problems that require you

to limit your activities?

Yes / No

Dp.Pr.04 Dependence Do you need someone to help

you on a regular basis? Yes / No

He.Pr.05 Health In general, do you have any

health problems that require you

to stay at home?

Yes / No

Dp.Pr.06 Dependence In case of need, can you count

on someone close to you? Yes / No

Mt.Pr.07 Motility Do you regularly use a stick,

walker or wheelchair to get

about?

Yes / No

8.2.6 Edmonton Frail Scale

8.2.6.1 Overview

The Edmonton Frail Scale (EFS) is intended as a simple valid measure of frailty, covering multiple

important Categories, with scores ranging from 0 (not frail) to 17 (very frail).

The EFS does not depend on formal medical training to administer, requires less than 5 minutes of the

patient’s time, and it has been shown to be a valid measure of frailty compared with the clinical

impression of geriatric specialists after their more comprehensive assessment.

8.2.6.2 References

The paper that defines and validates the EFS Instrument is referred by the following Pubmed

URL http://www.ncbi.nlm.nih.gov/pubmed/16757522

A study that further validates EFS demonstrating that it is associated with increasing

comorbidity, hospital lengths of stay, lower use of invasive procedures, and increased

mortality in a known high-risk population (elderly patients with ACS) is referred by Pubmed

URL http://www.ncbi.nlm.nih.gov/pubmed/24183299







8.2.6.3 Structure


Vi.Ed.01 Visuospatial Clock diagram: Place the

numbers in the correct positions

then place the hands to indicate

a time of “10 after 11”

No errors /

Minor spacing errors /

Other errors.

He.Ed.02 Health Hospital admissions in past year 0 /

1-2. /

>=2.

He.Ed.03 Health General health description Excellent, very good, or good /

Fair /

Poor.

Ia.Ed.04 IADLs Requires assistance with

activities such as meal

preparation, shopping,

transportation, dialling

telephone, housekeeping,

laundry, managing money,

taking medications

0-1 /

2-4 /

5-8.

Dp.Ed.05 Dependence Availability of individuals who

are willing and able to support

patient needs

Always /

Sometimes /

Never.

He.Ed.06 Health Five or more different

prescription medications on a

regular basis

Yes / No

Me.Ed.07 Medication Forgetfulness about taking

prescription medications Yes / No

We.Ed.08 Weight Weight loss Yes / No

Mo.Ed.09 Mood Reported feelings of sadness or

depression Yes / No

He.Ed.10 Health Unexpected urinary

incontinence Yes / No

Mt.Ed.11 Motility Patient begins by sitting in a

chair with back and arms

resting, then stands up and

walks approximately 3 m, and

returns to the chair and sits

down

0-10s /

11s-20s /

>= 20s

8.2.7 Tilburg Frailty Indicator

8.2.7.1 Overview

As most frailty assessments Instruments are dominated by biomedical indicators, the Tilburg Frailty

Indicator (TFI) was developed in order to extend this view, and to consider frailty from a live course

perspective, expressing relationships between life-course determinants, diseases, frailty and adverse

outcomes.

Based on this model, TFI addresses the measurement of frailty in community-dwelling older persons,

along three domains: physical, psychological and social.



TFI has been shown to have good predictive validity for quality of life and adverse outcomes such as

disability and receiving personal care, nursing, and informal care.

8.2.7.2 References

The TFI Instrument can be found on the website of the Tilburg University, at the URL

https://www.tilburguniversity.edu/upload/ac3c1079-6188-4bea-b4af-

8f552c07a1d2_tfieng.pdf

A study that assesses the reliability and validity of the TFI Instrument is referred by the


8.2.7.3 Structure


He.Ti.01 Health Do you feel physically healthy? Yes / No

We.Ti.02 Weight Have you lost a lot of weight

recently without wishing to do

so? (‘a lot’ is: 6 kg or more

during the last six months, or 3

kg or more during the last

month)

Yes / No

Mt.Ti.03 Motility Do you experience problems in

your daily life due to difficulty

in walking?

Yes / No

Mt.Ti.04 Motility Do you experience problems in

your daily life due to difficulty

maintaining your balance?

Yes / No

He.Ti.05 Health Do you experience problems in

your daily life due to poor

hearing?

Yes / No

He.Ti.06 Health Do you experience problems in

your daily life due to poor

vision?

Yes / No

Wk.Ti.07 Weakness Do you experience problems in

your daily life due to lack of

strength in your hands?

Yes / No

Ex.Ti.08 Exhaustion Do you experience problems in

your daily life due to physical

tiredness?

Yes / No

Mr.Ti.09 Memory Do you have problems with

your memory? Yes / Sometimes / No

Mo.Ti.10 Mood Have you felt down during the

last month? Yes / Sometimes / No

Mo.Ti.11 Mood Have you felt nervous or

anxious during the last month? Yes / Sometimes / No

Mo.Ti.12 Mood Are you able to cope with

problems well? Yes / No

Dp.Ti.13 Dependence Do you live alone? Yes / No

Dp.Ti.14 Dependence Do you sometimes miss having

people around you? Yes / Sometimes / No

Dp.Ti.15 Dependence Do you receive enough support

from other people? Yes / No

https://www.tilburguniversity.edu/upload/ac3c1079-6188-4bea-b4af-8f552c07a1d2_tfieng.pdf

https://www.tilburguniversity.edu/upload/ac3c1079-6188-4bea-b4af-8f552c07a1d2_tfieng.pdf




8.2.8 Comprehensive Frailty Assessment Instrument

8.2.8.1 Overview

The Comprehensive Frailty Assessment Instrument (CFAI) aims to broaden the body of knowledge

regarding the concept of frailty by introducing a multidimensional, self-administrated instrument

capturing 4 domains of frailty: physical, psychological, social and environmental.

In particular, CFAI was the first Instrument to include the environmental Category in the assessment of

frailty. It was developed on the basis of the TFI Instrument, that already added psychological and social

measurement to the traditional physical measurement.

The relationship of an aging individual with her/his spatial context is assumed to be essential and to

contribute to an aging individual’s quality of life. For instance, as authors report, evidence suggests that

the proximity of amenities and services may promote health either directly or indirectly through the

possibilities they provide for people to live healthy lives.

8.2.8.2 References

The paper that describes the development and initial assessment of the CFAI Instrument is

referred by the following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/23608069

A study that assesses the validity of the CFAI Instrument in comparison with the TFI

instrument is referred by the following URL:

http://www.europeangeriaticmedicine.com/article/S1878-7649(13)00045-4/abstract

8.2.8.3 Structure


Ac.Cf.01 Activity Indicate how long you have

been hampered by your health

status in performing less

demanding activities like

carrying shopping bags

Not at all \

3 months or less \

More than 3 months

Mt.Cf.02 Motility Indicate how long you have


status in walking up a hill/stairs

Not at all \

3 months or less \

More than 3 months



status in bending or lifting

Not at all \

3 months or less \

More than 3 months



status in going for a walk

Not at all \

3 months or less \

More than 3 months

Mo.Cf.05 Mood To what extent do you agree

with the statement “Feeling

unhappy”

Completely disagree /

Disagree /

Neither agree nor disagree /

Agree /

Completely agree.


http://www.europeangeriaticmedicine.com/article/S1878-7649(13)00045-4/abstract





with the statement “Losing self-

confidence”


Disagree /


Agree /

Completely agree.


with the statement “Unable to

cope with problems”


Disagree /


Agree /

Completely agree.



pressure”


Disagree /


Agree /

Completely agree.



worth nothing anymore”


Disagree /


Agree /

Completely agree.


with the statement “I experience

a general sense of emptiness”


Disagree /


Agree /

Completely agree.


with the statement “I miss

having people around me”


Disagree /


Agree /

Completely agree.


with the statement “I often feel

rejected”


Disagree /


Agree /

Completely agree.

Dp.Cf.13 Dependence There are plenty of people I can

lean on when I have problems Yes / No

Dp.Cf.14 Dependence There are many people I can

trust completely Yes / No

Dp.Cf.15 Dependence There are enough people I feel

close to Yes / No

Dp.Cf.16 Dependence How many persons can you rely

on among partner, son and

daughter-in-law? (Social support

network 1)

Number of persons





on among daughter, son-in-law

and grandchildren? (Social

support network 2)

Number of persons


on among brother or sister (-in-

law), family, neighbours and

friends? (Social support network

3)

Number of persons

En.Cf.19 Environment My house is in a bad condition Yes / No

En.Cf.20 Environment My house is not comfortable Yes / No

En.Cf.21 Environment It is difficult to heath my house Yes / No

En.Cf.22 Environment There is insufficient comfort in

my house Yes / No

En.Cf.23 Environment I do not like the neighbourhood Yes / No

8.2.9 Groningen Frailty Indicator

8.2.9.1 Overview

The Groningen Frailty Indicator (GFI) has been developed to identify frailty of home-dwelling as well

as institutionalized elderly people. It comprises both a professional and a self-assessed version.

The GFI is widely used in clinical practice (i.e., geriatric centres, nursing homes, emergency

departments, traumatology, pulmonology, rheumatology, and surgical medicine), in outpatient settings,

and in clinical studies.

To obtain the self-reported version (presented here), the professional version of the GFI was modified

from a patient-orientated questionnaire (with Items such as “Has the patient recently felt downhearted

or sad?”) to an individual-oriented questionnaire (with Items such as “Have you recently felt

downhearted or sad?”) and, as a consequence, the formulations of all Items were adapted.

8.2.9.2 Reference

The paper that illustrates the development and initial test of the GFI Instrument can be found

at the University of Groningen website:

http://www.rug.nl/research/portal/publications/measuring-frailty(f91ecfcc-18e0-481f-84a7-

1230d62e032c).html

A couple of studies that address the predictive validity of GFI (and conclude that, for the

self-reported version, more work is needed) can be found at the following Pubmed URLs:

http://www.ncbi.nlm.nih.gov/pubmed/20353611 and


8.2.9.3 Structure


Sh.Gr.01 Shopping Are you able to carry out

shopping single-handedly and

without any help?

Yes / No

http://www.rug.nl/research/portal/publications/measuring-frailty(f91ecfcc-18e0-481f-84a7-1230d62e032c).html

http://www.rug.nl/research/portal/publications/measuring-frailty(f91ecfcc-18e0-481f-84a7-1230d62e032c).html






Mt.Gr.02 Motility Are you able to carry out

walking around outside (around

the house or to the neighbours)

single-handedly and without any

help?

Yes / No

Ad.Gr.03 ADLs Are you able to carry out

dressing and undressing single-

handedly and without any help?

Yes / No

Ad.Gr.04 ADLs Are you able to carry out going

to the toilet single-handedly and

without any help?

Yes / No

He.Gr.05 Health What mark do you give yourself

for physical fitness? Yes / No

He.Gr.06 Health Do you experience problems in

daily life due to poor vision? Yes / No

He.Gr.07 Health Do you experience problems in

daily life due to being hard of

hearing?

Yes / No

We.Gr.08 Weight During the last 6 months have

you lost a lot of weight

unwillingly? (3 kg in 1 month or

6 kg in 2 months)

Yes / No

He.Gr.09 Health Do you take 4 or more different

types of medicine? Yes / No

Mr.Gr.10 Memory Do you have any complaints

about your memory? Yes / No

Mo.Gr.11 Mood Do you sometimes experience

emptiness around yourself? Yes / No

Mo.Gr.12 Mood Do you sometimes miss people

around yourself? Yes / No

Mo.Gr.13 Mood Do you sometimes feel

abandoned? Yes / No

Mo.Gr.14 Mood Have you recently felt

downhearted or sad? Yes / No

Mo.Gr.15 Mood Have you recently felt nervous

or anxious? Yes / No

8.2.10 The Sherbrooke Postal Questionnaire

8.2.10.1 Overview

The Sherbrooke Postal Questionnaire is a simple six-Items Instruments, suitable for postal

administration, that has been developed in order to enact effective programmes of assessment and

surveillance in a context of secondary prevention. The Items’ Categories cover the physical (four

Items), social support (one Item), and cognitive (one Item) domains of functioning.

The authors found it valid for screening elderly individuals at risk for functional decline, although

another study, comparing it with the GFI and TFI Instruments, suggests that further research is needed.



8.2.10.2 References

The paper that illustrates the development of the SPQ Instruments, also assessing its

predictive validity, can be found at the following Pubmed URL:


The SPQ Instrument has been compared to the GFI and TFI instruments in the study referred

by the following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/20353611

8.2.10.3 Structure


Dp.Sb.01 Dependence Do you live alone? Yes / No

He.Sb.02 Health Do you take more than three

different medications every day? Yes / No

Mt.Sb.03 Motility Do you regularly use a cane, a

walker or a wheelchair to move

about?

Yes / No

He.Sb.04 Health Do you see well? Yes / No

He.Sb.05 Health Do you hear well? Yes / No

Mr.Sb.06 Memory Do you have problems with

your memory? Yes / No

8.2.11 Frailty Index

8.2.11.1 Overview

The Frailty Index is presented here as an example of the Cumulative Deficit model.

We present the version discussed by Mitnitski et al. in the paper referred to in the Reference subsection

below.

It is composed by 40 Items that represent corresponding relevant health deficits (e.g. symptoms, health

attitudes, illnesses, and impaired function), the accumulation of which is assumed to represent the

frailty of the subject.

8.2.11.2 References

The development of the Instrument here presented is illustrated in the paper referred by the

following Pubmed URL: http://www.ncbi.nlm.nih.gov/pubmed/15215283; the paper also

offer predictive validation of the Instrument, by demonstrating the association between it

and mortality

8.2.11.3 Structure


He.Fx.01 Health Eyesight 5 levels Likert scale

He.Fx.02 Health Hearing 5 levels Likert scale

Ad.Fx.03 ADLs Help to eat 3 levels Likert scale

Ad.Fx.04 ADLs Help to dress and undress 3 levels Likert scale

Ad.Fx.05 ADLs Ability to take care of

appearance 3 levels Likert scale

Mt.Fx.06 Motility Help to walk 3 levels Likert scale







Mt.Fx.07 Motility Help to get in and out of bed 3 levels Likert scale

Ad.Fx.08 ADLs Help to take a bath or shower 3 levels Likert scale

Ad.Fx.09 ADLs Help to go to the bathroom 3 levels Likert scale

Co.Fx.10 Communication Help to use the telephone 3 levels Likert scale

Tr.Fx..11 Transportation Help to get to place out of

walking distance 3 levels Likert scale

Sh.Fx.12 Shopping Help in shopping 3 levels Likert scale

Fo.Fx.13 Food Help to prepare own meals 3 levels Likert scale

Ho.Fx.14 Housekeeping Help to do housework 3 levels Likert scale

Me.Fx.15 Medication Ability to take medicine 3 levels Likert scale

Fi.Fx.16 Finances Ability to handle own money 3 levels Likert scale

He.Fx.17 Health Self-rating of health 5 levels Likert scale

He.Fx.18 Health Troubles prevent normal

activities 3 levels Likert scale

Dp.Fx.19 Dependence Living alone Yes / No

He.Fx.20 Health Having a cough Yes / No

Ex.Fx.21 Exhaustion Feeling tired Yes / No

He.Fx.22 Health Nose stuffed up or sneezing Yes / No

He.Fx.23 Health High blood pressure Yes / No

He.Fx.24 Health Heart and circulation problems Yes / No

He.Fx.25 Health Stroke or effects of stroke Yes / No

He.Fx.26 Health Arthritis or rheumatism Yes / No

He.Fx.27 Health Parkinson’s disease Yes / No

He.Fx.28 Health Eye trouble Yes / No

He.Fx.29 Health Ear trouble Yes / No

He.Fx.30 Health Dental problems Yes / No

He.Fx.31 Health Chest problems Yes / No

He.Fx.32 Health Trouble with stomach Yes / No

He.Fx.33 Health Kidney trouble Yes / No

He.Fx.34 Health Losing control of bladder Yes / No

He.Fx.35 Health Losing control of bowels Yes / No

He.Fx.36 Health Diabetes Yes / No

He.Fx.37 Health Trouble with feet or ankles Yes / No

He.Fx.38 Health Trouble with nerves Yes / No

He.Fx.39 Health Skin problems Yes / No

He.Fx.40 Health Fractures Yes / No



9 Annex: Analysis of Items

The following Table presents the result of the Analysis of Items, extracted from currently established Instruments, aimed to assess the risk of MCI and frailty and to

predict the onset of the conditions.

The Table contains a row for each Item surveyed (see sub-section 4.1.3).

The first column assigns to the Item a unique ID that identifies it, for easy reference.

The second column describes the Item, while the third column reports the set of possible values for it.

The fourth column specifies the measurement method used for gauging the Item, in the Instrument to which it belongs, with the following meaning (see sub-section

4.2.2 for a discussion of this aspect):

Report: self-report or report from an informant (e.g. caregiver, family member, etc.)

Observation: observation by medical personnel

Meter: reading of a measurement instrument

The fifth column reports about the geriatric validity of the Instrument to which the Item belongs, with the following meaning:

Diagnostic: the Instrument has been validated for diagnostic usage (i.e. detection of the condition)

Predictive: the Instrument has been validated as a predictor of the condition’s onset

Predictive (study): studies have shown predictive validity for the Instrument, but the need of more work is acknowledged

Research (diagnostic | predictive): the Instrument derives from a research effort, not currently part of routine clinical practice

The last column reports which Measures, specified by the City4Age data collection platform (WP3), have been planned for deployment at the Project Pilot sites in

order to assess the Item (this provides a preliminary assessment of technical feasibility). The different fonts have the following meaning:

BOLD: means that the Measure is deployed as a direct assessment for the Item

NORMAL: means that the Measure is deployed as a proxy for the Item

ITALICS: means that the Measure is deployed as a tentative proxy for the Item; however it is recognized that verification against actual experimental data

is needed and more work may be necessary (e.g. more sophisticated filtering, connection with the City4Age Human Activity Recognition System,

combining more measures and datasets, etc.)

UNDERLINE: the Measure is not unobtrusive, i.e. it is obtained by directly asking a question to the person or to an informant, or by requesting the

person to do some specific measurement actions, such as using a smart scale, etc. (see discussion in sub-section 6.2)



ID Item Values Current

measurement

Geriatric validity City4Age measurement at Pilots

Mt.Ne.01 Did you walk around

outside?

Not at all /

With help /


On your own

Report Predictive WALK_DISTANCE_OUTDOOR

WALK_STEPS_OUTDOOR

WALK_TIME_OUTDOOR

Mt.Cf.04 Indicate how long you have

been hampered by your

health status in going for a

walk

Not at all \

3 months or less \

More than 3 months

Report Diagnostic WALK_DISTANCE_OUTDOOR

WALK_STEPS_OUTDOOR

WALK_TIME_OUTDOOR

Mt.Gr.02 Are you able to carry out

walking around outside

(around the house or to the

neighbours) single-handedly

and without any help?

Yes / No Report Predictive (study) WALK_DISTANCE_OUTDOOR

WALK_STEPS_OUTDOOR

WALK_TIME_OUTDOOR

Mt.Fx.06 Help to walk 3 levels Likert scale Report Predictive WALK_DISTANCE_OUTDOOR

WALK_STEPS_OUTDOOR

WALK_TIME_OUTDOOR

Mt.Sh.04 Because of a health or

physical problem, do you

have any difficulty walking

100 metres? (Exclude any

difficulties that you expect

to last less than three

months)

Yes / No Report Predictive WALK_DISTANCE

WALK_STEPS

Mt.Fr.03 By yourself and not using

aids, do you have any

difficulty walking several

hundred yards?


WALK_STEPS

Mt.Ti.03 Do you experience problems

in your daily life due to

difficulty in walking?


WALK_STEPS




measurement


Mt.Fi.04 Time to walk 15 feet (4.57

meters) Number of seconds

beyond/not beyond

thresholds:

For men with height <= 173

cm: 7 secs /

For men with height > 173

cm: 6 secs /

For men with height <=

173cm: 7 secs /

For men with height >

173cm: 6 secs

Meter Diagnostic,

Predictive WALK_SPEED_OUTDOOR

WALK_SPEED_OUTDOOR_FAST

WALK_DISTANCE_OUTDOOR_FAST

WALK_DISTANCE_OUTDOOR_FAST_PERC

WALK_DISTANCE_OUTDOOR_SLOW_PERC

WALK_TIME_OUTDOOR_FAST

Mt.Gm.01 Time in seconds to walk

from a starting point to a

marker 15 feet away, turn,

and back at a normal casual

gait for a total of 30 feet

(9.14 meters).


all subjects /


men who are MCI

converters, at 14.2 yr prior to

MCI diagnosis /


women who are MCI

converters, at 6.0 yr prior to

MCI diagnosis

Meter Research

(predictive) WALK_SPEED_OUTDOOR






Mt.Ne.02 Did you climb stairs? Not at all /

With help /


On your own

Report Predictive STAIRS_FLOOR_CHANGES_UP




measurement


Mt.Sh.05 Because of a health or

physical problem, do you

have any difficulty climbing

one flight of stairs without

resting? (Exclude any

difficulties that you expect

to last less than three

months)

Yes / No Report Predictive STAIRS_FLOOR_CHANGES_UP

Mt.Fr.02 By yourself and not using

aids, do you have any

difficulty walking up 10

steps without resting?

Yes / No Report Predictive STAIRS_FLOOR_CHANGES_UP



health status in walking up a

hill/stairs

Not at all \

3 months or less \

More than 3 months

Report Diagnostic STAIRS_FLOOR_CHANGES_UP

Mt.Ne.03 Did you get in and out of a

car?

Not at all /

With help /


On your own

Report Predictive

Mt.Ne.04 Did you walk over uneven

ground?

Not at all /

With help /


On your own

Report Predictive

Mt.Ne.05 Did you cross roads? Not at all /

With help /


On your own

Report Predictive

Mt.So.02 Subject's inability to rise

from a chair 5 times without

using her arms

True / False Observation Predictive




measurement


Mt.Ed.11 Patient begins by sitting in a

chair with back and arms

resting, then stands up and

walks approximately 3 m,

and returns to the chair and

sits down

0-10s /

11s-20s /

>= 20s

Meter Diagnostic

Mt.Pr.07 Do you regularly use a stick,

walker or wheelchair to get

about?

Yes / No Report Diagnostic

Mt.Sb.03 Do you regularly use a cane,

a walker or a wheelchair to

move about?

Yes / No Report Predictive

Mt.Ti.04 Do you experience problems


difficulty maintaining your

balance?




health status in bending or

lifting

Not at all \

3 months or less \

More than 3 months

Report Diagnostic

Mt.Fx.07 Help to get in and out of bed 3 levels Likert scale Report Predictive

Ac.Ne.21 Did you manage your own

garden?

Not at all /

With help /


On your own

Report Predictive PHYSICALACTIVITY_INTENSE_TIME

PHYSICALACTIVITY_MODERATE_TIME

PHYSICALACTIVITY_SOFT_TIME




measurement


Ac.Fi.05 Physical expenditure on

activity scale per week on 18

items (Walking for exercise,

moderately strenuous

household chores, mowing

or raking the lawn,

gardening, hiking, jogging,

biking, exercise cycle,

dancing, aerobics, bowling,

golf, singles or doubles

tennis, racquetball,

calisthenics, swimming. To

compute kcals expended per

week, use the formula:

kcal/week = [activity-

specific MET (kcal/kg ×

hour) ] × [duration per

session (min) / 60 min] ×

[body weight (kg)] ×

[number of sessions in the

last 2 wk / 2] × [number of

months per year activity was

done])

kCal beyond / not beyond

threshold of 270 kCal

Report Diagnostic,

Predictive PHYSICALACTIVITY_CALORIES

PHYSICALACTIVITY_NUM

Ac.Sh.06 How often do you engage in

activities that require a low

or moderate level of energy

such as gardening, cleaning

the car, or doing a walk?

Hardly ever, or never /

One to three times a month /

Once a week /

More than once a week.

Report Predictive PHYSICALACTIVITY_MODERATE_TIME



Ac.Cf.01 Indicate how long you have


health status in performing

less demanding activities

like carrying shopping bags

Not at all \

3 months or less \

More than 3 months

Report Diagnostic PHYSICALACTIVITY_SOFT_TIME





measurement


Ad.Fx.08 Help to take a bath or

shower 3 levels Likert scale Report Predictive BATHROOMS_VISITS

BATHROOM_TIME

Ad.Gr.03 Are you able to carry out

dressing and undressing

single-handedly and without

any help?

Yes / No Report Predictive (study)

Ad.Fx.04 Help to dress and undress 3 levels Likert scale Report Predictive

Ad.Ne.07 Did you manage to feed

yourself?

Not at all /

With help /


On your own

Report Predictive

Ad.Fx.03 Help to eat 3 levels Likert scale Report Predictive

Ad.Da.15 The person, sitting at a table,

shows how she/he would cut

a steak, take a bite of it, eat

soup and pour water into a

glass and drink it

Correct / Incorrect

performance for each item

Observation Diagnostic

Ad.Da.13 The person is taken to a

bathroom and asked to take

the cap off a tube of

toothpaste, put toothpaste on

a toothbrush, turn on the tap,

brush teeth, dampen

washcloth, put soap on

washcloth, wash the face

and turn off the tap

Correct / Incorrect


Observation Diagnostic BATHROOMS_VISITS

BATHROOM_TIME

Ad.Da.14 The person is invited to use a

hairbrush, put on a coat,

button a coat (three buttons),

fasten a zip and tie shoelaces

Correct / Incorrect






measurement


Ad.Fx.05 Ability to take care of

appearance 3 levels Likert scale Report Predictive

Ad.Gr.04 Are you able to carry out

going to the toilet single-

handedly and without any

help?

Yes / No Report Predictive (study) RESTROOM_VISITS

RESTROOM_TIME

Ad.Fx.09 Help to go to the bathroom 3 levels Likert scale Report Predictive RESTROOM_VISITS

RESTROOM_TIME

Ia.Ed.04 Requires assistance with

activities such as meal

preparation, shopping,

transportation, dialling

telephone, housekeeping,

laundry, managing money,

taking medications

0-1 /

2-4 /

5-8.

Report Diagnostic

Fo.Li.03 Food Preparation Plans, prepares, and serves

adequate meals

independently /

Prepares adequate meals if

supplied with ingredients. /

Heats and serves prepared

meals or prepares meals but

does not maintain adequate

diet /

Needs to have meals

prepared and served

Report Predictive MEALS_NUM




measurement


Fo.Oa.04 Can you prepare your own

meals?

Without help (plan and cook

full meals yourself) /

With some help (can prepare

some things but unable to

cook full meals yourself) /

Completely unable to prepare

any meals

Report Diagnostic MEALS_NUM

Fo.Fx.13 Help to prepare own meals 3 levels Likert scale Report Predictive MEALS_NUM

Fo.Ne.08 Did you manage to make

yourself a hot drink?

Not at all /

With help /


On your own


Fo.Ne.09 Did you take hot drinks from

one room to another?

Not at all /

With help /


On your own

Report Predictive

Fo.Ne.11 Did you make yourself a hot

snack?

Not at all /

With help /


On your own


Ho.Ne.14 Did you do your own

housework?

Not at all /

With help /


On your own

Report Predictive




measurement


Ho.Li.04 Housekeeping Maintains house alone with

occasion assistance (heavy

work) /

Performs light daily tasks

such as dishwashing, bed

making /

Performs light daily tasks,

but cannot maintain

acceptable level of

cleanliness /

Needs help with all home

maintenance tasks /

Does not participate in any

housekeeping tasks.

Report Predictive

Ho.Oa.05 Can you do your housework? Without help (can clean

floors, etc.) /

With some help (can do light

housework but need help

with heavy work) /


housework

Report Diagnostic

Ho.Ne.10 Did you do the washing up? Not at all /

With help /


On your own

Report Predictive

Ho.Fx.14 Help to do housework 3 levels Likert scale Report Predictive




measurement


Ln.Li.05 Laundry Does personal laundry

completely /

Launders small items, rinses

socks, stockings, etc. /

All laundry must be done by

others

Report Predictive WASHINGMACHINE_SESSIONS

Ln.Ne.13 Did you wash small items of

clothing?

Not at all /

With help /


On your own


Ln.Ne.16 Did you do a full clothes

wash?

Not at all /

With help /


On your own


Co.Li.01 Ability to Use Telephone Operates telephone on own

initiative; looks up and dials

numbers /

Dials a few well-known

numbers /

Answers telephone, but does

not dial /

Does not use telephone at all

Report Predictive PHONECALLS_PLACED

PHONECALLS_RECEIVED

PHONECALLS_LONG_PLACED_PERC

PHONECALLS_LONG_RECEIVED_PERC

PHONECALLS_MISSED

PHONECALLS_PLACED_PERC

PHONECALLS_RECEIVED_PERC

PHONECALLS_SHORT_PLACED_PERC

PHONECALLS_SHORT_RECEIVED_PERC




measurement


Co.Oa.01 Can you use the telephone? Without help, including

looking up numbers and

dialling /

With some help (can answer

phone or dial operator in an

emergency, but need a

special phone or help in

getting the number or

dialling) /

Completely unable to use the

telephone

Report Diagnostic PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED





Co.Ne.18 Did you use the telephone? Not at all /

With help /


On your own

Report Predictive PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED





Co.Da.03 The person is invited to dial

the operator, dial from a list

of telephone numbers, dial

from oral presentation and

dial from written

presentation.

Correct / Incorrect


Observation Diagnostic PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED








measurement


Co.Da.04 The person is observed on

picking up the receiver,

dialling, hanging up and

operating the telephone in

the correct sequence.

Correct / Incorrect


Observation Diagnostic PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED





Co.Fx.10 Help to use the telephone 3 levels Likert scale Report Predictive PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED





Co.Da.05 The person is invited to fold

a letter in half, put it in an

envelope, seal the envelope,

put on a stamp, address it

(from a presented stimulus

card) and add a return

address (the person’s own

current address, without a

postal code).

Correct / Incorrect






measurement


Sh.Li.02 Shopping Takes care of all shopping

needs independently /

Shops independently for

small purchases /

Needs to be accompanied on

any shopping trip /

Completely unable to shop

Report Predictive SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS

SUPERMARKET_VISITS_WEEK

SUPERMARKET _TIME

SUPERMARKET _TIME_PERC

Sh.Oa.03 Can you go shopping for

groceries or clothes?

Without help (taking care of

all shopping needs yourself,

assuming you had

transportation) /

With some help (need

someone to go with you on

all shopping trips) /


shopping

Report Diagnostic SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME


Sh.Ne.15 Did you do your own

shopping?

Not at all /

With help /


On your own

Report Predictive SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME





measurement


Sh.Da.08 Before the preparation of the

letter (Co.Da.05), the

examiner instructs the

person that in 10 min she/he

will be going to a grocery

store to select six items:

orange juice, spaghetti,

cherry jam, tuna fish, rice

and tomatoes. After 1 min to

recall as many grocery items

as possible

Number of recalled items Observation Diagnostic SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME


Sh.Da.09 The person is taken to a

simulated grocery store to

pick out the items from a

total of 25

Number of recalled items Observation Diagnostic SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME


Sh.Da.10 The examiner then gives the

person a written grocery list

(milk, crackers, eggs and

laundry detergent) and asks

them to select the four items

and to hand them to the

examiner

Correctly / Incorrectly

picked items

Observation Diagnostic SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME


Sh.Gr.01 Are you able to carry out

shopping single-handedly

and without any help?





measurement


Sh.Fx.12 Help in shopping 3 levels Likert scale Report Predictive

Tr.Li.06 Mode of Transportation Travels independently on

public transportation or

drives own car /

Arranges own travel via taxi,

but does not otherwise use

public transportation /

Travels on public

transportation when assisted

or accompanied by another /

Travel limited to taxi or

automobile with assistance of

another /

Does not travel at all

Report Predictive PUBLICTRANSPORT_RIDES_MONTH

PUBLICTRANSPORT_DISTANCE_MONTH

PUBLICTRANSPORT_TIME

Tr.Ne.06 Did you travel on public

transport?

Not at all /

With help /


On your own

Report Predictive PUBLICTRANSPORT_RIDES_MONTH



Tr.Oa.02 Can you get to places out of

walking distance?

Without help (drive your own

car, or travel alone on buses,

or taxis) /

With some help (need

someone to help you or go

with you when traveling) /

Unable to travel unless

emergency arrangements are

made for a specialized

vehicle like an ambulance

Report Diagnostic TRANSPORT_TIME

PUBLICTRANSPORT_RIDES_MONTH






measurement


Tr.Fx..11 Help to get to place out of

walking distance 3 levels Likert scale Report Predictive TRANSPORT_TIME




Tr.Ne.22 Did you drive a car? Not at all /

With help /


On your own

Report Predictive

Tr.Da.12 The examiner asks the

person to identify a driver's

correct response to 13 road

signals


identified signals


Fi.Li.08 Ability to Handle Finances Manages financial matters

independently (budgets,

writes checks, pays rent and

bills, goes to bank); collects

and keeps track of income /

Manages day-to-day

purchases, but needs help

with banking, major

purchases, etc. /

Incapable of handling money

Report Predictive

Fi.Oa.07 Can you handle your own

money?

Without help (write checks,

pay bills, etc.) /

With some help (manage

day-to-day buying but need

help with managing your

check-book and paying your

bills) /

Completely unable to handle

money

Report Diagnostic




measurement


Fi.Ne.12 Did you manage your own

money when out?

Not at all /

With help /


On your own

Report Predictive

Fi.Da.06 The person is invited to

identify four different coins

and three notes

Correct / Incorrect



Fi.Da.07 The person is invited to

count four amounts of

money

Correct / Incorrect



Fi.Da.11 Given a note to pay for the

items, the person is invited

to make change

Change correctly / incorrectly

identified


Fi.Fx.16 Ability to handle own

money 3 levels Likert scale Report Predictive

Me.Li.07 Responsibility for Own

Medications

Is responsible for taking

medication in correct dosages

at correct time /

Takes responsibility if

medication is prepared in

advance in separate dosages /

Is not capable of dispensing

own medication

Report Predictive

Me.Oa.06 Can you take your own

medicine?

Without help (in the right

doses at the right time) /

With some help (able to take

medicine if someone

prepares it for you and/or

reminds you to take it) /

Completely unable to take

your medicines

Report Diagnostic




measurement


Me.Ed.07 Forgetfulness about taking

prescription medications Yes / No Report Diagnostic

Me.Fx.15 Ability to take medicine 3 levels Likert scale Report Predictive

So.Ne.20 Did you go out socially? Not at all /

With help /


On your own

Report Predictive VISITS_PAYED_WEEK

VISITS_RECEIVED_WEEK

VISITORS_WEEK

SENIORCENTER_VISITS


SENIORCENTER_VISITS_MONTH

SENIORCENTER_TIME

SENIORCENTER_TIME_OUT_PERC

SENIORCENTER_LONG_VISITS

FOODCOURT_VISITS_MONTH

FOODCOURT_VISITS_WEEK

FOODCOURT_TIME

PUBLICPARK_VISITS

PUBLICPARK _VISITS_MONTH

PUBLICPARK _TIME

OTHERSOCIAL_VISITS

OTHERSOCIAL _LONG_VISITS

OTHERSOCIAL _TIME

OTHERSOCIAL _TIME_OUT_PERC

RESTAURANTS_VISITS_WEEK

RESTAURANTS_VISITS_MONTH

RESTAURANTS_TIME




measurement


Cu.Ne.17 Did you read newspapers or

books?

Not at all /

With help /


On your own

Report Predictive CINEMA_VISITS

CINEMA_VISITS_MONTH

CINEMA_TIME

CULTUREPOI_VISITS_MONTH

CULTUREPOI_VISITS_TIME_PERC_MONTH

TVWATCHING_TIME

TVWATCHING_TIME_PERC

Cu.Ne.19 Did you write letters? Not at all /

With help /


On your own

Report Predictive

En.Cf.19 My house is in a bad

condition Yes / No Report Diagnostic

En.Cf.20 My house is not comfortable Yes / No Report Diagnostic

En.Cf.21 It is difficult to heath my

house Yes / No Report Diagnostic

En.Cf.22 There is insufficient comfort

in my house Yes / No Report Diagnostic

En.Cf.23 I do not like the

neighbourhood Yes / No Report Diagnostic

Dp.Ti.13 Do you live alone? Yes / No Report Predictive

Dp.Sb.01 Do you live alone? Yes / No Report Predictive

Dp.Fx.19 Living alone Yes / No Report Predictive

Dp.Pr.06 In case of need, can you

count on someone close to

you?


Dp.Ed.05 Availability of individuals

who are willing and able to

support patient needs

Always /

Sometimes /

Never.

Report Diagnostic




measurement


Dp.Cf.17 How many persons can you

rely on among daughter,

son-in-law and

grandchildren? (Social

support network 2)

Number of persons Report Diagnostic


rely on among partner, son

and daughter-in-law? (Social

support network 1)



rely on among brother or

sister (-in-law), family,

neighbours and friends?

(Social support network 3)


Dp.Cf.13 There are plenty of people I

can lean on when I have

problems


Dp.Cf.14 There are many people I can

trust completely Yes / No Report Diagnostic

Dp.Cf.15 There are enough people I

feel close to Yes / No Report Diagnostic

Dp.Ti.15 Do you receive enough

support from other people? Yes / No Report Predictive

Dp.Pr.04 Do you need someone to

help you on a regular basis? Yes / No Report Diagnostic

Dp.Ti.14 Do you sometimes miss

having people around you? Yes / Sometimes / No Report Predictive

He.Fx.17 Self-rating of health 5 levels Likert scale Report Predictive GP_VISITS_MONTH

GP_TIME_MONTH

He.Ed.03 General health description Excellent, very good, or good

/ Fair /

Poor.

Report Diagnostic GP_VISITS_MONTH

GP_TIME_MONTH




measurement


He.Ti.01 Do you feel physically

healthy? Yes / No Report Predictive GP_VISITS_MONTH

GP_TIME_MONTH

He.Gr.05 What mark do you give

yourself for physical fitness? Yes / No Report Predictive (study) GP_VISITS_MONTH

GP_TIME_MONTH

He.Ed.06 Five or more different

prescription medications on

a regular basis

Yes / No Report Diagnostic PHARMACY_VISITS

PHARMACY _VISITS_WEEK

PHARMACY _VISITS_MONTH

PHARMACY _TIME

He.Gr.09 Do you take 4 or more

different types of medicine? Yes / No Report Predictive (study) PHARMACY_VISITS



PHARMACY _TIME

He.Sb.02 Do you take more than three

different medications every

day?

Yes / No Report Predictive PHARMACY_VISITS



PHARMACY _TIME

He.Fr..04 Did a doctor ever tell you

that you have [illness]?

[where illness is:

hypertension, diabetes,

cancer (other than a minor

skin cancer), chronic lung

disease, heart attack,

congestive heart failure,

angina, asthma, arthritis,

stroke, and kidney disease

Number < 5

Number >= 5

Report Predictive

He.Ed.02 Hospital admissions in past

year 0 /

1-2. /

>=2.

Report Diagnostic




measurement


He.Pr.03 In general, do you have any

health problems that require

you to limit your activities?


He.Pr.05 In general, do you have any

health problems that require

you to stay at home?


He.Fx.02 Hearing 5 levels Likert scale Report Predictive

He.Fx.29 Ear trouble Yes / No Report Predictive

He.Sb.05 Do you hear well? Yes / No Report Predictive

He.Ti.05 Do you experience problems

in your daily life due to poor

hearing?


He.Gr.07 Do you experience problems

in daily life due to being

hard of hearing?


He.Fx.01 Eyesight 5 levels Likert scale Report Predictive

He.Fx.28 Eye trouble Yes / No Report Predictive

He.Sb.04 Do you see well? Yes / No Report Predictive

He.Ti.06 Do you experience problems

in your daily life due to poor

vision?


He.Gr.06 Do you experience problems

in daily life due to poor

vision?


He.Fx.18 Troubles prevent normal

activities 3 levels Likert scale Report Predictive

He.Fx.20 Having a cough Yes / No Report Predictive

He.Fx.22 Nose stuffed up or sneezing Yes / No Report Predictive

He.Ed.10 Unexpected urinary

incontinence Yes / No Report Diagnostic




measurement


He.Fx.23 High blood pressure Yes / No Report Predictive

He.Fx.24 Heart and circulation

problems Yes / No Report Predictive

He.Fx.25 Stroke or effects of stroke Yes / No Report Predictive

He.Fx.26 Arthritis or rheumatism Yes / No Report Predictive

He.Fx.27 Parkinson’s disease Yes / No Report Predictive

He.Fx.30 Dental problems Yes / No Report Predictive

He.Fx.31 Chest problems Yes / No Report Predictive

He.Fx.32 Trouble with stomach Yes / No Report Predictive

He.Fx.33 Kidney trouble Yes / No Report Predictive

He.Fx.34 Losing control of bladder Yes / No Report Predictive

He.Fx.35 Losing control of bowels Yes / No Report Predictive

He.Fx.36 Diabetes Yes / No Report Predictive

He.Fx.37 Trouble with feet or ankles Yes / No Report Predictive

He.Fx.38 Trouble with nerves Yes / No Report Predictive

He.Fx.39 Skin problems Yes / No Report Predictive

He.Fx.40 Fractures Yes / No Report Predictive

We.Fi.01 The person lost >10 pounds

unintentionally last year True / False Report Diagnostic,

Predictive

WEIGHT

We.So.01 Weight loss (irrespective of

intent to lose weight) of 5%

or more in a 2 years period

True / False Report Predictive WEIGHT




measurement


We.Ti.02 Have you lost a lot of weight

recently without wishing to

do so? (‘a lot’ is: 6 kg or

more during the last six

months, or 3 kg or more

during the last month)

Yes / No Report Predictive WEIGHT

We.Gr.08 During the last 6 months

have you lost a lot of weight

unwillingly? (3 kg in 1

month or 6 kg in 2 months)

Yes / No Report Predictive (study) WEIGHT

We.Ed.08 Weight loss Yes / No Report Diagnostic WEIGHT

We.Fr.05 How much do you weigh

with your clothes on but

without shoes? One year ago

in (MO, YR), how much did

you weigh without your

shoes and with your clothes

on?

Decrease > 5% /

Decrease < 5%

Report Predictive WEIGHT

We.Sh.02 What has your appetite been

like? Diminution in desire for food

and/or eating less than usual /

No change in desire for food

and/or eating the same as

usual /

Increase in desire for food

and/or eating more than

usual.

Report Predictive WEIGHT

Wk.Ti.07 Do you experience problems

in your daily life due to lack

of strength in your hands?

Yes / No Report Predictive WEAKNESS




measurement


Wk.Fi.06 Grip strength (average of 3

trials, dominant hand,

measured with Jamar hand

dynamometer)

Kg beyond/not beyond

thresholds:

For men, BMI <= 24: 29 kg /

For men, BMI 24.1-26: 30 kg

/

For men, BMI 26.1-28: 30 kg

/

For men, BMI > 28: 32 kg /

For women, BMI <= 23:

17kg /

For women, BMI 23.1-26:

17.3 kg /

For women, BMI 26.1-29: 18

kg /

For women, BMI > 29: 21 kg

Meter Diagnostic,

Predictive

WEAKNESS

Wk.Sh.03 Grip strength (highest

among four measures, two

for each hand, taken with a

dynamometer)

Kg (continuous measure)

Meter Predictive WEAKNESS

Ex.Fi.02 The person felt that

everything she/he did was an

effort in last week

Rarely or none of the time

(<1 day) /

Some or little of the time (1

to 2 days) /

Moderate amount of the time

(3 to 4 days) /

Most of the time

Report Diagnostic,

Predictive

EXHAUSTION




measurement


Ex.Fi.03 The person felt that she/he

could not get going in last

week

Rarely or none of the time

(<1 day) /

Some or little of the time (1

to 2 days) /


(3 to 4 days) /

Most of the time

Report Diagnostic,

Predictive

EXHAUSTION

Ex.So.03 Do you feel full of energy? Yes / No Report Predictive EXHAUSTION

Ex.Sh.01 In the last month, have you

had too little energy to do

the things you wanted to do?

Yes / No Report Predictive EXHAUSTION

Ex.Fr.01 How much of the time

during the past 4 weeks did

you feel tired?

Rarely or none of the time /

Some or little of the time /


/

Most of the time

Report Predictive EXHAUSTION

Ex.Ti.08 Do you experience problems


physical tiredness?

Yes / No Report Predictive EXHAUSTION

Ex.Fx.21 Feeling tired Yes / No Report Predictive EXHAUSTION

Ab.Sm.06 The examiner proposes three

pair of words and asks the

person to state the abstract

interpretation:

orange/banana, dog/horse,

table/bookcase (e.g.

dog/horse = animal)


abstracted interpretations

Observation Predictive WORD_PAIRS_TEST




measurement


Ab.Mo.12 The examiner asks the

subject to explain what each

pair of words has in

common, starting with the

example: “Tell me how an

orange and a banana are

alike”. If the subject answers

in a concrete manner, then

say only one additional time:

“Tell me another way in

which those items are alike”.

If the subject does not give

the appropriate response

(fruit), say, “Yes, and they

are also both fruit.” Do not

give any additional

instructions or clarification.

After the practice trial, say:

“Now, tell me how a train

and a bicycle are alike”.

Following the response,

administer the second trial,

saying: “Now tell me how a

ruler and a watch are alike”

Do not give any additional

instructions or prompts.


abstracted interpretations

Observation Predictive WORD_PAIRS_TEST

At.Mo.09 The examiner gives the

following instruction: “Now,

I will ask you to count by

subtracting seven from 100,

and then, keep subtracting

seven from your answer

until I tell you to stop.”

Correct / Incorrect

subtractions

Observation Predictive




measurement


At.Mm.04 The examiner asks the person

to spell the word WORLD.

Now spell it backwards. (in

alternative, the interviewer

counting down from one

hundred by sevens)


(position of the first error)

Observation Predictive (study)

At.Sm.03 The following digit spans of

increasing length are

presented in sequence to the

person, asking for repetition:

2-9-6-8-3, 5-7-1-9-4-6, 2-1-

5-9-3-6-2.

Length of correctly repeated

spans


At.Sm.05 The examiner proposes 4

arithmetic operations to be

computed: 5x13, 65-7, 58/2,

29+11


computed operations




going to say some numbers

and when I am through,

repeat them to me exactly as

I said them”. Then she/he

reads the five number

sequence at a rate of one

digit per second: 2-1-8-5-4.


repeated sequence





measurement



following instruction: “Now

I am going to say some more

numbers, but when I am

through you must repeat

them to me in the backwards

order.” Read the three

number sequence at a rate of

one digit per second: 7-4-2


repeated sequence


At.Mo.08 The examiner reads a list of

letters (F B A C M N A A J

K L B A F A K D E A A A J

A M O F A A B) at a rate of

one per second, after giving

the following instruction: “I

am going to read a sequence

of letters. Every time I say

the letter A, tap your hand

once. If I say a different

letter, do not tap your hand”.

Correctly / Incorrectly tapped

As


Mr.Gr.10 Do you have any complaints

about your memory? Yes / No Report Predictive (study) MEMORY

Mr.Sb.06 Do you have problems with

your memory? Yes / No Report Predictive MEMORY

Mr.Ti.09 Do you have problems with

your memory? Yes / Sometimes / No Report Predictive MEMORY




measurement


Mr.Mm.03 The examiner says “I am

going to name three objects.

When I am finished, I want

you to repeat them.

Remember what they are

because I am going to ask

you to name them again in a

few minutes.” The examiner

says the following words

slowly at 1‐second intervals ‐ ball/ car/ man


for each word

Observation Predictive (study) MEMORY

Mr.Mm.05 The examiner asks the person

what were the three objects

she/he asked to remember?


recalled words

Observation Predictive (study) MEMORY

Mr.Sm.04 The examiner tell four

unrelated words: “apple”,

“Mr. Johnson”, “charity”,

“tunnel”. The person is

requested to repeat all words.

Number of trial to repeat all

four words

Observation Predictive MEMORY

Mr.Sm.10 The examiner asks the person

to recall the words from

Mr.Sm.04

Correctly/incorrectly recalled

words





measurement


Mr.Mo.05 The examiner reads a list of

5 words (face, velvet,

church, daisy, red) at a rate

of one per second, giving the

following instructions: “This

is a memory test. I am going

to read a list of words that

you will have to remember

now and later on. Listen

carefully. When I am

through, tell me as many

words as you can remember.

It doesn’t matter in what

order you say them”. When

the subject indicates that

she/he has finished (has

recalled all words), or can

recall no more words, the

examiner reads the list a

second time with the


going to read the same list

for a second time. Try to

remember and tell me as

many words as you can,

including words you said the

first time.”. At the end of the

second trial, the examiner

informs the person that

she/he will be asked to recall

these words again”


for each word





measurement


Mr.Mo.13 The examiner gives the

following instruction: “I

read some words to you

earlier, which I asked you to

remember. Tell me as many

of those words as you can

remember.”


recalled words


Mo.Ed.09 Reported feelings of sadness

or depression Yes / No Report Diagnostic

Mo.Ti.10 Have you felt down during

the last month? Yes / Sometimes / No Report Predictive

Mo.Ti.11 Have you felt nervous or

anxious during the last

month?

Yes / Sometimes / No Report Predictive

Mo.Ti.12 Are you able to cope with

problems well? Yes / No Report Predictive

Mo.Cf.05 To what extent do you agree


unhappy”


Disagree /


Agree /

Completely agree.

Report Diagnostic


with the statement “Losing

self-confidence”


Disagree /


Agree /

Completely agree.

Report Diagnostic




measurement



with the statement “Unable

to cope with problems”


Disagree /


Agree /

Completely agree.

Report Diagnostic



pressure”


Disagree /


Agree /

Completely agree.

Report Diagnostic



worth nothing anymore”


Disagree /


Agree /

Completely agree.

Report Diagnostic


with the statement “I

experience a general sense

of emptiness”


Disagree /


Agree /

Completely agree.

Report Diagnostic


with the statement “I miss

having people around me”


Disagree /


Agree /

Completely agree.

Report Diagnostic




measurement



with the statement “I often

feel rejected”


Disagree /


Agree /

Completely agree.

Report Diagnostic

Mo.Gr.11 Do you sometimes

experience emptiness around

yourself?


Mo.Gr.12 Do you sometimes miss

people around yourself? Yes / No Report Predictive (study)

Mo.Gr.13 Do you sometimes feel

abandoned? Yes / No Report Predictive (study)

Mo.Gr.14 Have you recently felt

downhearted or sad? Yes / No Report Predictive (study)

Mo.Gr.15 Have you recently felt

nervous or anxious? Yes / No Report Predictive (study)

Ti.Da.01 The person is shown four

different times (0300 h, 0800

h, 1030 h and 1215 h) using a

large model of a clock and is

asked to tell the time


for each item


Ti.Da.02 The person is asked to state

the date, the day, the month

and the year.


for each item


Ti.Mm.01 The person is asked to state

the year, season, month, date

and day of the week.


for each element





measurement


Ti.Mo.14 The examiner gives the

following instructions: “Tell

me the date today”. If the

subject does not give a

complete answer, then

prompt accordingly by

saying: “Tell me the [year,

month, exact date, and day

of the week].”


for each element


Ti.Sm.02 The person is asked to state

date (day), month, year. Correct / Incorrect answer

for each item


Sp.Mm.02 The person is asked to state

the country, province,

city/town, address (home:

street address, in-facility:

building name) and location

(home: room name, in-

facility: floor)


for each element


Sp.Sm.01 The person is asked to state

name, address, current

location (building), city

state.


for each item


Sp.Mo.15 Then the examiner says:

“Now, tell me the name of

this place, and which city it

is in.”


for each element


Vi.Sm.07 The examiner asks the person

to draw a clock face showing

11:15.

Clock face correctly /

incorrectly drawn


Vi.Sm.08 The examiner shows the

person the drawing of a cube

and asks to copy it

Drawing correctly /

incorrectly copied





measurement


Vi.Mm.10 The examiner places a

drawing (interlocking

pentagons, i.e. two five-sided

figures intersecting to make a

four-sided figure), eraser and

pencil in front of the person

and asks “Copy this drawing

please.” (Allow multiple

tries; wait until person is

finished and hands it back.)

Drawing correctly /

incorrectly copied


Vi.Mo.01 The examiner instructs the

subject: "Please draw a line,

going from a number to a

letter in ascending order.

Begin here [point to (1)] and

draw a line from 1 then to A

then to 2 and so on. End here

[point to (E)].".


for each item


Vi.Mo.02 The examiner gives the

following instructions,

pointing to the cube: “Copy

this drawing as accurately as

you can, in the space

below”.

Drawing correctly /

incorrectly copied


Vi.Mo.03 The examiner give the

following instructions:

“Draw a clock. Put in all the

numbers and set the time to

10 past 11”.

Clock face correctly /

incorrectly drawn





measurement


Vi.Ed.01 Clock diagram: Place the

numbers in the correct

positions then place the

hands to indicate a time of

“10 after 11”

No errors /

Minor spacing errors /

Other errors.


La.Mm.06 The examiner show a

wristwatch and asks the

person “What is this called?”.

Repeats with a pencil


identified items


La.Mm.07 The examiner asks the person

to repeat this phrase “No ifs,

ands or buts.”

Correct / Incorrect repetition Observation Predictive (study)


to read the words on a page

and then do what it says.

Then hand the person the

page with “CLOSE YOUR

EYES” on it. If the subject

reads and does not close their

eyes, repeat up to three times.

Score only if subject closes

eyes

Closing / Not closing eyes Observation Predictive (study)

La.Mm.09 The examiner hands the

person a pencil and paper,

then says “Write any

complete sentence on that

piece of paper”

Writing a sentence that make

/ does not make sense (ignore

spelling errors)





measurement



if she/he is right or left‐handed, takes a piece of

paper and hold it up in front

of the person. Then says

“Take this paper in your

right/left hand [note:

whichever is non‐dominant],

fold the paper in half once

with both hands and put the

paper down on the floor”.

Correct / incorrect



La.Sm.09 The examiner asks the person

to provide three information

element: “first president”,

“define an island”, “number

of weeks per year”

Correct/incorrect information

provided for each item


La.Mo.04 The examiner points to three

animal figures (lion, rhino,

camel) in turn and says:

“Tell me the name of this

animal”.

Names correctly / incorrectly

repeated.





measurement


La.Mo.10 The examiner gives the


going to read you a sentence.

Repeat it after me, exactly as

I say it [pause]: I only know

that John is the one to help

today.” Following the

response, say: “Now I am

going to read you another

sentence. Repeat it after me,

exactly as I say it [pause]:

The cat always hid under the

couch when dogs were in the

room.”

Correct / Incorrect repetition Observation Predictive

La.Mo.11 The examiner gives the

following instruction: “Tell

me as many words as you

can think of that begin with

a certain letter of the

alphabet that I will tell you

in a moment. You can say

any kind of word you want,

except for proper nouns (like

Bob or Boston), numbers, or

words that begin with the

same sound but have a

different suffix, for example,

love, lover, loving. I will tell

you to stop after one minute.

Are you ready? [Pause]

Now, tell me as many words

as you can think of that

begin with the letter F. [time

for 60 sec]. Stop.”

Words generated Observation Predictive




measurement


De.Pr.01 Are you more than 85 years? Yes / No Report Diagnostic

De.Pr.02 Are you male? Yes / No Report Diagnostic

Table 9. Analysis of Items



10 Annex: Coverage of Geriatric Factors

For the meaning of the different fonts in the third column of the following Table, see note in Section 9

(in particular, regarding the last column of Table 9).

Geriatric factors Corresponding

Category

Measures proposed by Pilots for testbed

experiments

Motility Mt

Walking Mt WALK_DISTANCE

WALK_STEPS

WALK_DISTANCE_OUTDOOR

WALK_STEPS_OUTDOOR

WALK_TIME_OUTDOOR

WALK_SPEED_OUTDOOR






Climbing stairs Mt STAIRS_FLOOR_CHANGES_UP

Still/Moving n/a STILL_TIME

Moving across

rooms n/a ROOM_CHANGES

BEDROOM_VISITS

BEDROOM_TIME

BATHROOM_VISITS

BATHROOM_TIME

KITCHEN_VISITS

KITCHEN_TIME

LIVINGROOM_VISITS

LIVINGROOM _TIME

RESTROOM_VISITS

RESTROOM_TIME

Gait balance Mt

Physical activity Ac PHYSICALACTIVITY_CALORIES

PHYSICALACTIVITY_INTENSE_TIME

PHYSICALACTIVITY_MODERATE_TIME



Basic Activities of Daily

Living

Ad

Bathing and

showering Ad BATHROOMS_VISITS

BATHROOM_TIME

Dressing Ad

Self-feeding Ad

Personal hygiene

and grooming Ad BATHROOMS_VISITS

BATHROOM_TIME




Category


experiments

Toilet hygiene Ad RESTROOM_VISITS

RESTROOM_TIME

Going out Ad OUTDOOR_NUM

OUTDOOR_TIME

HOME_TIME

Instrumental Activities of

Daily Living

Ia

Ability to cook food Fo MEALS_NUM

Housekeeping Ho

Laundry Ln WASHINGMACHINE_SESSIONS

Phone usage Co PHONECALLS_PLACED

PHONECALLS_RECEIVED



PHONECALLS_MISSED





New media

communication n/a

Shopping Sh SHOPS_VISITS

SHOPS_VISITS_WEEK

SHOPS_TIME


SUPERMARKET_VISITS


SUPERMARKET _TIME


Transportation Tr TRANSPORT_TIME




Finance

Management Fi

Medication Me

Socialization So

Visits So VISITS_PAYED_WEEK

VISITS_RECEIVED_WEEK

VISITORS_WEEK

Attending senior

centers So SENIORCENTER_VISITS


SENIORCENTER_VISITS_MONTH

SENIORCENTER_TIME

SENIORCENTER_TIME_OUT_PERC

SENIORCENTER_LONG_VISITS




Category


experiments

Attending other

social places So FOODCOURT_VISITS_MONTH

FOODCOURT_VISITS_WEEK

FOODCOURT_TIME

PUBLICPARK_VISITS

PUBLICPARK _VISITS_MONTH

PUBLICPARK _TIME

OTHERSOCIAL_VISITS

OTHERSOCIAL _LONG_VISITS

OTHERSOCIAL _TIME

OTHERSOCIAL _TIME_OUT_PERC

Restaurant So RESTAURANTS_VISITS_WEEK

RESTAURANTS_VISITS_MONTH

RESTAURANTS_TIME

Cultural engagement Cu

Visit entertainment /

culture places Cu CINEMA_VISITS

CINEMA_VISITS_MONTH

CINEMA_TIME

CULTUREPOI_VISITS_MONTH

CULTUREPOI_VISITS_TIME_PERC_MONTH

Watching TV Cu TVWATCHING_TIME

TVWATCHING_TIME_PERC

Reading newspapers Cu

Reading books Cu

Environment En

Quality of housing En

Quality of

neighbourhood En

Dependence Dp

Health – Physical He

Falls n/a FALLS_MONTH

Weight We WEIGHT

Weakness Wk WEAKNESS

Exhaustion Ex EXHAUSTION

Pain n/a PAIN

Appetite n/a APPETITE

Quality of sleep n/a SLEEP_AWAKE_TIME

SLEEP_DEEP_TIME

SLEEP_LIGHT_TIME

SLEEP_REM_TIME

SLEEP_TIME

SLEEP_TOSLEEP_TIME

SLEEP_WAKEUP_NUM

Visit to doctors He GP_VISITS_MONTH

GP_TIME_MONTH




Category


experiments

Visit to health

related places He PHARMACY_VISITS



PHARMACY _TIME

Health – Cognitive n/a

Abstraction Ab WORD_PAIRS_TEST

Attention At

Memory Mr MEMORY

Mood Mo

Table 10. Coverage of City4Age geriatric factors from Pilots



11 Annex: Measures at City4Age Pilots

Payload property name Meaning Reporting

frequency

Payload

property type

Needed by (for what)

APPETITE Appetite reported in the day on a 1-5

scale

Daily integer Lecce (Health – physical,

Appetite)

BATHROOM_TIME Time in seconds spent in bathroom in

the day.

Daily integer Montpellier (Motility,

Moving across rooms; Basic

ADLs, Bathing and

showering), Singapore

(Motility, Moving across

rooms; Basic ADLs, Bathing

and showering)

BATHROOMS_VISITS Number of bathrooms entrances in

the day.

Daily integer Montpellier (Motility,


ADLs, Bathing and

showering), Singapore


rooms; Basic ADLs, Bathing

and showering)

BEDROOM_TIME Time in seconds spent in bedroom in

the day.

Daily integer Lecce (Motility, Moving

across rooms), Montpellier


rooms), Singapore (Motility,

Moving across rooms)




frequency

Payload

property type


BEDROOM_VISITS Number of bedroom entrances in the

day.






CINEMA_TIME Time in seconds spent in monitored

cinema/theatres in the day.

Daily integer Athens (Cultural

Engagement, Visiting

entertainment/culture places),

Montpellier (Cultural


entertainment/culture places)

CINEMA_VISITS Number of visits to monitored

cinema/theatres in the day.

Daily integer Athens (Cultural






CINEMA_VISITS_MONTH Number of visits to monitored

cinema/theatres in the month.

Monthly integer Athens (Cultural






CULTUREPOI_VISITS_MONTH Number of visits to monitored

cultural places in the month.

Monthly integer Madrid (Cultural






frequency

Payload

property type


CULTUREPOI_VISITS_TIME_PERC_MONTH Percentage of time spent in monitored

cultural places with respect to time

spent outside home in the month.

Monthly float Madrid (Cultural



EXHAUSTION Exhaustion measured along a 1-4

scale sampled in the month, according

to the following definition (from

Fried Frailty Index):

The person felt that everything

she/he did was an effort in last week

1. Rarely or none of the time (<1

day)

2. Some or little of the time (1 to 2

days)

3. Moderate amount of the time (3

to 4 days) /

4. Most of the time

Monthly float Madrid (Health – physical,

Weight)

FALLS_MONTH Number of falls detected in the

month.

Monthly integer Montpellier (Health –

physical, Fall), Singapore

(Health – physical, Fall)

FOODCOURT_TIME Time in seconds spent in FoodCourt

in the day.

Daily integer Singapore (Socialization,

Attending Other Social

Places)




frequency

Payload

property type


FOODCOURT_VISITS_MONTH Number of visits to FoodCourt in the

month.

Monthly integer Singapore (Socialization,


Places)

FOODCOURT_VISITS_WEEK Number of visits to FoodCourt in the

week.

Weekly integer Singapore (Socialization,


Places)

GP_TIME_MONTH Time spent at GP’s practice in the

month,

Monthly integer Lecce (Health – physical,

Visits to doctors)

GP_VISITS_MONTH Number of visits to GP in the month. Monthly integer Lecce (Health – physical,

Visits to doctors)

HEART_RATE Resting heart rate, measured in rpms

as well as the daily minutes in 5

ranges of heart rate in reference to the

maximum heart rate regime.

Daily float Madrid (Note: to be further

researched during the

experiment)

HOME_TIME Time in seconds spent at home in the

day.

Daily integer Athens (Basic ADLs, Going

out), Lecce (Basic ADLs,

Going out), Montpellier

(Basic ADLs, Going out),

Singapore (Basic ADLs,

Going out)




frequency

Payload

property type


KITCHEN_TIME Time in seconds spent in kitchen in

the day.




rooms; Instrumental ADLs,

Ability to cook food),

Singapore (Motility, Moving

across rooms; Instrumental

ADLs, Ability to cook food)

KITCHEN_VISITS Number of kitchen entrances in the

day.




rooms; Instrumental ADLs,

Ability to cook food),

Singapore (Motility, Moving

across rooms; Instrumental


LIVINGROOM _TIME Time in seconds spent in living room

in the day.






LIVINGROOM_VISITS Number of living room entrances in

the day.









frequency

Payload

property type


MEALS_NUM Meals prepared by the person in the

day.

Weekly integer Montpellier (Instrumental

ADLs, Ability to cook food),

Singapore (Instrumental


MEMORY Memory performance reported in the

day, as numerical indicator returned

from the CANTAB system

Monthly An object with a

float property for

each CANTAB

results test score

(the name of the

property is the

name of the

CANTAB test

code).

Lecce (Health – cognitive,

Memory)

OTHERSOCIAL _LONG_VISITS Number of long visits to

OtherSocialPlace in the day.

Daily integer Lecce (Socialization,


Places)

OTHERSOCIAL _TIME Time in seconds spent in

OtherSocialPlace in the day.

Daily integer Athens (Socialization,


Places), Lecce (Socialization,


Places)

OTHERSOCIAL _TIME_OUT_PERC Percentage of time spent in

OtherSocialPlace with respect to total

time spent outside home in the day.

Daily float Lecce (Socialization,


Places)




frequency

Payload

property type


OTHERSOCIAL_VISITS Number of visits to OtherSocialPlace

in the day.



Places), Lecce (Socialization,


Places)

OUTDOOR_NUM Total number of exits from home in

the day.

Daily integer Montpellier (Basic ADLs,

Going out), Singapore (Basic

ADLs, Going out)

OUTDOOR_TIME Time in seconds spent outdoor in the

day.

Note: as outdoor and indoor states

are complementary (i.e. outdoor

indoor) INDOOR_TIME can be

computed by subtracting

OUTDOOR_TIME from 86.400

(seconds in a day).

Daily integer Lecce (Basic ADLs, Going

out), Montpellier (Basic

ADLs, Going out), Singapore

(Basic ADLs, Going out)

PAIN Pain level reported in the day on a 1-5

scale

Daily integer Lecce (Health – physical,

Pain)

PERCEIVED_TEMPERATURE Average measure of the thermal

comfort (thermal sensation) during

the day.

Note: This measure can be collected

by hours as minimum time gap. We

have to agree what is the more

convenient and meaningful.

Daily float Madrid (Note: to be further

researched during the

experiment)




frequency

Payload

property type


PHARMACY _TIME Time in seconds spent in monitored

pharmacies in the day.

Daily integer Athens (Instrumental ADLs,

Shopping; Health – physical,

Visits to health related places)

PHARMACY _VISITS_MONTH Number of visits to monitored

pharmacies in the month.

Monthly integer Athens (Instrumental ADLs,


Visits to health related

places), Lecce (Instrumental

ADLs, Shopping; Health –

physical, Visits to health

related places), Montpellier

(Instrumental ADLs,



PHARMACY _VISITS_WEEK Number of visits to monitored

pharmacies in the week.

Weekly integer Athens (Instrumental ADLs,



places), Montpellier

(Instrumental ADLs,



PHARMACY_VISITS Number of visits to monitored

pharmacies in the day.




places), Montpellier

(Instrumental ADLs,






frequency

Payload

property type


PHONECALLS_LONG_PLACED_PERC Percentage of long phone calls placed

on total phone calls placed in the day.

Daily float Lecce (Instrumental ADLs,

Phone usage)

PHONECALLS_LONG_RECEIVED_PERC Percentage of long phone calls

received on total phone calls received

in the day.


Phone usage)

PHONECALLS_MISSED Number of phone calls missed in the

day.

Daily integer Lecce (Instrumental ADLs,

Phone usage)

PHONECALLS_PLACED Number of phone calls placed in the

day.


Phone usage), Montpellier

(Instrumental ADLs, Phone

usage)

PHONECALLS_PLACED_PERC Percentage of phone calls placed on

total phone calls in the day.


Phone usage)

PHONECALLS_RECEIVED Number of phone calls received in the

day.


Phone usage), Montpellier

(Instrumental ADLs, Phone

usage)

PHONECALLS_RECEIVED_PERC Percentage of phone calls received on

total phone calls in the day.


Phone usage)

PHONECALLS_SHORT_PLACED_PERC Percentage of short phone calls placed

on total phone calls placed in the day.


Phone usage)




frequency

Payload

property type


PHONECALLS_SHORT_RECEIVED_PERC Percentage of short phone calls

received on total phone calls received

in the day.


Phone usage)

PHYSICALACTIVITY_CALORIES Total calories in kcal burned in the

day.

Daily float Birmingham (Physical

Activity), Madrid (Physical

Activity)

PHYSICALACTIVITY_INTENSE_TIME Time in seconds spent in intense

activities in the day.

Daily integer Birmingham (Physical

Activity)

PHYSICALACTIVITY_MODERATE_TIME Time in seconds spent in moderate



Activity)

PHYSICALACTIVITY_NUM Number of physical activity sessions

attended in the day.

Daily integer

PHYSICALACTIVITY_SOFT_TIME Time in seconds spent in soft



Activity)

PUBLICPARK _TIME Time in seconds spent in PublicPark

in the day.



Places)

PUBLICPARK _VISITS_MONTH Number of visits to PublicPark in the

month.



Places)

PUBLICPARK_VISITS Number of visits to PublicPark in the

day.



Places)




frequency

Payload

property type


PUBLICTRANSPORT_DISTANCE_MONTH Distance in km travelled on the bus in

a month.

Monthly float Madrid (Instrumental ADLs,

Transportation)

PUBLICTRANSPORT_RIDES_MONTH Number of times the user gets on the

bus in a month.

Monthly integer Madrid (Instrumental ADLs,

Transportation)

PUBLICTRANSPORT_TIME Time in seconds spent in public

transportation in the day.

Daily integer

RESTAURANTS_TIME Time in seconds spent in monitored

restaurants in the day.

Weekly integer Athens (Socialization,

Restaurant), Montpellier

(Socialization, Restaurant)

RESTAURANTS_VISITS_MONTH Number of visits to monitored

restaurants in the month.

Monthly integer Montpellier (Socialization,

Restaurant)

RESTAURANTS_VISITS_WEEK Number of visits to monitored

restaurants in the week.

Weekly integer Athens (Socialization,

Restaurant), Montpellier

(Socialization, Restaurant)

RESTROOM_TIME Time in seconds spent in restroom in

the day.




rooms; Basic ADLs, Toilet

hygiene), Singapore (Motility,


ADLs, Toilet hygiene)




frequency

Payload

property type


RESTROOM_VISITS Number of restroom entrances in the

day.




rooms; Basic ADLs, Toilet

hygiene), Singapore (Motility,


ADLs, Toilet hygiene)

ROOM_CHANGES Number of room changes in the day. Daily integer Lecce (Motility, Moving





SENIORCENTER_LONG_VISITS Number of long visits to

SeniorCenter in the day.

Daily integer Lecce (Socialization,

Attending Senior Centers)

SENIORCENTER_TIME Time in seconds spent in

SeniorCenter in the day.


Attending Senior Centers),

Lecce (Socialization,


SENIORCENTER_TIME_OUT_PERC Percentage of time spent in

SeniorCenter with respect to total

time spent outside home in the day.

Daily float Lecce (Socialization,





frequency

Payload

property type


SENIORCENTER_VISITS Number of visits to SeniorCenter in

the day.



Lecce (Socialization,


Singapore (Socialization,


SENIORCENTER_VISITS_MONTH Number of visits to SeniorCenter in

the month.



SENIORCENTER_VISITS_WEEK Number of visits to SeniorCenter in

the week.

Weekly integer Singapore (Socialization,


SHOPS_OUTDOOR_TIME_PERC Percentage of time spent in monitored

shops with respect to time spent

outdoor in the day.

Daily float Athens (Instrumental ADLs,

Shopping), Lecce

(Instrumental ADLs,

Shopping)

SHOPS_TIME Time in seconds spent in monitored

shops in the day.


Shopping), Lecce

(Instrumental ADLs,

Shopping), Montpellier

(Instrumental ADLs,

Shopping), Singapore

(Instrumental ADLs,

Shopping)




frequency

Payload

property type


SHOPS_VISITS Number of visits to monitored shops

in the day.


Shopping), Lecce

(Instrumental ADLs,


(Instrumental ADLs,

Shopping), Singapore

(Instrumental ADLs,

Shopping)

SHOPS_VISITS_WEEK Number of visits to monitored shops

in the week.

Weekly integer Montpellier (Instrumental

ADLs, Shopping), Singapore

(Instrumental ADLs,

Shopping)

SLEEP_AWAKE_TIME Total time in seconds spent awake

while at rest in the day.

Daily integer Birmingham (Health –

physical, Quality of sleep),

Madrid (Health – physical,

Quality of sleep)

SLEEP_DEEP_TIME Total time in seconds spent in deep

sleeping in the day.




Quality of sleep)

SLEEP_LIGHT_TIME Total time in seconds spent in light





Quality of sleep)

SLEEP_REM_TIME Total time in seconds spent in REM


Daily integer Madrid (Health – physical,

Quality of sleep)




frequency

Payload

property type


SLEEP_TIME Total time in seconds spent sleeping

in the day.

Day integer Madrid (Health – physical,

Quality of sleep),

Montpellier (Health –


Singapore (Health – physical,

Quality of sleep)

SLEEP_TOSLEEP_TIME Total time in seconds the user spent

falling asleep in the day.


physical, Quality of sleep)

SLEEP_WAKEUP_NUM Number of times the user woke up in

the day.


physical, Quality of sleep)

STAIRS_FLOOR_CHANGES_UP Number of floor changes performed

in the day by climbing stairs upwards.

Daily integer

Lecce (Motility, Climbing

stairs)

STILL_TIME Time in seconds spent in the still state

in the day.

Note: as still and moving states are

complementary (i.e. still

moving) MOVING_TIME can be

computed by subtracting

STILL_TIME from 86.400 (seconds

in a day).

Daily integer Lecce (Motility,

Still/Moving)




frequency

Payload

property type


SUPERMARKET _TIME Time in seconds spent in monitored

supermarkets in the day.


Shopping), Lecce

(Instrumental ADLs,


(Instrumental ADLs,

Shopping)

SUPERMARKET _TIME_PERC Percentage of time spent in monitored

supermarkets on total time spent in

shops in the day.


Shopping)

SUPERMARKET_VISITS Number of visits to monitored

supermarkets in the day.


Shopping), Lecce

(Instrumental ADLs,


(Instrumental ADLs,

Shopping)

SUPERMARKET_VISITS_WEEK Number of visits to monitored

supermarkets in the week.

Weekly integer Athens (Instrumental ADLs,


(Instrumental ADLs,

Shopping)

TRANSPORT_TIME Time in seconds spent in

transportation (either public or

private) in the day.


Transportation)

TVWATCHING_TIME Time in seconds spent watching TV

in the day.

Daily integer Lecce (Cultural Engagement,

Watching TV)




frequency

Payload

property type


TVWATCHING_TIME_PERC Percentage of time spent watching

TV with respect to time spent inside

home in the day.

Daily float Lecce (Cultural Engagement,

Watching TV)

VISITORS_WEEK Number of visitors met in the week. Weekly integer Lecce (Socialization, Visits),

Singapore (Socialization,

Visits)

VISITS_PAYED_WEEK Number of visits payed in the week. Weekly integer Lecce (Socialization, Visits),

Montpellier (Socialization,

Visits)

VISITS_RECEIVED_WEEK Number of visits received in the

week.

Weekly integer Lecce (Socialization, Visits),

Montpellier (Socialization,

Visits), Singapore

(Socialization, Visits),

WALK_DISTANCE Total distance in meters walked in the

day.

Daily integer Birmingham (Motility,

Walking), Madrid (Motility,

Walking)

WALK_DISTANCE_OUTDOOR Total distance in meters walked

outdoor in the day.

Daily integer Athens (Motility, Walking),

Lecce (Motility, Walking),

Montpellier (Motility,

Walking), Singapore

(Motility, Walking)

WALK_DISTANCE_OUTDOOR_FAST Distance in meters walked outdoor at

fast speed in the day.

Daily float Lecce (Motility, Walking)




frequency

Payload

property type


WALK_DISTANCE_OUTDOOR_FAST_PERC Percentage of distance walked

outdoor at fast speed in the day on

total distance walked outdoor.

Daily float Athens (Motility, Walking),

Lecce (Motility, Walking)

WALK_DISTANCE_OUTDOOR_SLOW_PERC Percentage of distance walked

outdoor at slow speed in the day on

total distance walked outdoor.


Lecce (Motility, Walking)

WALK_SPEED_OUTDOOR Average outdoor walking speed in

meters/seconds in the day.



Madrid (Motility, Walking)

WALK_SPEED_OUTDOOR_FAST Average outdoor walking speed at

fast speed in meters/seconds in the

day.

Daily float Lecce (Motility, Walking)

WALK_STEPS Number of steps done in the day. Daily integer Birmingham (Motility,

Walking)

WALK_STEPS_OUTDOOR Number of steps done outdoor in the

day.

Daily integer


Walking), Singapore

(Motility, Walking)

WALK_TIME_OUTDOOR Time in seconds spent walking

outdoor in the day.

Daily integer Athens (Motility, Walking),


Madrid (Motility, Walking),


Walking), Singapore

(Motility, Walking)




frequency

Payload

property type


WALK_TIME_OUTDOOR_FAST Time in seconds spent walking at fast

speed outdoor in the day.

Daily integer Lecce (Motility, Walking)

WASHINGMACHINE_SESSIONS Washing machine usage sessions in

the day.


Laundry)

WEAKNESS <yet to be defined; following

definitions are currently considered>

Grip strength in kg sampled in the

month.

Or

The person experiences problems in

daily life due to lack of strength in

hands.

Monthly float

or

boolean


Weakness)

WEIGHT User weight in kg sampled in the

month.

Monthly float Madrid (Health – physical,

Weight)




frequency

Payload

property type


WORD_PAIRS_TEST The result of the word pair test (from

the “Short Test of Mental Status”

scale) sampled in the month

(replicated below for reader’s

convenience).

Word Pair Test: The examiner

proposes three pair of words and asks

the person to state the abstract

interpretation: orange/banana,

dog/horse, table/bookcase (e.g.

dog/horse = animal).

Monthly float Madrid (Health – cognitive,

Abstraction)



12 Annex: Correlation matrices for Pilots measures

The correlation values reported in the following subsections have been assessed in June 2018, based on the status of the City4Age Shared Repository at that time.

12.1 Athens Pilot

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

1 1 0,6 0 -0,01 0,04 0,08 -0,01 0 0 0 0 -0,01 -0,02 -0,01 -0,01 0,04 -0,01 0,04 0,03 0,05 0,05 -0,05 0,03 0,01

2 0,6 1 0 -0,05 0,02 0,06 -0,02 0,01 0 -0,01 -0,01 -0,01 -0,01 -0,01 -0,01 0,06 -0,01 0,06 0,04 0,12 0,07 -0,07 0,04 0,06

3 0 0 1 -0,02 0,04 0,04 0 0,03 0,2 0 0 0,02 0,01 -0,01 -0,01 0,01 -0,01 0,01 0 0,01 0 0 -0,01 0,02

4 -0,01 -0,05 -0,02 1 -0,09 -0,1 -0,24 -0,31 -0,04 -0,04 0,01 -0,23 -0,24 0,04 -0,18 -0,25 -0,18 -0,25 -0,13 -0,37 -0,06 0,06 -0,01 -0,84

5 0,04 0,02 0,04 -0,09 1 0,77 0,03 0,3 0,06 -0,02 -0,02 -0,04 -0,01 -0,03 -0,02 0,09 -0,02 0,09 0 0,06 0 0 0,07 0,04

6 0,08 0,06 0,04 -0,1 0,77 1 0,05 0,47 0,06 -0,02 -0,03 -0,03 0 -0,02 -0,01 0,18 -0,01 0,18 0,01 0,12 0,01 -0,01 0,07 0,08

7 -0,01 -0,02 0 -0,24 0,03 0,05 1 0,63 0,06 -0,01 0 0,18 0,26 0,07 0,27 0,18 0,27 0,18 -0,02 0,09 -0,03 0,03 0 0,09

8 0 0,01 0,03 -0,31 0,3 0,47 0,63 1 0,07 -0,01 0 0,14 0,26 0,06 0,19 0,38 0,19 0,38 0,01 0,22 -0,01 0,01 0,05 0,21

9 0 0 0,2 -0,04 0,06 0,06 0,06 0,07 1 0 0 0,01 0 0 0 0,03 0 0,03 0 0,01 -0,02 0,02 0 0,04

10 0 -0,01 0 -0,04 -0,02 -0,02 -0,01 -0,01 0 1 0,17 0,03 0,07 -0,01 -0,01 0,03 -0,01 0,03 0,02 0,03 0,02 -0,02 0,02 0,02

11 0 -0,01 0 0,01 -0,02 -0,03 0 0 0 0,17 1 0,01 0,06 0,02 0 0,01 0 0,01 0,02 0,01 0,02 -0,02 0,02 0

12 -0,01 -0,01 0,02 -0,23 -0,04 -0,03 0,18 0,14 0,01 0,03 0,01 1 0,68 -0,01 0,09 0,07 0,09 0,07 -0,03 0,01 -0,02 0,02 0,05 0

13 -0,02 -0,01 0,01 -0,24 -0,01 0 0,26 0,26 0 0,07 0,06 0,68 1 0,03 0,16 0,19 0,16 0,19 -0,04 0,06 -0,03 0,03 0,07 0,08

14 -0,01 -0,01 -0,01 0,04 -0,03 -0,02 0,07 0,06 0 -0,01 0,02 -0,01 0,03 1 0,75 0,41 0,75 0,41 -0,03 -0,04 0 0 0,03 -0,11

15 -0,01 -0,01 -0,01 -0,18 -0,02 -0,01 0,27 0,19 0 -0,01 0 0,09 0,16 0,75 1 0,55 1 0,55 -0,02 0,05 -0,02 0,02 0,01 0,04

16 0,04 0,06 0,01 -0,25 0,09 0,18 0,18 0,38 0,03 0,03 0,01 0,07 0,19 0,41 0,55 1 0,55 1 0,01 0,27 0,02 -0,02 0,07 0,2

17 -0,01 -0,01 -0,01 -0,18 -0,02 -0,01 0,27 0,19 0 -0,01 0 0,09 0,16 0,75 1 0,55 1 0,55 -0,02 0,05 -0,02 0,02 0,01 0,04

18 0,04 0,06 0,01 -0,25 0,09 0,18 0,18 0,38 0,03 0,03 0,01 0,07 0,19 0,41 0,55 1 0,55 1 0,01 0,27 0,02 -0,02 0,07 0,2

19 0,03 0,04 0 -0,13 0 0,01 -0,02 0,01 0 0,02 0,02 -0,03 -0,04 -0,03 -0,02 0,01 -0,02 0,01 1 0,31 0,61 -0,61 0,33 0,15

20 0,05 0,12 0,01 -0,37 0,06 0,12 0,09 0,22 0,01 0,03 0,01 0,01 0,06 -0,04 0,05 0,27 0,05 0,27 0,31 1 0,25 -0,25 0,18 0,39

21 0,05 0,07 0 -0,06 0 0,01 -0,03 -0,01 -0,02 0,02 0,02 -0,02 -0,03 0 -0,02 0,02 -0,02 0,02 0,61 0,25 1 -1 0,63 0,11

22 -0,05 -0,07 0 0,06 0 -0,01 0,03 0,01 0,02 -0,02 -0,02 0,02 0,03 0 0,02 -0,02 0,02 -0,02 -0,61 -0,25 -1 1 -0,63 -0,11

23 0,03 0,04 -0,01 -0,01 0,07 0,07 0 0,05 0 0,02 0,02 0,05 0,07 0,03 0,01 0,07 0,01 0,07 0,33 0,18 0,63 -0,63 1 -0,03

24 0,01 0,06 0,02 -0,84 0,04 0,08 0,09 0,21 0,04 0,02 0 0 0,08 -0,11 0,04 0,2 0,04 0,2 0,15 0,39 0,11 -0,11 -0,03 1



Legenda

1 cinema_time

2 cinema_visits

3 cinema_visits_month

4 home_time

5 othersocial_time

6 othersocial_visits

7 pharmacy_time

8 pharmacy_visits

9 pharmacy_visits_month

10 restaurants_time

11 restaurants_visits_week

12 seniorcenter_time

13 seniorcenter_visits

14 shops_outdoor_time_perc

15 shops_time

16 shops_visits

17 supermarket_time

18 supermarket_visits

19 transport_time

20 walk_distance_outdoor

21 walk_distance_outdoor_fast_perc

22 walk_distance_outdoor_slow_perc

23 walk_speed_outdoor

24 walk_time_outdoor



12.2 Birmingham Pilot

1 2 3 4 5 6 7 8 9 10 11

1 1 0,12 0,31 0,34 0,07 -0,05 0,05 0 0,13 0,51 0,44

2 0,12 1 0,14 0,07 -0,02 -0,03 0,08 0,01 -0,02 0,2 0,12

3 0,31 0,14 1 0,27 -0,07 0 0,03 0 -0,12 0,82 0,75

4 0,34 0,07 0,27 1 -0,16 -0,14 -0,05 -0,02 -0,2 0,56 0,61

5 0,07 -0,02 -0,07 -0,16 1 0,25 0,13 0,29 0,76 -0,08 -0,09

6 -0,05 -0,03 0 -0,14 0,25 1 -0,09 0,03 0,24 -0,06 -0,02

7 0,05 0,08 0,03 -0,05 0,13 -0,09 1 0,06 0,19 0,05 0

8 0 0,01 0 -0,02 0,29 0,03 0,06 1 0,04 0,01 0,01

9 0,13 -0,02 -0,12 -0,2 0,76 0,24 0,19 0,04 1 -0,09 -0,09

10 0,51 0,2 0,82 0,56 -0,08 -0,06 0,05 0,01 -0,09 1 0,96

11 0,44 0,12 0,75 0,61 -0,09 -0,02 0 0,01 -0,09 0,96 1

Legenda

1 physicalactivity_calories

2 physicalactivity_intense_time

3 physicalactivity_moderate_time

4 physicalactivity_soft_time

5 sleep_awake_time

6 sleep_deep_time

7 sleep_light_time

8 sleep_tosleep_time

9 sleep_wakeup_num

10 walk_distance

11 walk_steps



12.3 Lecce Pilot

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

1 1 0,29 -0,02 0,32 0 0 0,37 0,41 0,25 0,13 -0,05 0,02 0,54 0 0 0 0 0,46 0,15 0,09 -0,02 0,01

2 0,29 1 -0,08 0,17 0 0 0,23 0,28 0,21 0,19 -0,09 0,04 0,22 0 0 0 0 0,28 0,12 -0,03 -0,03 -0,1

3 -0,02 -0,08 1 0,39 0 0 0,23 0,06 0,29 0,37 0,31 0 0,52 0 0 0,01 0 0,16 0,39 0,36 -0,02 0,09

4 0,32 0,17 0,39 1 0 0 0,66 0,69 0,52 0,59 0,4 0,03 0,6 0 0 0,01 0 0,57 0,42 0,54 -0,04 0,27

5 0 0 0 0 1 0,34 0 0 0 0 0 0,41 0,01 0,27 0,05 0,36 0,05 0,01 0,01 0 -0,11 0

6 0 0 0 0 0,34 1 0 0 0 0 0 0,16 0 0,12 0,12 0,21 0,28 0,01 0 0 -0,03 0

7 0,37 0,23 0,23 0,66 0 0 1 0,72 0,71 0,55 0,25 0,01 0,65 0 0 0,02 0 0,36 0,52 0,59 -0,01 0,28

8 0,41 0,28 0,06 0,69 0 0 0,72 1 0,61 0,54 0,2 0,02 0,59 0 0 0,01 0 0,51 0,48 0,46 -0,04 0,34

9 0,25 0,21 0,29 0,52 0 0 0,71 0,61 1 0,52 0,17 0,02 0,56 0 0 0,01 0 0,4 0,47 0,49 0,01 0,38

10 0,13 0,19 0,37 0,59 0 0 0,55 0,54 0,52 1 0,44 0,04 0,47 0 0 0,01 0 0,48 0,34 0,49 -0,02 0,3

11 -0,05 -0,09 0,31 0,4 0 0 0,25 0,2 0,17 0,44 1 0,01 0,26 0 0 -0,01 0 0,18 0,16 0,38 -0,02 0,18

12 0,02 0,04 0 0,03 0,41 0,16 0,01 0,02 0,02 0,04 0,01 1 0,01 0,56 0,09 0,53 0,09 0,02 0,01 0 -0,36 0,03

13 0,54 0,22 0,52 0,6 0,01 0 0,65 0,59 0,56 0,47 0,26 0,01 1 0,01 0 0,02 0 0,53 0,47 0,43 -0,04 0,2

14 0 0 0 0 0,27 0,12 0 0 0 0 0 0,56 0,01 1 0,22 0,32 0,17 0 0,01 0 -0,16 0

15 0 0 0 0 0,05 0,12 0 0 0 0 0 0,09 0 0,22 1 0,18 0,96 0 0 0 -0,02 0

16 0 0 0,01 0,01 0,36 0,21 0,02 0,01 0,01 0,01 -0,01 0,53 0,02 0,32 0,18 1 0,22 0,01 0,02 0 -0,14 0

17 0 0 0 0 0,05 0,28 0 0 0 0 0 0,09 0 0,17 0,96 0,22 1 0 0 0 -0,02 0

18 0,46 0,28 0,16 0,57 0,01 0,01 0,36 0,51 0,4 0,48 0,18 0,02 0,53 0 0 0,01 0 1 0,17 0,31 -0,02 0,21

19 0,15 0,12 0,39 0,42 0,01 0 0,52 0,48 0,47 0,34 0,16 0,01 0,47 0,01 0 0,02 0 0,17 1 0,42 -0,03 0,03

20 0,09 -0,03 0,36 0,54 0 0 0,59 0,46 0,49 0,49 0,38 0 0,43 0 0 0 0 0,31 0,42 1 0,03 0,38

21 -0,02 -0,03 -0,02 -0,04 -0,11 -0,03 -0,01 -0,04 0,01 -0,02 -0,02 -0,36 -0,04 -0,16 -0,02 -0,14 -0,02 -0,02 -0,03 0,03 1 -0,02

22 0,01 -0,1 0,09 0,27 0 0 0,28 0,34 0,38 0,3 0,18 0,03 0,2 0 0 0 0 0,21 0,03 0,38 -0,02 1



23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44

1 0,37 0 0 0 0,01 0 0 0 0,01 0 0,01 0,16 0 -0,23 0,01 0,01 0 0 0 0,01 0 0,01

2 0,31 0 0 0 0,02 0 0,01 -0,01 -0,02 0,01 -0,01 0,14 -0,01 0,01 0,01 0,01 0 0 0 0,03 0,01 0,02

3 0,06 0 0 0 -0,01 -0,01 -0,01 0 0,01 -0,02 0 0,26 0,1 -0,19 0,01 0,01 0 0 0 -0,01 0 0,01

4 0,38 0 0 0 0,01 0 0 0 0,01 0 0 0,44 0,26 0,06 0,01 0,01 0 0 0 0,03 0,01 0,01

5 0 0 0,01 -0,01 0,05 0,33 0,22 0,31 0,13 0,25 0,24 0,01 0 0 0,18 0,22 0,1 0,05 0,16 0,17 0,1 0,15

6 0 0 0 0 0,03 0,13 0,08 0,13 0,05 0,09 0,09 0 0 0 0,08 0,12 0,32 0,01 0,05 0,06 0,06 0,06

7 0,52 0 0 0 0 0 0 0 0,01 -0,01 0,01 0,48 0,23 0,14 0,01 0,01 0 0 0 -0,01 0 0,01

8 0,82 0 0 0 0,01 0 0 0 0,01 0 0 0,4 0,31 0,28 0,01 0,01 0 0 0 0,02 0,01 0,01

9 0,41 0 0 0 -0,01 0 -0,01 0 0,01 -0,01 0,01 0,61 -0,02 0,16 0,02 0,02 0 0 0 0 0 0,01

10 0,49 0 0 0 0,01 0 0 0 0,01 0,01 0,01 0,4 0,2 0,26 0,01 0,01 0 0 0 0,04 0,01 0,01

11 0,06 0 0 0 0,01 0,01 0,01 0,01 0,01 0,01 0,01 0,21 0,17 0,1 0 0 0 0 0 0,02 0,01 -0,01

12 0,02 0 0,01 -0,01 0,19 0,4 0,35 0,41 0,26 0,41 0,38 0 -0,01 -0,02 0,28 0,33 0,12 0,17 0,42 0,67 0,41 0,57

13 0,48 0 0 0 0,01 0 0 0,01 0,02 0 0,01 0,34 0,07 -0,14 0 0 0 0 0 -0,01 0 0,02

14 0 -0,01 0 -0,01 0,15 0,39 0,28 0,4 0,22 0,34 0,35 0 0 -0,01 0,32 0,45 0,22 0,05 0,23 0,24 0,17 0,2

15 0 -0,01 0 0 0,03 0,07 0,05 0,07 0,04 0,06 0,06 0 0 0 0,03 0,06 0,97 0,01 0,06 0,04 0,25 0,04

16 0,01 0 -0,01 -0,01 0,12 0,24 0,19 0,27 0,16 0,23 0,25 0,01 -0,02 -0,01 0,12 0,23 0,23 0,04 0,18 0,18 0,15 0,27

17 0 0 0 0 0,02 0,05 0,04 0,07 0,04 0,05 0,06 0 0 0 0,02 0,05 0,99 0,01 0,06 0,04 0,25 0,05

18 0,38 0 0 0 0,01 0 0 0,01 0,02 0 0,01 0,56 0,03 0,02 -0,01 0 0 0 0 0,01 0 0,02

19 0,33 0 0 0 -0,01 0 -0,01 0,01 0,02 -0,01 0,01 0,25 0,17 0,17 0,02 0,01 0 0 0 -0,01 0 0,02

20 0,25 0 0 0 0 0 0 0 0 0 0 0,29 0,19 0,23 0,02 0,02 0 0 0 -0,01 0 0

21 -0,03 0 -0,05 -0,03 -0,16 -0,22 -0,2 -0,17 -0,08 -0,23 -0,16 -0,01 -0,03 -0,01 -0,04 -0,01 -0,02 -0,02 -0,18 -0,29 -0,18 -0,25

22 0,31 0 0 0 0,02 0 0,01 0 0 0,01 0 0,24 0,26 0,31 0,02 0,01 0 0 0 0,03 0,01 0



45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

1 0 0,02 0 0,01 0 0,02 0,56 0 0 0,23 0,41 0,32 0,02 0,02 -0,01 0,02 -0,01 0 0,01 0,02 0 0,16 0,04

2 0,01 0,04 0,01 0,01 0 0,05 0,3 0 0,02 0,46 -0,04 0,36 0,01 0,02 0,04 0,04 0,06 0,06 0 0,02 0 0,09 0,04

3 0 0,01 0 0 0 0,02 -0,1 -0,01 0,01 0,04 0,36 0,26 0 0 -0,01 -0,02 -0,02 -0,02 0,01 0 0 0,31 0,27

4 0 0,01 0,01 0 0 0,02 0,54 -0,01 0,01 0,34 0,62 0,54 0,01 0,01 0,04 0,04 0,04 0,05 0,01 0,01 0 0,28 0

5 0,11 0,12 0,22 0,01 -0,03 -0,02 0 0,22 0,1 0 0 0,01 0,18 0,17 0,17 0,17 0,17 0,2 0,15 0,17 -0,01 0,01 0

6 0,06 0,08 0,13 0,01 0 0,01 0 0,1 0,05 0 0 0 0,06 0,06 0,07 0,07 0,08 0,09 0,05 0,06 0,01 0 0

7 0,01 0,02 0 0,01 0 0,03 0,51 -0,01 0,02 0,3 0,38 0,36 0,01 0,01 0,04 0,05 0,04 0,05 0,01 0,01 0 0,42 -0,01

8 0 0,01 0,01 0 0 0,01 0,85 -0,01 0,01 0,33 0,4 0,52 0,01 0,01 0,05 0,04 0,05 0,05 0,01 0,01 0 0,17 -0,11

9 0,01 0,02 0 0,01 0 0,03 0,49 0 0,01 0,3 0,35 0,35 0,01 0,01 0,04 0,05 0,04 0,04 0,01 0,01 0 0,62 -0,05

10 0,01 0,02 0,01 0,01 0 0,03 0,39 -0,01 0,01 0,2 0,41 0,4 0,01 0,01 0,06 0,04 0,05 0,05 0,02 0,01 0 0,3 0,18

11 0 -0,02 0,01 -0,01 0 -0,03 0,03 -0,01 -0,02 0,16 0,21 0,21 0,01 0,01 0,02 0,01 0,01 0,01 0,01 0,01 0 0,08 0,07

12 0,41 0,48 0,38 0,29 0,01 0,29 0,01 0,42 0,2 0,03 0,02 0,02 0,14 0,15 0,4 0,35 0,43 0,45 0,03 0,15 0 0,02 0,02

13 0,01 0,03 0 0,01 0 0,04 0,43 -0,01 0,02 0,15 0,49 0,38 0,02 0,02 0,02 0,03 0,02 0,02 0,02 0,02 0 0,26 0,02

14 0,16 0,2 0,39 0,16 0,05 0,16 0 0,55 0,3 0 0,01 0,01 0,04 0,04 0,17 0,17 0,2 0,21 0 0,04 0,04 0,01 0

15 0,1 0,28 0,19 0,12 0,25 0,33 0 0,08 0,05 0 0 0 0,01 0,01 0,04 0,03 0,04 0,04 0 0,01 0,41 0 0

16 0,28 0,33 0,27 0,23 0,1 0,27 0,01 0,4 0,25 -0,01 0,01 0,01 0,09 0,09 0,19 0,18 0,19 0,22 0,05 0,09 0,03 0,02 0,01

17 0,11 0,29 0,19 0,12 0,24 0,34 0 0,07 0,04 0 0 0 0,02 0,02 0,04 0,03 0,04 0,04 0 0,02 0,41 0 0

18 0,01 0,03 0 0,01 0 0,05 0,54 -0,01 0 0,16 0,37 0,44 0,02 0,02 0,03 0,02 0,02 0,02 0,02 0,02 0 0,27 0,07

19 0,01 0,03 0 0,01 0 0,04 0,4 0 0,02 0,19 0,25 0,44 0,02 0,02 0,01 0 0,01 0,01 0,02 0,02 0 0,18 -0,01

20 0 0 0,01 0 0 0 0,3 -0,01 0 0,29 0,18 0,24 0 0 0,02 0,02 0,02 0,02 0 0 0 0,38 0,09

21 -0,18 -0,21 -0,09 -0,12 0 -0,13 -0,02 -0,1 0,05 -0,02 -0,04 -0,03 0,01 0 -0,02 -0,04 -0,05 -0,06 0,08 0 0 0,06 0,01

22 0 0,01 0,01 0 0 0,01 0,22 -0,01 -0,01 0,05 0,28 0,15 0,01 0,01 0,04 0 0,04 0,03 0,01 0,01 0 0,22 -0,13



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

23 0,37 0,31 0,06 0,38 0 0 0,52 0,82 0,41 0,49 0,06 0,02 0,48 0 0 0,01 0 0,38 0,33 0,25 -0,03 0,31

24 0 0 0 0 0 0 0 0 0 0 0 0 0 -0,01 -0,01 0 0 0 0 0 0 0

25 0 0 0 0 0,01 0 0 0 0 0 0 0,01 0 0 0 -0,01 0 0 0 0 -0,05 0

26 0 0 0 0 -0,01 0 0 0 0 0 0 -0,01 0 -0,01 0 -0,01 0 0 0 0 -0,03 0

27 0,01 0,02 -0,01 0,01 0,05 0,03 0 0,01 -0,01 0,01 0,01 0,19 0,01 0,15 0,03 0,12 0,02 0,01 -0,01 0 -0,16 0,02

28 0 0 -0,01 0 0,33 0,13 0 0 0 0 0,01 0,4 0 0,39 0,07 0,24 0,05 0 0 0 -0,22 0

29 0 0,01 -0,01 0 0,22 0,08 0 0 -0,01 0 0,01 0,35 0 0,28 0,05 0,19 0,04 0 -0,01 0 -0,2 0,01

30 0 -0,01 0 0 0,31 0,13 0 0 0 0 0,01 0,41 0,01 0,4 0,07 0,27 0,07 0,01 0,01 0 -0,17 0

31 0,01 -0,02 0,01 0,01 0,13 0,05 0,01 0,01 0,01 0,01 0,01 0,26 0,02 0,22 0,04 0,16 0,04 0,02 0,02 0 -0,08 0

32 0 0,01 -0,02 0 0,25 0,09 -0,01 0 -0,01 0,01 0,01 0,41 0 0,34 0,06 0,23 0,05 0 -0,01 0 -0,23 0,01

33 0,01 -0,01 0 0 0,24 0,09 0,01 0 0,01 0,01 0,01 0,38 0,01 0,35 0,06 0,25 0,06 0,01 0,01 0 -0,16 0

34 0,16 0,14 0,26 0,44 0,01 0 0,48 0,4 0,61 0,4 0,21 0 0,34 0 0 0,01 0 0,56 0,25 0,29 -0,01 0,24

35 0 -0,01 0,1 0,26 0 0 0,23 0,31 -0,02 0,2 0,17 -0,01 0,07 0 0 -0,02 0 0,03 0,17 0,19 -0,03 0,26

36 -0,23 0,01 -0,19 0,06 0 0 0,14 0,28 0,16 0,26 0,1 -0,02 -0,14 -0,01 0 -0,01 0 0,02 0,17 0,23 -0,01 0,31

37 0,01 0,01 0,01 0,01 0,18 0,08 0,01 0,01 0,02 0,01 0 0,28 0 0,32 0,03 0,12 0,02 -0,01 0,02 0,02 -0,04 0,02

38 0,01 0,01 0,01 0,01 0,22 0,12 0,01 0,01 0,02 0,01 0 0,33 0 0,45 0,06 0,23 0,05 0 0,01 0,02 -0,01 0,01

39 0 0 0 0 0,1 0,32 0 0 0 0 0 0,12 0 0,22 0,97 0,23 0,99 0 0 0 -0,02 0

40 0 0 0 0 0,05 0,01 0 0 0 0 0 0,17 0 0,05 0,01 0,04 0,01 0 0 0 -0,02 0

41 0 0 0 0 0,16 0,05 0 0 0 0 0 0,42 0 0,23 0,06 0,18 0,06 0 0 0 -0,18 0

42 0,01 0,03 -0,01 0,03 0,17 0,06 -0,01 0,02 0 0,04 0,02 0,67 -0,01 0,24 0,04 0,18 0,04 0,01 -0,01 -0,01 -0,29 0,03

43 0 0,01 0 0,01 0,1 0,06 0 0,01 0 0,01 0,01 0,41 0 0,17 0,25 0,15 0,25 0 0 0 -0,18 0,01

44 0,01 0,02 0,01 0,01 0,15 0,06 0,01 0,01 0,01 0,01 -0,01 0,57 0,02 0,2 0,04 0,27 0,05 0,02 0,02 0 -0,25 0



23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44

23 1 0 0 0 0,01 0 0 0 0 0 0 0,17 0,26 0,18 0 0 0 0 0 0,02 0,01 0,01

24 0 1 0 0 0 0 0 0 -0,01 0 -0,01 0 0 0 0 0 -0,01 0 0 0 -0,02 0

25 0 0 1 0 0 0,05 0,09 0,02 0 0,01 0,02 0 0 0 -0,01 0 0 0,08 0,08 0,03 0,02 0

26 0 0 0 1 -0,01 0,11 0,05 0,03 0,01 0,05 -0,01 0 0 0 -0,01 -0,01 0 0 0 -0,01 0 -0,01

27 0,01 0 0 -0,01 1 0,25 0,09 0,17 0,04 0,22 0,17 0 0 0 0,05 0,08 0,03 -0,03 0,05 0,1 0,05 0,09

28 0 0 0,05 0,11 0,25 1 0,68 0,58 0,17 0,74 0,52 0 0 0 0,19 0,22 0,08 -0,02 0,13 0,16 0,11 0,24

29 0 0 0,09 0,05 0,09 0,68 1 0,24 0,07 0,95 0,25 -0,01 0 0 0,16 0,16 0,06 0 0,13 0,18 0,15 0,21

30 0 0 0,02 0,03 0,17 0,58 0,24 1 0,71 0,44 0,87 0,01 -0,01 -0,01 0,21 0,28 0,09 0,01 0,13 0,15 0,1 0,25

31 0 -0,01 0 0,01 0,04 0,17 0,07 0,71 1 0,18 0,85 0,02 -0,02 -0,02 0,1 0,15 0,05 0,04 0,11 0,13 0,12 0,13

32 0 0 0,01 0,05 0,22 0,74 0,95 0,44 0,18 1 0,43 -0,01 0,01 0 0,19 0,21 0,07 0 0,15 0,2 0,17 0,25

33 0 -0,01 0,02 -0,01 0,17 0,52 0,25 0,87 0,85 0,43 1 0,01 -0,01 -0,01 0,16 0,23 0,08 0,04 0,15 0,17 0,13 0,23

34 0,17 0 0 0 0 0 -0,01 0,01 0,02 -0,01 0,01 1 0,15 0,18 0,03 0,02 0 0 0 -0,02 -0,01 0,02

35 0,26 0 0 0 0 0 0 -0,01 -0,02 0,01 -0,01 0,15 1 0,55 0,02 0,01 0 0 0 0 0 -0,02

36 0,18 0 0 0 0 0 0 -0,01 -0,02 0 -0,01 0,18 0,55 1 0,02 0,01 0 0 0 -0,01 0 -0,01

37 0 0 -0,01 -0,01 0,05 0,19 0,16 0,21 0,1 0,19 0,16 0,03 0,02 0,02 1 0,68 0,06 -0,01 0,06 0,11 0,04 0,05

38 0 0 0 -0,01 0,08 0,22 0,16 0,28 0,15 0,21 0,23 0,02 0,01 0,01 0,68 1 0,1 -0,05 0,08 0,09 0,03 0,1

39 0 -0,01 0 0 0,03 0,08 0,06 0,09 0,05 0,07 0,08 0 0 0 0,06 0,1 1 0,01 0,07 0,05 0,25 0,05

40 0 0 0,08 0 -0,03 -0,02 0 0,01 0,04 0 0,04 0 0 0 -0,01 -0,05 0,01 1 0,7 0,4 0,21 -0,07

41 0 0 0,08 0 0,05 0,13 0,13 0,13 0,11 0,15 0,15 0 0 0 0,06 0,08 0,07 0,7 1 0,67 0,53 0

42 0,02 0 0,03 -0,01 0,1 0,16 0,18 0,15 0,13 0,2 0,17 -0,02 0 -0,01 0,11 0,09 0,05 0,4 0,67 1 0,59 0,01

43 0,01 -0,02 0,02 0 0,05 0,11 0,15 0,1 0,12 0,17 0,13 -0,01 0 0 0,04 0,03 0,25 0,21 0,53 0,59 1 0,04

44 0,01 0 0 -0,01 0,09 0,24 0,21 0,25 0,13 0,25 0,23 0,02 -0,02 -0,01 0,05 0,1 0,05 -0,07 0 0,01 0,04 1



45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

23 0 0,01 0 0 0 0,01 0,77 0 0,01 0,24 0,3 0,39 0,01 0,01 0,04 0,02 0,03 0,03 0,01 0,01 0 0,06 -0,1

24 -0,01 -0,01 -0,01 -0,01 -0,03 -0,01 0 0 0 0 0 0 0 0 0,01 0 0 0 0 0 -0,05 0 0

25 0,01 0 -0,01 0 0 -0,01 0 -0,01 -0,01 0 0 0 0,01 0,01 0 0,02 0,01 0,01 0 0,01 0 0 0

26 0 0 -0,01 0 0 0 0 -0,01 -0,01 0 0 0 0 0 0,03 0,02 0,03 0,03 0 0 0 0 0

27 0,16 0,11 0,08 0,09 -0,01 0,07 0 0,13 0,09 0,02 0 0,01 0,02 0,02 0,13 0,13 0,14 0,14 -0,02 0,02 0 -0,01 0,01

28 0,19 0,21 0,31 0,13 0,03 0,14 0 0,29 0,19 0 0 0 0,08 0,08 0,25 0,21 0,24 0,27 0 0,08 0 0 0

29 0,14 0,21 0,32 0,11 0,13 0,16 0 0,2 0,17 0 0 0 0,09 0,1 0,32 0,19 0,28 0,28 0,01 0,1 0 -0,01 0

30 0,19 0,21 0,28 0,14 0,02 0,15 0 0,29 0,16 0 0,01 0,01 0,1 0,1 0,23 0,16 0,2 0,24 0,01 0,1 0 0 0

31 0,1 0,13 0,2 0,09 0,07 0,1 0 0,16 0,11 -0,01 0,02 0,02 0,11 0,11 0,22 0,11 0,17 0,18 0,03 0,11 0,01 0,01 0,01

32 0,18 0,24 0,35 0,14 0,12 0,18 0 0,24 0,19 0,01 0 0 0,1 0,11 0,35 0,21 0,31 0,32 0,01 0,11 0 -0,01 0

33 0,17 0,2 0,29 0,13 0,05 0,15 0 0,26 0,16 -0,01 0,01 0,01 0,12 0,12 0,26 0,17 0,23 0,25 0,02 0,12 0 0,01 0

34 0,01 0,03 0 0,01 0 0,04 0,26 -0,01 -0,02 0,36 0,37 0,39 0,01 0,01 0,01 0,02 0 0,01 0,02 0,01 0 0,38 0,24

35 -0,01 -0,03 0,01 -0,01 0 -0,04 0,2 -0,01 -0,01 0,2 0,31 0,33 0,02 0,02 0,01 -0,04 0 0 0,03 0,02 0 -0,06 0,2

36 -0,01 -0,02 0,01 -0,01 0 -0,03 0,27 -0,01 -0,02 0,18 -0,14 0,03 0,01 0,01 0,01 -0,03 0 0 0,02 0,01 0 -0,01 -0,08

37 0,07 0,05 0,2 0,04 -0,02 0,03 0,01 0,14 0,1 0,02 0,02 0,03 0,03 0,03 0,14 0,13 0,16 0,16 -0,02 0,03 -0,01 0,02 0,01

38 0,1 0,1 0,27 0,11 0 0,09 0,01 0,27 0,17 0,01 0,01 0,02 0,05 0,06 0,19 0,15 0,19 0,2 0,01 0,06 -0,02 0,02 0,01

39 0,11 0,29 0,21 0,12 0,24 0,33 0 0,1 0,06 0 0 0 0,02 0,02 0,05 0,05 0,05 0,06 0,01 0,02 0,4 0 0

40 0,02 0,01 0,03 0,07 0 0,04 0 -0,02 -0,03 0 0 0 0 0 0,09 0,09 0,1 0,1 0 0 0 0 0

41 0,22 0,2 0,19 0,25 0,03 0,19 0 0,1 0,02 0 0 0 0,02 0,02 0,19 0,17 0,2 0,21 -0,02 0,02 0 0 0

42 0,15 0,17 0,19 0,15 0 0,13 0,01 0,15 0,03 0,03 0,01 0,02 0,04 0,04 0,28 0,25 0,31 0,32 -0,02 0,04 0 0,01 0,02

43 0,21 0,31 0,2 0,22 0,19 0,3 0 0,1 0,02 0,01 0 0,01 0,03 0,03 0,24 0,14 0,21 0,2 -0,02 0,03 0,08 0 0,01

44 0,56 0,66 0,24 0,35 0,01 0,37 0 0,18 0,04 0,02 0,02 0 0,14 0,14 0,24 0,21 0,25 0,27 0,01 0,14 0 0,01 0,01



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

45 0 0,01 0 0 0,11 0,06 0,01 0 0,01 0,01 0 0,41 0,01 0,16 0,1 0,28 0,11 0,01 0,01 0 -0,18 0

46 0,02 0,04 0,01 0,01 0,12 0,08 0,02 0,01 0,02 0,02 -0,02 0,48 0,03 0,2 0,28 0,33 0,29 0,03 0,03 0 -0,21 0,01

47 0 0,01 0 0,01 0,22 0,13 0 0,01 0 0,01 0,01 0,38 0 0,39 0,19 0,27 0,19 0 0 0,01 -0,09 0,01

48 0,01 0,01 0 0 0,01 0,01 0,01 0 0,01 0,01 -0,01 0,29 0,01 0,16 0,12 0,23 0,12 0,01 0,01 0 -0,12 0

49 0 0 0 0 -0,03 0 0 0 0 0 0 0,01 0 0,05 0,25 0,1 0,24 0 0 0 0 0

50 0,02 0,05 0,02 0,02 -0,02 0,01 0,03 0,01 0,03 0,03 -0,03 0,29 0,04 0,16 0,33 0,27 0,34 0,05 0,04 0 -0,13 0,01

51 0,56 0,3 -0,1 0,54 0 0 0,51 0,85 0,49 0,39 0,03 0,01 0,43 0 0 0,01 0 0,54 0,4 0,3 -0,02 0,22

52 0 0 -0,01 -0,01 0,22 0,1 -0,01 -0,01 0 -0,01 -0,01 0,42 -0,01 0,55 0,08 0,4 0,07 -0,01 0 -0,01 -0,1 -0,01

53 0 0,02 0,01 0,01 0,1 0,05 0,02 0,01 0,01 0,01 -0,02 0,2 0,02 0,3 0,05 0,25 0,04 0 0,02 0 0,05 -0,01

54 0,23 0,46 0,04 0,34 0 0 0,3 0,33 0,3 0,2 0,16 0,03 0,15 0 0 -0,01 0 0,16 0,19 0,29 -0,02 0,05

55 0,41 -0,04 0,36 0,62 0 0 0,38 0,4 0,35 0,41 0,21 0,02 0,49 0,01 0 0,01 0 0,37 0,25 0,18 -0,04 0,28

56 0,32 0,36 0,26 0,54 0,01 0 0,36 0,52 0,35 0,4 0,21 0,02 0,38 0,01 0 0,01 0 0,44 0,44 0,24 -0,03 0,15

57 0,02 0,01 0 0,01 0,18 0,06 0,01 0,01 0,01 0,01 0,01 0,14 0,02 0,04 0,01 0,09 0,02 0,02 0,02 0 0,01 0,01

58 0,02 0,02 0 0,01 0,17 0,06 0,01 0,01 0,01 0,01 0,01 0,15 0,02 0,04 0,01 0,09 0,02 0,02 0,02 0 0 0,01

59 -0,01 0,04 -0,01 0,04 0,17 0,07 0,04 0,05 0,04 0,06 0,02 0,4 0,02 0,17 0,04 0,19 0,04 0,03 0,01 0,02 -0,02 0,04

60 0,02 0,04 -0,02 0,04 0,17 0,07 0,05 0,04 0,05 0,04 0,01 0,35 0,03 0,17 0,03 0,18 0,03 0,02 0 0,02 -0,04 0

61 -0,01 0,06 -0,02 0,04 0,17 0,08 0,04 0,05 0,04 0,05 0,01 0,43 0,02 0,2 0,04 0,19 0,04 0,02 0,01 0,02 -0,05 0,04

62 0 0,06 -0,02 0,05 0,2 0,09 0,05 0,05 0,04 0,05 0,01 0,45 0,02 0,21 0,04 0,22 0,04 0,02 0,01 0,02 -0,06 0,03

63 0,01 0 0,01 0,01 0,15 0,05 0,01 0,01 0,01 0,02 0,01 0,03 0,02 0 0 0,05 0 0,02 0,02 0 0,08 0,01

64 0,02 0,02 0 0,01 0,17 0,06 0,01 0,01 0,01 0,01 0,01 0,15 0,02 0,04 0,01 0,09 0,02 0,02 0,02 0 0 0,01

65 0 0 0 0 -0,01 0,01 0 0 0 0 0 0 0 0,04 0,41 0,03 0,41 0 0 0 0 0

66 0,16 0,09 0,31 0,28 0,01 0 0,42 0,17 0,62 0,3 0,08 0,02 0,26 0,01 0 0,02 0 0,27 0,18 0,38 0,06 0,22

67 0,04 0,04 0,27 0 0 0 -0,01 -0,11 -0,05 0,18 0,07 0,02 0,02 0 0 0,01 0 0,07 -0,01 0,09 0,01 -0,13



23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44

45 0 -0,01 0,01 0 0,16 0,19 0,14 0,19 0,1 0,18 0,17 0,01 -0,01 -0,01 0,07 0,1 0,11 0,02 0,22 0,15 0,21 0,56

46 0,01 -0,01 0 0 0,11 0,21 0,21 0,21 0,13 0,24 0,2 0,03 -0,03 -0,02 0,05 0,1 0,29 0,01 0,2 0,17 0,31 0,66

47 0 -0,01 -0,01 -0,01 0,08 0,31 0,32 0,28 0,2 0,35 0,29 0 0,01 0,01 0,2 0,27 0,21 0,03 0,19 0,19 0,2 0,24

48 0 -0,01 0 0 0,09 0,13 0,11 0,14 0,09 0,14 0,13 0,01 -0,01 -0,01 0,04 0,11 0,12 0,07 0,25 0,15 0,22 0,35

49 0 -0,03 0 0 -0,01 0,03 0,13 0,02 0,07 0,12 0,05 0 0 0 -0,02 0 0,24 0 0,03 0 0,19 0,01

50 0,01 -0,01 -0,01 0 0,07 0,14 0,16 0,15 0,1 0,18 0,15 0,04 -0,04 -0,03 0,03 0,09 0,33 0,04 0,19 0,13 0,3 0,37

51 0,77 0 0 0 0 0 0 0 0 0 0 0,26 0,2 0,27 0,01 0,01 0 0 0 0,01 0 0

52 0 0 -0,01 -0,01 0,13 0,29 0,2 0,29 0,16 0,24 0,26 -0,01 -0,01 -0,01 0,14 0,27 0,1 -0,02 0,1 0,15 0,1 0,18

53 0,01 0 -0,01 -0,01 0,09 0,19 0,17 0,16 0,11 0,19 0,16 -0,02 -0,01 -0,02 0,1 0,17 0,06 -0,03 0,02 0,03 0,02 0,04

54 0,24 0 0 0 0,02 0 0 0 -0,01 0,01 -0,01 0,36 0,2 0,18 0,02 0,01 0 0 0 0,03 0,01 0,02

55 0,3 0 0 0 0 0 0 0,01 0,02 0 0,01 0,37 0,31 -0,14 0,02 0,01 0 0 0 0,01 0 0,02

56 0,39 0 0 0 0,01 0 0 0,01 0,02 0 0,01 0,39 0,33 0,03 0,03 0,02 0 0 0 0,02 0,01 0

57 0,01 0 0,01 0 0,02 0,08 0,09 0,1 0,11 0,1 0,12 0,01 0,02 0,01 0,03 0,05 0,02 0 0,02 0,04 0,03 0,14

58 0,01 0 0,01 0 0,02 0,08 0,1 0,1 0,11 0,11 0,12 0,01 0,02 0,01 0,03 0,06 0,02 0 0,02 0,04 0,03 0,14

59 0,04 0,01 0 0,03 0,13 0,25 0,32 0,23 0,22 0,35 0,26 0,01 0,01 0,01 0,14 0,19 0,05 0,09 0,19 0,28 0,24 0,24

60 0,02 0 0,02 0,02 0,13 0,21 0,19 0,16 0,11 0,21 0,17 0,02 -0,04 -0,03 0,13 0,15 0,05 0,09 0,17 0,25 0,14 0,21

61 0,03 0 0,01 0,03 0,14 0,24 0,28 0,2 0,17 0,31 0,23 0 0 0 0,16 0,19 0,05 0,1 0,2 0,31 0,21 0,25

62 0,03 0 0,01 0,03 0,14 0,27 0,28 0,24 0,18 0,32 0,25 0,01 0 0 0,16 0,2 0,06 0,1 0,21 0,32 0,2 0,27

63 0,01 0 0 0 -0,02 0 0,01 0,01 0,03 0,01 0,02 0,02 0,03 0,02 -0,02 0,01 0,01 0 -0,02 -0,02 -0,02 0,01

64 0,01 0 0,01 0 0,02 0,08 0,1 0,1 0,11 0,11 0,12 0,01 0,02 0,01 0,03 0,06 0,02 0 0,02 0,04 0,03 0,14

65 0 -0,05 0 0 0 0 0 0 0,01 0 0 0 0 0 -0,01 -0,02 0,4 0 0 0 0,08 0

66 0,06 0 0 0 -0,01 0 -0,01 0 0,01 -0,01 0,01 0,38 -0,06 -0,01 0,02 0,02 0 0 0 0,01 0 0,01

67 -0,1 0 0 0 0,01 0 0 0 0,01 0 0 0,24 0,2 -0,08 0,01 0,01 0 0 0 0,02 0,01 0,01



45 1 0,72 0,22 0,76 0,1 0,55 0 0,11 0,02 0,01 0,01 0 0,09 0,09 0,19 0,15 0,19 0,2 0,01 0,09 0,02 0,01 0

46 0,72 1 0,3 0,57 0,34 0,76 0,01 0,16 0,04 0,03 0,04 0 0,11 0,11 0,26 0,17 0,23 0,24 0,01 0,11 0,08 0,02 0,01

47 0,22 0,3 1 0,18 0,2 0,24 0 0,28 0,2 0,01 0 0,01 0,07 0,08 0,38 0,18 0,31 0,29 -0,02 0,08 0,08 0 0

48 0,76 0,57 0,18 1 0,15 0,73 0 0,11 0,03 0,01 0,01 0 0,07 0,07 0,15 0,1 0,15 0,15 0,01 0,07 0,03 0,01 0

49 0,1 0,34 0,2 0,15 1 0,46 0 0,01 0 0 0 0 0 0 0,13 -0,06 0,02 0 0 0 0,06 0 0

50 0,55 0,76 0,24 0,73 0,46 1 0,01 0,1 0,03 0,04 0,05 0 0,06 0,06 0,18 0,09 0,15 0,14 0 0,06 0,1 0,03 0,01

51 0 0,01 0 0 0 0,01 1 0 0,01 0,29 0,34 0,46 0,01 0,01 0,03 0,03 0,03 0,03 0,01 0,01 0 0,19 -0,07

52 0,11 0,16 0,28 0,11 0,01 0,1 0 1 0,54 -0,02 -0,01 0 0,02 0,03 0,17 0,15 0,19 0,19 -0,02 0,03 0 0 0

53 0,02 0,04 0,2 0,03 0 0,03 0,01 0,54 1 -0,03 -0,01 0,02 0,05 0,05 0,19 0,21 0,23 0,22 0 0,05 0 0,02 0

54 0,01 0,03 0,01 0,01 0 0,04 0,29 -0,02 -0,03 1 0,14 0,29 0 0 0 0,01 0,01 0,01 -0,01 0 0 0,22 0,06

55 0,01 0,04 0 0,01 0 0,05 0,34 -0,01 -0,01 0,14 1 0,51 0,01 0,01 0 0,02 0 0,01 0,02 0,01 0 0,2 0,27

56 0 0 0,01 0 0 0 0,46 0 0,02 0,29 0,51 1 0,02 0,02 0,02 0 0,02 0,02 0,02 0,02 0 0,22 0,26

57 0,09 0,11 0,07 0,07 0 0,06 0,01 0,02 0,05 0 0,01 0,02 1 0,99 0,33 0,36 0,26 0,4 0,81 0,99 0 0,01 0,01

58 0,09 0,11 0,08 0,07 0 0,06 0,01 0,03 0,05 0 0,01 0,02 0,99 1 0,36 0,33 0,27 0,41 0,81 1 0 0,01 0,01

59 0,19 0,26 0,38 0,15 0,13 0,18 0,03 0,17 0,19 0 0 0,02 0,33 0,36 1 0,56 0,88 0,88 0,19 0,36 0,01 0,03 0,01

60 0,15 0,17 0,18 0,1 -0,06 0,09 0,03 0,15 0,21 0,01 0,02 0 0,36 0,33 0,56 1 0,72 0,79 0,21 0,33 0 0,04 0

61 0,19 0,23 0,31 0,15 0,02 0,15 0,03 0,19 0,23 0,01 0 0,02 0,26 0,27 0,88 0,72 1 0,95 0,02 0,27 0 0,03 0

62 0,2 0,24 0,29 0,15 0 0,14 0,03 0,19 0,22 0,01 0,01 0,02 0,4 0,41 0,88 0,79 0,95 1 0,23 0,41 0 0,03 0

63 0,01 0,01 -0,02 0,01 0 0 0,01 -0,02 0 -0,01 0,02 0,02 0,81 0,81 0,19 0,21 0,02 0,23 1 0,81 0 0,01 0,01

64 0,09 0,11 0,08 0,07 0 0,06 0,01 0,03 0,05 0 0,01 0,02 0,99 1 0,36 0,33 0,27 0,41 0,81 1 0 0,01 0,01

65 0,02 0,08 0,08 0,03 0,06 0,1 0 0 0 0 0 0 0 0 0,01 0 0 0 0 0 1 0 0

66 0,01 0,02 0 0,01 0 0,03 0,19 0 0,02 0,22 0,2 0,22 0,01 0,01 0,03 0,04 0,03 0,03 0,01 0,01 0 1 0,36

67 0 0,01 0 0 0 0,01 -0,07 0 0 0,06 0,27 0,26 0,01 0,01 0,01 0 0 0 0,01 0,01 0 0,36 1



Legenda

1 ability_to_cook_food 23 personal_hygiene_and_grooming 45 shops_time

2 abstraction 24 pharmacy_visits_month 46 shops_visits

3 appetite 25 phonecalls_long_placed_perc 47 still_time

4 bathing_and_showering 26 phonecalls_long_received_perc 48 supermarket_time

5 bedroom_time 27 phonecalls_missed 49 supermarket_time_perc

6 bedroom_visits 28 phonecalls_placed 50 supermarket_visits

7 dependence 29 phonecalls_placed_perc 51 toilet_hygiene

8 dressing 30 phonecalls_received 52 tvwatching_time

9 exhaustion 31 phonecalls_received_perc 53 tvwatching_time_perc

10 falls 32 phonecalls_short_placed_perc 54 visits

11 gait_balance 33 phonecalls_short_received_perc 55 visit_to_doctors

12 home_time 34 quality_of_sleep 56 visit_to_health_related_places

13 housekeeping 35 reading_books 57 walk_distance_outdoor

14 kitchen_time 36 reading_newspapers 58 walk_distance_outdoor_fast

15 kitchen_visits 37 restroom_time 59 walk_distance_outdoor_fast_perc

16 livingroom_time 38 restroom_visits 60 walk_distance_outdoor_slow_perc

17 livingroom_visits 39 room_changes 61 walk_speed_outdoor

18 medication 40 seniorcenter_long_visits 62 walk_speed_outdoor_fast

19 memory 41 seniorcenter_time 63 walk_time_outdoor

20 mood 42 seniorcenter_time_out_perc 64 walk_time_outdoor_fast

21 outdoor_time 43 seniorcenter_visits 65 washingmachine_sessions

22 pain 44 shops_outdoor_time_perc 66 weakness

67 weight



12.4 Madrid Pilot

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

1 1 -0,1 0,03 0 0 0 0 0,53 0,51 -0,02 0 0 0 0 -0,01 0,01 -0,01 -0,02

2 -0,1 1 -0,15 0,03 0,03 0,11 0 0,08 0,1 0,07 0,17 0,29 -0,19 -0,12 -0,09 0,03 -0,09 -0,05

3 0,03 -0,15 1 0,01 0 0,01 0 0,01 0 0 0,02 0,03 -0,05 0,06 -0,03 -0,07 -0,06 -0,06

4 0 0,03 0,01 1 0,36 0,22 0,05 0 0 0,04 0 0 -0,01 -0,01 -0,11 -0,09 -0,07 -0,06

5 0 0,03 0 0,36 1 0,05 -0,05 0,01 0,01 -0,08 0,05 -0,01 0,03 0,02 0,19 0,13 0,06 0,06

6 0 0,11 0,01 0,22 0,05 1 0,12 0 0 0,03 0,17 0,3 -0,01 -0,01 -0,01 -0,02 0 0,02

7 0 0 0 0,05 -0,05 0,12 1 0 0 0,06 -0,04 -0,03 -0,01 -0,01 0 -0,02 0,01 0,04

8 0,53 0,08 0,01 0 0,01 0 0 1 0,99 0,01 0 0 0 0 0 0,01 0 -0,01

9 0,51 0,1 0 0 0,01 0 0 0,99 1 0,01 0 0 0 0 0 0,01 0 -0,01

10 -0,02 0,07 0 0,04 -0,08 0,03 0,06 0,01 0,01 1 -0,03 0,03 0,04 0,02 -0,05 -0,11 0,07 0,19

11 0 0,17 0,02 0 0,05 0,17 -0,04 0 0 -0,03 1 0,63 0 0 -0,03 -0,02 0,03 -0,03

12 0 0,29 0,03 0 -0,01 0,3 -0,03 0 0 0,03 0,63 1 0 0 -0,03 -0,04 -0,01 -0,03

13 0 -0,19 -0,05 -0,01 0,03 -0,01 -0,01 0 0 0,04 0 0 1 0,33 0,06 -0,02 0,06 0,01

14 0 -0,12 0,06 -0,01 0,02 -0,01 -0,01 0 0 0,02 0 0 0,33 1 0,03 -0,02 0,15 -0,01

15 -0,01 -0,09 -0,03 -0,11 0,19 -0,01 0 0 0 -0,05 -0,03 -0,03 0,06 0,03 1 0,51 0,24 0,44

16 0,01 0,03 -0,07 -0,09 0,13 -0,02 -0,02 0,01 0,01 -0,11 -0,02 -0,04 -0,02 -0,02 0,51 1 -0,19 -0,04

17 -0,01 -0,09 -0,06 -0,07 0,06 0 0,01 0 0 0,07 0,03 -0,01 0,06 0,15 0,24 -0,19 1 0,31

18 -0,02 -0,05 -0,06 -0,06 0,06 0,02 0,04 -0,01 -0,01 0,19 -0,03 -0,03 0,01 -0,01 0,44 -0,04 0,31 1



Legenda

1 culturepoi_visits_month

2 othersocial_time

3 othersocial_visits

4 outdoor_num

5 outdoor_time

6 publicpark_time

7 publicpark_visits

8 publictransport_distance_month

9 publictransport_rides_month

10 publictransport_time

11 seniorcenter_time

12 seniorcenter_visits

13 supermarket_time

14 supermarket_visits

15 walk_distance

16 walk_speed_outdoor

17 walk_steps

18 walk_time_outdoor



12.5 Montpellier Pilot

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

1 1 0,27 -0,25 -0,07 -0,02 0 -0,06 -0,18 -0,12 -0,02 -0,03 0,14 0,02 0,06 -0,03 0 0 0 -0,02 -0,08 0 -0,01 -0,03 -0,04 -0,22 -0,01 0

2 0,27 1 -0,01 0,45 -0,12 -0,06 0,03 -0,22 -0,1 0,1 0,42 0,04 -0,01 -0,03 -0,08 -0,04 -0,02 0,13 0,02 -0,15 0,02 0,54 0,05 0,04 -0,22 0 0,03

3 -0,25 -0,01 1 0,5 -0,01 -0,02 -0,01 -0,24 -0,1 -0,61 -0,23 -0,24 -0,08 0,01 0,01 0 0,01 0,02 0,02 -0,3 -0,37 -0,03 -0,01 0,01 0,59 0,01 0,04

4 -0,07 0,45 0,5 1 -0,06 -0,04 0,06 -0,25 0,05 -0,21 0,35 -0,12 -0,07 -0,06 -0,04 -0,02 -0,01 0,09 0,02 -0,26 -0,25 0,57 0,05 0,04 0,14 0 0,07

5 -0,02 -0,12 -0,01 -0,06 1 0,01 -0,03 0,01 0,02 0,01 -0,06 0 0,02 0,03 -0,11 0 0,16 0,01 -0,14 0,03 0 -0,07 0,53 0,03 0,02 -0,07 0,03

6 0 -0,06 -0,02 -0,04 0,01 1 0 0 0 0,02 -0,04 0 0 0 0,13 0,7 0 -0,26 0 0 0,01 -0,05 0,01 0 0,01 0 0

7 -0,06 0,03 -0,01 0,06 -0,03 0 1 0,11 0,16 0,16 0,18 -0,04 -0,16 -1 0 0 0 0 0 0 0,12 0,18 -0,02 0 -0,01 0 0,01

8 -0,18 -0,22 -0,24 -0,25 0,01 0 0,11 1 0,5 -0,06 0,13 0,06 -0,08 -0,11 0,02 0 0,01 -0,01 0 0,03 0,26 0,13 -0,01 -0,01 -0,06 0,01 0,04

9 -0,12 -0,1 -0,1 0,05 0,02 0 0,16 0,5 1 0,09 0,65 0,11 -0,07 -0,15 0,02 0 0,01 -0,01 0 -0,05 0,36 0,67 -0,01 -0,01 -0,04 0,01 0,13

10 -0,02 0,1 -0,61 -0,21 0,01 0,02 0,16 -0,06 0,09 1 0,39 0,08 -0,04 -0,16 -0,04 -0,02 -0,05 0 -0,02 -0,02 0,3 0,19 0,04 0,03 -0,39 -0,01 0,01

11 -0,03 0,42 -0,23 0,35 -0,06 -0,04 0,18 0,13 0,65 0,39 1 0,09 -0,06 -0,18 -0,05 -0,03 -0,01 0,08 0,02 -0,09 0,33 0,92 0,02 0,04 -0,25 0 0,09

12 0,14 0,04 -0,24 -0,12 0 0 -0,04 0,06 0,11 0,08 0,09 1 0,1 0,04 0 0 0 0 0 0,22 0,22 0,08 0 0 -0,27 0 0,01

13 0,02 -0,01 -0,08 -0,07 0,02 0 -0,16 -0,08 -0,07 -0,04 -0,06 0,1 1 0,16 0,15 0 0,03 -0,01 -0,01 0,07 0,03 -0,07 -0,01 -0,01 -0,06 0 -0,03

14 0,06 -0,03 0,01 -0,06 0,03 0 -1 -0,11 -0,15 -0,16 -0,18 0,04 0,16 1 0 0 0 0 0 0 -0,12 -0,18 0,02 0 0,01 0 -0,01

15 -0,03 -0,08 0,01 -0,04 -0,11 0,13 0 0,02 0,02 -0,04 -0,05 0 0,15 0 1 0,18 -0,02 -0,02 0,01 0,02 -0,01 -0,05 -0,09 0,01 0 0,01 0

16 0 -0,04 0 -0,02 0 0,7 0 0 0 -0,02 -0,03 0 0 0 0,18 1 0 -0,19 0 0,01 -0,01 -0,03 0 0 0,01 0 0

17 0 -0,02 0,01 -0,01 0,16 0 0 0,01 0,01 -0,05 -0,01 0 0,03 0 -0,02 0 1 0 -0,2 0 0,01 -0,01 -0,27 -0,49 -0,01 -0,11 -0,06

18 0 0,13 0,02 0,09 0,01 -0,26 0 -0,01 -0,01 0 0,08 0 -0,01 0 -0,02 -0,19 0 1 0 -0,01 -0,02 0,09 0,01 0 -0,01 0 0

19 -0,02 0,02 0,02 0,02 -0,14 0 0 0 0 -0,02 0,02 0 -0,01 0 0,01 0 -0,2 0 1 0 0 0,02 -0,01 0,47 0 0,56 -0,06

20 -0,08 -0,15 -0,3 -0,26 0,03 0 0 0,03 -0,05 -0,02 -0,09 0,22 0,07 0 0,02 0,01 0 -0,01 0 1 0,33 -0,13 0 0,01 -0,13 0 -0,07

21 0 0,02 -0,37 -0,25 0 0,01 0,12 0,26 0,36 0,3 0,33 0,22 0,03 -0,12 -0,01 -0,01 0,01 -0,02 0 0,33 1 0,34 -0,04 0 -0,26 0 0,04

22 -0,01 0,54 -0,03 0,57 -0,07 -0,05 0,18 0,13 0,67 0,19 0,92 0,08 -0,07 -0,18 -0,05 -0,03 -0,01 0,09 0,02 -0,13 0,34 1 0,03 0,04 -0,17 0,01 0,12

23 -0,03 0,05 -0,01 0,05 0,53 0,01 -0,02 -0,01 -0,01 0,04 0,02 0 -0,01 0,02 -0,09 0 -0,27 0,01 -0,01 0 -0,04 0,03 1 0,39 0,01 -0,04 0,07

24 -0,04 0,04 0,01 0,04 0,03 0 0 -0,01 -0,01 0,03 0,04 0 -0,01 0 0,01 0 -0,49 0 0,47 0,01 0 0,04 0,39 1 0 0,11 0,08

25 -0,22 -0,22 0,59 0,14 0,02 0,01 -0,01 -0,06 -0,04 -0,39 -0,25 -0,27 -0,06 0,01 0 0,01 -0,01 -0,01 0 -0,13 -0,26 -0,17 0,01 0 1 0,02 0,01

26 -0,01 0 0,01 0 -0,07 0 0 0,01 0,01 -0,01 0 0 0 0 0,01 0 -0,11 0 0,56 0 0 0,01 -0,04 0,11 0,02 1 -0,08

27 0 0,03 0,04 0,07 0,03 0 0,01 0,04 0,13 0,01 0,09 0,01 -0,03 -0,01 0 0 -0,06 0 -0,06 -0,07 0,04 0,12 0,07 0,08 0,01 -0,08 1



Legenda

1 bathroom_time

2 bathroom_visits

3 bedroom_time

4 bedroom_visits

5 cinema_visits

6 cinema_visits_month

7 home_time

8 kitchen_time

9 kitchen_visits

10 livingroom_time

11 livingroom_visits

12 meals_num

13 outdoor_num

14 outdoor_time

15 pharmacy_visits

16 pharmacy_visits_month

17 pharmacy_visits_week

18 restaurants_visits_month

19 restaurants_visits_week

20 restroom_time

21 restroom_visits

22 room_changes

23 shops_visits

24 shops_visits_week

25 sleep_time

26 supermarket_visits_week

27 visits_received_week



12.6 Singapore Pilot

1 2 3 4 5 6 7 8 9 10 11

1 1 0,1 -0,01 0,01 0 -0,02 -0,03 -0,02 -0,03 -0,04 -0,02

2 0,1 1 0,01 0,08 0,04 0,04 -0,04 -0,03 -0,05 -0,06 0

3 -0,01 0,01 1 0,39 -0,11 -0,12 -0,17 -0,14 -0,18 -0,18 -0,1

4 0,01 0,08 0,39 1 -0,03 0,01 -0,11 -0,09 -0,13 -0,1 0,01

5 0 0,04 -0,11 -0,03 1 0,22 -0,08 -0,03 -0,14 0,15 0,13

6 -0,02 0,04 -0,12 0,01 0,22 1 0,48 0,67 -0,21 0,65 0,79

7 -0,03 -0,04 -0,17 -0,11 -0,08 0,48 1 0,74 -0,21 0,52 0,58

8 -0,02 -0,03 -0,14 -0,09 -0,03 0,67 0,74 1 -0,17 0,64 0,78

9 -0,03 -0,05 -0,18 -0,13 -0,14 -0,21 -0,21 -0,17 1 -0,1 -0,19

10 -0,04 -0,06 -0,18 -0,1 0,15 0,65 0,52 0,64 -0,1 1 0,71

11 -0,02 0 -0,1 0,01 0,13 0,79 0,58 0,78 -0,19 0,71 1

Legenda

1 bathroom_time

2 bathroom_visits

3 bedroom_time

4 bedroom_visits

5 kitchen_time

6 kitchen_visits

7 livingroom_time

8 livingroom_visits

9 restroom_time

10 restroom_visits

11 room_changes



13 Annex: FAI data from Pilots

13.1 Athens Pilot

Frailty score

Robustness score

Label Frailty score

Robustness score

Label Frailty score

Robustness score

Label Frailty score

Robustness score

Label

0 4 Robust

0 3 Robust

0 4 Robust

1 2 preFrail

1 3 Robust

0 3 Robust

1 2 preFrail

2 4 Robust

1 3 Robust

1 2 preFrail

3 3 postRrobust

0 3 Robust

0 4 Robust

1 3 Robust

2 3 Robust

1 3 Robust

0 4 Robust

0 3 Robust

0 4 Robust

0 3 Robust

0 3 Robust

2 2 preFrail

0 3 Robust

2 2 preFrail

1 2 preFrail

4 2 Frail

1 2 preFrail

3 3 postRrobust

0 4 Robust

0 3 Robust

0 4 Robust

1 4 Robust

0 3 Robust

3 3 postRrobust

0 3 Robust

3 4 postRrobust

2 3 Robust

0 4 Robust

1 2 preFrail

1 4 Robust

3 4 postRrobust

1 3 Robust

3 4 postRrobust

1 3 Robust

2 3 Robust

0 4 Robust

2 3 Robust

0 4 Robust

0 4 Robust

1 3 Robust

1 3 Robust

1 3 Robust

1 3 Robust

1 3 Robust

1 3 Robust

1 3 Robust

0 3 Robust

0 4 Robust

1 3 Robust

1 3 Robust

0 3 Robust

0 4 Robust

1 2 preFrail

0 4 Robust

1 4 Robust

5 0 Frail

3 2 Frail

4 0 Frail

2 2 preFrail

3 3 postRrobust

0 4 Robust

3 3 postRrobust

0 4 Robust

1 3 Robust

1 3 Robust

1 3 Robust

3 2 Frail

0 4 Robust

2 3 Robust

1 3 Robust



13.2 Birmingham Pilot

Frailty score

Robustness score Label

0 5 Robust

0 3 Robust

0 5 Robust

0 5 Robust

0 5 Robust

0 6 Robust

0 5 Robust

0 5 Robust

2 6 Robust

0 4 Robust

4 3 postRrobust

0 4 Robust

0 5 Robust

0 4 Robust

0 6 Robust



13.3 Madrid Pilot

Frailty score

Robustness score Label

1 5 Robust

1 5 Robust

0 5 Robust

0 5 Robust

0 4 Robust

0 4 Robust

0 5 Robust

0 4 Robust

2 3 Robust

3 3 postRrobust

0 4 Robust

0 6 Robust

0 4 Robust

0 4 Robust

1 5 Robust



14 Annex: Data pre-processing SQL for the Athens Pilot experiment

Note: for full data anonymization, the City4AgeID pseudonyms have been obscured in the query below

select

avg(cinema_time) as cinema_time_avg,

stddev(cinema_time) as cinema_time_sd,

percentile_cont(0.75) within group ( order by cinema_time ) as cinema_time_best,

avg(cinema_visits) as cinema_visits_avg,

stddev(cinema_visits) as cinema_visits_sd,

percentile_cont(0.75) within group ( order by cinema_visits ) as cinema_visits_best,

avg(home_time) as home_time_avg,

stddev(home_time) as home_time_sd,

percentile_cont(0.25) within group ( order by home_time ) as home_time_best,

avg(othersocial_time) as othersocial_time_avg,

stddev(othersocial_time) as othersocial_time_sd,

percentile_cont(0.75) within group ( order by othersocial_time ) as othersocial_time_best,

avg(othersocial_visits) as othersocial_visits_avg,

stddev(othersocial_visits) as othersocial_visits_sd,

percentile_cont(0.75) within group ( order by othersocial_visits ) as othersocial_visits_best,

avg(pharmacy_time) as pharmacy_time_avg,

stddev(pharmacy_time) as pharmacy_time_sd,



percentile_cont(0.25) within group ( order by pharmacy_time ) as pharmacy_time_best,

avg(pharmacy_visits) as pharmacy_visits_avg,

stddev(pharmacy_visits) as pharmacy_visits_sd,

percentile_cont(0.25) within group ( order by pharmacy_visits ) as pharmacy_visits_best,

avg(restaurants_time) as restaurants_time_avg,

stddev(restaurants_time) as restaurants_time_sd,

percentile_cont(0.75) within group ( order by restaurants_time ) as restaurants_time_best,

avg(restaurants_visits_week) as restaurants_visits_week_avg,

stddev(restaurants_visits_week) as restaurants_visits_week_sd,

percentile_cont(0.75) within group ( order by restaurants_visits_week ) as restaurants_visits_week_best,

avg(seniorcenter_time) as seniorcenter_time_avg,

stddev(seniorcenter_time) as seniorcenter_time_sd,

percentile_cont(0.75) within group ( order by seniorcenter_time ) as seniorcenter_time_best,

avg(seniorcenter_visits) as seniorcenter_visits_avg,

stddev(seniorcenter_visits) as seniorcenter_visits_sd,

percentile_cont(0.75) within group ( order by seniorcenter_visits ) as seniorcenter_visits_best,

avg(shops_outdoor_time_perc) as shops_outdoor_time_perc_avg,

stddev(shops_outdoor_time_perc) as shops_outdoor_time_perc_sd,

percentile_cont(0.75) within group ( order by shops_outdoor_time_perc ) as shops_outdoor_time_perc_best,

avg(shops_time) as shops_time_avg,

stddev(shops_time) as shops_time_sd,



percentile_cont(0.75) within group ( order by shops_time ) as shops_time_best,

avg(shops_visits) as shops_visits_avg,

stddev(shops_visits) as shops_visits_sd,

percentile_cont(0.75) within group ( order by shops_visits ) as shops_visits_best,

avg(transport_time) as transport_time_avg,

stddev(transport_time) as transport_time_sd,

percentile_cont(0.75) within group ( order by transport_time ) as transport_time_best,

avg(walk_distance_outdoor) as walk_distance_outdoor_avg,

stddev(walk_distance_outdoor) as walk_distance_outdoor_sd,

percentile_cont(0.75) within group ( order by walk_distance_outdoor ) as walk_distance_outdoor_best,

avg(walk_speed_outdoor) as walk_speed_outdoor_avg,

stddev(walk_speed_outdoor) as walk_speed_outdoor_sd,

percentile_cont(0.75) within group ( order by walk_speed_outdoor ) as walk_speed_outdoor_best,

avg(walk_time_outdoor) as walk_time_outdoor_avg,

stddev(walk_time_outdoor) as walk_time_outdoor_sd,

percentile_cont(0.75) within group ( order by walk_time_outdoor ) as walk_time_outdoor_best,

max(status) as status

from crosstab('

select

format(''%s%s'', date(time_interval.interval_start), care_recipient.id) as row_name,

case when time_interval.interval_start <= ''2018-01-31'' then 1 else 2 end as trimester,

date(time_interval.interval_start), care_recipient.id,



variable.detection_variable_name, min(measure.measure_value)

from

city4age_sr.variation_measure_value as measure inner join

city4age_sr.user_in_role as care_recipient on measure.user_in_role_id = care_recipient.id inner join

city4age_sr.pilot as pilot on care_recipient.pilot_code = pilot.pilot_code inner join

city4age_sr.cd_detection_variable as variable on measure.measure_type_id = variable.id inner join

city4age_sr.time_interval as time_interval on measure.time_interval_id = time_interval.id

where

pilot.pilot_code = ''ath'' and

date(time_interval.interval_start) >= ''2017-11-01'' and date(time_interval.interval_start) <= ''2018-04-30''

group by time_interval.interval_start, care_recipient.id, variable.detection_variable_name

order by 1

','

select cat

from (VALUES

(''cinema_time''),

(''cinema_visits''),

(''cinema_visits_month''),

(''home_time''),

(''othersocial_time''),

(''othersocial_visits''),

(''pharmacy_time''),

(''pharmacy_visits''),

(''pharmacy_visits_month''),

(''restaurants_time''),

(''restaurants_visits_week''),

(''seniorcenter_time''),

(''seniorcenter_visits''),



(''shops_outdoor_time_perc''),

(''shops_time''),

(''shops_visits''),

(''supermarket_time''),

(''supermarket_visits''),

(''transport_time''),

(''walk_distance_outdoor''),

(''walk_distance_outdoor_fast_perc''),

(''walk_distance_outdoor_slow_perc''),

(''walk_speed_outdoor''),

(''walk_time_outdoor'')) as v(cat)

order by 1

') as ct(

row_name text,

trimester numeric,

day_m date,

cr_m numeric,

cinema_time numeric,

cinema_visits numeric,

cinema_visits_month numeric,

home_time numeric,

othersocial_time numeric,

othersocial_visits numeric,

pharmacy_time numeric,

pharmacy_visits numeric,

pharmacy_visits_month numeric,

restaurants_time numeric,

restaurants_visits_week numeric,



seniorcenter_time numeric,

seniorcenter_visits numeric,

shops_outdoor_time_perc numeric,

shops_time numeric,

shops_visits numeric,

supermarket_time numeric,

supermarket_visits numeric,

transport_time numeric,

walk_distance_outdoor numeric,

walk_distance_outdoor_fast_perc numeric,

walk_distance_outdoor_slow_perc numeric,

walk_speed_outdoor numeric,

walk_time_outdoor numeric

) inner join (VALUES

([City4AgeID obscured], 1, 'preFrail'),

([City4AgeID obscured], 1, 'Robust'),





([City4AgeID obscured], 1, 'postRobust'),












([City4AgeID obscured], 1, 'Frail'),



































































([City4AgeID obscured], 2, 'Frail')

) as labels(cr_m, trimester, status)

on ct.cr_m = labels.cr_m and ct.trimester = labels.trimester

group by ct.cr_m, ct.trimester

having avg(home_time) != 86400



15 Annex: MultiSchemeAUC modifications

Note: modifications that have been implemented to convert Weka MultiScheme meta-learner into City4Age MultiSchemeAUC are shaded

public void buildClassifier(Instances data) throws Exception {

...

double AUC = evaluation.weightedAreaUnderROC();

if (m_Debug) {

System.err.println("Error rate: " + Utils.doubleToString(AUC, 6, 4)

+ " for classifier "

+ currentClassifier.getClass().getName());

}

if ((i == 0) || (AUC > bestPerformance)) {

bestClassifier = currentClassifier;

bestPerformance = AUC;

bestIndex = i;

}

...



16 Annex: Computing *_time_per_visits features

case when sum(cinema_visits) != 0 then sum(cinema_time) / sum(cinema_visits) else NULL end as cinema_time_per_visit,

case when sum(othersocial_visits) != 0 then sum(othersocial_time) / sum(othersocial_visits) else NULL end as othersocial_time_per_visit,

case when sum(pharmacy_visits) != 0 then sum(pharmacy_time) / sum(pharmacy_visits) else NULL end as pharmacy_time_per_visit,

case when sum(restaurants_visits_week) != 0 then sum(restaurants_time) / sum(restaurants_visits_week) else NULL end as restaurants_time_per_visit,

case when sum(seniorcenter_visits) != 0 then sum(seniorcenter_time) / sum(seniorcenter_visits) else NULL end as seniorcenter_time_per_visit,

case when sum(shops_visits) != 0 then sum(shops_time) / sum(shops_visits) else NULL end as shops_time_per_visit,

Documents

ec.europa.eu · Project City4Age Grant Agreement #689731 D2.11 City4Age frailty and MCI risk model 1/183 Elderly-friendly city services for active and healthy ageing City4Age frailty