Predicting Vasovagal Syncope from Heart Rate and Blood Pressure:

A Prospective Study in 140 Subjects

Short Title: Predicting Vasovagal Syncope

Nathalie Virag1, PhD, Mark Erickson2, BS, Patricia Taraborrelli3, RN, PhD, Rolf Vetter4,

PhD, Phang Boon Lim3,5, PhD, FRCP, Richard Sutton3,5, DSc, FRCP, FHRS

1Medtronic Europe, Tolochenaz, Switzerland, 2Medtronic Inc., Minneapolis, MN, USA,

3Imperial Healthcare NHS Trust, Hammersmith Hospital, London, United Kingdom,

4Bern University of Applied Sciences, Burgdorf, Switzerland, 5National Heart & Lung

Institute, Imperial College, London, United Kingdom

Corresponding Author: Nathalie Virag, Medtronic Europe

Route du Molliau, CH-1131 Tolochenaz, Switzerland

Phone: + 41 21 802 73 35

Email: nathalie.virag@medtronic.com

Manuscript word count: 3915Abstract word count: 248

Page 1 of 27

Abstract

Background: We developed a vasovagal syncope (VVS) prediction algorithm for use

during head-up tilt with simultaneous analysis of heart rate (HR) and systolic blood

pressure (SBP). We previously tested this algorithm retrospectively in 1155 subjects,

showing sensitivity 95%, specificity 93% and median prediction time of 59s.

Objective: This study was prospective, single center, on 140 subjects to evaluate this

VVS prediction algorithm and assess if retrospective results were reproduced and

clinically relevant. Primary endpoint was VVS prediction: sensitivity and specificity

>80%.

Methods: In subjects, referred for 60° head-up tilt (Italian protocol), non-invasive HR

and SBP were supplied to the VVS prediction algorithm: simultaneous analysis of RR

intervals, SBP trends and their variability represented by low-frequency power generated

cumulative risk which was compared with a predetermined VVS risk threshold. When

cumulative risk exceeded threshold, an alert was generated. Prediction time was duration

between first alert and syncope.

Results: Of 140 subjects enrolled, data was usable for 134. Of 83 tilt+ve (61.9%), 81

VVS events were correctly predicted and of 51 tilt-ve subjects (38.1%), 45 were correctly

identified as negative by the algorithm. Resulting algorithm performance was sensitivity

97.6%, specificity 88.2%, meeting primary endpoint. Mean VVS prediction time was

2min 26s±3min16s with median 1min 25s. Using only HR and HR variability (without

SBP) the mean prediction time reduced to 1min34s±1min45s with median 1min13s.

1

Page 2 of 27

Conclusion: The VVS prediction algorithm, is clinically-relevant tool and could offer

applications including providing a patient alarm, shortening tilt-test time, or triggering

pacing intervention in implantable devices.

Key Words: vasovagal syncope, syncope prediction study, tilt-test, autonomic nervous

system, heart rate, blood pressure.

Clinical Trials-gov Identifier: NCT02140567

List of Abbreviations

BPV: blood pressure variability

HRV: heart rate variability

HUT: head-up tilt

LF: low frequency

OI: orthostatic intolerance

RR: RR interval

SBP: systolic blood pressure

SPS: syncope prediction study

VVS: vasovagal syncope

2

Page 3 of 27

Introduction

Vasovagal syncope (VVS) is the common form of neurally-mediated reflex syncope,

which is marked by a sudden fall in blood pressure with an associated fall in heart rate

resulting in syncope. The diagnosis of VVS may be made from the patient’s history but a

historical diagnosis is not always possible. Therefore, head-up tilt (HUT) testing is

widely used to offer diagnostic information on the patient’s syncope using ECG and

blood pressure monitoring with medical observation.

Prediction of impending VVS is desirable because if the patient has sufficient warning it

may be possible to abort an attack by sitting/lying or by use of physical counter-pressure

maneuvers. Many, especially older subjects have little or no warning (or prodrome) (1,2).

Provision of warning for these patients has particular value. Several methods have been

created to predict VVS, based on the analysis of cardiovascular variables such as heart

rate and blood pressure (3,4). Approaches using heart rate or blood pressure alone have

limited predictive value (3-5). The main problem of most syncope prediction algorithms

is the lack of specificity and great variations in the results. We overcame this limitation

by developing an algorithm to predict VVS during HUT based on the simultaneous

analysis of heart rate (RR interval) and beat-to-beat systolic blood pressure (SBP). This

algorithm was tested retrospectively on 1155 subjects during HUT. Results were

promising both in terms of sensitivity (94.7%), specificity (92.7%) and median prediction

time (59 seconds) (6).

3

Page 4 of 27

The purpose of the Syncope Prediction Study (SPS) was to perform a prospective

evaluation of our VVS prediction algorithm during HUT tests.

Methods

Clinical Protocol

SPS was a prospective, single center, observational study. The study population consisted

of 140 patients referred to the Cardiology Department, Hammersmith Hospital, Imperial

College Healthcare National Health Service Trust, London, United Kingdom, for tilt

testing where a diagnosis of VVS could not be made from the history alone. This sample

size was chosen in such a way as to reproduce results obtained in the retrospective

analysis on 1155 subjects (6) with a clinically relevant sensitivity and specificity.

Informed consent was obtained from each subject and the study had institutional research

ethics approval. The subject then underwent a standard HUT according to the Italian

protocol (7). The HUT started with a 5 min baseline recording in supine position, after

which the subject was tilted to a 60° head-up position. If symptoms did not develop after

20 min of tilt, sublingual glyceryl trinitrate (GTN) 400 µg, as spray, was administered

and the patient remained upright for another 15 min. The subject was returned to supine

as soon as syncope developed or after a total of 35 min of tilt.

A commercially available laptop computer with the software for VVS prediction was

called Tilt Test Analyzer (Figure 1). It was connected to the existing HUT system as

4

Page 5 of 27

follows: the output of the surface ECG (Fukuda Denshi Inc. Tokyo, Japan) and the digital

photoplethysmographic blood pressure recorded noninvasively (Finometer Pro, Finapress

Medical System BV, Amsterdam, Netherlands) were connected to the Tilt Test Analyzer,

which computed VVS risk in real time. Before the tilt test, the nurse, who conducted the

test, entered study patient ID in the Analyzer. The computer responded by displaying a

button to start baseline data collection. When this button was pressed, a timer was shown

indicating time to head up tilt. When the time to tilt the patient was reached, the computer

indicated to the nurse to start tilting the patient. During tilt, the computer indicated

whether the recorded physiological data was adequate, i.e. not been disconnected. The

output of the VVS prediction algorithm was blinded to the subject and the syncope nurse.

All study data was acquired before and during HUT, no follow-up data was collected.

Syncope Prediction Algorithm

The VVS prediction algorithm is described in detail by Virag et al (6). The algorithm is

based on concurrent analysis of several signals, each with a potential predictive value

(Integration of information block shown in Figure 1):

1. A normalized trend of RR intervals (low-pass filtering at 0.01Hz)

2. A normalized trend of SBP (low-pass filtering at 0.01Hz)

3. An indicator of autonomic modulation extracted from RR intervals.

4. An indicator of autonomic modulation extracted from SBP.

5

Page 6 of 27

We used Heart Rate Variability (HRV) to include information regarding the patient’s

autonomic modulation into the prediction. We selected the low-frequency (LF) power of

RR (LFRR) and SBP (LFSBP). The LF oscillations from 0.04-0.15Hz are primarily an

indicator of sympathetic modulation. LF was computed via an autoregressive frequency

spectrum, evaluated in sliding windows of 360s, shifted in 10s increments.

RR trend, SBP trend, LFRR and LFSBP have different ranges, so the values were

normalized with respect to baseline to bring them to comparable levels. Baseline values

for RR and SBP trends (mean and standard deviation) were computed during the first

180s of HUT. However, LFRR and LFSBP baseline values were established during the 300s

before tilt because stable signals are needed. Normalization of these 4 variables allows a

direct comparison of their effect on the global risk of VVS. As shown in Figure 2, the

cumulative VVS risk is a weighted sum of the normalized trends of RR/SBP and

LFRR/LFSBP. RR trend has a positive contribution to VVS risk since an RR increase,

corresponding to a heart rate decrease, will induce a decrease in blood pressure and as

such increase the risk of syncope. SBP trend, LFRR and LFSBP have a negative

contribution. Optimal weights for each parameter were determined as described in Virag

et al (6).

The VVS cumulative risk (VVS risk) reflects the probability that a patient will experience

VVS (from 0 to 1). The assessed risk is compared with an empirically determined VVS

risk threshold, which itself was determined during an algorithm optimization phase based

6

Page 7 of 27

on retrospective data: the risk threshold and the weights of the cumulative risk were

chosen to obtain the best tradeoff between sensitivity and specificity on a training set of

50 tilt-positives and 50 tilt-negatives using receiver-operating characteristics (6). As a

result, a fixed threshold of 0.42 was used. In the current study, when the computed VVS

cumulative risk exceeds 0.42, the algorithm predicts an imminent VVS episode and an

alert is generated (red asterisk in Figure 3).

Primary Endpoint

The primary objective of the study was to evaluate the VVS prediction algorithm in a

prospective cohort in the tilt laboratory. The primary endpoint was the VVS prediction

algorithm performance: sensitivity and specificity values >80%. Correct/incorrect VVS

prediction was assessed by comparing the output of the VVS algorithm (VVS alert) with

relevant clinical symptoms noted by the syncope nurse during HUT. This led to values

for the sensitivity and specificity.

Prediction time was the duration between first VVS alert and syncope. This value informs

us about how long before the event, VVS can be predicted, independent of the duration of

the tilt. The diagnosis time was the duration between the start of tilt and the first alert.

Results are expressed as mean ± standard deviation and median, where appropriate.

7

Page 8 of 27

Results

One-hundred-forty subjects were enrolled in the study and underwent HUT. The

following 6 subjects were excluded from the efficacy analysis due to recording problems

that prevented the computation of VVS risk:

• Subject #23: Finapress did not record blood pressure correctly.

• Subject #33: Tilt Test Analyzer stopped recording during HUT (unknown

software error).

• Subject #42: Finapress did not record blood pressure correctly.

• Subject #85: Tilt Test Analyzer did not record signals correctly (unknown

software error).

• Subject #92: Tilt Test Analyzer did not record signals correctly (unknown

software error).

• Subject #121: Finapress did not record blood pressure correctly.

The remaining 134 subjects were included for the calculation of the primary endpoints:

42 males (31.3%) and 92 (68.7%) females, mean age 37.2±15.1 years. During the tilt

procedure, 77 subjects (57.5%) experienced loss of consciousness and were considered as

tilt-positives. Of the 57 subjects who did not experience loss of consciousness, 6 showed

vasovagal pre-syncope and were clinically considered as tilt-positives by the syncope

nurse. According to the VASIS classification of positive responses to HUT (6): these

patients were classified as follows:

• Subject #17: VASIS-2A.

8

Page 9 of 27

• Subject #18: VASIS-1.

• Subject #48: Delayed orthostatic intolerance (OI).

• Subject #96: VASIS-1, HUT was stopped due to subject distress.

• Subject #134: VASIS-1.

• Subject #138: VASIS-3.

Therefore, for the whole database, we observed 83 tilt-positives (61.9%) and 51 tilt-

negatives (38.1%). For comparison, the dataset of our retrospective study on 1155

subjects had the following baseline characteristics: 65.7% tilt-positives and 34.3% tilt-

negatives. For the 83 tilt positives, VVS occurred at a mean of 23min 37s±7min 14s after

tilt-up, median 25min22s (ranging from 4min00s to 31min51s).

Efficacy results are presented in Figure 2. Of the 83 tilt-positives, 81 were correctly

predicted, leading to a sensitivity of 97.6%. Of the 51 tilt-negatives, 45 were correctly

identified as negative by the algorithm, leading to a specificity of 88.2%. The VVS

prediction algorithm generated 2 false negatives (FN, subjects with a tilt-positive test that

was not be predicted by the Tilt Test Analyzer) and 6 false positives (FP, subjects with a

tilt-negative test that were incorrectly detected as positive by the Tilt Test Analyzer). The

clinical observations for the non-predicted/wrongly predicted subjects are the following:

• Subject #27: FP, OI/VVS driven by hypotensive medication plus beta-

blockers.

9

Page 10 of 27

• Subject #58: FP, after 35 minutes of tilt, subject performed a Valsalva

maneuver and reproduced symptoms of defecation syncope.

• Subject #61: FP, not VVS but early OI.

• Subject #68: FP, history suggestive of VVS and reported as such.

• Subject #73: FP, clinically strictly negative but tendency to early OI.

• Subject #96: FN, subject did not experience loss of consciousness and

HUT was stopped due to subject distress.

• Subject #126: FN, tilt positive with loss of consciousness.

• Subject #128: FP, history suggestive of VVS and reported as such.

For the 81 subjects that were correctly predicted by the Tilt Test Analyzer, the mean

prediction time was 2min 26s±3min16s, ranging from 0min00s (VVS detection) to 22min

47s (VVS prediction). Median prediction time was 1min 25s. Distribution of prediction

time is shown in Figure 3. For comparison purposes, in our retrospective study on 1155

subjects, mean prediction time was 2min08s±3min36s, median prediction time was 59

seconds.

We assessed the effect of age on prediction. As shown in Figure 3, no statistical

relationship was found, in this study, between the age of the subject and the prediction

time. Furthermore no statistical difference was found between mean prediction time in

younger subjects (<40 years, 2min21s±2min50s) versus older subjects (>40 years,

2min36s±3min57s).

10

Page 11 of 27

We assessed the effect of gender. Of the 2 false negatives one was male and one female.

Of the 6 false positives, 1 was male and 4 female. This means that for the male group we

have a sensitivity of 96.3% and specificity of 93.3% and for the female group a

sensitivity of 98.2% and specificity of 86.1%. Of the 81 tilt-positives who were included

for the computation of prediction time 26 (32.1%) were male and 55 (67.9%) female. No

significant difference was observed between mean prediction time in the male group

(1min40s±1min01s) and the female group (2min49s±3min52s).

We also assessed the effect of baseline medication. 8 subjects were on beta blockers, 4

tilt-positives and 4 tilt-negatives. No significant difference was observed between mean

prediction time in the group with beta blockers (2min13s±44s) and the group without

(2min27s±3min21s). Two of the 6 FP subjects were taking beta-blockers (subject#27 and

subject#73). 12 subjects were on hypotensive drugs (ACE inhibitor or angiotensin II

receptor blockers), 7 tilt-positives and 5 tilt-negatives. No significant difference was

observed between mean prediction time in the group with hypotensive drugs

(1min38s±40s) and the group without (2min31s±3min25s).

Finally we assessed the effect of GTN administration on prediction time. Of the 81 tilt-

positives including in the computation of prediction time, 16 (19.8%) had a time to VVS

shorter than 20min (without GTN) and 65 (80.2%) had a time to VVS longer than 20min

(with GTN administration). However, we could not observe any statistical difference in

11

Page 12 of 27

mean prediction time between the group without GNT (2min13s±2min11s) and the group

with GTN (2min30s±3min32s).

Figure 4 shows four examples of the evolution of RR and SBP together with computed

LFRR, LFSBP and VVS cumulative risk for true positive, false positive, true negative and

false negative subjects. Periods when the subject’s VVS cumulative risk was above

threshold are indicated by unfilled circles.

The positive predictive value (PPV) and negative predictive value (NPV) were computed

by combining sensitivity, specificity and prevalence. For a prevalence of 60% the PPV is

92.6%, reflecting the probability that the subject detected by the Tilt Test Analyzer truly

suffers from VVS. This corresponds to the prevalence of syncope for the patients referred

to the center (tilt positive rate of 61.9%). If we consider lower prevalence rates of 13%

and 40%, this leads to a PPV of 55.4% and 85.7% respectively. While a higher

prevalence of 80% leads to a PPV of 97.1%. NPV values are 99.6%, 98.2%, 96.1% and

90.1% for a prevalence of 13%, 40%, 60% and 80% respectively.

If SBP is suppressed from computation in the syncope prediction algorithm, retaining

only RR and HRV (Figure 1) sensitivity falls to 89.5%, specificity to 64.1%, mean

prediction time to 1min34s±1min45min and median prediction to 1min13s, p=non-

significant, (VVS risk threshold of 0.42). By varying the VVS risk threshold the tradeoff

between sensitivity and specificity can be altered. A VVS risk threshold of 0.46 leads to

12

Page 13 of 27

sensitivity of 85.5%, specificity 72.5%, mean prediction time 1min27s±1min34s and

median prediction time of 1min10s. A VVS risk threshold of 0.50 leads to sensitivity of

77.2%, specificity 75.3%, mean prediction time 1min11s±1min08s and median prediction

time of 1min06s.

Discussion

The VVS algorithm, using simultaneous heart rate and systolic blood pressure

measurements, offers a clinically-useful prospective prediction tool for impending

syncope with high sensitivity of 97.6% and specificity of 88.2%. The median prediction

time of 1min 25s could allow the patient sufficient time to take evasive action such as

sitting or lying down or to employ physical counter-measures to abort syncope (9).

Warning may be particularly valuable to patients who experience no or very brief

prodrome for syncope. Warning could be delivered by a patient monitoring device or via

a smart watch.

Tilt testing is a long procedure that physicians have been trying to shorten since its

inception. Currently, there is some international agreement, at least in Europe (10), on use

of the Italian protocol (7). Introduction of this warning system to tilt testing has sufficient

sensitivity and specificity to allow early termination of a test yet with a confident

diagnosis. This approach would obviate the need to induce full syncope, an unpleasant

13

Page 14 of 27

experience for patients, and reduce the recovery time following tilt testing. This would

shorten the entire testing time, potentially allowing more patients to be studied in a day.

In our study, of the 6/51 patients who had a FP prediction on Tilt Test Analyzer, one had

clear situational (defecation) syncope, 2 had a compelling clinical history for vasovagal

syncope, and 3 had an orthostatic intolerance diagnosis. Despite the strictly negative tilt

table test result, all of these patients were given a likely diagnosis, as we believe that a

full assessment of the patient involves both the clinical history, and tilt test data. We

would then give suggested management plans based on both, individualized to the patient

clinical history and circumstances. We do not propose that the algorithm replaces the

clinical history, but rather supports it, without the need to bring patients to complete LOC

on routine tilt testing. In all these 6 patients, the tilt test analyser data corroborated the

clinical diagnosis, ascertained from the clinical history, more closely than the full tilt test

result.

A further possible use of this algorithm might be inclusion in an implantable device to

trigger earlier pacing intervention. There is available evidence that earlier pacing

intervention in evolving vasovagal syncope provides benefit in aborting or ameliorating

an attack (11,12). It is possible that use of this prediction algorithm could provide more

effective pacing than the present bradycardia dependent rate hysteresis devices such as

the Rate Drop Response algorithm (13,14).

14

Page 15 of 27

There remain important challenges to inclusion of this algorithm in an implantable device

as systolic blood pressure is required for its proper function. Work is in progress to

achieve this by employment of a blood pressure surrogate for external and implantable

use. Another possible challenge is extension of the warning period by seeking heart rate

increase that precedes the decrease that is currently used. It is well accepted that in most

VVS cases blood pressure starts to decline before heart rate fall (15). However, heart rate

increase reflecting the ubiquitous epinephrine rise prior to syncope (16) might offer a

longer prediction time.

In our algorithm the use of heart rate only led to a sensitivity reduction from 97.6% to

89.5% and a specificity reduction from 88.2% to 64.1%. Mean prediction time was

reduced from 2min26s to 1min34s while median prediction time remained in a

comparable range. The use of blood pressure is therefore important to keep the false

alarm rate within an acceptable range. By varying the algorithm parameters (VVS risk

threshold) we could optimize the tradeoff between sensitivity and specificity to 77.2%

and 75.3%. These findings encourage us to pursue use of a surrogate of blood pressure in

the algorithm. An effect of patient age on the prediction time was considered possibly to

be relevant but no statistical relationship was found between these two parameters.

To our knowledge, this is the first numerical estimate in a substantial series of randomly

(by referral) selected patients of the different timing of onset of VVS comparing SBP

15

Page 16 of 27

with HR alone yielding about a 1min difference. Such figures have been anticipated (15)

but numerical support is offered here.

Diagnostic time from tilt-up to alert signal was recorded but the data has not been

presented in detail as it was felt not to be germane to this report but it is possible that it

might have importance in a future patient application.

Limitations of the study

At the outset, it was our intention to include consecutive patients but this proved

impractical because of refusal to enter the study and a few patients were judged ahead of

possible inclusion to be unsuitable, for example, a high likelihood of the patient having

orthostatic hypotension, postural orthostatic tachycardia or psychogenic pseudosyncope.

This is a single center study with all the accompanying limitations that such a study

imposes. We based our analysis on data from tilt tests and not from spontaneous attacks.

Our aim is to study data from implantable loop recorders in the future but limitations

apply without a surrogate of blood pressure at present.

Conclusions

A clinically relevant syncope prediction algorithm has been designed, tested

retrospectively and now tested prospectively with good results. Clinical applications are

possible and they are now being explored.

16

Page 17 of 27

Acknowledgments

The SPS study was sponsored by Medtronic.

Conflicts of interest

NV and ME are employees of Medtronic. RS is a consultant to Medtronic, is a member of

Abbott Laboratories Inc. Speakers’ bureau and a stock holder in AstraZeneca PLC,

Edwards Life Sciences Corp. and Boston Scientific Inc. RV is a consultant to Medtronic,

PT and PBL have no conflicts to disclose.

17

Page 18 of 27

References

[1] O’Dwyer C, Bennett K, Langan Y, Fan CW, Kenny RA. Amnesia for loss of

consciousness is common in vasovagal syncope. Europace 2011; 13: 1040-1045.

[2] Brignole M, Menozzi C, Moya A, Andresen D, Blanc JJ, Krahn AD, Wieling W,

Beiras X, Deharo JC, Russo V, Tomaino M, Sutton R. Pacemaker therapy in patients

with neurally-mediated syncope and documented asystole. Third international study

on syncope of unknown etiology (ISSUE-3): a randomized trial. Circulation 2012;

125: 2566-2571.

[3] Kochiadakis GE, Orfanakis A, Chryssostomakis SI, Manios EG, Kounali DK, Vardas

PE. Autonomic nervous system activity during tilt testing in syncopal patients,

estimated by power spectral analysis of heart rate variability. Pacing Clin

Electrophysiol 1995; 20: 1332-41.

[4] Furlan R, Piazza S, Dell’Orto S, Barbic F, Bianchi A, Mainardi L, Cerutti S, Pagani

A, Malliani A. Cardiac autonomic patterns preceding occasional vasovagal reactions

in healthy humans. Circulation 1998;98:1756-1761.

[5] Pitzalis M, Massari F, Guida P, Iacoviello M, Mastropasqua F, Rizzon B, Forleo C,

Rizzon P. Shortened head-up tilting test guided by systolic pressure reductions in

neurocardiogenic syncope. Circulation 2002;105:146-148.

[6] Virag N, Sutton R, Vetter R, Markowitz T, Erickson M. Prediction of vasovagal

syncope from heart rate and blood pressure trend and variability: Experience in 1155

patients. Heart Rhythm 2007; 4: 1375-1382.

18

Page 19 of 27

[7] Bartoletti A, Alboni P, Ammirati F, Brignole M, Del Rosso A, Foglia Manzillo G,

Menozzi C, Raviele A, Sutton R. 'The Italian Protocol': a simplified head-up tilt

testing potentiated with oral nitroglycerin to assess patients with unexplained

syncope. Europace 2000; 2: 339-342.

[8] Brignole M, Menozzi C, Del Rosso A, Costa S, Gaggioli G, Bottoni N, Bartoli P,

Sutton R. New classification of haemodynamics of vasovagal syncope: beyond the

VASIS classification. Analysis of the pre-syncopal phase of the tilt test without and

with nitroglycerin challenge. Vasovagal Syncope International Study. Europace 2000;

2: 66-76.

[9] van Dijk N, Quartieri F, Blanc JJ, Garcia-Civera R, Brignole M, Moya A et al.

Effectiveness of physical counterpressure maneuvers in preventing vasovagal

syncope: the Physical Counterpressure Maneuvers Trial (PC-Trial). J Am Coll

Cardiol 2006;48:1652–7.

[10] Moya A, Sutton R, Ammirati F, Blanc J-J, Brignole M, Dahm JB, De Haro J-C,

Gajek J, Gjesdal K, Krahn A, Massin M, Pepi M, Pezawas T, Granell R, Sarasin F,

Ungar A, van Dijk J, Walma EP, Wieling W. Guidelines for the diagnosis and

treatment of syncope (version 2009). Eur Heart J 2009; 30: 2631-2671

[11] Sutton R, Petersen MEV. Invasive tilt testing: the search for a new sensor to

permit earlier pacing therapy in vasovagal syncope. In Cardiac Arrhythmias 1995 Ed.

A Raviele Springer-Verlag Italia Milano Italy 1996 pp132-133.

[12] Palmisano P, Dell’Era G, Russo V, Zaccaria M, Mangia R, Bortnik M, De Vecchi

F,Giubertoni A, Patti F, Magnani A, Nigro G, Rago A, Occhetta E, Accogli M.

19

Page 20 of 27

Effects of closed-loop stimulation vs. DDD pacing on haemodynamic variations and

occurrence of syncope induced by head-up tilt test in older patients with refractory

cardioinhibitory vasovagal syncope: the Tilt test-Induced REsponse in Closed-loop

Stimulation multicentre, prospective, single blind, randomized study. Europace 2017;

19: doi:10.1093/europace/eux015

[13] Sutton R, Petersen ME. First steps toward a pacing algorithm for vasovagal

syncope. Pacing Clin Electrophysiol 1997; 20: 827-828.

[14] Benditt DG, Sutton R, Gammage MD, Markowitz T, Gorski J, Nygaard GA,

Fetter J. Clinical experience with Thera DR rate-drop response pacing algorithm in

carotid sinus syndrome and vasovagal syncope. The International Rate-Drop

Investigators Group. Pacing Clin Electrophysiol 1997; 20: 832-839.

[15] Jardine DL, Wieling W, Brignole M, Lenders JWM, Sutton R, Stewart J. The

pathophysiology of the vasovagal response. (Part 2). Heart Rhythm 2018: DOI

10.1016/j.hrthm.2017.12.013

[16] Benditt DG, Detloff BL, Adkisson WO, Lu F, Sakaguchi S, Schussler S, Austin

E, Chen L. Age dependence of relative change in circulating epinephrine and

norepinephrine concentrations during tilt induced vasovagal syncope. Heart Rhythm

2012;9:1847-1852.

20

Page 21 of 27

Figure Legends

Figure 1. Tilt Test Analyzer.

Figure 2. Summary of VVS Prediction Efficacy Results for the 140 Subjects.

Figure 3. Prediction times: (a) histogram of prediction times, (b) distribution of

prediction times as a function of age.

Figure 4. Examples of VVS prediction: (a) true positive subject, (b) false positive

subject, (c) true negative subject, (d) false negative subject. Each panel shows the

following signals: RR intervals (RR), systolic blood pressure (SBP), heart rate and blood

pressure variability (HRV and BPV represented by the low frequency power LFRR and

LFSBP), and risk of VVS. The time of tilt and syncope (faint) are indicated as vertical bars.

The time, during which baseline computation is performed and no VVS risk is computed,

is indicated in grey. The amplitudes of LFRR and LFSBP have been scaled so that they can

be represented on the same graph. VVS alarms are represented by red asterisks on the

VVS risk signal.

21

Page 22 of 27

Figure 1

22

Page 23 of 27

Figure 2

23

Page 24 of 27

(a) (b)

Figure 3

24

Page 25 of 27

(a) (b)

(c) (d)Figure 4

25