Puponu

TeLa

Univ

International Journal of Nursing Studies 50 (2013) 1351–1358

A R

Artic

Rece

Rece

Acce

Keyw

Post

Puls

Nur

Nur

*

Unit

002

http

lse oximetry desaturation alarms on a generalstoperative adult unit: A prospective observational study ofrse response time

rri Voepel-Lewis *, Mary Lynn Parker, Constance N. Burke, Jennifer Hemberg,uren Perlin, Salim Kai, Satya Krishna Ramachandran

ersity of Michigan Health Systems, Ann Arbor, MI, United States

T I C L E I N F O

le history:

ived 4 June 2012

ived in revised form 16 November 2012

pted 8 February 2013

ords:

operative monitoring

e oximetry monitoring

se response

se surveillance

A B S T R A C T

Background: Continuous pulse oximetry monitoring is recommended to improve safety

during postoperative opioid use, however concerns with monitoring on general care units

remain, given potential system barriers to alarm transmission, recognition, and nursing

response.

Methods: This prospective, observational study evaluated unit and hospital-level factors

affecting nurses’ response to monitor desaturation alarms in postoperative patients on a

general postoperative unit. With exemption and waiver of consent granted from the

Institutional Review Board, monitoring data were downloaded from bedside monitors of

postoperative patients. Alarm notification data and response times were recorded from

the continuous capture of institutional surveillance data. Paging notifications were coded

as clinically relevant (i.e., true oxygen desaturation with SpO2< 89 for > 15 s) or irrelevant

(i.e., artifact, inappropriate alarm threshold, or failure to delay page). Linear mixed models,

and correlation coefficients were used to examine the relationships between unit staffing,

shift, paging burden and response time. Means and [95% confidence intervals] are

presented.

Results: 1616 monitoring hours in 103 patients yielded 342 desaturation events (duration

23.6 s [20.99, 26.1]) and 710 notification pages, 36% of which were for clinically relevant

desaturation. Nursing response time was 52.1 s [46.4, 57.7], which was longer at night

(63.8 [51.2, 76.35]; p = 0.035), but not related to unit staffing. Missed alarm events (i.e., no

notification page transmitted) occurred for 26% of the clinically relevant events, and were

associated with higher paging burden (p = 0.04), lower SpO2 values (81.8 [80.5, 83.0] vs.

83.2 [82.6, 83.8]; p = 0.026), and higher odds of intervention (OR 3.5 [1.38, 8.9]). 65% of

patients with desaturation events received interventions which correlated with the

number of pages (rho = 0.422; p < 0.01) and events (rho = 0.57; p < 0.01), desaturation

duration (rho = 0.505; p < 0.01), and SpO2 (rho = �0.324; p < 0.01).

Conclusions: One-third of pulse oximetry alarm notifications were for clinically relevant

oxygen desaturation, facilitating timely nursing response and intervention for most

patients. Unit staffing and false alarm frequency were not associated with response time,

suggesting a high level of attention on this unit. The nature and degree of missed alarm

events suggests patient safety concerns posed by hospital-level transmission systems

warranting further strategies to ensure monitoring safety.

� 2013 Elsevier Ltd. All rights reserved.

Corresponding author at: University of Michigan Health Systems, 4917 Mott Children’s Hospital 1540 E. Hospital Drive, Ann Arbor, MI 48109-4245,

ed States. Tel.: +1 734 936 0747; fax: +1 734 763 6651.

E-mail address: terriv@umich.edu (T. Voepel-Lewis).

Contents lists available at SciVerse ScienceDirect

International Journal of Nursing Studies

journal homepage: www.elsevier.com/ijns

0-7489/$ – see front matter � 2013 Elsevier Ltd. All rights reserved.

://dx.doi.org/10.1016/j.ijnurstu.2013.02.006

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–13581352

What is already known about this topic?Although previous studies have suggested that con-

tinuous pulse oximetry monitoring may reduce the needfor intensive care transfers, little is known about systemsfactors that affect safe monitoring practices for generalcare patients.

What the paper addsThis research presents original data showing that nurse

response to pulse oximetry alarms on a general care unitwas, on average, less than one minute, but longer on thenight shift. While unit workload was not related to nursingresponse time, a hospital level factor of paging burden wasassociated with missed alarm transmission, suggesting aneed for system strategies to ensure safe patient monitor-ing and response.

1. Introduction

It has been suggested that, despite technologicalshortfalls and a lack of evidence regarding safer patientoutcomes, the use of continuous pulse oximetry mon-itoring (POM) may reduce the risks associated withpostoperative opioid use as well as other high riskpatient situations (Weinger and Lee, 2011). However, thepotential for nurse desensitization due to high falsealarm rates, missed alarms or delayed response due toworkload, poor audibility, and delayed recognition ofrespiratory compromise may impede timely and effec-tive nursing response to patient deterioration (Edworthyand Hellier, 2005; Eichhorn, 2003; Imhoff et al., 2009;Phillips, 2006; Sobieraj et al., 2006). The quality of suchsurveillance in general care areas has therefore beenquestioned (Edworthy and Hellier, 2006, Eichhorn,1998), and alarm safety has been made a Joint Commis-sion Regulatory Agency patient safety goal with anemphasis on ensuring timely and effective nursingresponse (Clark, 2005).

1.1. Monitoring issues in non-specialist (i.e., general care)

areas

The primary concerns regarding non-specialty areapatient monitoring include the high potential for alarmfatigue due to high rates of false alarms, as well aschallenges related to notification and response of nurseswho may be in areas remote to the bedside alarm. Indeed,reports of a recent death on a general care unitdemonstrated a lack of nursing response to repeatedmonitor alarms at the central nurses’ station, suggestingeither a lack of an effective alarm notification system, or asthe author noted, a potential result of alarm fatigue (Bell,2010). Alarm fatigue resulting from excessive false orunnecessary alarms is a well-studied phenomenon incritical care, but only recently studied in non-specialtyareas (Gross et al., 2011). It has been suggested that asfalse alarm rates increase, response time decreases(Edworthy and Hellier, 2005; Phillips, 2006), and thatnuisance alarms disrupt patient care and lead to distrustand disabling of alarm devices potentially compromisingsafety (Imhoff and Kuhls, 2006; Korniewicz et al., 2008).

load of 42 (median 11.6) per hour on a general medical-surgical unit with an 80% patient occupancy, demonstrat-ing the potentially high alarm burden placed on generalcare nurses. These investigators further found high ratesof false alarms on general care, where only 34% of ‘‘criticalalarms’’ (i.e., pulse oximetry, apnea, etc.) were evaluatedas true. These investigators did not, however, address therelationship between false alarm rates and safety out-comes such as nursing response time, patient care oroutcomes.

Timely notification is an issue of great concern for non-specialty nurses given the unique burden of caring formultiple patients in various locations throughout the floor.Yet, this issue has been poorly addressed. Hallway alerts,paging interfaces, and tiered systems that provide a backupplan in the absence of initial response have beenrecommended as strategies to address nurse notificationand to facilitate prompt responses to alarms in non-criticalcare areas (ECRI, 2002; Phillips, 2006; Xiao and Seagull,2004). However, an evaluation of such systems has beenlimited. Recent studies have found that continuoussurveillance with a nurse notification component reducedtransfers to critical care (CC) (Taenzer et al., 2010) orimproved early identification of postoperative patientsrequiring CC admission (Ochroch et al., 2006). However,these studies did not evaluate factors or system shortfallsthat potentially impede nursing response, leaving sig-nificant gaps in our understanding of safe monitoringpractices.

This study evaluated nursing response and interven-tions prompted by a POM alarm surveillance system on ageneral postoperative care unit in a large, tertiary carecenter. The specific aims were to:

(1) Describe the nature of, and nurse response time to,POM desaturation alarms,

(2) Examine unit and hospital factors associated withlonger response times and missed alarms,

(3) Describe the association between response time,missed alarms, patient interventions and outcomes.

It was hypothesized that greater system burden (i.e.,census, nursing workload, paging burden) would beassociated with longer response times to and gaps intransmission of desaturation alarms.

2. Methods

2.1. Design, setting, and sampling

This prospective, observational study included the de-identified alarm and paging data from a conveniencesample of adults (>18 years of age) who were consecu-tively admitted to a single, 32 bed general care unitfollowing orthopedic surgery over a 3 month period in2009. Data from patients receiving opioids via patientcontrolled analgesia (PCA) with concomitant POM wereincluded, and demographics and perioperative informa-tion were recorded on admission. Patients who were notprescribed PCA with concomitant POM were excluded, aswere those who were assigned to other general care units.

Further, patients for whom POM data were not available Gross et al. (2011) recently observed an average alarm

(i.ewe

2.2.

wabetexpleas(95evethaperesti110pro(99

2.3.

incissusugwhleav

2.4.

conrecdurdiscto

tranbedbed

Tab

Des

Ag

W

Bo

Fe

AS

Pr

Du

Du

Data

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–1358 1353

., download problems or issues with bedside monitor)re also excluded.

Sample size determination

To ensure an adequate proportion of events, the samples determined a priori based on the assumption thatween 12% and 21% of adult patients on PCA woulderience oxygen desaturation (Overdyk et al., 2007). Att 87 subjects were needed to yield a proportion of 15%% � 7.5 CI) (Browner et al., 2007) with relevant alarmnts. Institutional paging data prior to the study showedt 11 alarm pages were generated from POM per patient day in a one month period on the study floor. Using thismate, a sample of 100 patients would yield approximately0 pages, deemed sufficient to detect an expectedportion of 50% relevant (i.e., oxygen desaturation) pages% � .05 CI) (Browner et al., 2007).

Resultant sample

One hundred twenty-five patients were screened forlusion; however 18 were excluded due to data retrievales at the POM level (e.g., lack of valid date/time stamps,gesting a swapping of monitors during the night) and 4ere paging reports or other relevant data were missing,ing 103 in the sample as described in Table 1.

Surveillance system

The tiered surveillance system was implemented injunction with the monitoring policy for patientseiving opioids via PCA that requires continuous POMing the first 24 h of treatment, or longer at theretion of the care provider. The system uses a monitor

hospital-paging interface whereby POM alarms aresmitted to paging via direct cable interface with theside nurse call system. The system is activated by theside monitor alarm following a decrease in oxygen

saturation (SpO2) below threshold, pre-defined in thissetting as <89%, or a pulse rate above or below threshold.These triggers initiate a bedside auditory alarm (internal),followed by a timed sequence of hallway call lights(visual), central nursing console (auditory and visual) andpaging notifications (primary and escalation paging alerts)as depicted in Fig. 1.

2.5. Procedures

The internal clocks of all pulse oximeters (Masimo Rad-8 Series, Irvine, CA) were synchronized to the nurse callsystem prior to the study start-up and daily throughout thestudy. Patients were placed on POM on arrival to the unitper routine practice. Immediately after discontinuation ofthe patient’s PCA or at 48 h (whichever came first), alarmsettings were recorded and all POM ASCII data (i.e., every2 s SpO2 and heart rate) were downloaded using the ProFoxSoftware (version PO Standard; Escondido, CA) with asummary report including; monitoring start, end, gaptimes, total sampling hours, lowest and mean SpO2,duration of time where SpO2< 89%, and the longestcontinuous desaturation (<89%).

2.6. Oxygen desaturation data and definitions

A thorough review of ASCII and summary data wasconducted to identify and code the number and duration ofdesaturation events (true SpO2< 89% lasting � 15 s thatshould trigger an alarm transmission), non-desaturationevents (artifact, disconnection, inappropriate alarmthreshold setting, or inadequate delay), and prolongedevents (lasting � 2 and � 3 min).

2.7. Nurse notification data

Paging notification data were extracted from institu-tional data that continuously and prospectively capturespaging information for each inpatient (Amcom Software,Eden Prarie, MN) including; the nature, date and time ofeach page and the time the page was canceled at thebedside (i.e., responded to). The number and duration ofalarm pages (i.e., initial and escalation) and response timeto each page (defined as time from paging notification tomonitor alarm cancelation) were recorded. All non-monitor-related pages were excluded as were 35 pagesrelated to insignificant heart rate changes unrelated to thestudy purpose.

2.8. Clinical event definitions

To examine the relationship between response timeand desaturation events, pages were matched to events bythe exact time of each occurrence. Events were then coded(based on the review of pulse oximeter ASCII data asdescribed above) according to the following definitions;clinically relevant notification defined as a nurse pageassociated with true SpO2< 89% for � 15 s, irrelevant

notifications associated with monitor artifact, disconnec-tion, inappropriate threshold on monitor (i.e., >89%),failures to delay page (i.e., <15 s duration of desaturation);

le 1

cription of the patients (n = 103).

e (years) 54.1 � 17.6 (19–92)

eight (kg) 84.7 � 24.2 (38–179)

dy mass index 29.5 � 7.6 (18.7–58.3)

male gender 59 (57%)

A physical status

I–II 64 (62%)

III–IV 39 (38%)

ocedure

Knee 25 (24%)

Hip 31 (30%)

Other extremity 26 (25%)

Spine 19 (18%)

Other 2 (2%)

ration of anesthesia (min) 218.5 � 78.7

ration of SpO2 monitoring (h) 15.7 � 6.5 (2–37.3)

Daytime monitoring (h) 7.1 � 4.4

Nighttime monitoring (h) 10.1 � 3.7

presented as mean � SD (range), or n (%).

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–13581354

and missed alarm events defined as desaturation eventsthat did not trigger transmission of a page. Interventions(e.g., supplemental oxygen, naloxone use) and outcomes(e.g., transfer to CC, need for rapid response team [RRT])were documented for each patient by the bedside nurse ondata collection sheets, and extracted from the medicalrecord for verification.

2.9. System data

To examine the association between the primaryoutcomes (i.e., oxygen desaturation duration, nurseresponse time, missed alarm events) and system (unitand hospital) factors, unit census (C) with patient transfersin and out (T), and number of nurses (RN) and nursing aides(NA) were recorded for each shift (Gabbay and Bukchin,2009; Minnick and Mion, 2009). Unit staffing factors werecalculated using RN hours per patient shift (RN-HPPS) andtotal (i.e., RNs and aides) hours per patient shift (totalHPPS) using the formulae; RN/(C*(C/T) and total/(C*(C/T),respectively. Hospital census and the total number ofsystem pages (i.e., paging burden) were recorded for eachday of the study to examine the potential relationship

between hospital level factors, response time and missedalarms.

3. Statistical analysis

Data were analyzed using PASW statistical software(formerly SPSS; version 17.0.3; Chicago, IL). Descriptivestatistics were used to analyze the outcomes of interestand data are presented as n (%), mean � standard deviationwith 95% confidence intervals (CI) where applicable. Boot-strapping was used to describe the mean values andconfidence intervals of skewed variables (R Statistical Soft-ware; R Project http://www.r-project.org/). Chi squareanalysis compared the rate of missed events between shifts,and unpaired t tests compared SpO2 and duration betweenthose with and without missed events. Spearman’s rhocorrelation coefficients were used to examine the relation-ships between outcomes, interventions, and systems factors,given the lack of normal distribution of data. Given thehierarchical nature of the collected data (i.e., patient, unit,hospital), linear mixed models were used to quantify therelationships between the outcome variables (response timeand duration of oxygen desaturation or response time) and

Fig. 1. Description of the pulse oximeter alarm tiered notification system in this setting. Notification includes audible oximeter alarm (at bedside), call light

activation (outside patient door), and nursing page (via hospital paging system).

varpagwhouthavnotto uEstiby

3.1.

Micthedat

4. R

4.1.

(15consignmohun20

expevethe

signHPPtimratishodes(e.grelameove33.res

4.2.

12

Tab

Tab

Des

tion

Av

Lo

Av

Ev

Data

app

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–1358 1355

ious predictors (shift, unit staffing, and total hospitales). Two-level random intercept models were employed,ere the day represented the random effect (cluster) andcomes were the units of analysis. Each day was allowed toe its own random intercept to partition out any variability

explained in the outcome variables, which could be duenderlying factors such as nurse or patient characteristics.mated marginal means (group means that are estimatedthe fitted models) and CIs are presented.

Ethics

The Institutional Review Board at the University ofhigan exempted this project and waived consent given

nature of de-identified monitoring, paging, and patienta.

esults

Oxygen desaturation data

A total of 1616 monitoring hours were recorded.7 � 6.5 h per patient). Ninety-one (88%) patients hadtinuous monitoring for the duration of PCA, whileificant gaps occurred in 12 (12%), 6 with <6 h of

nitoring that was sporadic during nighttime. Threedred forty-two desaturation events were identified, with

lasting � 2 min, and 11 � 3 min. Sixty-one patients (59%)erienced at least 1 event, 7 (7%) had 1 or more 2 minnts, and 5 (5%) 1 or more 3 min events. Table 2 describes

oxygen saturation data.Duration of desaturation events correlated weakly butificantly with RN-HPPS (rho = 0.176; p = 0.002), total-S (rho = 0.180; p = 0.001), and also with nurse response

e (rho = 0.280; p < 0.001). The REML-based likelihoodo test indicated that the random intercept for each dayuld be retained in the model describing duration ofaturation events, and none of the fixed effect factors., shift or unit staffing) showed a statistically significanttionship with this outcome. The estimated marginal

an durations of these events were 24.3 s [CI 18.9, 29.7],rall; 26.4 [18.2, 34.7] for the night shift and 24.8 [15.8,9] and 21.6 [14.2, 28.9] for the day and evening shifts,pectively.

Response time

Seven hundred ten notification pages were transmitted,(1.7%) of which were escalated (triggered after 3 min).le 3 describes the nature of pages and nursing response

times. There was a longer time to nurse response topatients with the highest numbers of notifications (i.e.,upper quartile for number of pages) compared to thosewith the lowest or medium number of pages (F 3.9 (df 2,708), p = 0.046). Nurse response time did not differbetween clinically relevant desaturation or non-desatura-tion alarm conditions (51.37 [40.9, 61.9] vs. 53.12 [46.6,59.7] s; p = 0.769), and did not correlate with eithermeasure of unit staffing (RN-HPPS rho = �0.01;p = 0.784; total-HPPS rho = �0.009; p = 0.804). The REML-based likelihood ratio test indicated that the randomintercept for each day should be retained in the modeldescribing response time. Shift was the only system factorhaving a significant contribution to response time (F 3.4;p = 0.035), and the estimated marginal response timemeans were; 47.9 s [39.3, 56.6] overall and 63.8 [51.2,76.35], 32.3 [�1.4, 65.99], and 47.7 [35.0, 60.44] for thenight, day, and evening shifts, respectively. Lastly, nurseresponse time correlated poorly, but significantly withtotal paging burden (rho = 0.111; p = 0.003), but not withhospital census.

4.3. Missed alarm events

Missed events occurred in 25 patients and are detailedin Table 3. Forty-three percent of these events clustered in3 patients, and 40% occurred in 4 days. Patients whoexperienced missed events had a higher overall frequencyof desaturation events compared to others (7.68 [4.4, 11.0]vs. 1.7 [1.0, 2.3]; <0.001). The number of missed eventswas not significantly different between the night (27%),day (33.3%), and evening shifts (22.2%; p > 0.100). SpO2

values were lower during missed events compared to thosetriggering a page (81.8 [80.5, 83.0] vs. 83.2 [82.6, 83.8],respectively; p = 0.026), however, the difference in dura-tion was not significant (53.5 [37.3, 69.8] vs. 41.9 [34.8,48.99] s, respectively, p = .135). Paging burden (i.e., num-ber of institutional pages per day) was greater whenmissed events occurred compared to when they did not(69.3 K [67.4 K, 71.1 K] vs. 67.5 K [66.6 K, 68.3 K];p = 0.043), suggesting an impact of a hospital level systemfactor.

4.4. Interventions

Forty of the 61 patients with desaturation events (66%)received an intervention, including increased or addition ofsupplemental O2 in the majority (Table 3). Use ofinterventions correlated with the number of per patientpages (rho = 0.422), desaturation events (rho = 0.570),missed events (rho = 0.283), duration of oxygen desatura-tion (rho = 0.505), and lowest SpO2 (rho = �0.324; allcomparisons significant at p < 0.001). Patients with missedevents were more likely to receive an interventioncompared to those without (65% vs. 27%, p = 0.007; oddsratio 3.5 [1.38, 8.90]).

Sixty-five percent of patients received supplemental O2

on admission to general care. Patients with and withoutsupplemental O2 experienced no differences in the totalduration of oxygen desaturation (0.38 [0.26, 0.49] vs. 0.38[0.22, 0.54] min/h monitoring, p = 0.959) or nursing

le 2

cription of oxygen saturation (SpO2) and clinically relevant desatura-

events (n = 342).

erage SpO2 (%) 95.8 [95.4, 96.2]

west SpO2 (%) during

desaturation events

82.8 [82.3, 83.4]

erage desaturation

event duration (s)

44.9 [38.2, 51.7]

ents per patient (n) 4 [2.5, 5.6]; 1 (0–53)

presented as mean [95% confidence interval]; median (range), as

ropriate.

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–13581356

response time (50.1 [35.4, 64.9] vs. 37.7 [28.7, 46.6] s,p = 0.235). All patients who experienced oxygen desatura-tion lasting 2 or 3 min were receiving O2 on admission, and71% and 100%, respectively required an intervention. Nopatient in the study required escalation of care or a rapidresponse team call.

5. Discussion

This study describes factors associated with nursingresponse to POM alarms in postoperative patients receiv-ing opioids on a general nursing care unit. Thirty-sixpercent of alarm notifications reflected clinically relevantoxygen desaturation prompting interventions for two-thirds of patients. Nurses’ response time was not differentbetween clinically relevant or irrelevant alarms and wasnot associated with unit staffing factors, but was longerduring the night shift. Importantly, one quarter ofdesaturation events did not trigger transmission of anotification page (i.e., were missed events), and wereassociated with higher hospital paging burden, suggestinga system level fault. Further, patients with these missedevents had lower oxygen saturation levels, a higherfrequency of events, and higher probability of receivingan intervention suggesting potential compromised safetythat may be attributed to ineffective nurse notification.

5.1. Alarm fatigue

Investigators recently described an average of 95.6

unit, suggesting a high alarm burden for non-specialtyareas (Gross et al., 2011). However, they did not describethe impact on nurse response time or patient safety.Although we did not directly assess nurses’ perceivedalarm burden, the noted gaps in monitoring for somepatients during the night shift suggests that perceivednuisance may have led to disabling the POM devices.Furthermore, while the overall nurse response to alarmswas less than one minute, there was a longer time toresponse for patients in the high alarm group comparedthose with a low or medium number of alarms, suggesting,perhaps, some impact of alarm fatigue. Importantly, therewas not a difference in response between clinicallyrelevant and irrelevant desaturation events, and two-thirds of patients with relevant events received anintervention, nearly all with prolonged events did so,and none required a rescue intervention, suggesting thatnurse responses were adequate to prevent patientdeterioration. While studies suggest that nursing workloadmay hinder response to alarms (ECRI, 2002; Korniewiczet al., 2008), the present study found no relationshipbetween staffing and response. However, overall staffing inthis setting was high, so it is important that this finding beinterpreted carefully.

5.2. Notification issues

Notification pages were not transmitted for 26% ofdesaturation events, attributable to either alarm silencingat the bedside or a potential system failure in alarm

Table 3

Description of alarm notification pages (n = 710).

Number of pages per patient 7 [4.6, 9.4]; 3 (0–80)

Duration of pages per patient (min) 6 [3.3, 8.7]; 2 (0–115)

Nursing response time (s) 52.1 [46.4, 57.7]; 32 (18–1221)

Response to patients with low frequency of alarmsa 45.4 [33.7, 56.1]

Response to those with medium frequency 42.3 [36.1, 48.6]

Response to those with high frequency 57.6 [49.2, 65.9]a

Description of alarm notification episodes

Clinically relevant notification (i.e., true desaturation event) 253 (36%)

Irrelevant notification

Monitor artifact 329 (46%)

Inappropriate alarm limit (i.e., SpO2 limit 91% or pulse

limit high for patient’s baseline)

76 (11%)

Desaturation event <15 s (i.e., failure to delay page) 42 (6%)

Unable to determine 10 (1%)

Missed alarm eventb 89 (26%)

Missed 2 min eventb 8 (40%)

Missed 3 min eventb 3 (27%)

Duration of missed events (s) 53.5 [37.3, 69.8]; 30 (24–678)

Clinical interventions (in 40 patients)

Addition or escalation of supplemental oxygen 27 (68%)

Stimulate patient 9 (23%)

Continuous positive airway pressure (all but one case were

patients using a device pre-operatively)

8 (20%)

Other interventionsc 10 (25%)

Data presented as mean � SD [confidence interval], median (range); or n (%).a Patients with notifications split into quartiles (below 25th percentile, median percentiles, and above 75th percentile for number of notifications);

patients in the highest quartile had a significantly longer time to nurse response compared to those in the other groups (p = 0.045).b Percentages calculated from the number of true desaturation events, 2-min events, and 3-min events, respectively.c Other = encourage patient to cough and deep breathe, use incentive spirometry, reminded patient to leave oxygen in place.

transmission through the communication system. The

alarms per patient per day on a general medical surgical

freqof rmisdureveburmisfailmofindsugcenresin

relitrancen

hasirre201andof wof enot(No(Spset201thrsinsetinteeffehalartiof eaffepotset

5.3.

resunimaHowposthelimor

posassandinteproactno

resthis

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–1358 1357

uency of events was higher, SpO2 lower, and the oddseceiving an intervention were greater in patients withsed events, suggesting greater clinical deteriorationing non-notification times. Nearly half of missednts clustered in a few patients and days, and pagingden was significantly greater on the days whensed events occurred, suggesting sporadic systemures due to hospital level burden and the lack ofnitor alarm prioritization at the system level. Theing that no patient required a rescue interventiongests that the system safety net of hallway andtral console alerts may have facilitated a secondaryponse, despite longer duration of oxygen desaturationthis group. However, the potential for nurses’ over-ance on paging may create a gap in safety duringsmission failure, paging downtime or delays within

tralized systems.Use of longer filtering epochs and lower alarm limits

been suggested as a strategy to reduce the number oflevant alarms (Chambrin, 2001; Graham and Cvach,0; Gross et al., 2011). Establishing appropriate delays

thresholds, however, demands a tradeoff between useide settings that minimize false alarms at the expensearly notification, vs. a narrow range to provide earlyification at the expense of considerable false positivesrris and Dawant, 2001). A lower thresholdO2 < 80%) with a 30 s delay was employed in oneting to minimize irrelevant events (Taenzer et al.,0), however, it was difficult to discern whether such

esholds delayed nurses’ recognition of deterioration,ce there was no difference in patient mortality in thatting. A higher threshold (i.e., 89%) and shorter delayrval (15 s) were employed in the present system,ctively reducing the number of pages by more than

f over no delay, but leaving a considerable number offact-induced alarms. A 30 s delay reduced the numbervents by more than half again in a post hoc analysis,cting relevant and irrelevant alarms equally. Theential effect of lowering the SpO2 threshold in ourting remains unknown.

Strengths and limitations

This prospective study is the first to examine nursingponse time to POM alarms in a non-specialty nursingt. The study found important notification issues thaty impact timely response to patient deterioration.

ever, given the sampling technique which includedtoperative patients from a single unit at one institution,

possibility of selection bias cannot be overlooked,iting the ability to generalize findings to other settingspopulations. The potential for observer bias, whilesible, was minimized by use of two well-trainedistants (J.H. and L.P.) who downloaded and coded data,

cross-checked the work of each other. Lastly, the use ofrventions was not prescribed or regulated by a studytocol, thereby limiting our understanding of theirual necessity or relevance. However, the findings thatpatient experienced escalation of care or failure to

cue infers appropriate recognition and intervention in

6. Clinical implications

This study has several implications for nursingpractice. Importantly, despite a potentially high numberof clinically irrelevant or false alarms from POM on non-specialty care units, timely nursing response to all eventsis imperative for prompt intervention and to preventdeterioration. Nurses and administrators should examinestrategies to reduce the burden of alarms while ensuringsafety, such as adjusting alarm thresholds or addingdelays to notification. Further, implementation of tieredsystems, such as that described here, is imperative toensure timely response to alarm events on units where theassigned nurse may be involved with another patient.Within such systems, it is important that nurses’ attentionbe directed to all tiers (i.e., hall lights and central console,as well as paging) to enhance effective recognition ofpatient compromise.

Acknowledgement

The authors would like to thank Giselle Kolenic, M.A.,Statistical Consultant at The University of Michigan, for herassistance with the analyses.

Conflict of interest: The authors have no conflicts toreport.

Funding: The authors received no external funding forthis work.

Ethical approval: This study was conducted withapproval from the Institutional Review Board at theUniversity of Michigan, with waiver of consent.

References

Bell, L., 2010. Alarm fatigue linked to patient’s death. Interview by LauraWallis. American Journal of Nursing 110 (7), 16.

Browner, W., Newman, T., Hulley, S., 2007. Estimating sample size andpower: applications and examples. In: Hulley, S., Cummings, S., Br-owner, W., Grady, D., Newman, T. (Eds.), Designing Clinical Research.3rd ed. Lippincott, Williams & Wilkins, Philadelphia, pp. 65–94.

Chambrin, M.C., 2001. Alarms in the intensive care unit: how can thenumber of false alarms be reduced? Critical Care 5 (4), 184–188.

Clark, T., 2005. Clinical alarms remain an area for patient safetyimprovement. Biomedical Instrumentation & Technology 39 (5),357–359.

ECRI, 2002. Critical alarms and patient safety. ECRI’s guide to developingeffective alarm strategies and responding to JCAHO’s alarm-safetygoal. Health Devices 31 (11), 397–417.

Edworthy, J., Hellier, E., 2005. Fewer but better auditory alarms willimprove patient safety. Quality and Safety in Health Care 14 (3),212–215.

Edworthy, J., Hellier, E., 2006. Alarms and human behaviour: implicationsfor medical alarms. British Journal of Anaesthesia 97 (1), 12–17.

Eichhorn, J.H., 1998. Pulse oximetry monitoring and late postoperativehypoxemia on the general care floor. Journal of Clinical Monitoringand Computing 14 (1), 49–55.

Eichhorn, J.H., 2003. Recognizing and preventing hypoxemic injury risk onthe general care floor. Journal of Healthcare Risk Management 23 (1),17–22.

Gabbay, U., Bukchin, M., 2009. Does daily nurse staffing match wardworkload variability? Three hospitals’ experiences. InternationalJournal of Health Care Quality Assurance 22 (6), 625–641.

Graham, K.C., Cvach, M., 2010. Monitor alarm fatigue: standardizing useof physiological monitors and decreasing nuisance alarms. AmericanJournal of Critical Care 19 (1), 28–34.

Gross, B., Dahl, D., Nielsen, L., 2011. Physiologic monitoring alarm load on

medical/surgical floors of a community hospital. Biomedical Instru-mentation and Technology Supplement 29–36. small sample.

T. Voepel-Lewis et al. / International Journal of Nursing Studies 50 (2013) 1351–13581358

Imhoff, M., Kuhls, S., 2006. Alarm algorithms in critical care monitoring.Anesthesia and Analgesia 102 (5), 1525–1537.

Imhoff, M., Kuhls, S., Gather, U., Fried, R., 2009. Smart alarms from medicaldevices in the OR and ICU. Best Practice and Research. ClinicalAnaesthesiology 23 (1), 39–50.

Korniewicz, D.M., Clark, T., David, Y., 2008. A national online survey on theeffectiveness of clinical alarms. American Journal of Critical Care 17(1), 36–41.

Minnick, A.F., Mion, L.C., 2009. Nurse labor data: the collection andinterpretation of nurse-to-patient ratios. Journal of Nursing Admin-istration 39 (9), 377–381.

Norris, P., Dawant, B., 2001. Closing the loop in ICU decision support:physiologic event detection, alerts, and documentation. In:Paper Presented at the American Medical Informatics AssociationSymposium.

Ochroch, E.A., Russell, M.W., Hanson 3rd, W.C., Devine, G.A., Cucchiara,A.J., Weiner, M.G., Schwartz, S.J., 2006. The impact of continuouspulse oximetry monitoring on intensive care unit admissionsfrom a postsurgical care floor. Anesthesia and Analgesia 102 (3),868–875.

Overdyk, F.J., Carter, R., Maddox, R.R., Callura, J., Herrin, A.E., Henriquez, C.,2007. Continuous oximetry/capnometry monitoring reveals frequentdesaturation and bradypnea during patient-controlled analgesia.Anesthesia and Analgesia 105 (2), 412–418.

Phillips, J., 2006. Clinical alarms: complexity and common sense. CriticalCare Nursing Clinics of North America 18 (2), 145–156, ix.

Sobieraj, J., Ortega, C., West, I., Voepel, L., Battle, S., Robinson, D., 2006.Audibility of patient clinical alarms to hospital nursing personnel.Military Medicine 171 (4), 306–310.

Taenzer, A.H., Pyke, J.B., McGrath, S.P., Blike, G.T., 2010. Impact of pulseoximetry surveillance on rescue events and intensive care unit trans-fers: a before-and-after concurrence study. Anesthesiology 112 (2),282–287.

Weinger, M.G., Lee, L.A., 2011. No patient shall be harmed by opioid-induced respiratory depression. APSF Newsletter: Official Journal ofthe Anesthesia Patient Safety Foundation 26 (2), 21–28.

Xiao, Y., Seagull, J., 2004. Organizational-historical analysis of the ‘‘failureto respond to alarm’’ problems. IEEE Transactions on Systems Man,and Cybernetics Part A: Systems and Humans 34 (6), 772–778.