Miriam Zander, BA

Eligibility Criteria

With help from an experienced biomedical librarian (C.D.S.), we searched PubMed, the Cumulative Index to Nursing and Allied Health Literature, Scopus, Cochrane Library, ClinicalTrials.gov, and Google Scholar from January 1980 through April 2015 (see Supporting Information in the online version of this article for the search terms and queries). We hand searched the reference lists of included articles and reviewed our personal libraries to identify additional relevant studies.

We included peer‐reviewed, original research studies published in English, Spanish, or French that addressed the questions outlined above. Eligible patient populations were children and adults admitted to hospital inpatient units and emergency departments (EDs). We excluded alarms in procedural suites or operating rooms (typically responded to by anesthesiologists already with the patient) because of the differences in environment of care, staff‐to‐patient ratio, and equipment. We included observational studies reporting the actionability of physiologic monitor alarms (ie, alarms warranting special attention or clinical intervention), as well as nurse responses to these alarms. We excluded studies focused on the effects of alarms unrelated to patient safety, such as families' and patients' stress, noise, or sleep disturbance. We included only intervention studies evaluating pragmatic interventions ready for clinical implementation (ie, not experimental devices or software algorithms).

Selection Process and Data Extraction

First, 2 authors screened the titles and abstracts of articles for eligibility. To maximize sensitivity, if at least 1 author considered the article relevant, the article proceeded to full‐text review. Second, the full texts of articles screened were independently reviewed by 2 authors in an unblinded fashion to determine their eligibility. Any disagreements concerning eligibility were resolved by team consensus. To assure consistency in eligibility determinations across the team, a core group of the authors (C.W.P, C.P.B., E.E., and V.V.G.) held a series of meetings to review and discuss each potentially eligible article and reach consensus on the final list of included articles. Two authors independently extracted the following characteristics from included studies: alarm review methods, analytic design, fidelity measurement, consideration of unintended adverse safety consequences, and key results. Reviewers were not blinded to journal, authors, or affiliations.

Synthesis of Results and Risk Assessment

Given the high degree of heterogeneity in methodology, we were unable to generate summary proportions of the observational studies or perform a meta‐analysis of the intervention studies. Thus, we organized the studies into clinically relevant categories and presented key aspects in tables. Due to the heterogeneity of the studies and the controversy surrounding quality scores,[5] we did not generate summary scores of study quality. Instead, we evaluated and reported key design elements that had the potential to bias the results. To recognize the more comprehensive studies in the field, we developed by consensus a set of characteristics that distinguished studies with lower risk of bias. These characteristics are shown and defined in Table 1.

For the purposes of this review, we defined nonactionable alarms as including both invalid (false) alarms that do not that accurately represent the physiologic status of the patient and alarms that are valid but do not warrant special attention or clinical intervention (nuisance alarms). We did not separate out invalid alarms due to the tremendous variation between studies in how validity was measured.

Study Selection

Search results produced 4629 articles (see the flow diagram in the Supporting Information in the online version of this article), of which 32 articles were eligible: 24 observational studies describing alarm characteristics and 8 studies describing interventions to reduce alarm frequency.

Observational Study Characteristics

Characteristics of included studies are shown in Table 1. Of the 24 observational studies,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30] 15 included adult patients,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21] 7 included pediatric patients,[22, 23, 24, 25, 26, 27, 28] and 2 included both adult and pediatric patients.[29, 30] All were single‐hospital studies, except for 1 study by Chambrin and colleagues[10] that included 5 sites. The number of patient‐hours examined in each study ranged from 60 to 113,880.[7, 8, 9, 10, 11, 13, 14, 15, 16, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 29, 30] Hospital settings included ICUs (n = 16),[9, 10, 11, 13, 14, 16, 17, 18, 19, 22, 23, 24, 25, 26, 27, 29] general wards (n = 5),[12, 15, 20, 22, 28] EDs (n = 2),[7, 21] postanesthesia care unit (PACU) (n = 1),[30] and cardiac care unit (CCU) (n = 1).[8] Studies varied in the type of physiologic signals recorded and data collection methods, ranging from direct observation by a nurse who was simultaneously caring for patients[29] to video recording with expert review.[14, 19, 22] Four observational studies met the criteria for lower risk of bias.[11, 14, 15, 22]

Intervention Study Characteristics

Of the 8 intervention studies, 7 included adult patients,[31, 32, 33, 34, 35, 36, 37] and 1 included pediatric patients.[38] All were single‐hospital studies; 6 were quasi‐experimental[31, 33, 34, 35, 37, 38] and 2 were experimental.[32, 36] Settings included progressive care units (n = 3),[33, 34, 35] CCUs (n = 3),[32, 33, 37] wards (n = 2),[31, 38] PACU (n = 1),[36] and a step‐down unit (n = 1).[32] All except 1 study[32] used the monitoring system to record alarm data. Several studies evaluated multicomponent interventions that included combinations of the following: widening alarm parameters,[31, 35, 36, 37, 38] instituting alarm delays,[31, 34, 36, 38] reconfiguring alarm acuity,[35, 37] use of secondary notifications,[34] daily change of electrocardiographic electrodes or use of disposable electrocardiographic wires,[32, 33, 38] universal monitoring in high‐risk populations,[31] and timely discontinuation of monitoring in low‐risk populations.[38] Four intervention studies met our prespecified lower risk of bias criteria.[31, 32, 36, 38]

Proportion of Alarms Considered Actionable

Results of the observational studies are provided in Table 2. The proportion of alarms that were actionable was <1% to 26% in adult ICU settings,[9, 10, 11, 13, 14, 16, 17, 19] 20% to 36% in adult ward settings,[12, 15, 20] 17% in a mixed adult and pediatric PACU setting,[30] 3% to 13% in pediatric ICU settings,[22, 23, 24, 25, 26] and 1% in a pediatric ward setting.[22]

Relationship Between Alarm Exposure and Response Time

Whereas 9 studies addressed response time,[8, 12, 17, 18, 20, 21, 22, 27, 28] only 2 evaluated the relationship between alarm burden and nurse response time.[20, 22] Voepel‐Lewis and colleagues found that nurse responses were slower to patients with the highest quartile of alarms (57.6 seconds) compared to those with the lowest (45.4 seconds) or medium (42.3 seconds) quartiles of alarms on an adult ward (P = 0.046). They did not find an association between false alarm exposure and response time.[20] Bonafide and colleagues found incremental increases in response time as the number of nonactionable alarms in the preceding 120 minutes increased (P < 0.001 in the pediatric ICU, P = 0.009 on the pediatric ward).[22]

Interventions Effective in Reducing Alarms

Results of the 8 intervention studies are provided in Table 3. Three studies evaluated single interventions;[32, 33, 36] the remainder of the studies tested interventions with multiple components such that it was impossible to separate the effect of each component. Below, we have summarized study results, arranged by component. Because only 1 study focused on pediatric patients,[38] results from pediatric and adult settings are combined.

Widening alarm parameter default settings was evaluated in 5 studies:[31, 35, 36, 37, 38] 1 single intervention randomized controlled trial (RCT),[36] and 4 multiple‐intervention, quasi‐experimental studies.[31, 35, 37, 38] In the RCT, using a lower SpO₂ limit of 85% instead of the standard 90% resulted in 61% fewer alarms. In the 4 multiple intervention studies, 1 study reported significant reductions in alarm rates (P < 0.001),[37] 1 study did not report preintervention alarm rates but reported a postintervention alarm rate of 4 alarms per patient‐day,[31] and 2 studies reported reductions in alarm rates but did not report any statistical testing.[35, 38] Of the 3 studies examining patient safety, 1 study with universal monitoring reported fewer rescue events and transfers to the ICU postimplementation,[31] 1 study reported no missed acute decompensations,[38] and 1 study (the RCT) reported significantly more true hypoxemia events (P = 0.001).[36]

Alarm delays were evaluated in 4 studies:[31, 34, 36, 38] 3 multiple‐intervention, quasi‐experimental studies[31, 34, 38] and 1 retrospective analysis of data from an RCT.[36] One study combined alarm delays with widening defaults in a universal monitoring strategy and reported a postintervention alarm rate of 4 alarms per patient.[31] Another study evaluated delays as part of a secondary notification pager system and found a negatively sloping regression line that suggested a decreasing alarm rate, but did not report statistical testing.[34] The third study reported a reduction in alarm rates but did not report statistical testing.[38] The RCT compared the impact of a hypothetical 15‐second alarm delay to that of a lower SpO₂ limit reduction and reported a similar reduction in alarms.[36] Of the 4 studies examining patient safety, 1 study with universal monitoring reported improvements,[31] 2 studies reported no adverse outcomes,[35, 38] and the retrospective analysis of data from the RCT reported the theoretical adverse outcome of delayed detection of sudden, severe desaturations.[36]

Reconfiguring alarm acuity was evaluated in 2 studies, both of which were multiple‐intervention quasi‐experimental studies.[35, 37] Both showed reductions in alarm rates: 1 was significant without increasing adverse events (P < 0.001),[37] and the other did not report statistical testing or safety outcomes.[35]

Secondary notification of nurses using pagers was the main intervention component of 1 study incorporating delays between the alarms and the alarm pages.[34] As mentioned above, a negatively sloping regression line was displayed, but no statistical testing or safety outcomes were reported.

Disposable electrocardiographic lead wires or daily electrode changes were evaluated in 3 studies:[32, 33, 38] 1 single intervention cluster‐randomized trial[32] and 2 quasi‐experimental studies.[33, 38] In the cluster‐randomized trial, disposable lead wires were compared to reusable lead wires, with disposable lead wires having significantly fewer technical alarms for lead signal failures (P = 0.03) but a similar number of monitoring artifact alarms (P = 0.44).[32] In a single‐intervention, quasi‐experimental study, daily electrode change showed a reduction in alarms, but no statistical testing was reported.[33] One multiple‐intervention, quasi‐experimental study incorporating daily electrode change showed fewer alarms without statistical testing.[38] Of the 2 studies examining patient safety, both reported no adverse outcomes.[32, 38]

Limitations of This Review and the Underlying Body of Work

There are limitations to this systematic review and its underlying body of work. With respect to our approach to this systematic review, we focused only on monitor alarms. Numerous other medical devices generate alarms in the patient‐care environment that also can contribute to alarm fatigue and deserve equally rigorous evaluation. With respect to the underlying body of work, the quality of individual studies was generally low. For example, determinations of alarm actionability were often made by a single rater without evaluation of the reliability or validity of these determinations, and statistical testing was often missing. There were also limitations specific to intervention studies, including evaluation of nongeneralizable patient populations, failure to measure the fidelity of the interventions, inadequate measures of intervention safety, and failure to statistically evaluate alarm reductions. Finally, though not necessarily a limitation, several studies were conducted by authors involved in or funded by the medical device industry.[11, 15, 19, 31, 32] This has the potential to introduce bias, although we have no indication that the quality of the science was adversely impacted.

Moving forward, the research agenda for physiologic monitor alarms should include the following: (1) more intensive focus on evaluating the relationship between alarm exposure and response time with analysis of important mediating factors that may promote or prevent alarm fatigue, (2) emphasis on studying interventions aimed at improving alarm management using rigorous designs such as cluster‐randomized trials and trials randomized by individual participant, (3) monitoring and reporting clinically meaningful balancing measures that represent unintended consequences of disabling or delaying potentially important alarms and possibly reducing the clinicians' ability to detect true patient deterioration and intervene in a timely manner, and (4) support for transparent academicindustry partnerships to evaluate new alarm technology in real‐world settings. As evidence‐based interventions emerge, there will be new opportunities to study different implementation strategies of these interventions to optimize effectiveness.

Eligibility Criteria

With help from an experienced biomedical librarian (C.D.S.), we searched PubMed, the Cumulative Index to Nursing and Allied Health Literature, Scopus, Cochrane Library, ClinicalTrials.gov, and Google Scholar from January 1980 through April 2015 (see Supporting Information in the online version of this article for the search terms and queries). We hand searched the reference lists of included articles and reviewed our personal libraries to identify additional relevant studies.

We included peer‐reviewed, original research studies published in English, Spanish, or French that addressed the questions outlined above. Eligible patient populations were children and adults admitted to hospital inpatient units and emergency departments (EDs). We excluded alarms in procedural suites or operating rooms (typically responded to by anesthesiologists already with the patient) because of the differences in environment of care, staff‐to‐patient ratio, and equipment. We included observational studies reporting the actionability of physiologic monitor alarms (ie, alarms warranting special attention or clinical intervention), as well as nurse responses to these alarms. We excluded studies focused on the effects of alarms unrelated to patient safety, such as families' and patients' stress, noise, or sleep disturbance. We included only intervention studies evaluating pragmatic interventions ready for clinical implementation (ie, not experimental devices or software algorithms).

Selection Process and Data Extraction

First, 2 authors screened the titles and abstracts of articles for eligibility. To maximize sensitivity, if at least 1 author considered the article relevant, the article proceeded to full‐text review. Second, the full texts of articles screened were independently reviewed by 2 authors in an unblinded fashion to determine their eligibility. Any disagreements concerning eligibility were resolved by team consensus. To assure consistency in eligibility determinations across the team, a core group of the authors (C.W.P, C.P.B., E.E., and V.V.G.) held a series of meetings to review and discuss each potentially eligible article and reach consensus on the final list of included articles. Two authors independently extracted the following characteristics from included studies: alarm review methods, analytic design, fidelity measurement, consideration of unintended adverse safety consequences, and key results. Reviewers were not blinded to journal, authors, or affiliations.

Synthesis of Results and Risk Assessment

Given the high degree of heterogeneity in methodology, we were unable to generate summary proportions of the observational studies or perform a meta‐analysis of the intervention studies. Thus, we organized the studies into clinically relevant categories and presented key aspects in tables. Due to the heterogeneity of the studies and the controversy surrounding quality scores,[5] we did not generate summary scores of study quality. Instead, we evaluated and reported key design elements that had the potential to bias the results. To recognize the more comprehensive studies in the field, we developed by consensus a set of characteristics that distinguished studies with lower risk of bias. These characteristics are shown and defined in Table 1.

For the purposes of this review, we defined nonactionable alarms as including both invalid (false) alarms that do not that accurately represent the physiologic status of the patient and alarms that are valid but do not warrant special attention or clinical intervention (nuisance alarms). We did not separate out invalid alarms due to the tremendous variation between studies in how validity was measured.

Study Selection

Search results produced 4629 articles (see the flow diagram in the Supporting Information in the online version of this article), of which 32 articles were eligible: 24 observational studies describing alarm characteristics and 8 studies describing interventions to reduce alarm frequency.

Observational Study Characteristics

Characteristics of included studies are shown in Table 1. Of the 24 observational studies,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30] 15 included adult patients,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21] 7 included pediatric patients,[22, 23, 24, 25, 26, 27, 28] and 2 included both adult and pediatric patients.[29, 30] All were single‐hospital studies, except for 1 study by Chambrin and colleagues[10] that included 5 sites. The number of patient‐hours examined in each study ranged from 60 to 113,880.[7, 8, 9, 10, 11, 13, 14, 15, 16, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 29, 30] Hospital settings included ICUs (n = 16),[9, 10, 11, 13, 14, 16, 17, 18, 19, 22, 23, 24, 25, 26, 27, 29] general wards (n = 5),[12, 15, 20, 22, 28] EDs (n = 2),[7, 21] postanesthesia care unit (PACU) (n = 1),[30] and cardiac care unit (CCU) (n = 1).[8] Studies varied in the type of physiologic signals recorded and data collection methods, ranging from direct observation by a nurse who was simultaneously caring for patients[29] to video recording with expert review.[14, 19, 22] Four observational studies met the criteria for lower risk of bias.[11, 14, 15, 22]

Intervention Study Characteristics

Of the 8 intervention studies, 7 included adult patients,[31, 32, 33, 34, 35, 36, 37] and 1 included pediatric patients.[38] All were single‐hospital studies; 6 were quasi‐experimental[31, 33, 34, 35, 37, 38] and 2 were experimental.[32, 36] Settings included progressive care units (n = 3),[33, 34, 35] CCUs (n = 3),[32, 33, 37] wards (n = 2),[31, 38] PACU (n = 1),[36] and a step‐down unit (n = 1).[32] All except 1 study[32] used the monitoring system to record alarm data. Several studies evaluated multicomponent interventions that included combinations of the following: widening alarm parameters,[31, 35, 36, 37, 38] instituting alarm delays,[31, 34, 36, 38] reconfiguring alarm acuity,[35, 37] use of secondary notifications,[34] daily change of electrocardiographic electrodes or use of disposable electrocardiographic wires,[32, 33, 38] universal monitoring in high‐risk populations,[31] and timely discontinuation of monitoring in low‐risk populations.[38] Four intervention studies met our prespecified lower risk of bias criteria.[31, 32, 36, 38]

Proportion of Alarms Considered Actionable

Results of the observational studies are provided in Table 2. The proportion of alarms that were actionable was <1% to 26% in adult ICU settings,[9, 10, 11, 13, 14, 16, 17, 19] 20% to 36% in adult ward settings,[12, 15, 20] 17% in a mixed adult and pediatric PACU setting,[30] 3% to 13% in pediatric ICU settings,[22, 23, 24, 25, 26] and 1% in a pediatric ward setting.[22]

Relationship Between Alarm Exposure and Response Time

Whereas 9 studies addressed response time,[8, 12, 17, 18, 20, 21, 22, 27, 28] only 2 evaluated the relationship between alarm burden and nurse response time.[20, 22] Voepel‐Lewis and colleagues found that nurse responses were slower to patients with the highest quartile of alarms (57.6 seconds) compared to those with the lowest (45.4 seconds) or medium (42.3 seconds) quartiles of alarms on an adult ward (P = 0.046). They did not find an association between false alarm exposure and response time.[20] Bonafide and colleagues found incremental increases in response time as the number of nonactionable alarms in the preceding 120 minutes increased (P < 0.001 in the pediatric ICU, P = 0.009 on the pediatric ward).[22]

Interventions Effective in Reducing Alarms

Results of the 8 intervention studies are provided in Table 3. Three studies evaluated single interventions;[32, 33, 36] the remainder of the studies tested interventions with multiple components such that it was impossible to separate the effect of each component. Below, we have summarized study results, arranged by component. Because only 1 study focused on pediatric patients,[38] results from pediatric and adult settings are combined.

Widening alarm parameter default settings was evaluated in 5 studies:[31, 35, 36, 37, 38] 1 single intervention randomized controlled trial (RCT),[36] and 4 multiple‐intervention, quasi‐experimental studies.[31, 35, 37, 38] In the RCT, using a lower SpO₂ limit of 85% instead of the standard 90% resulted in 61% fewer alarms. In the 4 multiple intervention studies, 1 study reported significant reductions in alarm rates (P < 0.001),[37] 1 study did not report preintervention alarm rates but reported a postintervention alarm rate of 4 alarms per patient‐day,[31] and 2 studies reported reductions in alarm rates but did not report any statistical testing.[35, 38] Of the 3 studies examining patient safety, 1 study with universal monitoring reported fewer rescue events and transfers to the ICU postimplementation,[31] 1 study reported no missed acute decompensations,[38] and 1 study (the RCT) reported significantly more true hypoxemia events (P = 0.001).[36]

Alarm delays were evaluated in 4 studies:[31, 34, 36, 38] 3 multiple‐intervention, quasi‐experimental studies[31, 34, 38] and 1 retrospective analysis of data from an RCT.[36] One study combined alarm delays with widening defaults in a universal monitoring strategy and reported a postintervention alarm rate of 4 alarms per patient.[31] Another study evaluated delays as part of a secondary notification pager system and found a negatively sloping regression line that suggested a decreasing alarm rate, but did not report statistical testing.[34] The third study reported a reduction in alarm rates but did not report statistical testing.[38] The RCT compared the impact of a hypothetical 15‐second alarm delay to that of a lower SpO₂ limit reduction and reported a similar reduction in alarms.[36] Of the 4 studies examining patient safety, 1 study with universal monitoring reported improvements,[31] 2 studies reported no adverse outcomes,[35, 38] and the retrospective analysis of data from the RCT reported the theoretical adverse outcome of delayed detection of sudden, severe desaturations.[36]

Reconfiguring alarm acuity was evaluated in 2 studies, both of which were multiple‐intervention quasi‐experimental studies.[35, 37] Both showed reductions in alarm rates: 1 was significant without increasing adverse events (P < 0.001),[37] and the other did not report statistical testing or safety outcomes.[35]

Secondary notification of nurses using pagers was the main intervention component of 1 study incorporating delays between the alarms and the alarm pages.[34] As mentioned above, a negatively sloping regression line was displayed, but no statistical testing or safety outcomes were reported.

Disposable electrocardiographic lead wires or daily electrode changes were evaluated in 3 studies:[32, 33, 38] 1 single intervention cluster‐randomized trial[32] and 2 quasi‐experimental studies.[33, 38] In the cluster‐randomized trial, disposable lead wires were compared to reusable lead wires, with disposable lead wires having significantly fewer technical alarms for lead signal failures (P = 0.03) but a similar number of monitoring artifact alarms (P = 0.44).[32] In a single‐intervention, quasi‐experimental study, daily electrode change showed a reduction in alarms, but no statistical testing was reported.[33] One multiple‐intervention, quasi‐experimental study incorporating daily electrode change showed fewer alarms without statistical testing.[38] Of the 2 studies examining patient safety, both reported no adverse outcomes.[32, 38]

Limitations of This Review and the Underlying Body of Work

There are limitations to this systematic review and its underlying body of work. With respect to our approach to this systematic review, we focused only on monitor alarms. Numerous other medical devices generate alarms in the patient‐care environment that also can contribute to alarm fatigue and deserve equally rigorous evaluation. With respect to the underlying body of work, the quality of individual studies was generally low. For example, determinations of alarm actionability were often made by a single rater without evaluation of the reliability or validity of these determinations, and statistical testing was often missing. There were also limitations specific to intervention studies, including evaluation of nongeneralizable patient populations, failure to measure the fidelity of the interventions, inadequate measures of intervention safety, and failure to statistically evaluate alarm reductions. Finally, though not necessarily a limitation, several studies were conducted by authors involved in or funded by the medical device industry.[11, 15, 19, 31, 32] This has the potential to introduce bias, although we have no indication that the quality of the science was adversely impacted.

Moving forward, the research agenda for physiologic monitor alarms should include the following: (1) more intensive focus on evaluating the relationship between alarm exposure and response time with analysis of important mediating factors that may promote or prevent alarm fatigue, (2) emphasis on studying interventions aimed at improving alarm management using rigorous designs such as cluster‐randomized trials and trials randomized by individual participant, (3) monitoring and reporting clinically meaningful balancing measures that represent unintended consequences of disabling or delaying potentially important alarms and possibly reducing the clinicians' ability to detect true patient deterioration and intervene in a timely manner, and (4) support for transparent academicindustry partnerships to evaluate new alarm technology in real‐world settings. As evidence‐based interventions emerge, there will be new opportunities to study different implementation strategies of these interventions to optimize effectiveness.

Eligibility Criteria

With help from an experienced biomedical librarian (C.D.S.), we searched PubMed, the Cumulative Index to Nursing and Allied Health Literature, Scopus, Cochrane Library, ClinicalTrials.gov, and Google Scholar from January 1980 through April 2015 (see Supporting Information in the online version of this article for the search terms and queries). We hand searched the reference lists of included articles and reviewed our personal libraries to identify additional relevant studies.

We included peer‐reviewed, original research studies published in English, Spanish, or French that addressed the questions outlined above. Eligible patient populations were children and adults admitted to hospital inpatient units and emergency departments (EDs). We excluded alarms in procedural suites or operating rooms (typically responded to by anesthesiologists already with the patient) because of the differences in environment of care, staff‐to‐patient ratio, and equipment. We included observational studies reporting the actionability of physiologic monitor alarms (ie, alarms warranting special attention or clinical intervention), as well as nurse responses to these alarms. We excluded studies focused on the effects of alarms unrelated to patient safety, such as families' and patients' stress, noise, or sleep disturbance. We included only intervention studies evaluating pragmatic interventions ready for clinical implementation (ie, not experimental devices or software algorithms).

Selection Process and Data Extraction

First, 2 authors screened the titles and abstracts of articles for eligibility. To maximize sensitivity, if at least 1 author considered the article relevant, the article proceeded to full‐text review. Second, the full texts of articles screened were independently reviewed by 2 authors in an unblinded fashion to determine their eligibility. Any disagreements concerning eligibility were resolved by team consensus. To assure consistency in eligibility determinations across the team, a core group of the authors (C.W.P, C.P.B., E.E., and V.V.G.) held a series of meetings to review and discuss each potentially eligible article and reach consensus on the final list of included articles. Two authors independently extracted the following characteristics from included studies: alarm review methods, analytic design, fidelity measurement, consideration of unintended adverse safety consequences, and key results. Reviewers were not blinded to journal, authors, or affiliations.

Synthesis of Results and Risk Assessment

Given the high degree of heterogeneity in methodology, we were unable to generate summary proportions of the observational studies or perform a meta‐analysis of the intervention studies. Thus, we organized the studies into clinically relevant categories and presented key aspects in tables. Due to the heterogeneity of the studies and the controversy surrounding quality scores,[5] we did not generate summary scores of study quality. Instead, we evaluated and reported key design elements that had the potential to bias the results. To recognize the more comprehensive studies in the field, we developed by consensus a set of characteristics that distinguished studies with lower risk of bias. These characteristics are shown and defined in Table 1.

For the purposes of this review, we defined nonactionable alarms as including both invalid (false) alarms that do not that accurately represent the physiologic status of the patient and alarms that are valid but do not warrant special attention or clinical intervention (nuisance alarms). We did not separate out invalid alarms due to the tremendous variation between studies in how validity was measured.

Study Selection

Search results produced 4629 articles (see the flow diagram in the Supporting Information in the online version of this article), of which 32 articles were eligible: 24 observational studies describing alarm characteristics and 8 studies describing interventions to reduce alarm frequency.

Observational Study Characteristics

Characteristics of included studies are shown in Table 1. Of the 24 observational studies,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30] 15 included adult patients,[7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21] 7 included pediatric patients,[22, 23, 24, 25, 26, 27, 28] and 2 included both adult and pediatric patients.[29, 30] All were single‐hospital studies, except for 1 study by Chambrin and colleagues[10] that included 5 sites. The number of patient‐hours examined in each study ranged from 60 to 113,880.[7, 8, 9, 10, 11, 13, 14, 15, 16, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 29, 30] Hospital settings included ICUs (n = 16),[9, 10, 11, 13, 14, 16, 17, 18, 19, 22, 23, 24, 25, 26, 27, 29] general wards (n = 5),[12, 15, 20, 22, 28] EDs (n = 2),[7, 21] postanesthesia care unit (PACU) (n = 1),[30] and cardiac care unit (CCU) (n = 1).[8] Studies varied in the type of physiologic signals recorded and data collection methods, ranging from direct observation by a nurse who was simultaneously caring for patients[29] to video recording with expert review.[14, 19, 22] Four observational studies met the criteria for lower risk of bias.[11, 14, 15, 22]

Intervention Study Characteristics

Of the 8 intervention studies, 7 included adult patients,[31, 32, 33, 34, 35, 36, 37] and 1 included pediatric patients.[38] All were single‐hospital studies; 6 were quasi‐experimental[31, 33, 34, 35, 37, 38] and 2 were experimental.[32, 36] Settings included progressive care units (n = 3),[33, 34, 35] CCUs (n = 3),[32, 33, 37] wards (n = 2),[31, 38] PACU (n = 1),[36] and a step‐down unit (n = 1).[32] All except 1 study[32] used the monitoring system to record alarm data. Several studies evaluated multicomponent interventions that included combinations of the following: widening alarm parameters,[31, 35, 36, 37, 38] instituting alarm delays,[31, 34, 36, 38] reconfiguring alarm acuity,[35, 37] use of secondary notifications,[34] daily change of electrocardiographic electrodes or use of disposable electrocardiographic wires,[32, 33, 38] universal monitoring in high‐risk populations,[31] and timely discontinuation of monitoring in low‐risk populations.[38] Four intervention studies met our prespecified lower risk of bias criteria.[31, 32, 36, 38]

Proportion of Alarms Considered Actionable

Results of the observational studies are provided in Table 2. The proportion of alarms that were actionable was <1% to 26% in adult ICU settings,[9, 10, 11, 13, 14, 16, 17, 19] 20% to 36% in adult ward settings,[12, 15, 20] 17% in a mixed adult and pediatric PACU setting,[30] 3% to 13% in pediatric ICU settings,[22, 23, 24, 25, 26] and 1% in a pediatric ward setting.[22]

Relationship Between Alarm Exposure and Response Time

Whereas 9 studies addressed response time,[8, 12, 17, 18, 20, 21, 22, 27, 28] only 2 evaluated the relationship between alarm burden and nurse response time.[20, 22] Voepel‐Lewis and colleagues found that nurse responses were slower to patients with the highest quartile of alarms (57.6 seconds) compared to those with the lowest (45.4 seconds) or medium (42.3 seconds) quartiles of alarms on an adult ward (P = 0.046). They did not find an association between false alarm exposure and response time.[20] Bonafide and colleagues found incremental increases in response time as the number of nonactionable alarms in the preceding 120 minutes increased (P < 0.001 in the pediatric ICU, P = 0.009 on the pediatric ward).[22]

Interventions Effective in Reducing Alarms

Results of the 8 intervention studies are provided in Table 3. Three studies evaluated single interventions;[32, 33, 36] the remainder of the studies tested interventions with multiple components such that it was impossible to separate the effect of each component. Below, we have summarized study results, arranged by component. Because only 1 study focused on pediatric patients,[38] results from pediatric and adult settings are combined.

Widening alarm parameter default settings was evaluated in 5 studies:[31, 35, 36, 37, 38] 1 single intervention randomized controlled trial (RCT),[36] and 4 multiple‐intervention, quasi‐experimental studies.[31, 35, 37, 38] In the RCT, using a lower SpO₂ limit of 85% instead of the standard 90% resulted in 61% fewer alarms. In the 4 multiple intervention studies, 1 study reported significant reductions in alarm rates (P < 0.001),[37] 1 study did not report preintervention alarm rates but reported a postintervention alarm rate of 4 alarms per patient‐day,[31] and 2 studies reported reductions in alarm rates but did not report any statistical testing.[35, 38] Of the 3 studies examining patient safety, 1 study with universal monitoring reported fewer rescue events and transfers to the ICU postimplementation,[31] 1 study reported no missed acute decompensations,[38] and 1 study (the RCT) reported significantly more true hypoxemia events (P = 0.001).[36]

Alarm delays were evaluated in 4 studies:[31, 34, 36, 38] 3 multiple‐intervention, quasi‐experimental studies[31, 34, 38] and 1 retrospective analysis of data from an RCT.[36] One study combined alarm delays with widening defaults in a universal monitoring strategy and reported a postintervention alarm rate of 4 alarms per patient.[31] Another study evaluated delays as part of a secondary notification pager system and found a negatively sloping regression line that suggested a decreasing alarm rate, but did not report statistical testing.[34] The third study reported a reduction in alarm rates but did not report statistical testing.[38] The RCT compared the impact of a hypothetical 15‐second alarm delay to that of a lower SpO₂ limit reduction and reported a similar reduction in alarms.[36] Of the 4 studies examining patient safety, 1 study with universal monitoring reported improvements,[31] 2 studies reported no adverse outcomes,[35, 38] and the retrospective analysis of data from the RCT reported the theoretical adverse outcome of delayed detection of sudden, severe desaturations.[36]

Reconfiguring alarm acuity was evaluated in 2 studies, both of which were multiple‐intervention quasi‐experimental studies.[35, 37] Both showed reductions in alarm rates: 1 was significant without increasing adverse events (P < 0.001),[37] and the other did not report statistical testing or safety outcomes.[35]

Secondary notification of nurses using pagers was the main intervention component of 1 study incorporating delays between the alarms and the alarm pages.[34] As mentioned above, a negatively sloping regression line was displayed, but no statistical testing or safety outcomes were reported.

Disposable electrocardiographic lead wires or daily electrode changes were evaluated in 3 studies:[32, 33, 38] 1 single intervention cluster‐randomized trial[32] and 2 quasi‐experimental studies.[33, 38] In the cluster‐randomized trial, disposable lead wires were compared to reusable lead wires, with disposable lead wires having significantly fewer technical alarms for lead signal failures (P = 0.03) but a similar number of monitoring artifact alarms (P = 0.44).[32] In a single‐intervention, quasi‐experimental study, daily electrode change showed a reduction in alarms, but no statistical testing was reported.[33] One multiple‐intervention, quasi‐experimental study incorporating daily electrode change showed fewer alarms without statistical testing.[38] Of the 2 studies examining patient safety, both reported no adverse outcomes.[32, 38]

Limitations of This Review and the Underlying Body of Work

There are limitations to this systematic review and its underlying body of work. With respect to our approach to this systematic review, we focused only on monitor alarms. Numerous other medical devices generate alarms in the patient‐care environment that also can contribute to alarm fatigue and deserve equally rigorous evaluation. With respect to the underlying body of work, the quality of individual studies was generally low. For example, determinations of alarm actionability were often made by a single rater without evaluation of the reliability or validity of these determinations, and statistical testing was often missing. There were also limitations specific to intervention studies, including evaluation of nongeneralizable patient populations, failure to measure the fidelity of the interventions, inadequate measures of intervention safety, and failure to statistically evaluate alarm reductions. Finally, though not necessarily a limitation, several studies were conducted by authors involved in or funded by the medical device industry.[11, 15, 19, 31, 32] This has the potential to introduce bias, although we have no indication that the quality of the science was adversely impacted.

Moving forward, the research agenda for physiologic monitor alarms should include the following: (1) more intensive focus on evaluating the relationship between alarm exposure and response time with analysis of important mediating factors that may promote or prevent alarm fatigue, (2) emphasis on studying interventions aimed at improving alarm management using rigorous designs such as cluster‐randomized trials and trials randomized by individual participant, (3) monitoring and reporting clinically meaningful balancing measures that represent unintended consequences of disabling or delaying potentially important alarms and possibly reducing the clinicians' ability to detect true patient deterioration and intervene in a timely manner, and (4) support for transparent academicindustry partnerships to evaluate new alarm technology in real‐world settings. As evidence‐based interventions emerge, there will be new opportunities to study different implementation strategies of these interventions to optimize effectiveness.

Study Definitions

The alarm classification scheme is shown in Figure 1. Note that, for clarity, we have intentionally avoided using the terms true and false alarms because their interpretations vary across studies and can be misleading.

Figure 1

Setting and Subjects

We performed this study between August 2012 and July 2013 at a freestanding children's hospital. We evaluated nurses caring for 2 populations: (1) PICU patients with heart and/or lung failure (requiring inotropic support and/or invasive mechanical ventilation), and (2) medical patients on a general inpatient ward. Nurses caring for heart and/or lung failure patients in the PICU typically were assigned 1 to 2 total patients. Nurses on the medical ward typically were assigned 2 to 4 patients. We identified subjects from the population of nurses caring for eligible patients with parents available to provide in‐person consent in each setting. Our primary interest was to evaluate the association between nonactionable alarms and response time, and not to study the epidemiology of alarms in a random sample. Therefore, when alarm data were available prior to screening, we first approached nurses caring for patients in the top 25% of alarm rates for that unit over the preceding 4 hours. We identified preceding alarm rates using BedMasterEx (Excel Medical Electronics, Jupiter, FL).

Human Subjects Protection

This study was approved by the institutional review board of The Children's Hospital of Philadelphia. We obtained written in‐person consent from the patient's parent and the nurse subject. We obtained a Certificate of Confidentiality from the National Institutes of Health to further protect study participants.[14]

Monitoring Equipment

All patients in the PICU were monitored continuously using General Electric (GE) (Fairfield, CT) solar devices. All bed spaces on the wards include GE Dash monitors that are used if ordered. On the ward we studied, 30% to 50% of patients are typically monitored at any given time. In addition to alarming at the bedside, most clinical alarms also generated a text message sent to the nurse's wireless phone listing the room number and the word monitor. Messages did not provide any clinical information about the alarm or patient's status. There were no technicians reviewing alarms centrally.

Physicians used an order set to order monitoring, selecting 1 of 4 available preconfigured profiles: infant <6 months, infant 6 months to 1 year, child, and adult. The parameters for each age group are in Supporting Figure 1, available in the online version of this article. A physician order is required for a nurse to change the parameters. Participating in the study did not affect this workflow.

Primary Outcome

The primary outcome was the nurse's response time to potentially critical monitor alarms that occurred while neither they nor any other clinicians were in the patient's room.

Primary Exposure and Alarm Classification

The primary exposure was the number of nonactionable alarms in the same patient over the preceding 120 minutes (rolling and updated each minute). The alarm classification scheme is shown in Figure 1.

Due to technical limitations with obtaining time‐stamped alarm data from the different ventilators in use during the study period, we were unable to identify the causes of all ventilator alarms. Therefore, we included ventilator alarms that did not lead to clinical interventions as nonactionable alarm exposures, but we did not evaluate the response time to any ventilator alarms.

Data Collection

We combined video recordings with monitor time‐stamp data to evaluate the association between nonactionable alarms and the nurse's response time. Our detailed video recording and annotation methods have been published separately.[15] Briefly, we mounted up to 6 small video cameras in patients' rooms and recorded up to 6 hours per session. The cameras captured the monitor display, a wide view of the room, a close‐up view of the patient, and all windows and doors through which staff could visually assess the patient without entering the room.

Video Processing, Review, and Annotation

The first 5 video sessions were reviewed in a group training setting. Research assistants received instruction on how to determine alarm validity and actionability in accordance with the study definitions. Following the training period, the review workflow was as follows. First, a research assistant entered basic information and a preliminary assessment of the alarm's clinical validity and actionability into a REDCap (Research Electronic Data Capture; Vanderbilt University, Nashville, TN) database.[16] Later, a physician investigator secondarily reviewed all alarms and confirmed the assessments of the research assistants or, when disagreements occurred, discussed and reconciled the database. Alarms that remained unresolved after secondary review were flagged for review with an additional physician or nurse investigator in a team meeting.

Data Analysis

We summarized the patient and nurse subjects, the distributions of alarms, and the response times to potentially critical monitor alarms that occurred while neither the nurse nor any other clinicians were in the patient's room. We explored the data using plots of alarms and response times occurring within individual video sessions as well as with simple linear regression. Hypothesizing that any alarm fatigue effect would be strongest in the highest alarm patients, and having observed that alarms are distributed very unevenly across patients in both the PICU and ward, we made the decision not to use quartiles, but rather to form clinically meaningful categories. We also hypothesized that nurses might not exhibit alarm fatigue unless they were inundated with alarms. We thus divided the nonactionable alarm counts over the preceding 120 minutes into 3 categories: 0 to 29 alarms to represent a low to average alarm rate exhibited by the bottom 50% of the patients, 30 to 79 alarms to represent an elevated alarm rate, and 80+ alarms to represent an extremely high alarm rate exhibited by the top 5%. Because the exposure time was 120 minutes, we conducted the analysis on the alarms occurring after a nurse had been video recorded for at least 120 minutes.

We further evaluated the relationship between nonactionable alarms and nurse response time with Kaplan‐Meier plots by nonactionable alarm count category using the observed response‐time data. The Kaplan‐Meier plots compared response time across the nonactionable alarm exposure group, without any statistical modeling. A log‐rank test stratified by nurse evaluated whether the distributions of response time in the Kaplan‐Meier plots differed across the 3 alarm exposure groups, accounting for within‐nurse clustering.

Accelerated failure‐time regression based on the Weibull distribution then allowed us to compare response time across each alarm exposure group and provided confidence intervals. Accelerated failure‐time models are comparable to Cox models, but emphasize time to event rather than hazards.[17, 18] We determined that the Weibull distribution was suitable by evaluating smoothed hazard and log‐hazard plots, the confidence intervals of the shape parameters in the Weibull models that did not include 1, and by demonstrating that the Weibull model had better fit than an alternative (exponential) model using the likelihood‐ratio test (P<0.0001 for PICU, P=0.02 for ward). Due to the small sample size of nurses and patients, we could not adjust for nurse‐ or patient‐level covariates in the model. When comparing the nonactionable alarm exposure groups in the regression model (029 vs 3079, 3079 vs 80+, and 029 vs 80+), we Bonferroni corrected the critical P value for the 3 comparisons, for a critical P value of 0.05/3=0.0167.

Nurse Questionnaire

At the session's conclusion, nurses completed a questionnaire that included demographics and asked, Did you respond more quickly to monitor alarms during this study because you knew you were being filmed? to measure if nurses would report experiencing a Hawthorne‐like effect.[19, 20, 21]

Overall Response Times for Out‐of‐Room Alarms

We first evaluated response times without excluding alarms occurring prior to the 120‐minute mark. Of the 2445 clinical condition alarms, we excluded the 315 noncritical arrhythmia types from analysis of response time because they did not meet our definition of potentially critical alarms. Of the 2130 potentially critical alarms, 1185 (55.6%) occurred while neither the nurse nor any other clinician was in the patient's room. We proceeded to analyze the response time to these 1185 alarms (307 in the PICU and 878 on the ward). In the PICU, median response time was 3.3 minutes (IQR, 0.814.4). On the ward, median response time was 9.8 minutes (IQR, 3.222.4).

Response‐Time Association With Nonactionable Alarm Exposure

Next, we analyzed the association between response time to potentially critical alarms that occurred when the nurse was not in the patient's room and the number of nonactionable alarms occurring over the preceding 120‐minute window. This required excluding the alarms that occurred in the first 120 minutes of each session, leaving 647 alarms with eligible response times to evaluate the exposure between prior nonactionable alarm exposure and response time: 219 in the PICU and 428 on the ward. Kaplan‐Meier plots and tabulated response times demonstrated the incremental relationships between each nonactionable alarm exposure category in the observed data, with the effects most prominent as the Kaplan‐Meier plots diverged beyond the median (Figure 3 and Table 2). Excluding the extreme outlier patient had no effect on the results, because 1378 of the 1430 alarms occurred with the nurse present at the bedside, and only 2 of the remaining alarms were potentially critical.

Figure 3

Accelerated failure‐time regressions revealed significant incremental increases in the modeled response time as the number of preceding nonactionable alarms increased in both the PICU and ward settings (Table 2).

Hawthorne‐like Effects

Four of the 36 nurses reported that they responded more quickly to monitor alarms because they knew they were being filmed.

Study Definitions

The alarm classification scheme is shown in Figure 1. Note that, for clarity, we have intentionally avoided using the terms true and false alarms because their interpretations vary across studies and can be misleading.

Figure 1

Setting and Subjects

We performed this study between August 2012 and July 2013 at a freestanding children's hospital. We evaluated nurses caring for 2 populations: (1) PICU patients with heart and/or lung failure (requiring inotropic support and/or invasive mechanical ventilation), and (2) medical patients on a general inpatient ward. Nurses caring for heart and/or lung failure patients in the PICU typically were assigned 1 to 2 total patients. Nurses on the medical ward typically were assigned 2 to 4 patients. We identified subjects from the population of nurses caring for eligible patients with parents available to provide in‐person consent in each setting. Our primary interest was to evaluate the association between nonactionable alarms and response time, and not to study the epidemiology of alarms in a random sample. Therefore, when alarm data were available prior to screening, we first approached nurses caring for patients in the top 25% of alarm rates for that unit over the preceding 4 hours. We identified preceding alarm rates using BedMasterEx (Excel Medical Electronics, Jupiter, FL).

Human Subjects Protection

This study was approved by the institutional review board of The Children's Hospital of Philadelphia. We obtained written in‐person consent from the patient's parent and the nurse subject. We obtained a Certificate of Confidentiality from the National Institutes of Health to further protect study participants.[14]

Monitoring Equipment

All patients in the PICU were monitored continuously using General Electric (GE) (Fairfield, CT) solar devices. All bed spaces on the wards include GE Dash monitors that are used if ordered. On the ward we studied, 30% to 50% of patients are typically monitored at any given time. In addition to alarming at the bedside, most clinical alarms also generated a text message sent to the nurse's wireless phone listing the room number and the word monitor. Messages did not provide any clinical information about the alarm or patient's status. There were no technicians reviewing alarms centrally.

Physicians used an order set to order monitoring, selecting 1 of 4 available preconfigured profiles: infant <6 months, infant 6 months to 1 year, child, and adult. The parameters for each age group are in Supporting Figure 1, available in the online version of this article. A physician order is required for a nurse to change the parameters. Participating in the study did not affect this workflow.

Primary Outcome

The primary outcome was the nurse's response time to potentially critical monitor alarms that occurred while neither they nor any other clinicians were in the patient's room.

Primary Exposure and Alarm Classification

The primary exposure was the number of nonactionable alarms in the same patient over the preceding 120 minutes (rolling and updated each minute). The alarm classification scheme is shown in Figure 1.

Due to technical limitations with obtaining time‐stamped alarm data from the different ventilators in use during the study period, we were unable to identify the causes of all ventilator alarms. Therefore, we included ventilator alarms that did not lead to clinical interventions as nonactionable alarm exposures, but we did not evaluate the response time to any ventilator alarms.

Data Collection

We combined video recordings with monitor time‐stamp data to evaluate the association between nonactionable alarms and the nurse's response time. Our detailed video recording and annotation methods have been published separately.[15] Briefly, we mounted up to 6 small video cameras in patients' rooms and recorded up to 6 hours per session. The cameras captured the monitor display, a wide view of the room, a close‐up view of the patient, and all windows and doors through which staff could visually assess the patient without entering the room.

Video Processing, Review, and Annotation

The first 5 video sessions were reviewed in a group training setting. Research assistants received instruction on how to determine alarm validity and actionability in accordance with the study definitions. Following the training period, the review workflow was as follows. First, a research assistant entered basic information and a preliminary assessment of the alarm's clinical validity and actionability into a REDCap (Research Electronic Data Capture; Vanderbilt University, Nashville, TN) database.[16] Later, a physician investigator secondarily reviewed all alarms and confirmed the assessments of the research assistants or, when disagreements occurred, discussed and reconciled the database. Alarms that remained unresolved after secondary review were flagged for review with an additional physician or nurse investigator in a team meeting.

Data Analysis

We summarized the patient and nurse subjects, the distributions of alarms, and the response times to potentially critical monitor alarms that occurred while neither the nurse nor any other clinicians were in the patient's room. We explored the data using plots of alarms and response times occurring within individual video sessions as well as with simple linear regression. Hypothesizing that any alarm fatigue effect would be strongest in the highest alarm patients, and having observed that alarms are distributed very unevenly across patients in both the PICU and ward, we made the decision not to use quartiles, but rather to form clinically meaningful categories. We also hypothesized that nurses might not exhibit alarm fatigue unless they were inundated with alarms. We thus divided the nonactionable alarm counts over the preceding 120 minutes into 3 categories: 0 to 29 alarms to represent a low to average alarm rate exhibited by the bottom 50% of the patients, 30 to 79 alarms to represent an elevated alarm rate, and 80+ alarms to represent an extremely high alarm rate exhibited by the top 5%. Because the exposure time was 120 minutes, we conducted the analysis on the alarms occurring after a nurse had been video recorded for at least 120 minutes.

We further evaluated the relationship between nonactionable alarms and nurse response time with Kaplan‐Meier plots by nonactionable alarm count category using the observed response‐time data. The Kaplan‐Meier plots compared response time across the nonactionable alarm exposure group, without any statistical modeling. A log‐rank test stratified by nurse evaluated whether the distributions of response time in the Kaplan‐Meier plots differed across the 3 alarm exposure groups, accounting for within‐nurse clustering.

Accelerated failure‐time regression based on the Weibull distribution then allowed us to compare response time across each alarm exposure group and provided confidence intervals. Accelerated failure‐time models are comparable to Cox models, but emphasize time to event rather than hazards.[17, 18] We determined that the Weibull distribution was suitable by evaluating smoothed hazard and log‐hazard plots, the confidence intervals of the shape parameters in the Weibull models that did not include 1, and by demonstrating that the Weibull model had better fit than an alternative (exponential) model using the likelihood‐ratio test (P<0.0001 for PICU, P=0.02 for ward). Due to the small sample size of nurses and patients, we could not adjust for nurse‐ or patient‐level covariates in the model. When comparing the nonactionable alarm exposure groups in the regression model (029 vs 3079, 3079 vs 80+, and 029 vs 80+), we Bonferroni corrected the critical P value for the 3 comparisons, for a critical P value of 0.05/3=0.0167.

Nurse Questionnaire

At the session's conclusion, nurses completed a questionnaire that included demographics and asked, Did you respond more quickly to monitor alarms during this study because you knew you were being filmed? to measure if nurses would report experiencing a Hawthorne‐like effect.[19, 20, 21]

Overall Response Times for Out‐of‐Room Alarms

We first evaluated response times without excluding alarms occurring prior to the 120‐minute mark. Of the 2445 clinical condition alarms, we excluded the 315 noncritical arrhythmia types from analysis of response time because they did not meet our definition of potentially critical alarms. Of the 2130 potentially critical alarms, 1185 (55.6%) occurred while neither the nurse nor any other clinician was in the patient's room. We proceeded to analyze the response time to these 1185 alarms (307 in the PICU and 878 on the ward). In the PICU, median response time was 3.3 minutes (IQR, 0.814.4). On the ward, median response time was 9.8 minutes (IQR, 3.222.4).

Response‐Time Association With Nonactionable Alarm Exposure

Next, we analyzed the association between response time to potentially critical alarms that occurred when the nurse was not in the patient's room and the number of nonactionable alarms occurring over the preceding 120‐minute window. This required excluding the alarms that occurred in the first 120 minutes of each session, leaving 647 alarms with eligible response times to evaluate the exposure between prior nonactionable alarm exposure and response time: 219 in the PICU and 428 on the ward. Kaplan‐Meier plots and tabulated response times demonstrated the incremental relationships between each nonactionable alarm exposure category in the observed data, with the effects most prominent as the Kaplan‐Meier plots diverged beyond the median (Figure 3 and Table 2). Excluding the extreme outlier patient had no effect on the results, because 1378 of the 1430 alarms occurred with the nurse present at the bedside, and only 2 of the remaining alarms were potentially critical.

Figure 3