Physiologic monitor alarms occur frequently in the hospital environment, with average rates on pediatric wards between 42 and 155 alarms per monitored patient-day.1 However, average rates do not depict the full story, because only 9%–25% of patients are responsible for most alarms on inpatient wards.1,2 In addition, only 0.5%–1% of alarms on pediatric wards warrant action.3,4 Downstream consequences of high alarm rates include interruptions5,6 and alarm fatigue.3,4,7
Alarm customization, the process of reviewing individual patients’ alarm data and using that data to implement patient-specific alarm reduction interventions, has emerged as a potential approach to unit-wide alarm management.8-11 Potential customizations include broadening alarm thresholds, instituting delays between the time the alarm condition is met and the time the alarm sounds, and changing electrodes.8-11 However, the workflows within which to identify the patients who will benefit from customization, make decisions about how to customize, and implement customizations have not been delineated.
Safety huddles are brief structured discussions among physicians, nurses, and other staff aiming to identify and mitigate threats to patient safety.11-13 In this study, we aimed to evaluate the influence of a safety huddle-based alarm intervention strategy targeting high alarm pediatric ward patients on (a) unit-level alarm rates and (b) patient-level alarm rates, as well as to (c) evaluate implementation outcomes. We hypothesized that patients discussed in huddles would have greater reductions in alarm rates in the 24 hours following their huddle than patients who were not discussed. Given that most alarms are generated by a small fraction of patients,1,2 we hypothesized that patient-level reductions would translate to unit-level reductions.
Human Subject Protection
The Institutional Review Board of Children’s Hospital of Philadelphia approved this study with a waiver of informed consent. We registered the study at ClinicalTrials.gov (identifier NCT02458872). The original protocol is available as an Online Supplement.
Design and Framework
For our secondary effectiveness outcome evaluating the effect of the intervention on the alarm rates of the individual patients discussed in huddles, we used a cohort design embedded within the trial to analyze patient-specific alarm data collected only on randomly selected “intensive data collection days,” described below and in Figure 1.
Setting and Subjects
All patients hospitalized on 8 units that admit general pediatric and medical subspecialty patients at Children’s Hospital of Philadelphia between June 15, 2015 and May 8, 2016 were included in the primary (unit-level) analysis. Every patient’s bedside included a General Electric Dash 3000 physiologic monitor. Decisions to monitor patients were made by physicians and required orders. Default alarm settings are available in Supplementary Table 1; these settings required orders to change.
All 8 units were already convening scheduled safety huddles led by the charge nurse each day. All nurses and at least one resident were expected to attend; attending physicians and fellows were welcome but not expected to attend. Huddles focused on discussing safety concerns and patient flow. None of the preexisting huddles included alarm discussion.
For each nonholiday weekday, we generated customized paper-based alarm huddle data “dashboards” (Supplementary Figure 1) displaying data from the patients (up to a maximum of 4) on each intervention unit with the highest numbers of high-acuity alarms (“crisis” and “warning” audible alarms, see Supplementary Table 2 for detailed listing of alarm types) in the preceding 4 hours by reviewing data from the monitor network using BedMasterEx v4.2 (Excel Medical Electronics). Dashboards listed the most frequent types of alarms, alarm settings, and included a script for discussing the alarms with checkboxes to indicate changes agreed upon by the team during the huddle. Patients with fewer than 20 alarms in the preceding 4h were not included; thus, sometimes fewer than 4 patients’ data were available for discussion. We hand-delivered dashboards to the charge nurses leading huddles, and they facilitated the multidisciplinary alarm discussions focused on reviewing alarm data and customizing settings to reduce unnecessary alarms.
The study had 3 periods as shown in Supplementary Figure 2: (1) 16-week baseline data collection, (2) phased intervention implementation during which we serially spent 2-8 weeks on each of the 4 intervention units implementing the intervention, and (3) 16-week postimplementation data collection.
The primary effectiveness outcome was the change in unit-level alarms per patient-day between the baseline and postimplementation periods in intervention versus control units, with all patients on the units included. The secondary effectiveness outcome (analyzed using the embedded cohort design) was the change in individual patient-level alarms between the 24 hours leading up to a huddle and the 24 hours following huddles in patients who were versus patients who were not discussed in huddles.
Implementation outcomes included adoption and fidelity measures. To measure adoption (defined as “intention to try” the intervention),16 we measured the frequency of discussions attended by patients’ nurses and physicians. We evaluated 3 elements of fidelity: adherence, dose, and quality of delivery.17 We measured adherence as the incorporation of alarm discussion into huddles when there were eligible patients to discuss. We measured dose as the average number of patients discussed on each unit per calendar day during the postimplementation period. We measured quality of delivery as the extent to which changes to monitoring that were agreed upon in the huddles were made at the bedside.
To surveil for unintended consequences of reduced monitoring, we screened the hospital’s rapid response and code blue team database weekly for any events in patients previously discussed in huddles that occurred between huddle and hospital discharge. We reviewed charts to determine if the events were related to the intervention.
Prior to randomization, the 8 units were divided into pairs based on participation in hospital-wide Joint Commission alarm management activities, use of alarm middleware that relayed detailed alarm information to nurses’ mobile phones, and baseline alarm rates. One unit in each pair was randomized to intervention and the other to control by coin flip.
We used Research Electronic Data Capture (REDCap)18 database tools.
Data for Unit-Level Analyses
We captured all alarms occurring on the study units during the study period using data from BedMasterEx. We obtained census data accurate to the hour from the Clinical Data Warehouse.
Data Captured in All Huddles
During each huddle, we collected the number of patients whose alarms were discussed, patient characteristics, presence of nurses and physicians, and monitoring changes agreed upon. We then followed up 4 hours later to determine if changes were made at the bedside by examining monitor settings.
Data Captured Only During Intensive Data Collection Days
We randomly selected 1 day during each of the 16 weeks of the postimplementation period to obtain additional patient-level data. On each intensive data collection day, the 4 monitored patients on each intervention and control unit with the most high-acuity alarms in the 4 hours prior to huddles occurring — regardless of whether or not these patients were later discussed in huddles — were identified for data collection. On these dates, a member of the research team reviewed each patient’s alarm counts in 4-hour blocks during the 24 hours before and after the huddle. Given that the huddles were not always at the same time every day (ranging between 10:00 and 13:00), we operationally set the huddle time as 12:00 for all units.
We used Stata/SE 14.2 for all analyses.
Unit-Level Alarm Rates
To compare unit-level rates, we performed an interrupted time series analysis using segmented (piecewise) regression to evaluate the impact of the intervention.19,20 We used a multivariable generalized estimating equation model with the negative binomial distribution21 and clustering by unit. We bootstrapped the model and generated percentile-based 95% confidence intervals. We then used the model to estimate the alarm rate difference in differences between the baseline data collection period and the postimplementation data collection period for intervention versus control units.
Patient-Level Alarm Rates
In contrast to unit-level analysis, we used an embedded cohort design to model the change in individual patients’ alarms between the 24 hours leading up to huddles and the 24 hours following huddles in patients who were versus patients who were not discussed in huddles. The analysis was restricted to the patients included in intensive data collection days. We performed bootstrapped linear regression and generated percentile-based 95% confidence intervals using the difference in 4-hour block alarm rate between pre- and posthuddle as the outcome. We clustered within patients. We stratified by unit and preceding alarm rate. We modeled the alarm rate difference between the 24-hour prehuddle and the 24-hour posthuddle for huddled and nonhuddled patients and the difference in differences between exposure groups.
We summarized adoption and fidelity using proportions.
Alarm dashboards informed 580 structured alarm discussions during 353 safety huddles (huddles often included discussion of more than one patient).
Unit-Level Alarm Rates
Visually, alarm rates over time on each individual unit appeared flat despite the intervention (Supplementary Figure 3). Using piecewise regression, we found that intervention and control units had small increases in alarm rates between the baseline and postimplementation periods with a nonsignificant difference in these differences between the control and intervention groups (Table 1).
Patient-Level Alarm Rates
We then restricted the analysis to the patients whose data were collected during intensive data collection days. We obtained data from 1974 pre-post pairs of 4-hour time periods.
The patient’s nurse attended 482 of the 580 huddle discussions (83.1%), and at least one of the patient’s physicians (resident, fellow, or attending) attended 394 (67.9%).
In addition to the 353 huddles that included alarm discussion, 123 instances had no patients with ≥20 high acuity alarms in the preceding 4 hours therefore, no data were brought to the huddle. There were an additional 30 instances when a huddle did not occur or there was no alarm discussion in the huddle despite data being available. Thus, adherence occurred in 353 of 383 huddles (92.2%).
During the 112 calendar day postimplementation period, 379 patients’ alarms were discussed in huddles for an average intervention dose of 0.85 discussions per unit per calendar day.
Fidelity: Quality of Delivery
In 362 of the 580 huddle discussions (62.4%), changes were agreed upon. The most frequently agreed upon changes were discontinuing monitoring (32.0%), monitoring only when asleep or unsupervised (23.8%), widening heart rate parameters (12.7%), changing electrocardiographic leads/wires (8.6%), changing the pulse oximetry probe (8.0%), and increasing the delay time between when oxygen desaturation was detected and when the alarm was generated (4.7%). Of the huddle discussions with changes agreed upon, 346 (95.6%) changes were enacted at the bedside.
There were 0 code blue events and 26 rapid response team activations for patients discussed in huddles. None were related to the intervention.
Our main finding was that the huddle strategy was effective in safely reducing the burden of alarms for the high alarm pediatric ward patients whose alarms were discussed, but it did not reduce unit-level alarm rates. Implementation outcomes explained this finding. Although adoption and adherence were high, the overall dose of the intervention was low.
We also found that 36% of alarms had technical causes, the majority of which were related to the pulse oximetry probe detecting that it was off the patient or searching for a pulse. Although these alarms are likely perceived differently by clinical staff (most monitors generate different sounds for technical alarms), they still represent a substantial contribution to the alarm environment. Minimizing them in patients who must remain continuously monitored requires more intensive effort to implement other types of interventions than the main focus of this study, such as changing pulse oximetry probes and electrocardiographic leads/wires.
In one-third of huddles, monitoring was simply discontinued. We observed in many cases that, while these patients may have had legitimate indications for monitoring upon admission, their conditions had improved; after brief multidisciplinary discussion, the team concluded that monitoring was no longer indicated. This observation may suggest interventions at the ordering phase, such as prespecifying a monitoring duration.22,23
This study’s findings were consistent with a quasi-experimental study of safety huddle-based alarm discussions in a pediatric intensive care unit that showed a patient-level reduction of 116 alarms per patient-day in those discussed in huddles relative to controls.11 A smaller quasi-experimental study of implementing a nighttime alarm “ward round” in an adult intensive care unit showed a significant reduction in unit-level alarms/patient-day from 168 to 84.9 In a quality improvement report, a monitoring care process bundle that included discussion of alarm settings showed a reduction in unit-level alarms/patient-day from 180 to 40.10 Our study strengthens the body of literature using a cluster-randomized design, measuring patient- and unit-level outcomes, and including implementation outcomes that explain effectiveness findings.
On a hypothetical unit similar to the ones we studied with 20 occupied beds and 60 alarms/patient-day, an average of 1200 alarms would occur each day. We delivered the intervention to 0.85 patients per day. Changes were made at the bedside in 60% of those with the intervention delivered, and those patients had a difference in differences of 119 fewer alarms compared with the comparison patients on control units. In this scenario, we could expect a relative reduction of 0.85 x 0.60 x 119 = 61 fewer alarms/day total on the unit or a 5% reduction. However, that estimated reduction did not account for the arrival of new patients with high alarm rates, which certainly occurred in this study and explained the lack of effect at the unit level.
As described above, the intervention dose was low, which translated into a lack of effect at the unit level despite a strong effect at the patient level. This result was partly due to the manual process required to produce the alarm dashboards that restricted their availability to nonholiday weekdays. The study was performed at one hospital, which limited generalizability. The study hospital was already convening daily safety huddles that were well attended by nurses and physicians. Other hospitals without existing huddle structures may face challenges in implementing similar multidisciplinary alarm discussions. In addition, the study design was randomized at the unit (rather than patient) level, which limited our ability to balance potential confounders at the patient level.
A safety huddle intervention strategy to drive alarm customization was effective in safely reducing alarms for individual children discussed. However, unit-level alarm rates were not affected by the intervention due to a low dose. Leaders of efforts to reduce alarms should consider beginning with passive interventions (such as changes to default settings and alarm delays) and use huddle-based discussion as a second-line intervention to address remaining patients with high alarm rates.
We thank Matthew MacMurchy, BA, for his assistance with data collection.
This study was supported by a Young Investigator Award (Bonafide, PI) from the Academic Pediatric Association.
Role of the Funder/Sponsor
The Academic Pediatric Association had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit for publication.
No relevant financial activities, aside from the grant funding from the Academic Pediatric Association listed above, are reported.