Abstract
Keywords
Introduction
It has been well established in the medical literature that an excessive number of alerts in the electronic health record (EHR) contributes to alert fatigue and provider burnout.1,2 Despite their sometimes frustrating nature, alerts have been proven to work. They reduce duplicate laboratory orders and antibiotic use. 3 They can prevent medication errors and help providers adhere to current guidelines.4,5
Many alerts are suboptimal due to a lack of standardization and failure to adhere to previously established design recommendations.6–8 Numerous studies have shown that alerts are most effective when presented through an existing system, at the right moment in the workflow—when the provider is ready to make a decision.9–13 These recommendations align with the “Five Pillars of Clinical Decision Support”: right information, right person, right intervention, right format, and right time. 14
While alerts in the EHR itself tend to be the target of these discussions, another type exists: having an automated system from the EHR send an alert directly to the provider’s pager or smart phone-based text-message platform. Automated paging systems—distinct from the ubiquitous pop-up alerts in the medical record—have been in use for over 25 years for critical laboratory results.15–17 They have been shown to be an effective intervention to hasten patients’ transfer from the emergency department and to improve time to stat radiology studies and respiratory treatments.18,19 Not only do they simplify communication between providers, but a previous study found they improved time to provider action and that providers were satisfied with the more efficient communication. 18
Simplifying communication within healthcare teams is an increasingly important part of workflows in hospital in the era of burnout; the nursing-provider relationship is particularly vulnerable and critical to patient care and safety. 20 As part of a quality improvement effort, we interviewed several staff focus groups to evaluate where our hospital could streamline communication pain-points. We received feedback on the nursing burden of manually paging providers to renew restraint orders across the hospital.
To keep patients and staff safe during severe episodes of agitation, physical restraints are sometimes used. 21 However, restraints can lead to physical and psychological patient harms. 22 In order to minimize harms and ensure that patients are not placed in restraints excessively, our institution requires re-evaluation of restrained patients after a fixed time interval that varies by patient age and restraint type.
Historically, restraint renewal at our hospital depended on nursing to notice the expired order, alert the provider, and have them call back for a discussion. These extra steps put an additional burden on already overworked nursing staff to find the provider; creating a delay when nurses and providers could not connect immediately and adding unnecessary stress to the provider-nurse relationship.20,23
Though one previous study examined the use of an in-EHR alert for improvement of restraint order times, none to our knowledge have used an automatic alerting system (neither page nor text based) to improve these orders. 24
Based on this feedback, we developed an automated text-based alerting system to notify providers when a restraint order lapsed in our hospital, trialing the project in our pediatric intensive care unit (PICU) with the support of unit-based nursing and provider leadership. The PICU was chosen given the higher number of restraint orders needed per patient volume than in other areas of the Children’s Hospital. Previous efforts to reduce PICU restraint order times have shown that web-based dashboards displaying information about lapsed restraint orders do not change management. 25
Through clinical decision support tools, we hoped to improve patient safety and improve the nurse-provider relationship by implementing a text-based alerting system. We hypothesized that a direct EHR to text system would help providers reassess the situation and write any necessary orders more quickly, thus reducing the time without active restraint orders.
Methods
Our institution uses Epic Systems (Verona, WI) and a smart phone-based text alerting system (Voalte) across multiple adult and pediatric hospitals. We implemented our text-based alerting system at one facility’s pediatric intensive care unit in April 2023. Five minutes before an active restraint order lapsed, a text message was automatically sent to the responding provider who was signed into the patient. The message read “ACTION – Re: [Patient Name] – [Patient Bed Number] Restraint order is expiring or has expired. Please re-order if restraints should remain active.” Patient confidentiality and data security were maintained by embedding all alerts within the Epic and Voalte EHR systems, ensuring that no patient information left the secure, HIPAA-compliant medical record environment.
Data on all restraint orders in the PICU, pediatric wards, and adult ICU was pulled from the Epic Clarity database for 2 years pre- and post-intervention. Variables included order placement time by the user, order start time, patient ID, encounter ID, and order department.
Orders with a start time of more than 5 minutes in the past were considered backorders and were excluded from the analysis. These typically occur when staff realize that a restraint order was never previously entered and place it retroactively to ensure accurate documentation. Our study focused specifically on active restraint orders—those intended to take effect immediately—because the goal was to prevent lapses in coverage and improve timeliness in accordance with Joint Commission standards. Our institutional EHR does not allow for future standing orders for restraints.
Time without active restraint orders was calculated per patient. If an order was placed more than 24 h later than the previous order on the same patient, it was considered a new restraint session and did not count towards the average hours between orders. Orders were tallied among clinical sites pre- and post-intervention.
Average number of hours between restraint orders was calculated and an unpaired t-test was used to assess significance. All analysis was conducted using R. 26
Per our institutional policy, Institutional Review Board (IRB) approval was not required for this study as it was classified as a quality improvement (QI) project.
Results
There were 168 backorders that were excluded from the analysis. A total of 1394 orders were included.
Number of orders pre and post intervention across units in our facility with average number of hours between restraint orders.
*Unpaired t-test.
Following implementation of the intervention, time without an active restraint order decreased in the PICU by 39% (2 h 23 min vs 1 h 27 min hours), p = .24 (Figure 1). The pediatric wards showed a 15% increase in hours between active restraint orders (2 h 25 min vs 2 h 46 min), p = .63. In contrast, the adult ICU had a slight 4% decrease (4 h 59 min to 4 h 43 min), p = .69. Hours between active restraint orders before and after the intervention by clinical location.
Discussion
This study demonstrates that an automated text-based alerting system may improve the time without active restraint orders in the PICU. Although the reduction in time was not statistically significant, the 39% decrease brought practice more in line with Joint Commission standards and likely reduced the number of calls to physicians for order renewals. The gross reduction of 56 min between restraint orders in the PICU pilot spurred our alerts committee to implement this change across the rest of our institution.
While it was not possible to quantify the number of phone calls and pages previously made by nurses before the intervention and thus examine the number of pages a provider received, the intent of this intervention was not to increase the total number of alerts but to streamline communication by removing the nurse from the workflow of initially notifying providers. By automating this process, we hoped to reduce the burden on nursing staff while maintaining timely provider action. Importantly, removing the onus of the initial page from the nurse does not remove them from the clinical care conversation, as the purpose of the alert is to remind the provider to check on the patient at bedside, at which point they are able to confer with nursing if necessary.
Widescale implementation of this system could ease nursing workload and provider communication, as well as improve lapses in restraint orders in other areas of the hospital. This is increasingly relevant as the number of pediatric patients who are hospitalized for mental health reasons has been going up for years, and was further exacerbated by the COVID-19 pandemic.27–30 As they await psychiatric placement, many of these children are placed on hospitalist services which are ill-equipped to manage behavioral health outbursts, resulting in frequent use of restraints. 31 This messaging system may be helpful in guiding clinicians to re-evaluate their patients and remove the restraints.
This work builds upon previous restraint order research by Griffy et al. (2009), which demonstrated an improvement in time without restraint order lapses after implementation of an interruptive alert in their EHR. A ‘soft-stop’ renewal prompt in the emergency department decreased time to restraint order renewal to 125 min from the baseline of 189 min (34%). With a ‘hard-stop,’ the time without active orders was 133 min (30%). 24 Considering that providers may not be sitting at a computer the moment an order expires, we imagine that the EHR to text message based system would be more efficient.
Lessons learned
In reflecting on the implementation of the automated alerting system, three key lessons emerged that can guide future interventions.
First, actively gathering feedback across the hospital from both providers and nursing focus groups allowed us to highlight specific pain points in patient workflow and explore the utility for this project, which was a pain point for all team members.
Second, automated alerting systems outside of the EHR can function to simplify communication to be more closely aligned with the five rights of clinical decision support. By removing nursing from the workflow, the new alerting system was more closely aligned with the “right provider” and “right time” ideal.
Finally, initiating this system within a confined, controlled environment such as the PICU allowed for a focused pilot trial. This small-scale implementation provided valuable insights that could be compared against both the adult ICU and pediatric wards, and this served to set a precedent for scaling the intervention across the institution.
Limitations
Our study did not examine which patients in our PICU had restraint orders placed, and whether patient demographics had any influence on how often they were renewed. Historical evidence suggests that patients of color and with comorbid psychiatric conditions (particularly Autism Spectrum Disorder) are more likely to have restraints ordered, but this was outside the scope of this initial pilot. 32 Additionally, the study may have been influenced by the Hawthorne effect, as providers could have been more attentive knowing they were being observed. We also did not account for differences in patient age, which affect renewal requirements, or for which specific patients were present in the PICU at the time—both of which could have impacted the observed outcomes. Due to limitations with the alert software, we were also unable to track the exact number of alerts that were generated. The absence of this quantitative alarm frequency data means we cannot assess the potential implications for monitoring provider alarm fatigue.
Our findings were not statistically significant, likely due to the low sample size of our PICU in terms of the number of beds. Yet the reduction in time without active restraint orders is an encouraging trend, suggesting that with a larger sample size, statistical significance might be achieved.
Another potential limitation is the unique environment of the PICU, which may limit generalizability. PICUs have more providers readily available: there is a higher provider to patient ratio, a smaller physical footprint than the wards, and a smaller provider team. While we believe this messaging system could be effective in a wards setting, it is possible that providers will be otherwise overwhelmed with alerts given their larger patient census. Considering that providers who receive more alerts are more likely to dismiss them out of hand, it is possible this implementation would not work for a busy service.33,34
Future directions
Future studies should consider implementing a text-based alerting system in other units of the hospital and for other types of orders. Given the large number of restraint orders placed in the adult ICU, this is the best next site to deploy this intervention, and our institution plans to do so in the next year.
Considering that the study by Jacobs et al. (2010) demonstrated improved provider satisfaction with stat radiology and respiratory treatment pages, interviews should also be conducted with nurses and providers to evaluate their perception of the new alerting system.
Efforts can also be made to see if the text-based alerting system reduces the overall number of restraint orders. Previous work has demonstrated reduced pediatric restraint use on the hospitalist service through improvements in team based communication, medication order sets, and individualized therapy plans, but not on whether more frequent provider notification reduces restraint use. 35
Finally, examining variations between provider types in ordering restraints could provide insights into potential improvements in provider training.
Conclusion
Automated alerting systems for lapsed orders may improve provider response time as compared to having a nurse page and speak to the provider over the phone. Institutions may consider automatic alerting systems for similar routine orders that need provider attention while carefully balancing this with provider alarm fatigue and satisfaction.
Footnotes
Acknowledgements
We thank the AC3 committee at our institution for reviewing our work and implementing changes.
Author contributions
Jessica Pourian: Data curation, Formal Analysis, Software, Visualization, Writing – original draft, Writing – review & editing; Aris Oates: Conceptualization, Data curation, Formal Analysis, Supervision, Writing – original draft, Writing – review & editing.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The data underlying this study contain identifying patient information linked to clinical orders and are not publicly available due to privacy and confidentiality concerns. Access to the data is restricted in accordance with institutional and ethical guidelines to protect patient privacy.