Eliciting Design Requirements for a Computer-Aided Monitoring and Alerting System to Improve PPE Compliance among Healthcare Workers

Abstract

We examine the use of personal protective equipment (PPE) in an emergency medical setting to inform the design of a computer-aided PPE monitoring and alerting system. We reviewed videos of 19 pediatric resuscitations from the early months of the COVID-19 pandemic, finding that PPE violations were associated with specific phases in the patient care process (e.g., before patient arrival), specific tasks (e.g., medication preparation), and specific locations (e.g., at the room entrance, bedside). Our results suggest that alert design will require considering the urgency of correcting PPE violations vs. the urgency of the task being performed. We conclude by discussing our initial observations and the next steps.

1. Introduction

During the COVID-19 pandemic, front-line healthcare workers (HCWs) have had a higher than three-fold increase in infection risk compared to the general public [13]. Between March 1, 2020, and March 5, 2023, as many as 1,123,878 healthcare workers have become infected with SARS-CoV-2 in the United States alone [5]. Despite widely adopted guidance on the proper use of personal protective equipment (PPE) issued by the U.S. Centers for Disease Control and Prevention (CDC), compliance with PPE guidelines among healthcare personnel has been low [12] [19]. Several factors can affect the proper use of PPE in healthcare settings, including frequent changes in public health guidance and local hospital approaches [14], the perceived risk of infection [12], workplace stress and fatigue [18], and PPE availability [1]. For example, changes in institutional policies on PPE use in response to non-routine events such as pandemics are often poorly communicated to healthcare workers, leading to misunderstanding, confusion, and poor compliance [14].

To ensure compliance with local PPE guidelines, the CDC has recommended deploying trained observers to guide the donning and doffing of PPE by individual HCWs [4]. In response, many hospitals have deployed infection control nurses and “PPE spotters” to conduct audits of PPE use through observation [14]. Auditors are instructed to move around units and document whether gloves and gowns are worn for the appropriate tasks. These methods, however, are labor-intensive, susceptible to human error, and challenging to implement in crowded and time-critical settings [2].

As part of our long-term research, we are studying PPE use and monitoring practices in settings at high risk for virus transmission to inform the design of a computer-aided PPE monitoring and alerting system (Figure 1). We hypothesize that (1) HCWs are mostly unaware of episodes of PPE noncompliance, and (2) alerts provided by computer-aided monitoring will reduce noncompliance by increasing awareness of these episodes. The system will use machine learning and inputs from multiple cameras to track individuals over time and detect absent or inadequate PPE. A remote human operator observing the scene can verify the system detections before notifying the healthcare worker (Figure 1).

Figure 1:

A sketch of our computer-aided PPE monitoring system with a human-in-the-loop that will identify and alert noncompliant individuals.

In this paper, we focus on eliciting design requirements for potential alerting mechanisms initiated by the human operator. We describe the results from a video review of 19 emergency medical events from the early months of the COVID-19 pandemic. PPE use in emergency medical settings is more challenging than in other hospital units because of the urgency of care and dynamic team membership. A study of PPE compliance in the first few months of the pandemic found higher compliance with PPE guidelines in the COVID-19 wards (85% total compliance) than in the non-COVID-19 wards (76% total compliance) [12]. Even before the pandemic, compliance with recommended PPE use was suboptimal in settings such as resuscitation areas and intensive care units [11]. The goal of our video review was to obtain initial insights into noncompliant PPE use and the approaches used to correct PPE violations. These preliminary findings will inform our subsequent interviews and ideation sessions with users to discuss and co-design potential alerting mechanisms.

2. Related work: Noncompliance Alerts in Clinical Setting

Prior work has evaluated audiovisual monitoring and overhead auditory alerts by remote human operators for enhancing PPE compliance [6] [16]. These initial solutions involved a remote observer watching the donning and doffing process and alerting HCWs of PPE violations using a simple two-way audio-visual communication system equipped with microphones and cameras. This approach, however, excludes the intervals between donning and doffing and is prone to human error.

Existing hand hygiene surveillance systems that provide alerts to HCWs when hand hygiene is skipped are also of interest. These systems use different alert modalities, including personal bracelets that vibrate when noncompliance occurs [10], as well as visual and auditory reminders like beeps, red-flashing lights, and pre-recorded voice prompts [17]. Evaluation studies assessing the impact of these systems have shown that any alert mechanisms that target noncompliance must be highly accurate to succeed. Similar suggestions emerged from the literature on alert fatigue. A healthcare worker may be exposed to over 1,000 alarms per shift, up to 99% of which may be non-actionable, leading to cognitive overload and desensitization [15]. A recent study found that the likelihood of an HCW accepting best-practice reminders dropped with increases in the number of reminders, number of repeated reminders for the same patient or noncompliance, and patient complexity [3].

Most PPE monitoring systems have been evaluated in industrial and healthcare contexts, focusing on approaches for PPE detection (e.g., RFID technology, computer vision) and detection accuracy. To our knowledge, no current PPE monitoring system has implemented computer-aided, real-time alerts.

3. Initial Video Review to Analyze PPE Use Practices

3.1. Methods

We conducted a video review of 19 pediatric resuscitations performed in the emergency department (ED) of a level 1 trauma center in the mid-Atlantic region during the first three months of the COVID-19 pandemic (March-June 2020). Resuscitations were video recorded with three cameras, each providing a different view (overhead view, head view, and foot-of-the-bed view). Parental or guardian consent was required for reviewing videos for research purposes. The study was approved by the hospital’s Institutional Review Board.

We selected all available resuscitations with one or more aerosol generating procedures (AGPs) because these procedures generate aerosols that expose HCWs to respiratory pathogens causing acute respiratory infections [9]. On average, the resuscitations were 40 minutes long (SD = 18.14). Before the pandemic, PPE requirements for ED resuscitations at our site included only gloves and a gown. After the pandemic started, hospital guidelines for PPE use in the ED also required headwear (e.g., scrub cap), eyewear (e.g., goggles, face shield), and a N-95 facemask or powered air purifying respirator (PAPR) [2]. Trained coders on our team identified PPE items in all 19 videos and assigned a compliance level (complete, incomplete, absent) to each. This ground-truth coding was based on a PPE data dictionary developed by our team.

For all instances of PPE noncompliance (incomplete or absent PPE), we noted the following contextual information: timestamp, team role of the noncompliant HCW (e.g., charge nurse, bedside provider), activity or task performed at the time of noncompliance, if the HCW performed any action to correct their PPE violation, if another team member or PPE observer notified the noncompliant HCW of the PPE violation, and any other changes in PPE use. Two researchers reviewed the videos using a two-step process. First, both researchers independently reviewed five of the 19 videos. We compared the initial observations as a group, resolving any discrepancies by re-reviewing the videos and correcting the notes. We used this process to establish a shared understanding of PPE practices in resuscitations and to develop a note-taking protocol for video review. In the second step, one researcher reviewed the remaining 14 videos following the established note-taking protocol. We used a deductive approach to qualitative content analysis, coding all notes based on the predetermined categories of contextual information. We compared the contents of all categories and identified several major themes and associations.

3.2. Findings

Three themes emerged from our preliminary analysis: (1) PPE readiness level, (2) types of PPE violations, and (3) types of PPE violation correcting behaviors.

PPE Readiness Level.

After being summoned to treat a severely sick or ill patient in the ED, most HCWs entered the resuscitation room with properly worn masks, eyewear, and headwear. Some providers had their masks pulled down below the chin while waiting for the patient but pulled them up when the patient arrived. We did not observe many HCWs wearing gowns or gloves upon entering the room. They instead started donning gowns in the entrance area and then put on their gloves after their gowns. Most HCWs properly secured their gowns, covering the entire torso, arms, and upper legs. Some, however, did not tie their gown at the neck, leaving their shoulders exposed. The following excerpts from our notes illustrate these practices:

“(22:31:30) woman (pink scrubs) enters with no gloves or gown, goes to get gown and puts it on, only ties at neck; takes off the mask to double-mask and puts on gloves at 22:36:26.”

“(16:51:40) nurse enters with no gloves; leaves briefly and returns still putting on gloves at 16:51:53.”

Types of PPE Violations.

We observed several types of PPE violations during resuscitations. The most frequent PPE violation was associated with inadequate gown use. Because HCWs usually donned gowns in a rush after entering the room, they did not always tie their gowns at the neck and waist before approaching the patient bed. During patient care, these untied gowns were slipping off shoulders or sliding down the arms, obstructing HCWs’ work. For example:

“(17:54:54) Male nurse lifts falling gown off shoulders but does not tie it, gown falls off again.”

Second, we observed HCWs taking off one or both gloves when preparing medications at the counter and interacting with the medication dispensing system. Most HCWs put on new gloves after finishing work at the counter, but some did not. Third, the use of eyewear and headwear was inconsistent. Some providers put on head- and eyewear and some did not, especially if also wearing glasses. Finally, we observed multiple instances of masks being pulled down or not worn at the bedside.

Types of PPE Violation Correcting Behaviors.

We identified three approaches to addressing PPE noncompliance: self-correction, peer intervention, and no correction. First, after realizing their PPE use was inadequate, some HCWs attempted correcting the PPE items, like lifting their gowns or pulling up the masks:

“(17:58:02) Nurse [right] gown completely falls off shoulder, she lifts it back up after a moment but does not tie; lifts other falling shoulder at 18:03:49.”

Second, we observed instances of another HCW noticing noncompliant PPE use and helping to correct it. For example:

“(20:14:16) Man (blue scrubs) gown falling off shoulders; the woman next to him lifts one side of gown onto his one shoulder; gown falls off and the nurse makes another attempt to pull it up for again, but then walks away; man lifts his gown but doesn’t tie it and it just falls again.”

These assisting behaviors mostly occurred during brief idle moments, when other HCWs were taking over tasks or working on the patient. However, most instances of inadequate gown use were left uncorrected, as HCWs rushed to complete their tasks and attend to patients. Inadequate use of masks and gloves was usually corrected. In one case, a nurse took a box of N-95 masks and handed them out to several HCWs who entered the room without masks.

3.3. Discussion

Our initial observations suggest several implications for designing a computer-aided monitoring and alerting system to address PPE noncompliance among HCWs in emergency medical settings. Although PPE violations will be detected using computer vision, our results provide contextual information for augmenting computer vision techniques. For example, we found that an incomplete or absent gown was associated with the pre-arrival phase of the care process and frequently occurred when HCWs were at the bedside. In contrast, incomplete or absent gloves were associated with the medication preparation task at the medication counter.

Designing effective alerting mechanisms will require alternative approaches. Our results show that some HCWs were not able to correct their own PPE violations after becoming aware of them. We observed many HCWs who kept working on their tasks, even as the falling gowns started limiting their movement. While other HCWs helped correct PPE violations by pulling up the gowns of their coworkers, these corrections were done in passing and did not always achieve full or lasting compliance. These results suggest that PPE alert designs will require considering the urgency of correcting PPE violations vs. the urgency of the task being performed. Audio-visual alerts at the time of donning or doffing or when hand hygiene is skipped are usually received before work and in quiet and less crowded areas. In contrast, PPE compliance alerts will be received during dynamic medical work and will have to compete with other alarms and noise in the room. Prior work has already considered alarms based on peripheral interaction, like beepless pedals for silencing audio alarms and visual indicators for providing the system status [8]. Another approach sonified the patient’s vital signs from the monitor into harmonious, more meaningful musical notes [7], potentially mitigating alarm fatigue and offloading clinicians’ alarm-related work. As we continue our research, we will explore solutions like these that consider alerts in clinical settings in new and innovative ways.

4. Next Steps

Our next steps are aimed at further understanding PPE use and monitoring practices and identifying effective alerting mechanisms for PPE noncompliance. We will (1) interview nurses and other PPE observers to supplement our initial observations and learn more about current practices by which HCWs are monitored and alerted for noncompliance, (2) conduct longitudinal video review of resuscitations during COVID-19 surges and retreats to analyze changes in PPE use over time, and (3) conduct a series of workshops with HCWs to identify and co-design different alerting mechanisms.

CCS CONCEPTS.

• Human-centered computing • Human-computer interaction • Empirical studies in HCI