A comparison of self-triage tools to nurse driven triage in the emergency department

INTRODUCTION

When patients present to the emergency department (ED), triage serves as their gateway to receiving care. Triage involves the patient being assessed by a specially trained registered nurse, and the process results in a score which assigns a level of acuity and urgency to a given patient’s presenting complaint. The Canadian Triage and Acuity Scale (CTAS) is the reference standard for this output in Canada. The current landscape of emergency medicine is mired by prolonged wait times, which are known to have negative patient outcomes. [1 – 2] Incorporating the patient into the triage process may serve to enhance its efficiency.

This notion of self-triage has been studied previously with success in specific populations. [3 – 5] However, this approach may be fraught with challenges owing to individual health literacy. Previous literature would suggest that laypersons may be overcautious in deciding their need to seek medical care, or miss identifying potential emergencies. [6] Indeed, laypersons have also been previously demonstrated to not follow prescribed advice when directed to either present, or not present, to an ED to seek care. [6 – 8] We previously sought to compare a computer-assisted self-triage tool to the traditional method of triage and found low rates of agreement and high rates of over triage, a result consistent with other literature. [4, 6, 9]

In this study, we aimed to build on the existing knowledge base and further our previous work by assessing the ability for patients to self-triage. We based this on four specific presenting complaints: chest pain, shortness of breath, abdominal pain and musculoskeletal pain. We tested this through two different methods – an algorithm-generated self-triage (AGST) score, and a visual analogue scale (VAS) based score. In an attempt to see if this is a viable option for the evolution of triage, we aimed to compare the accuracy of these tools against the gold standard of nurse driven triage with respect to their resulting CTAS score.

METHODS

RESULTS

A total of 279 participants were enrolled in the study and completed the self-triage questionnaire. Of this group, 39 participants were excluded as their self-reported chief complaint did not match their actual chief complaint. An additional 17 participants were excluded as they had left prior to the formal triage process, and so no actual CTAS score could be determined. Ultimately, 223 participants were included for analysis. The emergency department of Royal University Hospital provided the majority of the data, as the emergency department of Saint Paul’s Hospital was often empty during the months in which it was attended or devoid of patients whose conditions were appropriate for the study.

With respect to the different forms of our self-triage tools, a total of 159 participants completed the VAS based tool, and 142 participants completed the AGST survey. Of these participants, 78 completed both self-triage tools. Of the four study categories, chest concerns comprised 14.3% (n = 32) of the responses, breathing concerns comprised 11.2% (n = 25) of the responses, abdominal concerns comprised 33.4% (n = 75) of the responses and musculoskeletal concerns comprised 40.8% (n = 91) of the responses.

Table 2 outlines the descriptive statistics of our study. Table 3 displays the concordance of the AGST tool with the nurse-driven triage result, and Table 4 displays the same but with the VAS tool. Table 5 presents the data of the subgroup which completed both the AGST and VAS tools.

DISCUSSION

In this study, we endeavored to see if ambulatory patients who presented to the ED could use our proprietary self-triage tools to similar acuity levels in comparison to the reference standard of nurse-driven triage. Our intent with this study was to see if the triage process could evolve in a patient-centered format. Our results demonstrate that, in their current form, both the AGST survey and VAS tools are not appropriate for this purpose. In the case of participants who individually completed either tool, agreement was non-existent, minimal, or weak at best. Though the existing literature has demonstrated a high degree of over-triage, in our study patients both over- and under-triaged themselves. [4, 6, 9] The AGST survey tool appeared to slightly outperform the VAS scale, including in the group of participants who completed both self-triage tools. Though it is important to understand the subjective experience of the patient with respect to treating their symptoms and managing their expectations in a given ED visit, it is likely that the subjectivity of the responses from the VAS score influenced the VAS tool’s relative underperformance.

We had hoped that by this degree of patient involvement in triage, the triage component in ED-experience for patients could be optimized. Self-assessments have previously demonstrated some success with triage when they are honed down to specific complaints. [3,5,11] One particular ED based study looked at patients presenting with chest pain used a similar “yes” or “no” based approach like our AGST tool and had high levels of agreement with low risk to patient safety. [11] The success of these studies raises the question of how focused, complaint – specific tools may be utilized to optimize triage overall. Kiosk based triage, though largely focused on early registration of patients, has been studied in terms of optimizing efficiency in triage. [12-13] In considering this, it follows that combining kiosk based triage with the development of individual, complaint-focused tools may represent an area for how triage can evolve.

Innovation in triage is not limited to this, however, as other technologies have potential for use. Previously, online-symptom checkers have been studied with respect to their overall diagnostic and triage accuracy, but these have similarly been faced with challenges surrounding accuracy, and rates of over or under triage. [14 – 16] Artificial intelligence may also have uses in triage, with machine learning models having been demonstrated as having some success with specific presenting complaints. [17 – 19] The possibilities of how triage can evolve can take a variety of different forms, and these merit further investigation.

Although our study had a negative result, we are confident that it provides additional context to studying the future of ED triage. Our study design was strengthened by the fact that we tested two different self-triage tools and included a group where patients completed both tools, comparing their accuracy against each other with a controlled userbase. Further, we had involved patient and family advisors in the development process so as to capture their insights. We did, however, have some limitations. Bias was likely introduced through our recruitment as it was a convenience sample during daytime and weekday hours. Given how the majority of our patients came from the Royal University Hospital, bias was also inherently introduced as we did not capture a significant population from our inner-city hospital. As well, although we had stated that participation in our study would not result in any expedited assessments in the ED, patients may have still intentionally over-triaged themselves in order to avoid waiting for a prolonged period. Finally, though nurse-driven triage was our reference standard, this is an imperfect system with established variability. [20 – 21]

Ultimately, our study showed that self-triage through use of an AGST or VAS tool does not appear to be a viable option for use in the ED. Further studies, including those using focused questionnaires around specific presenting complaints, or those leveraging existing technology, are required to optimize the triage process for ED patients.