Patient-Reported Outcome+ Platform for Remote Symptom Management, Featuring Automated Triage System

Abstract

<h3>Purpose/Objective(s)</h3> Despite the promising results, symptom monitoring is not widely implemented in the clinic. This is largely due to the additional staffing that is needed to triage the increased patients' symptom reporting. We herein develop a PRO+ platform for remote symptom management, where the plus sign stands for automated triage pathways. This includes AI and evidence-based triage chatbot and severity classification algorithm, to reduce manual triage needs. <h3>Materials/Methods</h3> 22 most common symptoms were included in the triage system. For each symptom, triage pathways were built in chatbot format. Typically, a triage starts with onset questions, like "when did it start?", "how many episodes?", followed by PRO CTCAE evaluation, how daily life was affected, home medication and remedies patients used, associated symptoms, etc. For most questions, multiple choices were identified as potential answers for patients to choose from. I.e., 10 scales were provided for pain evaluation. Some questions offer sharing pictures or free text. A decision tree-based classification algorithm was developed to classify the reported symptoms to non-urgent, urgent, and emergent situations. Three metrics were used for ongoing evaluation during development. 1) Patient engagement rate. Volunteer patients used the system to report symptoms. If a chatbot conversation was initiated and finished, it was counted as a successful engagement. 2) Chatbot triage efficiency. Nurses reviewed the reported symptom details and marked the situations that enough information was collected to provide patients clinical advice. Corresponding percentage was calculated. Although we expect that there will always be cases that additional communication is needed to triage patients, we want to collect enough information to triage most straightforward cases. 3) Classification accuracy. The classification algorithm results were compared with manual results by nurses, and discrepancy was reported. <h3>Results</h3> The PRO+ platform was built. Patients can select any symptom to report. Once a symptom is selected, the chatbot will carry a conversation with patient to collect necessary information for triage. The chat bot conversation is adapted real-time to the reported symptoms and conversation. 150 volunteer patients reported symptoms, and the engagement rate was 91%. 80% of the time the chatbot had collected enough information for nurses to provide clinical advice. The severity classification discrepancy was 95%. <h3>Conclusion</h3> A PRO+ platform was built for symptom management. It has automated triage pathways that can potentially allow clinics to provide remote symptom monitoring with reduced manual triage needs. More study is needed to examine the chatbot questions, severity classification performance, and staffing model required before wide adoption in the clinic.