Natural language processing chatbot and continuous activity monitoring in a phase II randomized trial: Impact on hospitalizations and quality of life.

12132 Background: Toxicities associated with concurrent chemoradiotherapy (CRT) can lead to hospitalizations and impaired quality of life (QoL). This study reports on secondary endpoints investigating whether combining continuous activity monitoring (CAM) with an AI-driven chatbot can support symptom identification, triage, reduce hospitalizations, and improve QoL. Methods: Patients receiving CRT for head and neck (H&N), gastrointestinal (GI) or thoracic cancers were recruited from radiation oncology clinics at a single, urban institution. Subjects were randomized to: (1) CAM with nursing triage or (2) CAM with chatbot triage with as needed symptom assessments. A Fitbit device monitored the step count and heart rate (HR) relative to baseline. Triage was triggered when the daily average HR was >100 or 20% above their resting average or if there was a decrease of >1000 steps from baseline. In the control arm, triage prompted a nursing assessment within 1 business day. In the intervention arm, triage visits were triggered based on a low-, intermediate-, or high-risk symptom algorithm determined by the chatbot (next OTV, next business day, within 24 hrs, respectively). Hospitalizations during CRT and one month post treatment were counted. Logistic regression controlled for ECOG and sex. EORTC C30 QoL instruments were administered at baseline, during OTVs, and during follow-up. A mixed method for repeated measures model assessed global health status/QoL, fatigue, and physical, role, emotional, cognitive, and social functioning scales adjusting for ECOG, race, and stage. All statistical tests were completed on SAS 9.4. Results: From June 2023 to January 2025, 77 patients undergoing CRT for their H&N (n=30, 39%), GI (n=29, 38%), or thoracic cancers (n=18, 23%) were enrolled. The median age at consent was 59 years (range 28-81 years). Subjects were randomized to either nursing (N = 36, 50%) or chatbot triage (N = 36, 50%). There were no statistically significant demographic or clinical differences between arms. Subjects in the chatbot arm had significantly fewer hospitalizations (11%, n=4 chatbot; 39%, n=14 nursing triage, p=0.01). On logistic regression, subjects on the chatbot arm (p=0.0125) and those who had ECOG 0 (p=0.0131) were less likely to be hospitalized. In both arms, global QoL/health, fatigue, and physical, social, cognitive, and emotional functioning declined during CRT. Chatbot subjects had significantly inferior global QoL/Health status (p=0.049), cognitive functioning (p=0.005), and physical functioning (p=0.02) vs nursing triage. Conclusions: Patients undergoing CRT experience functional decline. While AI-driven chatbots may reduce hospitalizations, they are not a perfect substitute for a compassionate healthcare team. An analysis of alerts and triage visits will be reported with our primary endpoint.