Integrating Treatment Notes and Smart Home Sensor Data for Adverse Event Prediction

This session presents secondary findings from a large randomized clinical trial. Fifty eight older adults with disabilites living in rural Missouri were randomized to either a health education control or a self-management telehealth intervention for a year. All participants had a smart home sensor system installed in their home. Any adverse events experienced by study participants were collected via self-report at quarterly check ins. This analysis integrates data from the treatment notes formated using SOAP, sensor system data, and adverse events. Several large langauge models were integrated to analyze SOAP notes and adverse event correlations were calculated. For both groups, text from the Subjective component of the SOAP notes were highly correlated and most predictive of adverse events.