A clinician-in-the-loop smart environment to support management of chronic health conditions

PROJECT SUMMARY / ABSTRACT As the population of individuals 65+ grows, so too will the number of individuals who struggle to manage chronic health conditions. Exacerbations, or sudden symptom worsening, of chronic conditions place a burden on the health care system and cause decreased quality of life and accelerated decline. Forecasting health exacerbations can prevent life-threatening events or limit their impact. The long-term goal of this work is to create a clinician-in-the-loop (CIL) smart environment that empowers individuals in managing their chronic health conditions. Our proposed CIL framework partners clinical expertise with pervasive computing and machine learning. In this system, sensor data are collected continuously by ambient and wearable sensors. Our ML algorithms extract behavior markers to show to a clinician. Behavior changes are detected due to internal (i.e., condition exacerbation) or external (i.e., wildfire smoke, shutdown due to COVID-19 pandemic) events. Trained by clinicians, prediction of future condition exacerbations and health states are generated with corresponding reliability scores. Correspondingly, clinicians will provide summaries of observed behavior, reasons to confirm likely exacerbations, and recommendations to prevent health events. These summaries are used to train a language model that will provide interactive explanations of future data and ML predictions. Data are collected continuously by ambient sensors in the smart home and wearable sensors in Apple watches to validate the findings and provide more complete monitoring. From smartwatch sensors, we will analyze the predictive relationship between external events, social determinants of health, socialization, environment quality, behavior, and health state. The result of these steps is an automated technology that can assist with self-management of chronic conditions and monitor the impact of interventions. The proposed aims will be validated using data collected in our prior study for n=44 older adults and in a new data collection with n=20 older adults. Participants in the restrospective study with historic data and prospective study with new data will be older adults age 50+ living in independent homes who are managing one or more chronic health conditions. Expanding the smart environment to encompass ambient sensors and wearable sensors increases the accessibility of the technology for underserved communities and, when used in combination, improves model robustness through joint prediction. Although clinical oversight of patient health will always be valuable, the technology can provide an “informatics triage” that allows a clinician to remotely monitor a greater number of individuals at a time and provide valuable information to the individual to assist in self-management. Outcomes of this proposed study include open-source software for forecasting condition exacerbations, software to train language models from clinician input, and the corresponding trained models; results of analyses linking external influences, socialization, behavior, and health; apps for collecting ambient and wearable data; and data that will be prepped for integration into the NIH RWDP platform.