A mobile sensing system to monitor symptoms during chemotherapy

Abstract The aims of the parent R37 grant were (1) to develop and refine a mobile sensing system that applies machine learning to smartphone and wearable sensor data to passively detect symptom burden during chemotherapy and (2) to evaluate the feasibility and acceptability of using this system in an academic oncology clinic. Since the start of the R37 award on July 1, 2019, we have enrolled 154 (77% of our target n = 200) patients into a 90-day study assessing daily and weekly symptoms and continuous behavioral sensing via wearable device and smartphone during chemotherapy, conducted interim analyses demonstrating that next-day overall symptom burden can be predicted from passive sensor data, and developed infrastructure to deploy symptom prediction models in real-time. The next logical step for the original R37 research is to integrate sensor- and machine learning- based predictions into a patient-centered digital health intervention to personalize supportive cancer care. The two-year extension will enable us to design and develop a web application that shares real-time symptom predictions with patients and uses them to trigger personalized symptom management instructions. To provide useful recommendations, we also need to develop prediction models that can predict specific, clinically significant and actionable symptoms rather than overall symptom burden. In addition, we need to determine the best way to share machine learning-based predictions and recommendations in a way that fosters trust and that supports patient autonomy as well as patient-provider communication. The R37 extension will enable us to address two new aims: (1) develop additional machine learning models to predict specific clinically meaningful and actionable symptoms and (2) design and develop a web application that integrates these predictions and uses them to inform real-time symptom management. This work falls well within the scope of the parent grant and represents continued refinement of both the clinical utility of our machine learning models as well as our integration of these models into supportive cancer care within an academic oncology clinic.