EVENTS

Models for clinical risk prediction are important tools for clinicians to understand the risks of adverse event decision-making, stratify patient populations, and enable shared decision-making on treatments. However, conventional models represent patient risk using fixed time and space for inputs and outputs, and following these patients out the door with remote sensing has remained limited in success. Significant gaps exist in developing adaptable models that represent patient status and risk across the vast heterogeneity in available data, and in updating patient representations across the continual collection of new data. This is particularly true when incorporating the growing breadth of wearable sensing health data where obtaining ground truth labeling associating wearable signals with clinical data and outcomes is particularly challenging. This talk will cover methods designed to bridge the gap, following the course of a hospitalization and recovery typically found in patients with diagnosed heart failure. The talk will first cover the need for time-varying, multimodal models of care with clinical data, and then discuss the challenges in both design and modeling of wearable sensing data intended to help track trajectories of recovery after discharge.