|
Abstract:
|
By allowing the collection of objective repeated measurements concerning various aspects of patient behavior, such as mobility and sociability, digital phenotyping plays a remarkable role for researchers interested in studying post-operative surgical recovery without the primary use of patient reported outcome measures (PROMs). Using these measurements, researchers can find temporal correlations with typical findings associated with PROMs and can, more importantly, uncouple previously unknown traits regarding patient recovery. In this talk, we focus on using digital phenotyping in order to learn behavioral patterns and treatment strategies that improve patient recovery after a surgical or clinical intervention. The methods we will discuss leverage the use of data that has been sampled from smartphones sensors in order to retrieve metrics that capture a patient’s physical activity, mobility, and daily pain response. Specifically, we present our work in developing interpretable offline reinforcement learning methods in order to estimate recovery strategies for surgical patients and predicting post-operative recovery patterns in temporally dense, longitudinal data.
|
