Behavior Modeling Approach for Forecasting Physical Functioning of People with Multiple Sclerosis

Abstract

Forecasting physical functioning of people with Multiple Sclerosis (MS) can inform timely clinical interventions and accurate "day planning" to improve their well-being. However, people's physical functioning often remains unchecked in between infrequent clinical visits, leading to numerous negative healthcare outcomes. Existing Machine Learning (ML) models trained on in-situ data collected outside of clinical settings (e.g., in people's homes) predict which people are currently experiencing low functioning. However, they do not forecast if and when people's symptoms and behaviors will negatively impact their functioning in the future. Here, we present a computational behavior model that formalizes clinical knowledge about MS to forecast people's end-of-day physical functioning in advance to support timely interventions. Our model outperformed existing ML baselines in a series of quantitative validation experiments. We showed that our model captured clinical knowledge about MS using qualitative visual model exploration in different "what-if" scenarios. Our work enables future behavior-aware interfaces that deliver just-in-time clinical interventions and aid in "day planning" and "activity pacing".