Toward Healthy Aging: Temporal Regression for Disability Prediction and Warning Decision-Making.

Virtually all countries in the world are experiencing growth in the number and proportion of seniors in their population. Almost half of these seniors live with one or more disabling conditions. This highlights the concern about when, and how probably, a disability is likely to occur in aging people. In this paper, we mathematicize this concern as a prediction and a warning of the onset of disability. As such, we propose to start by transforming the fitting problem into a series of independent survival learning and prediction problems. Our approach can use all repeated measures of disability-specific factors and, more importantly, effectively quantify their different impacts on the onset of disability. We also present a new approach for estimating the time of onset and determining the better-timed warning of disability onset. To evaluate our time predictions and warning decisions, we develop four evaluation metrics based on the criteria we explore for the aging study. The results of comparative experiments and ablation studies on the elderly cohorts across Canada demonstrate the effectiveness of our approach.