Stepping Beyond Assessment: Fall Risk Prediction Models Among Older Adults from Cumulative Change in Gait Parameter Estimates.

Falls significantly affect the health of older adults. Injuries sustained through falls have long-term consequences on the ability to live independently and age in place, and are the leading cause of injury death in the United States for seniors. Early fall risk detection provides an important opportunity for prospective intervention by healthcare providers and home caregivers. In-home depth sensor technologies have been developed for real-time fall detection and gait parameter estimation including walking speed, the sixth vital sign, which has been shown to correlate with the risk of falling. This study evaluates the use of supervised classification for estimating fall risk from cumulative changes in gait parameter estimates as captured by 3D depth sensors placed within the homes of older adult participants. Using recall as the primary metric for model success rate due to the severity of fall injuries sustained by false negatives, we demonstrate an enhancement of assessing fall risk with univariate logistic regression using multivariate logistic regression, support vector, and hierarchical tree-based modeling techniques by an improvement of 18.80%, 31.78%, and 33.94%, respectively, in the 14 days preceding a fall event. Random forest and XGBoost models resulted in recall and precision scores of 0.805 compared to the best univariate regression model of Y-Entropy with a recall of 0.639 and precision of 0.527 for the 14-day window leading to a predicted fall event.