Classifying ambulation patterns in institutional settings

Abstract

Observational studies of older adults show pacing, lapping, and stationary ambulation patterns can be associated with an increased risks for falls or an early sign of an acute or other health event in long-term care. The aim of this study is to use classical machine learning algorithms to automate the process of recognizing these patterns with the goal of assisting health care staff in monitoring the health and well-being of their residents. This study utilized data from six residents whose movements were tracked with a real-time locating system while performing everyday activities of daily living for up to 1.9 years. No residents exhibited lapping patterns over the course of the study. Machine learning statistical techniques recognized stationary and pacing with accuracy≥70%, with indirect and direct patterns having an accuracy of around 50% due to environmental constraints. Study findings suggest automated methods may be used with real-time locating data to recognize ambulation patterns that have been associated with poor health in this population. Study findings may be utilized by health care staff to tailor resident care plans and develop timely interventions that may affect falls and provide for the earlier recognition of acute events in this population.