Predictive Models of Mortality Risk from Passive Monitors measuring Physical Activity
Principal Investigator:
Professor Bruce Schatz
Approved Research ID:
45178
Approval date:
October 29th 2019
Lay summary
Suppose it was possible to monitor daily living with clinical accuracy, without requiring that persons do anything special. Such facility would revolutionize the prediction and prevention of disease and mortality. There are clinical monitors that support passive monitoring, but they are expensive and hard to utilize. Various 'smart' devices offer the possibility of supporting this vision, such as smart watches or smart clothes. But today the only 'smart' device widely used is the smart phone. However, despite wide variety of health apps, smart phones rarely support passive monitoring with clinical quality. This is due to passive detection limited to motion sensors to rotate displays. Motion sensors could be used to detect walking patterns, which are a fundamental measure of clinical status. For example, a standard medical textbook states 'Watching a patient walk is the most important part of the neurological examination. Normal gait requires that many systems, including strength, sensation, and coordination, do function in an integrated fashion.' It is well known that gait speed is closely correlated with mortality, so the walking pace of a person slows down as they move closer to death. The UK Biobank mortality risk study concluded 'measures that can simply be obtained by verbal interview without physical examination are the strongest predictors of all-cause mortality in middle-aged to elderly individuals. Self-reported health and walking pace were the strongest predictors in both sexes and across different causes of deaths.' The Principal Investigator has pioneered identifying walking from passive monitors during daily living, then utilizing characteristic patterns with predictive models for health status of cardiopulmonary diseases. Here we propose to predict 5-year mortality using statistical models based only upon motion datasets from smart devices. We propose to validate such predictions with longitudinal datasets from the physical activity study of the UK Biobank with 100K participants. We have already successfully analyzed a similar dataset from the US Women's Health Initiative with 5K participants, predicting risk of falling, which core stability is closely correlated with mortality risk. Since such models could then be implemented at mass scale on cheap phones, this would revolutionize the practice of healthcare.