A machine learning model for frailty based on wearable device measurements.

Area of Science:

• Gerontology and Biomedical Engineering
• Machine Learning Applications in Healthcare

Background:

• Frailty is a significant aging factor linked to adverse outcomes.
• Current frailty metrics are time-consuming, limiting large-scale use.

Purpose of the Study:

• To apply machine learning for predicting frailty metrics and adverse outcomes using wearable activity data.
• To evaluate the efficacy of machine learning models in assessing health risks in geriatric populations.

Main Methods:

• Utilized Actigraphy wearable accelerometer sensors to collect movement data.
• Employed machine learning models to predict frailty metrics, risk factors, and adverse outcomes.
• Evaluated model performance using AUC, AUPRC, and various statistical tests on subsampled data.

Main Results:

• Machine learning models demonstrated strong predictive performance with limited accelerometry data.
• Models accurately predicted adverse outcomes like hospitalization and mortality in geriatric patients.
• The predictive power for adverse outcomes surpassed that of traditional frailty metrics.

Conclusions:

• Wearable activity data-based frailty prediction provides a scalable surrogate for traditional metrics.
• This approach enables the prediction of adverse outcomes, facilitating frailty assessment in broader studies and clinical practice.