Actigraphy features for predicting mobility disability in older adults.

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Actigraphy is widely used for physical activity assessment.
• Existing algorithms lack general models for predicting mobility function.
• Detecting mobility impairment and major mobility disability (MMD) in older adults requires targeted approaches.

Purpose of the Study:

• To develop a general model using actigraphy features to predict mobility impairment and MMD.
• To identify key actigraphy-derived features indicative of mobility decline in older adults.
• To assess the accuracy of these features in predicting mobility function.

Main Methods:

• Collected 8.4-day hip-worn tri-axial accelerometer data from 1135 older adults (70-89 years).
• Summarized data into 67 features and applied machine learning techniques for feature selection.
• Predicted mobility impairment (400m walk speed < 0.80 m/s) and MMD using selected features.

Main Results:

• Actigraphy models estimated 400m walk speed with ~0.07 m/s RMSE and R-squared of 0.37-0.41.
• Sensitivity and specificity for identifying slow walkers were ~70% and ~80%, respectively.
• Top features related to activity pace, amount, bout length, accumulation, and variability significantly improved MMD prediction.

Conclusions:

• A subset of actigraphy features from free-living conditions moderately identifies mobility-impaired individuals.
• These features significantly enhance the prediction of major mobility disability (MMD) in older adults.
• Combinations of actigraphy features are crucial for predicting mobility phenotypes, outperforming individual features.