Daily activity patterns from wearable accelerometry predict physical frailty and concern about falling.

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Physical frailty and concern about falling are significant geriatric conditions impacting mobility and quality of life in older adults.
• Current assessment methods (e.g., Fried Frailty Phenotype, Falls Efficacy Scale-International) are subjective, require professional administration, and do not capture the interplay between frailty and fall concern.

Purpose of the Study:

• To develop and validate a wearable-sensor-based framework for simultaneous, objective prediction of physical frailty and concern about falling.
• To analyze real-world activity patterns for insights into these interrelated geriatric conditions.

Main Methods:

• Collected 48-hour continuous chest acceleration data from 146 participants using a pendant sensor.
• Transformed activity sequences into barcodes and analyzed using global complexity metrics (entropy) and local sequential dynamics (bidirectional LSTM networks).
• Employed a multi-task deep learning model with attention mechanisms for joint prediction.

Main Results:

• Achieved high predictive performance for both physical frailty (F1 score: 93.12%) and concern about falling (F1 score: 86.27%).
• Demonstrated the feasibility of using long-term accelerometry for joint modeling of these conditions.

Conclusions:

• The proposed wearable-sensor framework offers an objective and accurate method for assessing physical frailty and concern about falling.
• This approach provides a novel way to understand the mutual influence of these conditions in older adults using real-world data.