Accelerometer-based predictive models of fall risk in older women: a pilot study

Area of Science:

• Gerontology
• Biomedical Engineering
• Physical Therapy

Background:

• Clinical fall risk screening in older adults faces challenges.
• Objective physical activity and cardiovascular health (OPACH) study data were analyzed.
• Existing methods lack precision in identifying individuals prone to falls.

Purpose of the Study:

• To evaluate the effectiveness of accelerometer-based gait analysis for fall risk screening in older women.
• To compare machine learning model performance using traditional and signal-based gait features.
• To identify key gait parameters predictive of fall risk.

Main Methods:

• Data from 67 women (mean age 77.5 years) from the OPACH study were used.
• Participants underwent the Short Physical Performance Battery (SPPB), fall history questionnaires, and a 400-m walk with hip-worn triaxial accelerometers.
• Random Forests models were trained to classify high fall risk (SPPB ≤9 and ≥1 fall) versus low fall risk (SPPB 10-12 and 0 falls).

Main Results:

• The best model, incorporating triaxial accelerometer data, cross-correlations, and traditional gait measures, achieved 78.9% accuracy, 84.4% precision, and 0.846 AUC.
• Top predictive features included mediolateral signal-based measures: coefficient of variance, cross-correlation with anteroposterior accelerations, and mean acceleration.
• Classification accuracy demonstrated the potential of accelerometer data for fall risk assessment.

Conclusions:

• Accelerometer-based gait analysis shows promise for screening older women for fall risk.
• Signal-based features from accelerometers can effectively augment traditional gait measures.
• Further application on a larger dataset with prospective fall data is underway to validate these findings.