Using nonlinear dynamic analysis to differentiate fall status in older women

Background: Falls are a significant health concern among older adults. Nonlinear dynamic (NLD) analysis of gait offers insight into fall risk by capturing variability and complexity, but variation in methodological approaches has limited translation. This study aimed to identify NLD measures and data sources that best differentiate fallers from non-fallers.
Methods: Thirty-four healthy older women (mean age 69.3 ± 5.7 years; 17 fallers, 17 non-fallers) walked on a treadmill at preferred and at ±20 %. Kinematic data were collected using motion capture and a lower-back inertial measurement unit (IMU). Gait complexity and stability were quantified using Multiscale Entropy and Lyapunov Exponents (LyE). Principal component analysis, logistic regression, multivariate tests, ROC curves, and linear discriminant analysis (LDA) identified discriminative features.
Results: Fallers reported at least one fall in the past year, walked more slowly, and had a greater chronic disease burden. Short-term LyE (SLyE) from trunk acceleration in the anterior–posterior (AP) direction and sagittal-plane ankle angles best discriminated fall status. ROC analyses showed ankle SLyE provided the highest accuracy (AUC up to 0.88), and AP trunk SLyE had moderate accuracy (AUC up to 0.77). The LDA model achieved 85 % cross-validated accuracy with 82 % sensitivity and 88 % specificity.
Conclusions: The short-term Lyapunov exponent from ankle angle sagittal-plane motion and trunk AP acceleration provide robust markers of fall history in older women. Comparable performance of IMU and motion capture supports IMU-based NLD metrics for scalable fall risk screening.