Traditional fall risk assessment in older adults relies on intrinsic indicators derived from demographic information, clinical assessments, and mobility tests. These measures encompass variables of mixed types, including continuous, categorical, and ordinal, and are predominantly skill-oriented assessments. Wearable sensors can provide objective gait descriptors, yet how to best integrate wearable data with intrinsic indicators for fall-risk prediction remains insufficiently studied. This study systematically compares intrinsic risk indicators with wearable-derived gait parameters for predicting fall risk in older adults. We analysed 163 participants (86 fallers, 77 non-fallers; mean age 82.6 ± 6.2 years) and evaluated thirteen feature-set combinations, namely intrinsic-only, wearable-only, and hybrid, using four classifiers logistic regression (LR), support vector machine (SVM), random forest (RF), and artificial neural network (ANN) under systematic cross-validation. Genetic algorithms were employed for feature selection. Mobility tests were the strongest intrinsic indicators (AUC 0.87 to 0.90, 95% CI 0.81; 0.94). Wearable-derived gait features alone yielded moderate discrimination (LR AUC 0.83, 95% CI 0.76; 0.88). Combining wearable features with intrinsic indicators consistently improved performance (AUC 0.87 to 0.93 across combinations). The optimal combined signature achieved an LR AUC of 0.94 (95% CI 0.91; 0.97; F1 = 0.871). These findings indicate that wearable-derived gait features complement, rather than replace, intrinsic risk indicators, and that their combination provides the most effective fall risk assessment.
Wu et al. (Fri,) studied this question.