Falls are the leading cause of injury and mortality in older adults; however, the relative contributions of different fall risk factor domains remain unclear. A past fall history strongly predicts future falls, making fall history classification critical for prospective risk assessment. This study compared three domains for classifying fall history status in older adults as the basis for fall risk assessment. Cross-sectional observational study. We analyzed the G-STRIDE dataset (163 older adults; mean age standard deviation = 82.6 6.2 years; 72.4% female; 52.8% fallers). Three domains were examined: demographic information (DGI), clinical scales and questionnaires (CSQ), and mobility tests and contexts (MTC). Four classifiers (logistic regression, support vector machine, random forest, and artificial neural network) were evaluated using 10-fold cross-validation, leave-one-out, and hold-out validation. Bootstrap 95% confidence intervals (CIs) and paired t-tests were used for area under the receiver operating characteristic curve (AUC) comparisons. MTC alone achieved AUC = 0.89 (95% CI: 0.83–0.94), significantly outperforming DGI (AUC = 0.76, P < 0.001). DGI plus MTC showed a marginal advantage over DGI plus CSQ ( P = 0.064). The evolutionary optimization identified a seven-variable subset dominated by mobility measures that matched the full-model performance (AUC = 0.90). A multi-method feature importance analysis identified the examination location, frailty index, and short Falls Efficacy Scale-International as the top predictors. The external validation of GAIT2CARE (N = 127) achieved an AUC of 0.802 for DGI plus MTC. Objective mobility tests combined with demographic data provided efficient fall risk assessment without extensive questionnaire-based assessments, supporting streamlined clinical screening.
Wu et al. (Sun,) studied this question.