Development and Validation of Machine Learning Models for Predicting Falls Among Hospitalized Older Adults: A Retrospective Study (Preprint)

BACKGROUND Falls are one of the leading causes of injury or death among older adults. Falls occurring in individuals during hospitalization, as an adverse event, are a key concern for healthcare institutions. Identifying older adults at high risk of falls in clinical settings enables early interventions, thereby reducing the incidence of falls. OBJECTIVE This study aims to develop and validate machine learning (ML) models to predict the risk of falls among hospitalized older adults. METHODS This study retrospectively analyzed data from a tertiary general hospital in China, including 342 older adults who experienced falls and 684 randomly matched non-fallers, between January 2018 and December 2024, encompassing demographic information, comorbidities, laboratory parameters, and medication use, among other variables. The dataset was randomly split into training and testing sets in a 7:3 ratio. Predictors were selected from the training set using stepwise regression, Lasso, and random forest-recursive feature elimination (RF-RFE). Seven ML algorithms were employed to develop predictive models in the training set, and their performance was compared in the testing set. The optimal model was interpreted using Shapley Additive Explanations (SHAP). RESULTS The Gradient Boosting Machine (GBM) model demonstrated the best predictive performance (C-index = 0.744 [95% CI 0.688–0.799]). The eight most important variables associated with fall risk were dizziness, epilepsy, fall history within the past 3 months, use of walking assistance, emergency admission, Morse Fall Scale scores, modified Barthel Index scores, and the number of indwelling catheters. The model was interpreted using SHAP to enhance the clinical utility of the predictive model. CONCLUSIONS The GBM model was identified as the optimal predictive model. The SHAP method enhanced its integration into clinical workflows.