Machine Learning in Falls Prediction; A cognition-based predictor of falls for the acute neurological in-patient population

Background Information: Falls are associated with high direct and indirect costs, and significant morbidity and mortality for patients. Pathological falls are usually a result of a compromised motor system, and/or cognition. Very little research has been conducted on predicting falls based on this premise. Aims: To demonstrate that cognitive and motor tests can be used to create a robust predictive tool for falls. Methods: Three tests of attention and executive function (Stroop, Trail Making, and Semantic Fluency), a measure of physical function (Walk-12), a series of questions (concerning recent falls, surgery and physical function) and demographic information were collected from a cohort of 323 patients at a tertiary neurological center. The principal outcome was a fall during the in-patient stay (n = 54). Data-driven, predictive modelling was employed to identify the statistical modelling strategies which are most accurate in predicting falls, and which yield the most parsimonious models of clinical relevance. Results: The Trail test was identified as the best predictor of falls. Moreover, addition of any others variables, to the results of the Trail test did not improve the prediction (Wilcoxon signed-rank p < .001). The best statistical strategy for predicting falls was the random forest (Wilcoxon signed-rank p < .001), based solely on results of the Trail test. Tuning of the model results in the following optimized values: 68% (+- 7.7) sensitivity, 90% (+- 2.3) specificity, with a positive predictive value of 60%, when the relevant data is available. Conclusion: Predictive modelling has identified a simple yet powerful machine learning prediction strategy based on a single clinical test, the Trail test. Predictive evaluation shows this strategy to be robust, suggesting predictive modelling and machine learning as the standard for future predictive tools.

💡 Research Summary

This study set out to determine whether a simple, cognition‑based assessment could be leveraged by machine‑learning techniques to predict inpatient falls among acute neurological patients. A prospective cohort of 323 adults admitted to a tertiary neurological centre was assembled; each participant completed three neuropsychological tests (Stroop, Trail Making Test, and Semantic Fluency), a self‑report physical function questionnaire (Walk‑12), and a brief interview covering recent falls, recent surgery, and functional status. Demographic data (age, sex, primary diagnosis) were also recorded. The primary outcome was the occurrence of at least one fall during the hospital stay, which affected 54 patients (≈16.7%).

Initial univariate analyses identified the Trail Making Test (TMT) completion time as the single variable most strongly associated with subsequent falls. Adding Stroop, Semantic Fluency, Walk‑12 scores, or any demographic covariates did not improve predictive performance (Wilcoxon signed‑rank p < .001). Consequently, the authors built several classification models—logistic regression, support‑vector machines, k‑nearest neighbours, and random forests—using only the TMT metric as input. Model training employed five‑fold cross‑validation, and comparative performance was assessed with the Wilcoxon signed‑rank test.

The random‑forest algorithm emerged as the superior approach, achieving an average sensitivity of 68 % (±7.7 %), specificity of 90 % (±2.3 %), and a positive predictive value of 60 % when the TMT result was available. These figures indicate that a single bedside test can separate high‑risk from low‑risk patients with high specificity, while maintaining a clinically acceptable sensitivity. The authors argue that such a parsimonious model is more feasible for routine clinical use than complex multivariate scores.

Nevertheless, several methodological limitations temper the enthusiasm for immediate implementation. First, the dataset originates from a single tertiary centre, raising concerns about external validity; the model has not been tested on independent cohorts or in different healthcare settings. Second, the patient population is highly specific—primarily individuals with acute neurological insults such as stroke, traumatic brain injury, or postoperative neurosurgery—so the findings may not generalize to medical or surgical wards with different risk profiles. Third, the definition of “fall” is binary (any fall versus none) without grading severity, mechanism, or contributing environmental factors, limiting the model’s capacity to inform targeted preventive strategies. Fourth, the manuscript provides scant detail on data preprocessing (handling of missing values, scaling, outlier treatment), which hampers reproducibility and raises the possibility of hidden bias.

Future work should address these gaps by (1) validating the TMT‑based random‑forest model on multi‑centre, heterogeneous datasets; (2) incorporating temporal dynamics (e.g., repeated TMT assessments) to capture changes in risk over the course of admission; (3) enriching the feature set with medication profiles, bed‑environment variables, and nursing staffing levels to improve the positive predictive value; and (4) reporting a transparent preprocessing pipeline to facilitate replication.

In summary, the paper demonstrates that the Trail Making Test, a brief measure of attention and executive function, can serve as a powerful single predictor of inpatient falls in an acute neurological cohort when coupled with a random‑forest classifier. The resulting model achieves high specificity and reasonable sensitivity, suggesting that machine‑learning‑driven, cognition‑focused tools could become a cornerstone of fall‑prevention programs, provided that broader validation and integration with other clinical risk factors are pursued.