Predicting Fall Rate in Neurorehabilitation: Combining Physical and Cognitive Parameters

Background Falls pose a significant health challenge in neurological populations. In Switzerland, approximately 286.000 people fall annually, with up to 50% of neurological inpatients experiencing falls. These incidents lead to increased morbidity, healthcare costs, and psychological consequences. Current fall-risk assessments have limitations in discriminative capabilities, often underestimating the impact of psychological and cognitive factors, and are frequently restricted to binary outcomes (fall/no fall). Additionally, these tools are predominantly developed for elderly cohorts, which may not generalize to younger or neurologically impaired populations. This study aims to develop a comprehensive fall-rate prediction model for neurorehabilitation patients integrating (1) inpatient fall prevalence quantification and (2) comparative analysis of risk factors distinguishing single versus recurrent fallers through clinical data.

Methods This retrospective cohort study was conducted at a Swiss neurorehabilitation centre over four years. Fall prevalence was assessed in all inpatients aged>18 years with neurological disorders using structured fall records and clinical data. The prediction model incorporated demographic factors, clinical diagnoses, medication use, functional assessments, cognitive measures, and activities of daily living among fallers. A zero-inflated Poisson regression model with Akaike Information Criterion was employed, complemented by bootstrap validation and machine-learning algorithms.

Results Preliminary data indicate fall prevalence of 16.2% (95% CI: 14.7–17.9) among 2,193 inpatients. Age and gender distributions were similar between fallers (mean age: 55 years, 63% male) and the overall hospital population (mean age: 56 years, 62% male). However, fallers had significantly longer hospital stays (98 vs. 61 days). Some variables associated with fall-risk aligned with previous models for≤4 falls. However, these risk-factors appeared less predictive for individuals with excessive falls (>5 falls), who demonstrated better motor and overall cognitive abilities but progressively impaired divided-attention.

Conclusions Existing models lack generalizability to this neurological population, particularly regarding patient age and domain-specific neuropsychological deficits especially in multiple fallers. By integrating multidisciplinary clinical evaluations and machine-learning techniques, this study aims to enhance the accuracy of fall-rate prediction in neurorehabilitation settings, bridging the gap between research and clinical practice.

Publikationsverlauf

Artikel online veröffentlicht:
23. Oktober 2025