A prediction model of falls for patients with neurological disorder in acute care hospital

Abstract For the prevention of falls, individual fall risk assessment is the necessary first step. Thus, we attempted to identify independent risk factors for falls and develop a prediction model using a scoring system for patients with neurological disorders in acute hospital settings. This study was a secondary analysis of a previous study performed to compare the reliability and validity of three well-known fall assessment tools in patients with neurological disorders. We considered comorbid diseases and potential medications in addition to variables included in the three tools. Multiple logistic regression analysis was used to develop a prediction model for falls. Predictive scores were calculated using the proportional odds ratio (OR) of each predictor. The discriminative power of this model was evaluated by receiver-operating characteristic (ROC) area under the curve (AUC) analysis. A total of 32 falls were noted among 1018 patients. History of falls (OR, 4.01; 95% CI, 1.61-9.98; p =.003), cerebrovascular disease (CVD) (OR, 2.61; 95% CI, 1.11-6.14; p =.028), severe impaired gait (OR, 7.28; 95% CI, 2.45-21.65; p <.001), and overestimate of one's own gait ability (OR, 9.14; 95% CI, 3.89-21.45; p <.001) were identified as meaningful predictors for falling after adjusting for age, diabetes, confusion or disorientation, up-and-go test, altered elimination, and antipsychotics by univariate analysis. The discriminative power of fall risk score calculated by the prediction model was 0.904 of AUC (p <.001). Our results suggest that in addition to fall history and the presence of CVD, neurological assessment for gait and insight into gait ability are imperative to predict falls in patients with neurological disorders.