Abstract
OBJECTIVE: This study aimed to develop and validate a dynamic online nomogram for predicting inpatient fall risk using data from a Dryad database cohort to strengthen fall prevention strategies and enhance patient safety in hospital settings. METHODS: We analyzed data from a cohort study conducted at Fukushima Medical University Hospital, with external validation using an independent dataset from Taizhou People's Hospital (2019-2023, n=2000). Following multiple imputation, 9470 cases were included and divided into training (n=6631) and validation (n=2839) sets. LASSO regression identified fall-associated factors, leading to development of two predictive models using binomial logistic regression. Model 1 incorporated all selected variables, while Model 2 emphasized clinically relevant factors. Discriminatory power, calibration, and clinical decision curve analysis were conducted for both models. RESULTS: LASSO regression identified 14 key variables, reduced to 11 in Model 1 and 6 clinically relevant variables in Model 2. Both models demonstrated comparable performance (Z=1.152, p=0.249), with Model 2 selected for clinical applicability. Bootstrap validation showed strong performance with AUC of 0.801 (training set) and 0.796 (validation set). Calibration was adequate (Hosmer-Lemeshow test p>0.05). Decision curve analysis indicated potential intervention benefit for predicted probabilities of 1-95.1% (training) and 1-89.2% (validation). External validation in 2000 patients demonstrated robust generalizability (AUC=0.87, 95% CI: 0.80-0.93). CONCLUSION: We developed a predictive model for assessing fall risk among hospitalized patients. This model supports individualized patient evaluations, assists in identifying high-risk patients, and may contribute to reducing fall incidence in hospital settings.