External validation of machine learning models for estimation of mortality 1, 3, 6, and 12 months after hip fracture on 5,055 consecutive patients.

background and purposeMany models for prediction of mortality after hip fracture have been published but few have undergone external validation, usually considered a prerequisite for assessing the actual precision before being put to clinical use. The aim of our study was to externally validate our previously published models in an independent Swedish cohort.

methodsWe retrospectively analyzed 5,055 consecutive patients with hip fractures from 2 hospitals in Stockholm, Sweden, between 2010 and 2020. Previously developed Random Forest (RF), eXtreme Gradient Boosting (XGB), and Generalized Linear Models (GLM) (Mosfeldt et al. 2024) were deployed to estimate mortality at 1, 3, 6, and 12 months. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration metrics, and decision curve analysis (DCA). Bootstrapped isotonic regression was used for recalibration.

resultsAll models showed acceptable performance, with XGB performing best (AUC 0.72, 0.74, 0.75, and 0.77 for 1-, 3-, 6-, and 12-month mortality). Mortality was lower in the validation cohort, so the models were recalibrated to adjust for this difference.

conclusionExternal validation of the previously published models confirmed the original findings, with the XGB models again demonstrating the best overall performance. Recalibration addressed cohort differences in mortality rates and resulted in well-aligned predictions. The updated models for 3- and 12-month mortality are available online (https://hipfx.shinyapps.io/hipfxswe/), allowing clinicians to input patient data and receive individualized mortality predictions to support clinical decision-making.