External Validation of Ten Prediction Models for 30-day Mortality Following Hip Fracture

Overview

This study aims to externally validate ten existing prediction models with a low risk of bias for 30-day mortality following hip fracture. Data will be collected from the Dutch Hip Fracture Audit (DHFA) and supplemented with structured and unstructured data extracted through text mining using CTcue. Approximately 35 clinical variables will be used, including factors consistently associated with short-term mortality. The primary outcome is all-cause mortality within 30 days after hip fracture. Predictive performance will be assessed through discrimination (AUC), explained variance (R²), and calibration analysis. Clinical usefulness will be evaluated using Net Benefit and Decision Curve Analysis. This study seeks to identify models with strong predictive performance and practical applicability to support shared decision-making between clinicians and patients.

Hip fractures are a major health concern, especially among older adults, and are associated with substantial morbidity, mortality, and healthcare costs. While surgical intervention is standard practice for most patients, a growing number of cases require careful consideration of operative versus non-operative management based on individual risk profiles and patient preferences. Several prediction models have been developed to estimate the risk of short-term mortality after hip fracture, but many have shown only moderate predictive performance or lacked clinical applicability. In 2024, a systematic review identified ten models with a low risk of bias, based on methodological criteria such as adequate sample size, proper handling of missing data, internal validation, and assessment of calibration. This study aims to externally validate these ten prediction models using data from the Dutch Hip Fracture Audit (DHFA) combined with additional structured and unstructured clinical information extracted through CTcue, a text-mining software tool. Approximately 35 variables, including key preoperative factors such as age, sex, ASA score, institutionalization, and metastatic cancer, will be analyzed. Missing data will be addressed through multiple imputation. The primary outcome is 30-day all-cause mortality following a hip fracture. Validation of the models will involve evaluation of predictive performance through discrimination (area under the curve \[AUC\]), explained variance (R²), and calibration curves. The DeLong test will be used to statistically compare model AUCs. Clinical usefulness will be assessed by calculating Net Benefit and conducting Decision Curve Analysis. By rigorously validating these models in a large, real-world cohort, the study aims to identify which models offer both strong predictive accuracy and practical feasibility for supporting shared decision-making between clinicians and patients.