Abstract
OBJECTIVES: To investigate blood indicators associated with short-term mortality risk in older patients with hip fracture, and further evaluate the incremental predictive value of incorporating these indicators into existing clinical models. METHODS: Data from 1881 patients in our institutional hip fracture database between January 2013 and December 2023 were retrospectively analyzed. The study outcome was all-cause mortality within 90 days of admission. Stepwise logistic regression, the Boruta algorithm, and Lasso regression were performed to identify features associated with mortality risk. Following feature selection, two predictive models were developed: Model A (clinical indicators only) and Model B (both clinical and blood indicators). Predictive performance was assessed using the area under the curve (AUC), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). RESULTS: Of the 1881 patients, 217 (11.5%) died within 90 days. Stepwise logistic regression identified 12 significant features associated with mortality risk, the Boruta algorithm identified 25 important features, and Lasso regression analysis selected 18 features with non-zero coefficients (all P < 0.05). Model B significantly outperformed Model A across all feature selection methods (all P < 0.001): stepwise logistic regression (AUC: 0.822 vs. 0.753), the Boruta algorithm (AUC: 0.820 vs. 0.749), and Lasso regression (AUC: 0.826 vs. 0.756). Model B also showed significant reclassification improvements (all P < 0.001): stepwise logistic regression (NRI: 0.733, IDI: 0.107), the Boruta algorithm (NRI: 0.762, IDI: 0.113), and Lasso regression (NRI: 0.725, IDI: 0.112). CONCLUSION: Various blood indicators were associated with 90-day mortality in older patients with hip fracture, and significantly enhanced the predictive ability of clinical models for short-term mortality risk. By utilizing these blood indicators, clinicians can comprehensively and objectively assess the physiological status of hip fracture patients at admission, thereby facilitating the early identification of high-risk patients and guiding personalized treatment strategies.