Development and internal validation of prediction model for 90-day mortality in older patients with hip fracture using admission blood parameters

Abstract

To develop and validate an early prediction model for short-term mortality risk in older patients with hip fracture using admission laboratory parameters. In this retrospective cohort study, data from 1881 older patients with hip fracture (2013.01‑2023.12) were analyzed. All‑cause mortality within 90 days of admission was the primary outcome. From 29 candidate predictors, feature selection was performed on a training set (70%) using stepwise logistic regression, the Boruta algorithm, and LASSO regression. A consensus predictor set was determined by majority voting and used to construct a multivariable logistic regression model, presented as a nomogram. The model was evaluated on an independent testing set (30%) for discrimination, calibration, and clinical utility. The 90‑day mortality rate was 11.5% (217/1881). Ten predictors were selected: age, Charlson Comorbidity Index, lymphocyte, albumin, adenosine deaminase, direct bilirubin, blood urea nitrogen, prothrombin time, white blood cell, and β2-microglobulin. The nomogram showed good discrimination, with AUCs of 0.812 (95% CI: 0.779‑0.846) in the training set and 0.826 (0.775‑0.878) in the testing set. Calibration was satisfactory (Hosmer‑Lemeshow P = 0.191 and 0.778, respectively). Decision curve analysis demonstrated a higher net clinical benefit than treat‑all or treat‑none strategies across a wide threshold range. We developed and validated an early prediction model for 90-day mortality following hip fracture using admission blood parameters. The model demonstrated good discriminative ability and calibration, supporting its potential clinical utility in early risk stratification. Although it may aid in guiding individualized patient management, it should be used to complement, not replace, clinical judgment. Further external validation is warranted.