What question did this study set out to answer?

To develop and validate a multidimensional machine learning model combining biomarkers, clinical characteristics, and geriatric assessment to predict major postoperative complications in older adults with hip fracture.

August 22, 2026BMC Geriatrics0 citationsOpen Access

A multimodal model integrating biomarkers, clinical features, and geriatric assessment to predict post-hip-fracture complications

YTYa Tuo MYMeng Yang CWCaixia Wang

Key Points

To develop and validate a multidimensional machine learning model combining biomarkers, clinical characteristics, and geriatric assessment to predict major postoperative complications in older adults with hip fracture.
Retrospectively analyzed 342 older hip fracture patients divided into a training cohort (n = 240) and a validation cohort (n = 102) in a 7:3 ratio.
Selected predictive variables via univariate and LASSO regression, constructed three machine learning algorithms, and evaluated performance using receiver operating characteristic curves, calibration curves, and decision curve analysis.
Identified seven independent predictors: modified Frailty Index-5, C-reactive protein-to-albumin ratio, C-reactive protein-albumin-lymphocyte index, lymphocyte-to-calcium ratio, neutrophil-to-lymphocyte ratio, systemic immune-inflammation index, and Mini-Mental State Examination score.
The random forest model demonstrated the best predictive performance, achieving an AUC of 0.877 in the training set and 0.833 in the validation set, alongside favorable calibration and clinical net benefit across threshold probabilities of 0.1 to 0.8.

Abstract

Major postoperative complications in older adults with hip fracture vary widely among individuals. Reliable tools for identifying high-risk patients are lacking, especially multidimensional risk models combining clinical features, biomarkers and geriatric assessment. This study aimed to develop and validate a multidimensional predictive model for individualized risk stratification of major postoperative complications in older adults with hip fracture. A total of 342 older adults with hip fracture were retrospectively enrolled and split into training ( n = 240) and validation ( n = 102) sets at 7:3. Predictors were screened by univariate and LASSO regression. Three machine learning models were constructed and compared. Model performance was evaluated by ROC, calibration curves and decision curve analysis. Seven independent factors were identified: modified Frailty Index-5, C-reactive protein-to-albumin ratio, C-reactive protein-albumin-lymphocyte index, lymphocyte-to-calcium ratio, neutrophil-to-lymphocyte ratio, systemic immune-inflammation index and Mini-Mental State Examination score. The random forest model achieved optimal AUC (training: 0.877; validation: 0.833), with good calibration and high clinical net benefit at 0.1–0.8 threshold. A multimodal prediction model for postoperative complications in older adults with hip fracture was successfully established. It shows good discrimination, calibration and clinical utility, supporting early risk identification and perioperative strategy optimization.

Ask AI

Helpful

Bookmark

View Full Paper