A Machine Learning-Based Clinical Decision Support Tool for Intertrochanteric Hip Fracture Patients to Predict Postoperative Anemia Risk: A Retrospective Cohort Study

Abstract

Background: Postoperative anemia associated with intertrochanteric hip fracture is a detrimental complication that detrimentally impairs patients’ outcomes. This study is designed to develop an online predictive tool to assist physicians in developing surgical blood preparation strategies to prevent the occurrence of postoperative anemia. Methods: This study included data collected from June 2017 to June 2025 on intertrochanteric hip fracture patients at Tangdu Hospital, including demographic information, comorbidities, vital signs, and laboratory results. LASSO regression was used to select predictive variables, and seven machine learning techniques: Logistic Regression, Support Vector Machine, Decision Tree, LightGBM, XGBoost, Neural Networks, and Random Forest, were compared to identify the best tool for predicting postoperative anemia risk. We created a patient-specific risk prediction tool with SHAP-driven interpretability for clinical decision support. Results: A total of 815 patients were included in the analysis, of whom 208 (25.5%) presented with postoperative anemia. Eight variables were selected to build seven machine learning models. Among these, the SVM model exhibited the best predictive performance in terms of discrimination, calibration, and clinical applicability, with an AUC range of 0.827–0.831. In test sets encompassing diverse population characteristics, SVM achieved the highest sensitivity (72.73%), accuracy (77.78%), and F1 score (57.14%). Conclusions: We established an online prediction platform for clinical practice, enabling clinicians to assess anemia risk in intertrochanteric hip fracture patients and support early prevention of postoperative anemia.

Ähnliche Arbeiten