Predicting Postoperative Blood Transfusion in Elderly Patients Undergoing Total Hip and Knee Arthroplasty Using Machine Learning Models.

Liang, Dan; Pang, Yiming; Huang, Jingrui; Che, Xianda; Zhou, Raorao; Ding, Xueting; Wang, Chunfang; Zhao, Litao; Han, Yichen; Rong, Xueqin; Li, Pengcui

Purpose: With the aging population, the demand for total hip arthroplasty (THA) and total knee arthroplasty (TKA) has risen significantly. Elderly patients, especially those over 70 years, face a higher risk of perioperative bleeding and transfusion, increasing morbidity and mortality. Accurate transfusion risk prediction is vital for optimizing perioperative blood management. Traditional models often fail to capture complex factor interactions, whereas machine learning enhances predictive accuracy. This study aimed to develop predictive models for postoperative transfusion in elderly patients undergoing THA or TKA, identify key risk factors, and create an online prediction tool. Patients and Methods: We retrospectively analyzed 1,520 elderly patients who underwent THA (659) or TKA (861). The Least Absolute Shrinkage and Selection Operator (LASSO) method was used for variable selection. The dataset was randomly split into training (70%) and testing (30%) sets. Five models—Logistic Regression (LR), Random Forest (RF), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Naive Bayes (NB)—were developed and validated. Ten-fold cross-validation and grid search optimized model parameters. Model performance was evaluated using AUC, accuracy, precision, sensitivity, specificity, and F1 score. SHapley Additive exPlanations (SHAP) were applied to assess variable importance. An online tool was developed based on the models. Results: Nineteen variables were retained. RF, LR, and SVM showed superior performance with AUC values exceeding 0.90. RF achieved the best results, with an accuracy of 0.86, precision of 0.80, specificity of 0.91, F1-score of 0.78, and sensitivity of 0.76. SHAP analysis highlighted intraoperative blood loss, hypertension, and postoperative drainage volume as major predictors. Conclusion: The developed models and online tool support personalized transfusion risk assessment, optimizing perioperative management, optimizing blood utilization, and enhancing patient outcomes.

TOTAL hip replacement; MACHINE learning; TOTAL knee replacement; SURGICAL blood loss; OLDER patients

Risk Management & Healthcare Policy, 2025, Vol 18, p1697

1179-1594

Academic Journal

10.2147/RMHP.S503286