Machine Learning Approaches for Predicting Intraoperative Blood Transfusion in Partial Hip Arthroplasty

Abstract

<b>Objective:</b> Partial hip arthroplasty (PHA) procedures are often associated with significant blood loss, particularly in elderly patients with comorbidities. Predicting the need for intraoperative transfusion in advance is crucial for patient safety and surgical planning. Machine learning (ML) algorithms offer data-driven solutions to support clinical decision-making in such scenarios. <b>Methods:</b> This retrospective, single-center cohort study evaluated data from 202 patients who underwent PHA between December 2023 and July 2025. Demographic data, as well as preoperative and intraoperative variables, were collected. Six ML algorithms-Logistic Regression, Decision Tree, Support Vector Machines (SVM), Artificial Neural Network (ANN), Random Forest, and Gradient Boosting-were trained and tested to predict intraoperative blood transfusion. Model performance was assessed using accuracy, F1-score, and area under the ROC curve (AUC). SHAP (SHapley Additive exPlanations) analysis was used to evaluate model interpretability. <b>Results:</b> Among the 202 patients, 85 (42.1%) received intraoperative blood transfusions. Significant predictors included low preoperative hemoglobin, high ASA score, prolonged operative time, increased intraoperative blood loss, and elevated INR (all <i>p</i> < 0.05). The Random Forest and Decision Tree models achieved the highest accuracy (95.1%) and F1-score (0.960), while the SVM model yielded the highest AUC (0.992). SHAP analysis identified hemoglobin, age, ASA score, INR, and operative time as the most influential features in model decision-making. <b>Conclusions:</b> Machine learning models-particularly Random Forest, Decision Tree, and SVM-demonstrated high performance in predicting intraoperative transfusion needs during PHA. The incorporation of explainable AI techniques such as SHAP enhanced the clinical interpretability of model outputs, supporting personalized patient management. These findings provide a strong foundation for integrating such models into clinical decision support systems, though external validation through multicenter and prospective studies is warranted.

Ähnliche Arbeiten