Predicting intraoperative blood transfusion risk in hip fracture patients using explainable machine learning models
10.13303/j.cjbt.issn.1004-549x.2026.02.005
- VernacularTitle:基于可解释的机器学习模型预测髋部骨折患者术中输血风险
- Author:
Fengting LU
1
;
Xiaoming LI
1
;
Dekui LI
1
;
Xianyuan XIE
1
;
Jiazhong WANG
1
;
Qing YU
1
;
Gan HUANG
2
;
Jun SHEN
1
Author Information
1. Department of Anesthesiology, Affiliated Hospital of Western Anhui Health Vocational College, Lu'an 237000, China
2. Department One of Orthopedics, Affiliated Hospital of Western Anhui Health Vocational College, Lu'an 237000, China
- Publication Type:Journal Article
- Keywords:
hip fracture;
blood transfusion;
machine learning;
Boruta algorithm;
SVM model;
SHAP
- From:
Chinese Journal of Blood Transfusion
2026;39(2):196-202
- CountryChina
- Language:Chinese
-
Abstract:
Objective: To investigate the factors influencing intraoperative blood transfusion in patients with hip fractures and to develop a machine learning (ML) model for predicting this risk. Methods: A total of 424 patients with hip fractures who underwent surgical treatment between November 2022 and March 2025 in our hospital were selected. Key feature variables of intraoperative blood transfusion risk were identified using the Boruta algorithm. Four different ML algorithms—support vector machine (SVM), linear discriminant analysis (LDA), mixed discriminant analysis (MDA), and extreme gradient boosting (XGBoost)—were used to develop predictive models for intraoperative blood transfusion risk. The predictive performance of the four ML models were evaluated using accuracy, precision, receiver operating characteristic (ROC) curves, precision-recall curves (PRC), precision-recall gain curves (PRGC), and F1 scores. Shapley additive interpretation (SHAP) was used to interpret the final model. Results: Among the 424 patients, 77(18.2%) received intraoperative blood transfusion. The Boruta algorithm identified albumin (ALB), activated partial thromboplastin time (APTT), types of anesthesia, types of fracture, and hemoglobin (Hb) as key feature variables for predicting intraoperative blood transfusion risk. In model evaluation, the SVM model outperforms the other three models across multiple metrics, including the area under the receiver operating characteristic curve (AUC), recall, recall gain, accuracy, precision, F1 score, and the area under the precision-recall curve (PRC-AUC). The SVM model, interpreted and visualized based on SHAP values, effectively predicted intraoperative blood transfusion risk in patients with hip fracture. A visual online application was developed based on the SVM model (https://pbo-nomogram.shinyapps.io/blood/). Conclusion: Preoperative low ALB and Hb levels, prolonged APTT, general anesthesia, and intertrochanteric fractures are risk factors for intraoperative blood transfusion in hip fracture patients. The risk prediction model for intraoperative blood transfusion constructed based on the SVM algorithm has optimal performance, which provides new ideas and methods for the clinical early identification of hip fracture patients with high transfusion risk and the implementation of targeted interventions.
