Interpretable machine learning model predicts intraoperative acquired stress injury risk in patients undergoing posterior lumbar interbody fusion

YAN Hui-xia; LU Yao; ZHANG Yong-jun; TONG Mao-qi; MA Chen-xi; ZHANG Tao

doi:10.3969/j.issn.2095-252X.2025.06.012

Abstract：Objective To develop a model predicting the risk of intraoperative acquired pressure injuries(IAPI)in patients undergoing posterior lumbar interbody fusion(PLIF)based on Boruta algorithm,Shapley additive explanations(SHAP)and machine learning(ML).Methods A total of 271 patients who came to our hospital for PLIF treatment from January 2021 to April 2024 were selected and screened for IAPI important feature variables by Boruta algorithm.A 3:2 ratio was randomly divided into a training set(n=163)and a test set(n=108)to train nine ML models.ML model prediction performance was evaluated using receiver operating characteristic curve(ROC).ML models were additionally interpreted and visualized by SHAP values.Results The prevalence of IAPI in 271 PLIF patients was 12.2%.Boruta's algorithm screened fasting time,albumin(ALB),age,smoking,blood oxygen saturation(SaO2),body mass index(BMI),haemoglobin(Hb),gender and high blood pressure as important characteristic variables for IAPI.Among the 4 ML algorithms,the training set and test set ROC confirmed that the XGBoost model had the highest performance in predicting IAPI risk.SHAP visualisation plots showed that the rank order contributing IAPI risk was age,ALB,fasting time,smoking,gender,SaO2,BMI,Hb and surgical position.SHAP summary plots showed that the 9 IAPI risk profile variables were in the order of"age","ALB","fasting time","smoking","gender","BMI","Hb"and"high blood pressure".The SHAP values were"split",indicating effective diagnosis of IAPI,and the SHAP power plot showed that the XGBoost model predicted the risk of IAPI with close to 100%accuracy.Conclusions Age,ALB,fasting time,smoking,gender,SaO2,BMI,Hb,and high blood pressure are IAPI risk characteristic variables,and an interpretable XGBoost model based on SHAP values can accurately predict the risk of IAPI in patients with PLIF,and improve the early prevention and intervention of IAPI.