Development of a risk prediction model for intraoperative hypotension in patients undergoing lung cancer resection

ZHANG Jie; LIU Chang

doi:10.3969/j.issn.1009-0460.2026.01.006

Abstract：Objective To explore the risk factors for intraoperative hypotension(IOH)in lung cancer resection,and to construct a prediction model and risk scoring system based on the integration of causal inference and machine learning.Methods This study enrolled 160 patients who underwent lung cancer resection.Based on single-center retrospective data from January 2020 to December 2022,a training set(n=100)and an internal validation set(n=30)were constructed.Additionally,30 patients from the same center were prospectively collected between January 2023 and December 2023 to form an external validation set.Univariate analysis and Least Absolute Shrinkage and Selection Operator(LASSO)regression were applied to identify risk factors for IOH.Directed acyclic graphs were used to identify confounding factors,and propensity score matching was employed for causal inference.Subsequently,machine learning models such as Random Forest,XGBoost,and logistic regression were utilized to compare predictive performance,and a clinical risk scoring system was established.Results Through baseline analysis of patients in the training,internal validation,and external validation sets,it was confirmed that except for albumin levels and crystalloid infusion volume,the three groups were comparable in terms of major baseline indicators(P＞0.05).Univariate analysis and LASSO regression identified a history of hypertension,prolonged surgical duration,use of beta-blockers,pneumonectomy,low hemoglobin levels,and advanced age as key risk factors for intraoperative hypotension.Causal inference analysis further verified that a history of hypertension and surgical duration had significant average causal effects on intraoperative hypotension,with effect values of 0.405(P＜0.001)and 0.252(P=0.016),respectively.Based on these findings,and by integrating machine learning modeling results with expert consensus,an intraoperative hypotension risk scoring system comprising eight variables was constructed.Among multiple machine learning models,the random forest model demonstrated the best predictive performance,with an area under the receiver operating characteristic curve(AUC)of 0.795.In external validation,this scoring system exhibited good discriminative ability(AUC=0.692),calibration(Brier score=0.186),and clinical applicability(the decision curve showed higher net benefit).Using a cutoff score of ≥8 to define the high-risk group,the system achieved significant risk stratification(the incidence of intraoperative hypotension was 76.9％in the high-risk group,compared to 0％in the low-risk group).Conclusion The predictive model and risk scoring system constructed by integrating causal inference and machine learning demonstrate strong risk stratification capabilities and can be utilized to guide intraoperative management in lung cancer resection surgery.