Construction and comparison of risk models for post-PCI nausea and vomiting based on five machine-learning algorithms

Lu Yanxia; Ma Liang

doi:10.26939/j.cnki.CN11-9353/R.2025.09.005

Abstract：Objective To identify peri-operative factors associated with post-procedural nausea and vomiting(PONV)after percutaneous coronary intervention(PCI)and to develop a rigorously validated prediction system.Methods Clinical data of 250 patients who underwent PCI at The First People's Hospital of Lianyungang from September 2023 to August 2025 were retrospectively collected.A comprehensive model and a predictive evaluation framework were implemented in Python.In the model construction process,five machine learning methods were employed,namely Support Vector Machine(SVM),Decision Tree(DT),Random Forest(RF),Logistic Regression(LR),and Adaptive Boosting(Adaboost).The dataset was split so that 90％of observations were randomly allocated to the training set and the remaining 10％constituted the independent validation set.Model performance was evaluated by ten-fold cross-validation,with the area under the receiver-operating characteristic curve(AUC)serving as the primary metric.Results Seventy of the 250 patients(28.0％)developed PONV within 24 h after PCI.Univariable analysis revealed significant between-group differences(all P<0.05)in sex,smoking history,prior motion sickness or PONV,contrast volume>100 mL,procedure duration>120 min,intra-procedural hypotension,post-procedural pain score>3 on the FPS scale,and comorbidity count>3.Among the five algorithms,the LR-based model achieved the highest discriminative accuracy(mean AUC=0.871).Conclusion The LR model demonstrates superior predictive performance for post-PCI PONV.Integrating this model into user-friendly clinical software may facilitate individualized risk stratification and guide targeted prophylactic strategies to reduce the incidence of PONV.