Building a prediction model for wound infection after snakebite using machine learning algo-rithms:a multicenter retrospective clinical study

Ren Yan; Chen Fang; Xian Xiaoyan; Xia Yiqin; Zhou Yiwu

doi:10.3969/j.issn.1009-4237.2026.01.00

Abstract：Objective To develop a machine learning model for predicting wound infection in snakebite pa-tients,aiding clinicians in early risk identification.Methods Clinical data of snakebite patients treated at three medical facilities(West China Hospital,West China Tian Fu Hospital,Chengdu Shang Jin Nan Fu Hospital)be-tween Jan.2020 and Sep.2022 were retrospectively analyzed.Patients were categorized into infected(n=115)and non-infected(n=408)groups based on wound infection outcomes.There were291 males and232 females,aged50-68 years,median 59 years.The dataset was randomly split into training and validation sets at a 7∶3 ratio.Eleven machine learning models,such as logistic regression(LR),decision tree,etc.,were developed to predict the infection risk,with their performance evaluated using ROC curves to select the optimal model.SHapley Additive exPlanations was used to rank predictor variable importance and visually explain the model,identifying key factors influencing out-comes.Results LASSO regression identified six significant predictors of wound infection,i.e.,white blood cell count(WBC),platelet count(PLT),neutrophil-to-lymphocyte ratio(NLR),presence of blisters,antibiotic usage,and timing of antivenom administration.Among the 11 machine learning models evaluated,LR demonstrated the best performance,with an AUC of 0.858(95%CI:0.808-0.907)in the training set and 0.855(95%CI:0.789-0.921)in the validation set,indicating stable and reliable predictive ability.Conclusion WBC,PLT,NLR,presence of blisters,antibiotic administration,and timing of antivenom administration are independent risk factors for wound in-fection following snakebites.Among the 11 machine learning models assessed,LR exhibits superior performance,making it a promising tool for clinical decision-making.