不同机器学习模型对开放性胫腓骨骨折患者术后创伤性骨髓炎发生的预测价值OA

Objective To investigate the predictive value of different machine learning models for postoperative traumatic osteomyelitis in patients with open tibiofibular fractures.Methods A total of 282 patients with open tibiofibular fractures admitted to The People's Hospital of Guangxi Zhuang Autonomous Region from January 2019 to September 2025 were selected as the training set,and another 104 patients with open tibiofibular fractures admitted to The First People's Hospital of Nanning during the same period were selected as the validation set.Patients in the training set were divided into a traumatic osteomyelitis group(51 cases)and a non-traumatic osteomyelitis group(231 cases)based on whether they developed traumatic osteomyelitis postoperatively.Preoperative general clinical data,preoperative laboratory indicators,and surgery-related indicators were compared between the two groups.Multivariate logistic regression analysis was used to identify independent influencing factors for the occurrence of postoperative traumatic osteomyelitis in patients with open tibiofibular fractures.Based on the independent influencing factors,a nomogram model,a support vector machine model,and a random forest model were constructed respectively.The predictive performance and net benefit of the three constructed machine learning models were evaluated in the validation set.Receiver operating characteristic(ROC)curves were plotted,and the predictive performance of the different machine learning models was assessed using the area under the curve(AUC),sensitivity,specificity,accuracy,precision,and F1-score.Decision curve analysis was employed to evaluate the clinical utility of the different machine learning models.Results In the training set,there were statistically significant differences between the traumatic osteomyelitis group and the non-traumatic osteomyelitis group in terms of cause of injury,albumin,procalcitonin,systemic inflammation response index(SIRI),CD4+/CD8+ratio,Gustilo-Anderson classification,and fracture type(P<0.05).Multivariate logistic regression analysis showed that cause of injury,SIRI,CD4+/CD8+ratio,Gustilo-Anderson classification,and fracture type were independent influencing factors for the occurrence of postoperative traumatic osteomyelitis in patients with open tibiofibular fractures(P<0.05).Based on the independent influencing factors,a nomogram model,a support vector machine model,and a random forest model were constructed respectively.ROC curve analysis showed that the AUC values for predicting postoperative traumatic osteomyelitis in patients with open tibiofibular fractures in the validation set were 0.883(95%CI:0.832-0.908)for the nomogram model,0.892(95%CI:0.845-0.912)for the support vector machine model,and 0.923(95%CI:0.878-0.965)for the random forest model.Decision curve analysis showed that within the high-risk threshold range of 0-1.0,all three machine learning models in the validation set demonstrated high net benefits for predicting postoperative traumatic osteomyelitis in patients with open tibiofibular fractures.The sensitivity,specificity,accuracy,precision,and F1-score of the random forest model in the validation set were all higher than those of the nomogram model and the support vector machine model.Conclusion The nomogram model,support vector machine model,and random forest model all exhibit good predictive performance and net benefit for the occurrence of postoperative traumatic osteomyelitis in patients with open tibiofibular fractures,with the random forest model demonstrating the highest predictive efficacy.