摘要Background:Sepsis-associated acute kidney injury(S-AKI)is a common complication of sepsis,and early identification can improve patient prognosis.This study incorporated novel inflammatory markers as features to construct a model and employed 6 machine learning methods to predict the occurrence of S-AKI.Methods:A total of 3613 patients with sepsis were included in this study.Novel inflammatory markers,including neutrophil-to-lymphocyte ratio,monocyte-to-lymphocyte ratio,systemic inflammatory response index,systemic immune-inflammation index,systemic inflammatory aggregate index,lactate-to-albumin ratio,and prognostic nutritional index,along with demographic characteristics,clinical conditions,and routine laboratory results,were used to construct the model.The machine learning methods employed included logistic regression,support vector machine,random forest(RF),extreme gradient boosting(XGBoost),and ensemble methods(RF+XGBoost).Model performance and stability were evaluated using 5-fold cross-validation.Model performance was assessed using the area under the receiver-operating characteristic curve,sensitivity,specificity,accuracy,precision,recall,and F1 score.Additionally,SHapley Additive exPlanations values were used to interpret the predictive model.Results:In the final algorithm group,the ensemble model of RF and XGBoost(0.843;95％confidence interval:0.820-0.866)was higher than those of other models.Among the single models,the XGBoost model exhibited the highest sensitivity(0.856)and F1 score(0.780),indicating its stronger ability to identify patients who will develop S-AKI,albeit at the expense of lower specificity(0.667).The 4 most influential features for XGBoost were mechanical ventilation,mean arterial pressure,blood urea nitrogen level,and sequential organ failure assessment score.Among the 3 novel inflammatory markers,lactate-to-albumin ratio showed the greatest effect.Conclusion:We successfully developed machine learning methods to predict S-AKI,highlighting the importance of novel inflammatory markers in model construction.This breakthrough offers novel perspectives for feature selection in the future development of related predictive models.