脓毒症患者住院期间死亡风险的随机森林分析与预测模型评估OA

Objective To analyze risk factors for in-hospital mortality in sepsis patients using different algorithms and to construct a predictive model based on the findings.Methods A retrospective study was conducted on 316 sepsis patients admitted to the Department of Critical Care Medicine at Xiamen Fifth Hospital from January 2021 to June 2025.Patients were divided into a mortality group(n=61)and a survival group(n=255)based on whether they died during hospitalization.This study first employed univariate and multivariate logistic regression analyses to screen for relevant risk factors,and a logistic regression prediction model was built accordingly.Subsequently,the Boruta algorithm and a random forest model were used for variable importance ranking and modeling.Receiver operating characteristic(ROC)curve analysis was applied to compare the predictive performance of the logistic regression and random forest models,and the superior-performing model was selected for further development.A nomogram was constructed using the significant variables identified by the optimal algorithm.Internal validation was performed using the bootstrap method.The overall performance of the final model was comprehensively evaluated using the ROC curve,concordance index(C-index),Hosmer-Lemeshow goodness-of-fit test,and decision curve analysis(DCA).Results The logistic regression model identified the following risk factors in descending order of importance:blood lactate,peripheral perfusion index,serum albumin,blood urea nitrogen,serum creatinine,alanine aminotransferase(ALT),and aspartate aminotransferase(AST).The random forest model selected blood lactate,peripheral perfusion index,serum albumin,serum creatinine,blood urea nitrogen,and ALT as key predictive variables.ROC curve analysis showed that the area under the curve(AUC)was 0.835(95%CI:0.784-0.879)for the logistic regression model and 0.901(95%CI:0.867-0.934)for the random forest model,with a difference of 0.072(95%CI:0.016-0.128,Z=2.480,P=0.013).The nomogram prediction model,constructed based on the key variables from the random forest model,demonstrated AUCs of 0.816 and 0.872 in the training and validation sets,respectively.The concordance indices(C-index)were 0.823 and 0.886,respectively.Hosmer-Lemeshow goodness-of-fit tests yielded P-values>0.05 for both sets,indicating good model calibration.Decision curve analysis revealed that,across clinically relevant threshold ranges(0.20-0.92 for training,0.10-0.94 for validation),the model provided stable and superior net benefits compared to traditional strategies,demonstrating good clinical utility.Conclusions Blood lactate,peripheral perfusion index,serum albumin,serum creatinine,blood urea nitrogen,and ALT are important factors influencing in-hospital mortality in sepsis patients.Compared to the logistic regression model,the random forest model demonstrates superior predictive performance.The nomogram prediction model derived from its key variables showed discriminative ability,calibration,and clinical utility,holding promise for clinical application.