Construction and Predictive Value of XGBoost Model for Cardiac Dysfunction in MODS of Patients with Emergency Severe Multiple Trauma

ZHAI Hanjing; LI Li; XU Wenjing; WANG Xin

doi:10.3969/j.issn.1004-437X.2026.14.009

Abstract：Objective To construct an extreme gradient boosting(XGBoost)model for cardiac dysfunction of multiple organ dysfunction syndrome(MODS)in patients with emergency severe multiple trauma,and to evaluate and verify the value of the prediction model.Methods The patients with emergency severe multiple trauma of Henan Provincial People's Hospital from January 2022 to June 2025 were retrospectively selected as the research objects.According to the ratio of 7∶3,they were divided into training set and validation set.The incidence of cardiac dysfunction in MODS was counted.According to whether they were complicated with cardiac dysfunction,they were divided into occurrence group and non-occurrence group.The clinical data of two groups in validation set and training set were collected.LASSO-logistic regression analysis was used to screen the influencing factors of cardiac dysfunction in MODS of patients with emergency severe multiple trauma.The prediction model was constructed based on XGBoost algorithm.The calibration curve,receiver operating characteristic(ROC)curve,and precision-recall(PR)curve were used to evaluate the predictive performance of the XGBoost prediction model.Results The incidence of cardiac dysfunction in MODS in 360 patients with severe multiple trauma was 31.11％(112/360).In the training set and validation set,the age,number of injured sites,proportion of patients with shock,proportion of patients with sepsis,injury severity score(ISS),acute physiology and chronic health evaluation Ⅱ(APACHE Ⅱ)score,D-dimer(D-D),C-reactive protein(CRP),lactate dehydrogenase(LDH),cardiac troponin Ⅰ(cTnI),B-type brain natriuretic peptide(BNP)and lactate(Lac)levels in the occurrence group were higher than those in the non-occurrence group,and the product of reverse shock index and Glasgow coma scale score(rSIG)was lower than that in the non-occurrence group,the differences were statistically significant(P＜0.05).LASSO-logistic regression analysis showed that age,combined shock,ISS,APACHE Ⅱ score,rSIG,D-D,CRP and Lac were independent influencing factors of cardiac dysfunction in MODS of patients with emergency severe multiple trauma(P＜0.05).The importance ranking of the top 8 factors screened by the XGBoost algorithm was APACHE Ⅱscore,Lac,ISS,age,rSIG,CRP,D-D,combined shock,and the prediction performance of the XGBoost model was the best.In the training set,the calibration curve showed that the calibration degree of the model was 0.862.ROC curve analysis showed that the area under the curve(AUC)of the model for predicting cardiac dysfunction in MODS of patients with emergency severe multiple trauma was 0.889(95％CI:0.846-0.932),the sensitivity was 85.19％,and the specificity was 80.12％.PR curve analysis showed that the average accuracy of the model was 0.819(95％CI:0.738-0.908).In the validation set,the calibration curve showed that the calibration degree of the model was 0.860.ROC curve analysis showed that the AUC of this model in predicting cardiac dysfunction in MODS of patients with emergency severe multiple trauma was 0.880(95％CI:0.799-0.961),with a sensitivity of 83.87％and a specificity of 80.52％.PR curve analysis showed that the average accuracy of the model was 0.837(95％CI:0.757-0.923),which confirmed that the model had good robustness.Conclusion The prediction model of APACHE Ⅱ score,Lac,ISS,age,rSIG,CRP,D-D and combined shock based on XGBoost algorithm is constructed,it can accurately predict the risk of cardiac dysfunction in MODS of patients with emergency severe multiple trauma,and provide a reliable decision support tool for early identification of high-risk patients and individualized intervention.