Prognostic value of global longitudinal strain measured by speckle-tracking echocardiography in patients with sepsis

Li Meiling; Luo Yannian; Mao Wenjie; He Nannan; Cao Wen

doi:10.3969/j.issn.1002-1949.2026.03.005

Abstract：Objective This study aims to evaluate the prognostic value of global longitudinal strain(GLS)measured by speckle-tracking echocardiography in patients with sepsis,to compare it with conventional left ventricular ejection fraction(LVEF),and its predictive performance was further validated using machine learning approaches.Methods A prospective cohort of 119 sepsis patients admitted to the intensive care unit(ICU)underwent transthoracic echocardiography within 24 hours of admission for LVEF and GLS measurements.Patients were followed for 3 months.Clinical characteristics between survivors and non-survivors were compared using t-tests,analysis of variance,or rank-sum tests as appropriate.Multivariable Logistic regression was conducted to assess the independent predictive value of GLS and LVEF for mortality,and receiver operating characteristic(ROC)curves were constructed to compare predictive value.In addition,multiple machine learning algorithms,including random forest(RF),support vector machine(SVM),light gradient boosting machine(Light GBM),decision tree(DT),and k-nearest neighbors(KNN),were applied to validate the regression findings.Results In multivariable Logistic regression,septic shock was associated with a 25.625-fold increased risk of mortality compared with non-shock patients(95%CI 3.882-497.752,P=0.005),and this association remained stable after including EF in the model(OR=25.416,95%CI 3.862-489.217,P=0.005).Vasopressin use was consistently associated with elevated mortality risk across all models(Model 1:OR=11.803,95%CI 2.824-63.514,P=0.002;Model 3:OR=12.118,95%CI 2.860-67.301,P=0.002).Myocardial injury was also significantly associated with higher mortality(Model 1:OR=7.806,95%CI 1.449-72.705,P=0.034;Model 3:OR=8.586,95%CI 1.555-82.209,P=0.029).ROC curve analysis demonstrated that GLS exhibited superior predictive performance(AUC=0.911)compared with LVEF(AUC=0.906),and the combination of GLS and LVEF achieved the highest predictive accuracy(AUC=0.916).Among machine learning models,the KNN classifier incorporating both GLS and LVEF performed best,with an AUC of 0.968,a Kappa coefficient of 0.77,and balanced sensitivity(0.91)and specificity(0.91).Conclusions GLS is an independent predictor of 3-month mortality in sepsis patients,outperforming conventional LVEF.The combined application of GLS and LVEF further improves predictive performance,particularly when integrated into a KNN machine learning model,highlighting its potential for enhancing clinical risk stratification and guiding personalized management in sepsis.