The role of triglyceride-glucose (TyG) index in predicting mortality risk in critically ill patients with sepsis: a machine learning model in a cohort study.

BACKGROUND

The TyG index has been associated with mortality. However, its predictive value for death in critically ill septic patients remains unclear. Thus, this study aims to investigate the role of the TyG index in predicting mortality among critically ill septic patients.

METHODS

The data source was MIMIC- IV 3.0 and Zibo Central Hospital. After the inclusion exclusion process, 6445 patients with sepsis from the intensive care unit remained. The outcome variables were death in hospital death and death within one year of discharge. For two outcomes, four machine learning algorithms were employed to construct models, including linear models, ensemble tree models, gradient boosting tree models, and neural network models. The performance of the models is assessed using the Receiver Operating Characteristic (ROC) curve and the area under the curve (AUC). Additionally, the importance of the TyG index is evaluated by plotting the feature importance ranking of variables selected by each model. DISCUSSION: Among the sepsis patient in-hospital death prediction models, the Random Forest model demonstrated the highest internal validation prediction accuracy, with an average AUC of 0.778, and the TyG index was ranked 11th in feature importance. For the prediction of death within one year after discharge from sepsis, the Random Survival Forest model also performed better, with an average AUC of 0.776, and the TyG index was ranked 3th in feature importance. Our study found that the TyG index was more important in predicting death one year after discharge than in predicting in-hospital death. We suggest that the role of the TyG index should be emphasized when predicting the risk of long-term mortality, with attention to the differences between different prediction models. CLINICAL TRIAL NUMBER: Not applicable.