Machine Learning-Based Early Risk Stratification for Sepsis-Related Troponin-Defined Myocardial Injury Using Routine...

Area of Science:

• Critical Care Medicine
• Cardiology
• Machine Learning in Healthcare

Background:

• Early recognition of sepsis-related myocardial injury is challenging due to a lack of harmonized echocardiographic data in retrospective studies.
• Routine metabolic indicators, such as the triglyceride-glucose (TyG) index and triglyceride-to-HDL cholesterol (TG/HDL) ratio, show potential for early risk stratification.
• The utility of these metabolic indices within a multicenter machine-learning framework for sepsis-related myocardial injury requires further definition.

Purpose of the Study:

• To develop and validate a machine-learning model for early risk stratification of sepsis-related myocardial injury using routine metabolic indicators.
• To assess the performance of machine-learning classifiers in predicting troponin-defined myocardial injury in sepsis patients.
• To investigate the added value of the TyG index and TG/HDL ratio in predicting sepsis-related myocardial injury.

Main Methods:

• Retrospective multicenter study involving 2,588 sepsis patients from MIMIC-IV and eICU databases (development) and 504 from Xiangya ICU (validation).
• Primary endpoint: operational sepsis-related troponin-defined myocardial injury.
• Five machine-learning classifiers were compared, with a random forest model selected for interpretability. TyG and TG/HDL indices were incorporated into the models.

Main Results:

• The random forest model incorporating TyG and TG/HDL indices achieved an AUC of 0.725 in the internal test set and 0.619 in the external validation cohort.
• External calibration indicated underprediction, and decision-curve analysis did not support immediate clinical use.
• Joint stratification identified a subgroup with elevated metabolic indices and a higher risk (OR 1.59) of troponin-defined myocardial injury.

Conclusions:

• An interpretable random forest model using routine laboratory data, including metabolic indices, may aid exploratory early risk stratification for sepsis-related troponin-defined myocardial injury.
• The model is not a diagnostic tool for echocardiography-confirmed septic cardiomyopathy.
• Prospective validation is necessary before clinical implementation.