Triglyceride–high-density lipoprotein cholesterol–glucose–body mass index predicts 28-day all-cause mortality in critically ill patients with heart failure

The prognostic value of the triglyceride–high-density lipoprotein cholesterol–glucose–body mass index (TyHGB) in critically ill heart failure (HF) patients remains unclear. This study evaluated the association between TyHGB and 28-day mortality and developed machine learning models incorporating TyHGB. A total of 3,528 critically ill HF patients were included, comprising 836 in the primary cohort and 2,692 in the external validation cohort. Kaplan–Meier, Cox regression, and restricted cubic spline analyses were performed. Incremental predictive value beyond SOFA, SAPS II, and OASIS was assessed using net reclassification improvement (NRI) and decision curve analysis (DCA). Machine learning models were developed and validated, and an online prediction tool was constructed. Among 836 patients, 141 (16.9%) died within 28 days. Mortality increased across TyHGB quartiles (12.4%, 14.4%, 18.2%, and 22.5%). The highest TyHGB quartile was associated with increased mortality (HR = 2.10, 95% CI: 1.21–3.64). A linear association was observed. Findings were consistent in external validation (HR = 1.69, 95% CI: 1.22–2.34). Adding TyHGB improved risk reclassification (NRI = 0.086–0.094 across conventional scores). Machine learning models incorporating TyHGB demonstrated stable predictive performance across datasets, with XGBoost achieving AUCs of 0.810, 0.793, and 0.714 in the training, internal validation, and external validation cohorts, respectively, and showing modest incremental clinical utility. An interactive web-based calculator was further established to facilitate individualized risk estimation in clinical practice ( https://ketong-hf.shinyapps.io/tyhgb-hf-risk/ ). TyHGB is independently associated with 28-day mortality in critically ill HF patients and provides incremental prognostic information beyond conventional scores. Machine learning models incorporating TyHGB show good performance, and the web-based tool may support individualized risk assessment in clinical practice.