JMIR Publications

ABSTRACT

Background:

Background Heart failure (HF) is a refractory disease with a global public health issue that is continuously increasing. Metabolic syndrome plays a crucial role in prevalence and mortality of HF. The triglyceride-glucose (TyG)-related obesity indices, such as body mass index (BMI), a body shape index (ABSI), and waist-to-height ratio (WHtR), have been recognized as a significant predictor of cardiovascular disease risk. Nevertheless, the predictive value of these makers for HF prevalence and their association between all-cause mortality in general populations remains unclear.

Objective:

in this study, we aimed to evaluate their association with prevalence and all-cause mortality among HF patients using machine learning techniques.

Methods:

The U.S. National Health and Nutrition Examination Survey (NHANES) (2001-2018) database provided all the data for this study. The status of the participants was followed through December 31, 2019. Participants were categorized into a non-HF group and a HF group. Weighted binary logistic regression was performed to evaluate the independent associations between the TyG-related obesity indices and HF. Meanwhile, subgroup analysis was performed to confirm the reliability of the associations observed among different population. Restricted cubic spline (RCS) models were utilized to delineate whether the relationship is non-linear. Random forest analysis and Boruta algorithm were adopted to assess the predictive value of each biomarker for the prevalence of HF. Receiver operating characteristic (ROC) curves were generated to assess the predictive performance. Additionally, those biomarkers were categorized into two groups based on threshold derived from the maximally selected rank statistics (MSRS). Kaplan-Meier survival analysis and weighted Cox regression models were employed to explore the association between each TyG-related obesity indices and all-cause mortality among HF patients.

Results:

40,908 participants (1,174 HF patients) were encompassed in this retrospective study. In the fully adjusted model, TyG-BMI, TyG-ABSI, and TyG-WHtR exhibited higher odds ratio (OR) than TyG alone. TyG-ABSI exhibited the strongest association both as a continuous variable and across quartiles, demonstrating a significant near-linear positive dose-response relationship with HF risk. RCS analysis further confirmed a linear relationship between TyG-related obesity indices and HF risk. The ROC curve analysis demonstrated that TyG-ABSI had the best predictive performance for HF risk (AUC: 0.721, 95% CI: 0.690–0.736). Random forest analyses and Boruta algorithm identified those biomarkers as an important clinical feature. Subgroup analysis revealed no significant interactions across all subgroups, except for age. During a median follow-up of 9 years, a total of 566 deaths were documented, when stratified by the MSRS-derived optimal cutoff value, Kaplan-Meier survival analysis and Cox regression model demonstrated significantly worse overall survival for the higher TyG-ABSI group (HR:1.44, 95% CI=1.11-1.86, P=0.006), each standard deviation increment in TyG-ABSI was associated with an 11% increment all-cause mortality risk among HF patients.

Conclusions:

Our study suggests that TyG-BMI, TyG-ABSI and TyG-WHtR are associated with increased odds of HF in the U.S. TyG-ABSI demonstrate the best predicted performance and expect to become more effective metrics for improving risk stratification. TyG-ABSI is independently associated with increased all-cause mortality risk in HF patients, highlighting its potential as a useful tool in aiding personalized management.