Mortality associations of the cholesterol-high-density lipoprotein-glucose index and the triglyceride-glucose index in adults with heart-disease-related conditions: parallel analyses of two national cohorts

The cholesterol-high-density lipoprotein-glucose (CHG) index combines lipid and glucose measures, but its prognostic value in older adults with heart-disease-related conditions remains uncertain. This study assessed linear and non-monotonic associations between CHG and mortality, compared CHG with the triglyceride-glucose (TyG) index, and evaluated their incremental predictive value beyond clinical variables. Two national datasets were analysed in parallel, not as a validation exercise. The China Health and Retirement Longitudinal Study (CHARLS) contributed adults who reported a physician diagnosis within a heart-disease-related composite at the 2011 baseline (myocardial infarction, coronary heart disease, angina, congestive heart failure, or other heart disease). The National Health and Nutrition Examination Survey (NHANES), 1999–2018, contributed adults aged 45 years or older with coronary heart disease. Because CHARLS offers no comparable cause-of-death endpoint, cardiovascular mortality was examined only as a secondary NHANES analysis. Analyses comprised adjusted Cox models (survey-design models for NHANES), tertile comparisons, restricted cubic splines, sensitivity analyses, and prediction metrics that allow for incomplete follow-up. The CHARLS sample comprised 1,393 participants (252 deaths) and the NHANES sample 991 participants (449 deaths, of which 179 were cardiovascular). All-cause mortality was unrelated to continuous CHG in either cohort (CHARLS: hazard ratio [HR], 1.26; 95% confidence interval [CI], 0.93–1.71; P = 0.133; NHANES: HR, 1.16; 95% CI, 0.78–1.71; P = 0.468). TyG after CHARLS participant-identification reconstruction was also non-significant (HR, 1.18; 95% CI, 0.95–1.46; P = 0.128). Estimates were lowest in the middle CHG tertile in both cohorts, although the middle-versus-lowest contrast reached significance only in NHANES (HR, 0.62; 95% CI, 0.48–0.81; P < 0.001). Evidence of CHG nonlinearity was observed in CHARLS (P = 0.027) and NHANES (P < 0.001). In the secondary NHANES-only cardiovascular analysis, the Fine-Gray estimate was non-significant (subdistribution HR, 1.53; 95% CI, 0.82–2.86; P = 0.179). Adding CHG produced little change beyond the clinical model: the C-index changed by 0.004, and the 5-year inverse-probability-of-censoring-weighted area under the curve changed from 0.737 to 0.744 in CHARLS. The corresponding changes in NHANES were − 0.0003 and 0.719 to 0.718, respectively, with integrated discrimination improvement changes of -0.0003 and − 0.0002. CHG showed a context-dependent, non-monotonic pattern, which was clearest in the middle range in NHANES. The cardiovascular-mortality analysis was secondary, available only in NHANES, and had no directly comparable CHARLS endpoint. CHG and TyG added little beyond the clinical model and should not be used as stand-alone mortality indicators. These findings support more equitable risk stratification by avoiding the unnecessary use of weakly informative indices, thereby directing resources toward established preventive strategies in aging populations. The cholesterol-high-density lipoprotein-glucose (CHG) index has been evaluated in diabetes and several cardiovascular settings. The triglyceride-glucose (TyG) index is an insulin-resistance surrogate, but its associations vary by population. In older adults, low lipid levels may coexist with frailty or reverse causation. Does CHG show a consistent linear relation with mortality in older adults with heart-disease-related conditions? Continuous CHG and TyG showed no robust independent linear mortality association. The middle CHG tertile had the lowest estimate; the T2-versus-T1 contrast was significant only in NHANES. When incomplete follow-up was accounted for, adding CHG changed 5-year prediction very little beyond clinical variables. Adding CHG or TyG produced little change beyond the clinical model.