Abstract / Summary
Low-cost metabolic indices could support cardiovascular screening, but their utility may vary across populations. We evaluated between-cohort variation in the association, incremental predictive value, and absolute-risk performance of the triglyceride-glucose waist circumference index (TyG-WC) among ageing adults in China, England, and the United States. We harmonized three longitudinal ageing cohorts. Weighted discrete-time models estimated associations with incident self-reported heart disease or stroke, and overlap weighting aligned measured case mix. Repeated participant-level cross-validation compared a flexible clinical model with TyG-WC and its source measurements. Six unrefitted cross-country transfers assessed discrimination and calibration under cohort-specific follow-up schedules; because exact common horizons could not be reconstructed from interval-censored interviews, these analyses were interpreted as non-equivalent-horizon calibration stress tests rather than direct validation of a common-time absolute-risk model. Association analyses included 3,511 participants with 749 events in China, 2,358 with 689 events in England, and 1,865 with 249 events in the United States. Hazard ratios per within-cohort standard deviation of TyG-WC were 1.24 (95% confidence interval 1.14–1.35), 1.19 (1.09–1.29), and 1.01 (0.85–1.21), respectively. Overlap weighting balanced measured case mix but did not make associations uniform. TyG-WC did not consistently improve discrimination, prediction error, or net benefit. Across six transfers, calibration slopes ranged from 0.08 to 0.76 and observed-to-expected risk ratios from 0.33 to 2.71. TyG-WC showed between-cohort variation in its association with cardiovascular disease; evidence for effect modification depended on the exposure scale and should be interpreted cautiously. Its incremental predictive value was small and inconsistent, and absolute-risk performance was sensitive to cohort composition, outcome ascertainment, and prediction horizon. This study did not directly evaluate whether a specific TyG-WC cut-off could be transported across populations; local validation and calibration are required before implementation.