Abstract / Summary
Abstract Background Famine studies repeatedly report window-specific effects on adult disease risk, but because exposure windows are defined by date of birth, window-specific estimates are structurally confounded by birth cohort and by chronological age. We examined whether exposure timing is differentially associated with advanced cardiovascular-kidney-metabolic (CKM) syndrome and whether any window effect is separable from a smooth birth-cohort gradient. Methods Cross-sectional secondary analysis of the baseline wave of the China Health and Retirement Longitudinal Study 2011. Exposure was assigned from exact age at the onset of the 1959–1961 Chinese famine; advanced CKM syndrome (stage 3–4) was derived from measured variables using American Heart Association criteria. The primary sample comprised 7,761 participants aged 45 or older, analysed alongside a school-age anchor group exposed at 6 to 11 years ( n = 3,108), which lies outside every posited window. Because exposure group is almost perfectly collinear with chronological age, the primary logistic regression model adjusted for sex, residence, marital status, education, smoking status and alcohol use but not for age. The decisive analysis tested whether exposure-window indicators improved model fit over a restricted cubic spline of age at famine onset. Results Advanced CKM syndrome affected 23.9% of the primary sample. With the anchor group included ( n = 10,869), odds relative to the unexposed group rose monotonically with age at famine onset: 1.55 (95% confidence interval, 1.27–1.91) prenatal, rising to 3.84 (3.16–4.66) preschool and 5.74 (4.78–6.91) school-age, the largest odds of all. Each year of age at famine onset carried an odds ratio of 1.14 (1.13–1.16; P = 3.6 × 10 −78 ). Once a smooth spline of age at famine onset was allowed for, exposure-window indicators added nothing to model fit ( P = 0.48). Conclusions Odds of advanced CKM syndrome rose smoothly with age at famine onset and beyond every developmental window; no window added risk once that gradient was modelled. These data do not support a critical developmental window; an apparent one can arise where the gradient is truncated. Because a single cross-sectional wave cannot separate birth cohort from chronological age, the gradient is described as indistinguishable from age rather than demonstrably an age artefact.