Dietary deterioration from midlife to late life, intrinsic capacity decline, and mortality: a 10-year CLHLS study

Background Intrinsic capacity (IC), the WHO composite of physical and mental capacities, is a central endpoint for healthy ageing. Whether dietary patterns measured across midlife and late life predict IC decline and mortality in very old populations remains unclear. Methods We analysed 16,068 Chinese adults aged ≥65 years from the CLHLS (2008–2018; 9,359 deaths over 63,132 person-years). Eight harmonised food groups were scored from recalled midlife (≈age 60) and late-life (2008 baseline) reports, cross-classified into stable-high, improved, worsened, and stable-low groups, and separately clustered into data-driven latent classes. Outcomes were all-cause mortality, repeated IC trajectories (IPCW-weighted GEE), incident IC impairment, and observed 10-year IC status. Results Relative to stable-high, the worsened (HR 1.135, 95% CI 1.055–1.222) and stable-low (HR 1.056, 1.008–1.107) groups had higher mortality after disease burden was entered categorically; each 10-point higher late-life score was associated with lower mortality (HR 0.969, 0.943–0.996), whereas recalled midlife score showed no independent association. The stable-low group declined faster in IC (group-by-year p = 0.028), and data-driven low-quality classes declined four- to eightfold faster than the diverse class. No dietary pattern independently predicted incident IC impairment once death was handled as a competing event. By 2018, mortality showed the clearest gradient (28–30% dead in diverse classes vs. 47–48% in poor classes). Conclusion In this nationwide cohort of older adults, varied dietary profiles were associated with modestly lower mortality and slower IC decline. The mortality association was more evident for late-life diet than for recalled midlife diet, and this pattern was retained in complete-diet-item sensitivity analyses; however, these estimates are not a formal test of a difference between life-stage associations. Associations with incident IC impairment were not consistently supported in the competing-event analysis.