Eating-timing patterns derived from K-means clustering and mortality risk in US adults with chronic diseases: a cohort study.

Whether joint eating-timing configurations show comparable mortality associations across chronic diseases is uncertain. We analysed 26 712 adults aged 20 years or older from the National Health and Nutrition Examination Survey (NHANES) 2005-2016 who had two reliable 24-h dietary recalls and mortality linkage. Participant-level mean first eating time and eating-window duration were standardised and classified by K-means. Survey-weighted Cox models incorporated strata and primary sampling units within six chronic disease domains; missing covariates were multiply imputed in thirty datasets and prespecified Benjamini-Hochberg false-discovery-rate families addressed multiplicity. Three patterns represented 30·2 % (early-start), 57·2 % (intermediate) and 12·6 % (late-start) of the weighted population. During 224 837·9 person-years, 3279 deaths occurred. None of the six all-cause mortality global tests met the prespecified false-discovery-rate threshold (all q ≥ 0·281). In the cancer domain, the late-start pattern yielded a higher estimated hazard of cancer mortality than early-start (hazard ratio 1·88, 95 % CI 1·15-3·06; P = 0·012), but the global q was 0·052 and pairwise q was 0·073. This association attenuated after exclusion of early deaths and was not reproduced in an independent Day 1 analysis. Proportional hazards were not supported in three models, for which estimates weakened or crossed the null over follow-up. Eating-timing patterns were not universal mortality predictors across chronic diseases. Their nutritional relevance appears context-dependent, and repeated timing assessments with clinical measures of disease severity are needed before these patterns can inform personalised nutritional care.