A machine-learning diet score identifies eating patterns associated with lower aging-related mortality
1. Researchers used machine learning on 191,689 UK Biobank participants to develop and validate the Machine-learning YouTHful (MYTH) Diet, a dietary score associated with lower aging-related mortality in both an internal validation cohort and an external NHANES cohort
2. The MYTH Diet performed comparably to, rather than consistently better than, established indices such as the Healthy Eating Index and DASH, and because the study is observational and machine-learning derived, it identifies food combinations that predict mortality rather than establishing that following the diet will extend lifespan.
Diet is one of the few modifiable determinants of aging, and an npj Science of Food study published August 7th analyzed 191,689 UK Biobank participants (mean age 56.1 years; 55.1% female) to develop a dietary pattern built around aging-related mortality rather than individual diseases. The investigators grouped 206 food items into 34 food groups and identified 18 that were significantly associated with mortality after false discovery rate correction. A LightGBM model ranked their importance, and the final score comprises ten components (range 0 to 10): poultry, nuts, and refined grains are encouraged, processed meat, red meat, butter, and sweetened beverages are discouraged, and legumes, coffee, and alcohol are scored within a moderate intake range. Over a median 12.2 years, 13,652 participants (7.1%) experienced aging-related death. Higher scores were associated with lower mortality in the UK Biobank (top vs. bottom quartile HR 0.79) and in NHANES (HR 0.68). Performance was comparable to the Healthy Eating Index, which showed a stronger association when scores were modeled continuously in NHANES, and results were consistent after excluding early deaths to address reverse causation. A 50-protein signature of the score explained an estimated 26.7% of its association with mortality, with TNFRSF4 and CD74 among the leading mediators. Higher scores were also linked to slower aging on plasma protein-derived clocks for the lung, pancreas, kidney, liver, and arteries, and to lower risk of 15 of 49 age-related diseases tested.
These findings carry important limitations. The mediation analyses are hypothesis-generating, since they rest on untestable assumptions and single-time-point omics measurements, and dietary intake came from 24-hour recalls that may not reflect habitual intake. The model’s discriminative performance was modest (AUC 0.689), both cohorts are predominantly of Western or White European background, and UK Biobank participants tend to be healthier and of higher socioeconomic status than the general population. Some thresholds are also counterintuitive: refined grains were encouraged and alcohol was scored within a moderate range, patterns that reflect statistical associations in these cohorts rather than recommendations to begin drinking or to substitute refined for whole grains. For clinicians, the practical message is less about adopting a branded longevity diet and more about the consistency of the underlying pattern, since limiting processed and red meat, butter, and sweetened beverages while including nuts, legumes, and minimally processed foods aligns with existing dietary guidance. In practice, this supports counselling on overall diet quality and a few achievable changes rather than scoring systems, individualizing advice to cultural food practices, allergies, and comorbidities, and not recommending alcohol for health benefit. Whether adherence to this pattern extends lifespan or slows biological aging will require randomized trials and validation in more diverse populations.
©2026 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.
