Applying non-negative matrix factorization to 148 foods and beverages reported by 87,162 UK Biobank participants, the analysis extracted eight stable intake factors. Over a median 9.4 years, a high sugar and processed foods factor raised type 2 diabetes (HR 1.19) and chronic kidney disease risk (HR 1.15). A plant-based and fish factor lowered CKD risk (HR 0.90), and a cheese and whole grains factor lowered coronary disease risk (HR 0.94). The factors improved discrimination beyond diet quality scores and reclassified 8-16% more people correctly. Metabolomics linked the sugar factor to depleted omega-3 fatty acids.
The central message is that a single healthy-eating score flattens meaningful differences in what people actually eat. The cheese-and-whole-grains result echoes earlier cohort work suggesting fermented dairy is neutral to mildly protective, a finding that sits awkwardly with older saturated-fat guidance. Still, effect sizes are modest (the largest C-statistic gain was under 0.005), so clinical utility for individual prediction remains unproven. UK Biobank participants are healthier and less diverse than the general population, and the dietary data rest on self-report, with residual confounding from lifestyle and socioeconomic factors likely in any observational design. Associations are not proof that changing these foods changes outcomes. For readers, the practical takeaway is reassuringly familiar: less sugar and ultra-processed food, more plants, fish and whole grains. This is a preprint that has not been peer reviewed, so the numbers may shift. Overall, it is an incremental refinement of dietary assessment rather than a paradigm shift.