Summary
Introduction. Diet quality scores are well-established predictors of cardiometabolic health but compress an individuals food intake into a single dimension, potentially masking dietary variation relevant to disease risk. Whether this hidden variation carries prognostic information has not been tested at scale. Methods. We applied non-negative matrix factorization (NMF) to the reported intake of 148 individual foods and beverages from 87,162 UK Biobank participants. We tested the association of NMF-derived intake factors with incident coronary artery disease (CAD), type 2 diabetes (T2D) and chronic kidney disease (CKD) over a median 9.4 years of follow-up using Cox proportional hazards models. We evaluated the prognostic value of the factors beyond diet quality scores using stepwise Cox models and net reclassification improvement (NRI), and characterized their metabolic profiles through metabolome-wide association analyses (MWAS) of 249 circulating metabolites. Results. NMF identified eight reproducible intake factors that were stable across geographically distinct samples. Four factors were significantly associated with incident cardiometabolic disease: a high sugar and processed foods factor was adversely associated with T2D (HR 1.19; 95% CI 1.11-1.27) and CKD (HR 1.15; 1.07-1.23), while a plant-based and fish factor was protective for CKD (HR 0.90; 0.84-0.96) and a cheese and whole grains factor was protective for CAD (HR 0.94; 0.90-0.98). The factors improved disease discrimination beyond both diet quality scores ( {Delta}C up to +0.0049) and correctly reclassified 8-16% more individuals, whereas adding either score to models containing the factors produced negligible further improvement. The high sugar and processed foods factor was associated with depletion of omega-3 fatty acids and the plant-based and fish factor showed enrichment in polyunsaturated fatty acids. Conclusions. Food-level dietary intake profiling captures clinically relevant variation that established diet quality scores mask, offering a path toward more personalized dietary assessment for cardiometabolic disease prevention.
Regional applicability
This study was conducted in the United Kingdom using UK Biobank data, making the findings directly applicable to UK populations and healthcare policy. The large, geographically diverse sample and long follow-up period provide robust evidence relevant to UK dietary guidelines and cardiometabolic risk assessment.
Key measures
Incident coronary artery disease, type 2 diabetes, and chronic kidney disease; hazard ratios for eight NMF-derived food intake factors; net reclassification improvement; C-statistic improvements; metabolome-wide association with 249 circulating metabolites
Outcomes reported
Introduction. Diet quality scores are well-established predictors of cardiometabolic health but compress an individuals food intake into a single dimension, potentially masking dietary variation relevant to disease risk. Whether this hidden variation carries prognostic information has not been tested at scale. Methods. We applied non-negative matrix factorization (NMF) to the reported intake of 148 individual foods and beverages from 87,162 UK Biobank participants. We tested the association of NMF-derived intake factors with incident coronary artery disease (CAD), type 2 diabetes (T2D) and chronic kidney disease (CKD) over a median 9.4 years of follow-up using Cox proportional hazards models. We evaluated the prognostic value of the factors beyond diet quality scores using stepwise Cox models and net reclassification improvement (NRI), and characterized their metabolic profiles through metabolome-wide association analyses (MWAS) of 249 circulating metabolites. Results. NMF identified eight reproducible intake factors that were stable across geographically distinct samples. Four factors were significantly associated with incident cardiometabolic disease: a high sugar and processed foods factor was adversely associated with T2D (HR 1.19; 95% CI 1.11-1.27) and CKD (HR 1.15; 1.07-1.23), while a plant-based and fish factor was protective for CKD (HR 0.90; 0.84-0.96) and a cheese and whole grains factor was protective for CAD (HR 0.94; 0.90-0.98). The factors improved disease discrimination beyond both diet quality scores ( {Delta}C up to +0.0049) and correctly reclassified 8-16% more individuals, whereas adding either score to models containing the factors produced negligible further improvement. The high sugar and processed foods factor was associated with depletion of omega-3 fatty acids and the plant-based and fish factor showed enrichment in polyunsaturated fatty acids. Conclusions. Food-level dietary intake profiling captures clinically relevant variation that established diet quality scores mask, offering a path toward more personalized dietary assessment for cardiometabolic disease prevention.
Topic tags
Dig deeper with Pulse AI.
Ask about this record, its theme or its relevance to UK farming and policy. Pulse AI uses selected catalogue evidence and cites the sources it draws on.