Background:
Ultra-processed foods (UPFs) are a heterogeneous class of foods encompassing multiple subgroups. We investigated how risk estimates for cancer, cardiovascular diseases (CVDs), type 2 diabetes (T2D), and overall mortality associated with UPF consumption are affected when accounting for subgroup- and outcome-specific risks across 32 UPF consumption profiles.
Methods:
A total of 226,334 participants from the European Prospective Investigation into Cancer and Nutrition (EPIC) study, free of cancer, CVD, and T2D at recruitment, were included in this analysis. Dietary habits of all participants were assessed at baseline using validated food frequency questionnaires (FFQs), and all food items were classified according to their degree of processing using Nova. UPFs were further categorized into five subgroups. UPF consumption was modelled using two approaches: (1) a novel subgroup-weighted UPF consumption score, where each UPF subgroup was weighted according to its association with disease outcomes, and (2) an unweighted score, calculated as the unweighted sum of UPF subgroups. Cox proportional hazards models were applied to estimate hazard ratios (HRs), 95% confidence intervals (CIs), and Harrell’s C-index for both scores, which were compared across UPF consumption profiles.
Results:
After a median follow-up time of 10.9 years, 30,493 cancer cases, 10,463 cases of CVD, 10,595 cases of T2D and 22,267 cases of all-cause mortality were ascertained. Both UPF consumption scores showed associations with disease outcomes. Predicted HRs and C-index were consistently higher in models using the weighted score. The differences were most pronounced for T2D, where HRs were estimated at 1.22 (95% CI:1.19-1.26) and 1.11 (1.08-1.15) while the C-Index was estimated at 0.605 (0.596-0.614) and 0.597 (0.588-0.606) for the weighted and unweighted score, respectively. The weighted score showed substantial differences between UPF subgroup associated risks, with sweetened beverages and meat products as the main drivers of disease risk across all profiles and outcomes.
Conclusions:
Our findings indicate that disease risk associated with UPF consumption can be estimated more accurately when accounting for subgroup specific risks.