Investigators analyzed data from 461,682 US adults aged 50–71 who were cancer‑free at baseline and enrolled in the NIH‑AARP Diet and Health Study. They used a food frequency questionnaire to estimate intake and followed participants for cancer incidence through 2018, during which 10,075 colorectal adenocarcinoma cases were diagnosed. Median ultra-processed food (UPF) intake in the cohort was 293 g/1000 kcal/day, or about 43% of daily energy intake.
The primary result was null. Using multivariable Cox proportional hazards models, adjusted for demographic, lifestyle, and health-related factors, investigators found no association between total UPF intake and colorectal cancer incidence: hazard ratio for the highest versus lowest sex‑specific quintile was 0.97 (95% CI 0.91–1.03), with a p‑trend of 0.55. Short statement: total UPF, as categorized in this cohort, did not predict CRC incidence.
The study decomposed UPF into constituent dietary components and examined those by processing source. Data indicated divergent associations depending on the nutrient and whether it originated from UPF or non‑UPF items.
Whole grain, fiber, dairy, and calcium intakes that were derived from UPF sources were associated with lower CRC incidence in this analysis. By contrast, meat intake from UPF sources was associated with higher CRC incidence. The authors noted that some UPF items—examples include whole wheat breads or dairy products—can contain nutrients previously linked to reduced CRC risk, while others—such as processed meat products—have been tied to higher risk. This mixture of opposing associations helps explain why total UPF may show heterogeneous relationships with CRC in different cohorts.
Investigators assigned disaggregated FFQ items to the Nova classification to categorize foods by processing level. UPF intake (g/1000 kcal/day) was divided into sex‑specific quintiles for the main exposure. The multivariable models adjusted for age, sex, race/ethnicity, education, physical activity, smoking, alcohol drinking, and other lifestyle and health‑related covariates. The analytic approach therefore accounted for major measured confounders, while relying on self‑reported dietary assessment for exposure quantification.
The source article reports subgroup and secondary analyses that parsed dietary components by UPF and non‑UPF source and tested robustness across model specifications. Those component‑level analyses produced the contrasting signals described above—protective associations for whole grain, fiber, dairy, and calcium when consumed via UPF items, and higher risk associated with processed‑meat UPF. The study team emphasized that opposing component associations were a plausible explanation for the null association observed for total UPF intake. The source did not report biological equivalence testing comparing the same nutrients from UPF versus non‑UPF sources.
The authors acknowledged key limitations inherent to the study design. This is an observational cohort; as such, residual confounding cannot be excluded. Dietary intake was self‑reported via FFQ, which introduces potential exposure misclassification from systematic or random error. The UPF classification aggregates heterogeneous products and thus may mask divergent effects of specific foods or nutrients.
Data and code are available under the terms described by the investigators: underlying data are maintained by the National Cancer Institute and can be accessed through proposal approval to the NIH‑AARP Diet and Health Study Steering Committee; analysis code is available on the authors’ GitHub.
Unanswered clinical and research questions remain. The study did not determine whether nutrients or food components from UPF are biologically equivalent to those from minimally processed sources, and it did not provide mechanistic data to explain the opposite directions of association observed for different UPF‑sourced nutrients. Future work will need to resolve whether classification by processing alone is an adequate exposure for cancer risk prediction, or whether component‑level metrics better capture etiologic relations.
Limitations aside, the authors stated that their findings align with existing dietary recommendations: prioritize whole grains, fiber, dairy, and calcium and avoid processed meat to reduce colorectal cancer risk. The report ends on the concrete caveats above and points readers to the publicly available data and code for further scrutiny.