Ultra-processed food intake did not predict colorectal cancer in a massive US cohort
In a cohort of older US adults followed for more than two decades, total ultra-processed food intake did not track with incident colorectal cancer. The null overall association matters because the category itself is often treated as uniformly harmful, yet this analysis shows why that framing can be too blunt for cancer epidemiology: ultra-processed foods contain both components linked to lower colorectal cancer risk and components linked to higher risk.
Investigators analyzed data from the NIH-AARP Diet and Health Study, a large prospective cohort of adults aged 50 to 71 years at enrollment in 1995–96. After exclusions for prior cancer, end-stage renal disease, poor self-rated health, proxy response, implausible energy intake, and short follow-up, the final sample included 461,682 participants. Over follow-up through 2018, 10,075 colorectal adenocarcinoma cases were diagnosed. Diet came from a food frequency questionnaire and was mapped to Nova categories, with ultra-processed food intake expressed primarily as sex-specific quintiles of energy-adjusted grams per 1,000 kcal per day.
The main result was straightforward: comparing the fifth with the first quintile of ultra-processed food intake, the hazard ratio was 0.97, with a 95% confidence interval of 0.91 to 1.03 and a trend p value of 0.55. No dose-response signal emerged. For a question this large, that null result is not a technical footnote; it is the center of the paper.
Why a single food-processing label can mask opposite biologic directions
Ultra-processed food is not a biologically uniform exposure. Some foods in that category carry nutrients and structures that align with lower colorectal cancer risk, while others cluster with ingredients already established as carcinogenic or protective. Whole grain, fiber, dairy, and calcium were each inversely associated with colorectal cancer when they came from ultra-processed sources, while meat intake from ultra-processed sources was positively associated with incidence.
The direction of those associations is biologically plausible. Fiber and whole grains alter transit time, support colonic microbial metabolism, and reduce contact between the mucosa and carcinogenic compounds. Calcium has long been linked to reduced colorectal neoplasia risk, likely through binding bile acids and fatty acids in the gut lumen. Dairy intake often travels with calcium in dietary pattern analyses. By contrast, processed meat contributes compounds associated with colorectal carcinogenesis, including nitrosation-related products and cooking-derived mutagens. When those forces coexist inside one broad category, a null overall association becomes easier to understand.
One food label can hide two different risk profiles.
The analysis also addressed the practical tension between food processing and diet-quality scoring. A high proportion of ultra-processed intake does not automatically produce the same nutrient pattern in every person. Earlier work has shown that a diet rich in ultra-processed items can still earn a relatively high diet-quality score, and this cohort study reinforces why broader diet quality cannot be inferred from processing status alone.
Observational design leaves room for misclassification, but the cohort was large enough to test the signal
The exposure assessment relied on self-reported diet, which is always vulnerable to error, and the study explicitly recognized the possibility of residual confounding and exposure misclassification. The food-frequency questionnaire captured 124 food and beverage items, and researchers reconstructed intake through USDA food codes, Nova classification, and ingredient databases. Calibration work in this cohort suggested moderate validity for ultra-processed intake, with energy-adjusted gram-weight coefficients reaching 0.66 for both men and women. Even so, any dietary cohort is limited by what participants remember, report, and the questionnaire can decode.
Multivariable Cox models adjusted for demographic, lifestyle, and health-related factors including age, sex, race or ethnicity, education, physical activity, smoking, and alcohol drinking. Those adjustments are essential because colorectal cancer risk is shaped by clustered behaviors rather than a single dietary exposure. Still, residual confounding can linger, especially when diet quality, smoking, body mass index, screening behavior, and other factors travel together over decades.
The sample size and follow-up duration strengthen the interpretation of the null finding. This was not a small or short study. It was a long-running cohort with enough colorectal cancer events to examine both total ultra-processed intake and the contributions of specific dietary components by source. The design is well suited to identifying broad population-level associations, even if it cannot prove causality.
The result aligns more closely with component-specific guidance than with a blanket warning about processed foods
The study’s conclusion fits current evidence-based guidance better than a simplistic anti-processing message would. Public health recommendations already support higher intake of whole grains, fiber, dairy, and calcium, while discouraging processed meat. Those recommendations are reinforced here, but in a more nuanced way: the risk may sit in the ingredients and food subtypes, not in the processing label alone.
For clinicians, the practical implication is to counsel patients on dietary patterns and specific food groups rather than reducing counseling to a single avoided category. A patient eating a broad array of ultra-processed foods may still differ substantially in colorectal cancer risk depending on whether the pattern is built around processed meat and refined products or around grain, dairy, and fiber-containing items. In other words, processing status is not a surrogate for the entire nutritional profile.
The study’s authors also examined the joint effect of ultra-processed food intake and diet quality measured with the Healthy Eating Index-2020, though the excerpt provided here does not give the detailed joint-effect estimates. Even without those numbers, the framing is important: processing and diet quality are related but not interchangeable constructs.
The remaining uncertainty is not whether dietary prevention matters. It does. The unresolved issue is how best to separate the effects of processing from the effects of the foods embedded within that process, especially when epidemiology depends on self-reported intake and long latency outcomes. For now, the cleanest clinical message is the one supported by this cohort: total ultra-processed food intake was not associated with colorectal cancer incidence, but component-specific effects moved in opposite directions, and those opposing forces likely diluted the overall signal.