Vegetarian patterns and colorectal cancer: the vegan subgroup was uncertain
The Adventist cohort’s combined vegetarian estimate should not be relabeled as a proven vegan effect.

Illustrative food image, not a photograph of the study intervention.
How the cohort defined the comparison
The Adventist Health Study 2 analysis examined vegetarian dietary patterns and incident colorectal cancer in a North American cohort. After exclusions, 77,659 participants remained from the larger recruited population. Diet was assessed at baseline with a validated quantitative food frequency questionnaire. Researchers separated vegan, lacto-ovo vegetarian, pescovegetarian, and semivegetarian patterns from a nonvegetarian pattern. The presence of these distinct categories matters: the combined vegetarian label did not mean that everyone in that group ate a strictly vegan diet. This was a prospective dietary association analysis using Cox regression with demographic and lifestyle adjustment, not a randomized assignment to different eating patterns.
Orlich MJ, Singh PN, Sabaté J, Fan J et al.. JAMA internal medicine, 2015. PMID 25751512 · DOI
The combined result and the cancer sites
Over a mean follow-up of 7.3 years, investigators documented 380 colon cancer cases and 110 rectal cancer cases, identified primarily through state cancer registry linkages. Comparing all vegetarian patterns combined with the nonvegetarian pattern, the adjusted hazard ratio for colorectal cancers was 0.78, with a 95% confidence interval of 0.64 to 0.95. The colon cancer estimate was 0.81, with a confidence interval of 0.65 to 1.00; the rectal cancer estimate was 0.71, with a confidence interval of 0.47 to 1.06. The combined colorectal finding therefore should not erase the different precision of the separate site estimates.
Orlich MJ, Singh PN, Sabaté J, Fan J et al.. JAMA internal medicine, 2015. PMID 25751512 · DOI
The vegan estimate was not the combined estimate
For the vegan subgroup, the adjusted hazard ratio for colorectal cancer was 0.84 compared with nonvegetarians, with a 95% confidence interval of 0.59 to 1.19. That interval includes 1.00, so this subgroup did not provide a clear lower-risk result on its own. The lacto-ovo vegetarian estimate was 0.82, with a confidence interval of 0.65 to 1.02; the semivegetarian estimate was 0.92, with a confidence interval of 0.62 to 1.37. These subgroup figures are important on a vegan website: the pooled vegetarian association cannot honestly be presented as a demonstrated benefit specific to the vegan category.
Orlich MJ, Singh PN, Sabaté J, Fan J et al.. JAMA internal medicine, 2015. PMID 25751512 · DOI
The strongest subgroup was not vegan
The pescovegetarian category had a colorectal cancer hazard ratio of 0.57, with a 95% confidence interval of 0.40 to 0.82, compared with nonvegetarians. The authors highlighted this category in their conclusion, while noting that prevention implications depend on whether the associations are causal. They also reported similar effect estimates for men and women and for black and nonblack participants. None of these comparisons isolates a zero-oil eating pattern. The defensible interpretation is narrower than a universal dietary claim: combined vegetarian patterns were associated with lower incidence in this cohort, while individual categories, including vegan diets, differed in their estimates and uncertainty.
Orlich MJ, Singh PN, Sabaté J, Fan J et al.. JAMA internal medicine, 2015. PMID 25751512 · DOI