Maryann Miller

Here's a Great Example of Big headlines about Vegetarian Diets being complete BS 5 headlines (including Oxford's own press release) claimed this new Oxford paper shows a vegetarian diet “slashes” the risk of 5 cancers. THIS IS NOT TRUE. Read on... The paper: nature.com/articles/s41416-0… Oxford release: ndph.ox.ac.uk/news/largest-s… Sky: news.sky.com/story/vegetaria… Independent: the-independent.com/news/uk/… Independent alt: the-independent.com/news/hea… BBC: bbc.com/news/articles/c1w5g1… What the study Actually found in its main adjusted models: *Vegetarians had lower HRs for pancreatic, breast, prostate, kidney cancer, and multiple myeloma. *But vegetarians also had a HIGHER risk of esophageal squamous cell carcinoma. So even before any adjustments, this was never a simple “vegetarian protects against cancer” paper. The central issue is not that the models were “unadjusted”, the real issue is multiple comparisons. This paper tested many associations across diet groups and cancer sites, which raises the odds of false positives if you focus only on nominal P<0.05 findings. To the authors’ credit, they did address this. They explicitly say they report nominally significant findings and also mark which survive false discovery rate correction. News outlets used to have Science Editors would have known that the “5 cancers” claim Disappears once you apply the paper’s own stricter filters. In the GI cancer section, the paper says the FDR-significant findings were: • lower colorectal cancer risk in pescatarians • higher colorectal cancer risk in vegans • higher esophageal squamous cell carcinoma risk in vegetarians Not “vegetarians are robustly protected from 5 cancers.” Then the sensitivity analyses narrow it much further. The authors themselves say the most consistent findings were: • HIGHER risk of esophageal squamous cell carcinoma in vegetarians! • LOWER risk of kidney cancer in vegetarians. The other 9 nominally significant associations were not statistically significant in one or both sensitivity analyses. The strongest vegetarian signal was not protective. It was harmful: Esophageal squamous cell carcinoma in vegetarians HR 1.93 95% CI 1.30–2.87 And that remained statistically significant in the sensitivity analyses. Kidney cancer was the most durable protective vegetarian signal in the sensitivity analyses. But the broad media claim that vegetarian diets “slash risk of 5 cancers by up to 30%” leans on nominal associations that did not survive the paper’s own tougher checks. This is why science reporting so often goes off the rails. The press release highlights the attractive nominal findings. Then ignorant reporters amplify them. Then you blindly believe the headlines and try to be a vegetarian. The caveats about false discovery rate correction, sensitivity analyses, small case numbers, and residual confounding all get buried. At best, this is an interesting observational paper. It is hypothesis-generating, but it does NOT robustly show that vegetarian diets protect against 5 cancers. If anything, the most statistically durable vegetarian finding in the paper was an increased risk of one cancer subtype, not a reduced risk. So, the most accurate headline for this paper is... VEG DIET INCREASES RISK OF ESOPHAGEAL CANCER Also see great tweet about this from @AdamRochussen – at Tennessee, USA