A study says eating oranges and carrots three times a week may help prevent mouth cancer, thanks to their curcumin. Curcumin is the yellow in turmeric, and the survey the study used recorded four mouth cancers where the paper counts 59

Some studies are worth reading for what they find. This one is worth reading for what it shows about how a finding gets made. It appeared on 29 September 2026 in Oral Health & Preventive Dentistry, a peer-reviewed dental journal, and its conclusion is the kind that travels well: eat oranges or carrots at least three times a week and you may lower your risk of mouth cancer.

The authors, six dentists at hospitals in Shijiazhuang and Baoding in China's Hebei province, did not recruit anyone. They used NHANES, the long-running US government health survey, and took the 2003 to 2004 round. Their sample was 2,762 people, of whom 59 were counted as having had cancer of the mouth, tongue or lip. Everyone had also been asked how often they ate oranges and carrots.

On the face of it the gap was wide. Nearly half of the cancer group, 28 of 59, said they never ate oranges, against 12% of everyone else. For carrots it was 39% against 9%. Turned around: about 8 in 100 people who never ate oranges were in the cancer group, and about 1 in 100 of those who ate them three or more times a week. The paper's models put the odds of mouth cancer in frequent carrot eaters at less than a tenth of the odds in people who never ate them.

Then comes the explanation, and the first problem. The paper calls oranges and carrots 'curcumin-containing foods' throughout, and says curcumin is 'widely present in foods such as turmeric, oranges, and carrots'. Curcumin is the yellow pigment of turmeric, a root in the ginger family. Carrots get their colour from carotenoids such as beta-carotene, which is a different family of compounds altogether. The paper itself says the survey has no data on how much curcumin anyone ate.

It works out a dose anyway. A serving of carrots or oranges, it says, has 'trace amounts' of curcuminoids, about 1 to 5 mg, and it gives no reference for that. Three servings a week would then come to 3 to 15 mg. The paper's figure is 150 to 300 mg a week, or 20 to 40 mg a day, which is what you get if you use the number it quotes one line earlier for a serving of curry, 50 to 100 mg. That 20 to 40 mg a day then goes into the abstract as advice for clinical practice.

The second problem is the 59. NHANES publishes a codebook for every questionnaire, with the number of people who gave each answer. The paper names the question it used, the age at which mouth, tongue or lip cancer was first diagnosed. In the 2003 to 2004 codebook, four people in the whole survey answered it. The companion question, on what kind of cancer a person had, also shows four mentions of mouth, tongue or lip. We could not find anything in the paper that explains how four became 59. It does say that missing values were filled in using software, without saying which values or how. We have not put this to the authors.

The 59 look odd in other ways. They would make up 2.1% of the sample, when the US National Cancer Institute puts the lifetime risk of a wider group of cancers, mouth and throat together, at about 1.2%. Their typical age when surveyed was 46 and almost two thirds were women. In US cancer registries the typical age at diagnosis is 65, and men are diagnosed about twice as often as women.

Set that aside and take the table as printed. The cancer group was much poorer, with a household income only just above the poverty line while the rest sat at well over twice it. They were also far more likely to smoke: 47.5% were current smokers against 19.8%. Smoking is one of the main causes of mouth cancer, and people with less money tend to eat less fresh fruit and vegetables. The paper says its odds ratios were adjusted for 'potential confounding factors' but does not list them. And for oranges there is no dose pattern: people who ate them rarely had the lowest odds, not people who ate them often.

There is also the order of events, which is our point and not one the paper raises. Everyone was asked about their diet at the time of the survey, which for the cancer group was after the diagnosis. Surgery or radiotherapy to the mouth can make an acidic fruit or a hard raw vegetable difficult to eat, sometimes for a long time. Cancer could have changed the diet as easily as the other way round. The authors do say, at the very end, that a one-off survey like this cannot show cause.

Last, the prediction model. The abstract says it scored above 0.96 for precision and recall and so has 'strong predictive ability'. The paper's own chart of the test shows 829 people, 18 of them with cancer. The model found three of the 18. It also flagged nine people who did not have cancer. Labelling all 829 as healthy would have got 811 right, six more than the model managed. When one group is 2% of the total, a weighted score near 97% says almost nothing.

So what should you take from it? Not a curcumin dose, and not a reason to buy turmeric capsules; the US National Institutes of Health's own summary says there is not enough evidence to conclude that turmeric or curcumin helps any health condition. Oranges and carrots are good food for plenty of well-supported reasons, and nothing here changes that. What has solid evidence behind it for mouth cancer is not smoking or chewing tobacco, keeping alcohol low, and getting any mouth ulcer or patch that has not healed in about three weeks looked at by a dentist or doctor. The other lesson is about reading: a confident abstract in a peer-reviewed journal can still rest on a compound that isn't in the food and a count the source data doesn't show.