The Journal of the National Cancer Institute has just published a very large study which finds that eating fruit and vegetables has, at the most, a very weak effect on cancer risk (see BBC report). This could be seen as yet another example of lifestyle epidemiology contradicting itself—only a few days ago a study reported that eating a fried breakfast was a good way of combatting obesity—but there are good reasons to take this particular paper seriously.
Firstly, it is a very large study, involving over 400,000 subjects. Secondly, it is a cohort (or prospective) study, ie. it follows people over a period of years rather than interviewing people who are already ill. In these two respects, it trumps most of the studies that have found lower cancer risk amongst those who eat their 'five-a-day'.
The JNCI study found a relative risk for those with a high intake of fruit and vegetables to be 0.97 (95% CI = 0.96 to 0.99), ie. a 3% reduction, and there are serious doubts over whether even this extremely modest reduction is genuine or the result of confounding factors. There are few areas of science where a 3% reduction would be taken seriously.
Walter Willett, a prominent figure in the epidemiology of diet, has written a frank editorial to accompany the study, calling the association "very weak" and noting that there is not a single type of cancer that is significantly reduced by eating fruit and veg (full free text). The history of how such a belief came about bears repeating:
During the 1990s, enthusiasm swelled for increasing consumption of fruits and vegetables with the expectation that this would substantially reduce the risk of many cancers. Potential reductions as large as 50% were suggested...However, the evidence for a large preventive effect of fruits and vegetables came primarily from case–control studies, which can be readily biased by differences in recall of past diet by patients with cancer and healthy control subjects...
In the late 1990s, the results of large prospective cohort studies of diet and cancer began to accrue, and these did not confirm the strong inverse associations found in most case–control studies. Furthermore, a series of analyses that pooled the data from prospective studies for specific cancer sites confirmed the weak and non-statistically significant associations.
There are clear parallels with the evidence for passive smoking and lung cancer here. In both cases, the largest risks were reported when research was in its infancy (eg. Hirayama, 1981) and most of the evidence came from case-control, rather than cohort, studies. As the years went on, the reported risks diminished, falling from over 2.0 (100% increase) to less than 0.3 (30%).
Just as the World Cancer Research Fund used a meta-analysis of questionable studies in 2007 to condemn almost everything except fruit and vegetables are carcinogenic, so the EPA and SCOTH conducted meta-analyses based on shaky science to condemn secondhand smoke as carcinogenic.
In both cases, larger and more reliable studies found no risk. In the case of secondhand smoke, one of the most important null studies came from the World Health Organisation's IARC with Paolo Boffetta as lead author (1998). It found no statistically significant association with lung cancer despite one of the largest sample groups every studied. In the case of fruit and vegetables, the lead author is, again, Paolo Boffetta, and he finds a significant, but very weak, association.
In 1998, the WHO went to the unprecedented lengths of issuing a press release to contradict one of its own studies, so important was passive smoking (and thereby, smoking bans) to the battle against active smoking. It will be interesting to see if there is any backlash against this new study.
As it is a less heated area, possibly not. Willett concludes his editorial by calling for "heightened efforts to reduce smoking and obesity" which remain the key battlegrounds, but there are many food faddists and vegetarians who will not be happy to hear that their lifestyles are not as healthy as they believed. There will also be many epidemiologists who will (justifiably) feel their work has been discredited.
Ultimately, after a brief period of controversy, the IARC's secondhand smoke report was forgotten about and attention shifted back to the grab-bag of smaller studies which had been favorable to the passive smoking theory. This new study may meet the same fate; already the tiny association it reported is being taken as fact, with all the caveats and doubts ignored:
In any event, a reduced risk of 2.5% should not be dismissed out of hand, the World Cancer Research Fund argues.
"For the UK, this works out as about 7,000 cases a year, which is a significant number," says Dr Rachel Thompson from the charity, which in a major 1997 report said there was "convincing evidence" of the protective effect of fruit and vegetables.
In Velvet Glove, Iron Fist, I described the World Cancer Research Fund's report as "a veritable encyclopedia of weak associations and questionable meta-analyses" (p. 310). Today's JNCI study only reinforces that view. Whatever the truth about this particular issue, basing policy on statistical studies that change like the weather is a fool's errand.