Methodologic considerations for investigating the diet-cancer link.

Evidence from human ecological studies and experimental animal studies suggest that a number of dietary factors may have a role in the etiology of cancers of various sites. When associations are examined within populations on the level of the individual, they often weaken or disappear. Although in some cases the suspect nutrient may have no real carcinogenic effect, it is proposed that there are at least three important methodologic problems that could prevent the observation of a true association between dietary factors and human cancer. First, diet assessment methods are inadequate to estimate true exposure with sufficient accuracy and precision especially over long periods. Second, use of retrospective diet assessment methods in case-control study designs can often introduce an important bias. Third, sufficient within-study-group contrasts are often lacking. These problems are discussed in interpreting recent cancer studies of diet, and recommendations are made for future research.