Invited commentary: comparison of food frequency questionnaires.

In this issue, Wirfalt et al. (1) provide a useful comparison of three dietary assessment methods among a population of 101 obese women. The essential findings were mean correlations of 0.60 between a modified Willett/Harvard questionnaire and the Block/ National Cancer Institute (NCI) questionnaire, 0.34 between the Block/NCI questionnaire and 3-day recalls, and 0.36 between the Willett/Harvard questionnaire and 3-day recalls. The Block questionnaire tended to underestimate energy sources (which was not surprising because of its length), and the Willett questionnaire tended to overestimate nutrients from fruits and vegetables (which was not surprising because these items had been substantially expanded in the modification). In general, these findings support the reproducibility and robustness of food frequency methods, as the two food frequency methods varied substantially in format and were administered at different times and under different circumstances. Beyond this basic conclusion, though, the authors may be overinterpreting the findings of statistically significant differences between the two methods (compared with dietary recalls, higher correlations for nutrient-density and energy-adjusted intakes of fat and carbohydrate with the Block/NCI questionnaire, and higher correlations for energy-adjusted and unadjusted intakes of vitamin A and calcium and energy-adjusted intakes of alcohol and polyunsaturated fat with the Willett/Harvard questionnaire). Specifically, the study was not designed to compare questionnaire methods; such a study would have randomized the order of questionnaires and held the conditions of administration constant. In addition, the 3 days of surprise telephone 24-hour recalls that served as the comparison method provided a weak gold standard, although this was probably ideal for assessing compliance in the