Reproducibility of a food frequency questionnaire and stability of dietary habits determined from five annually repeated measurements.

OBJECTIVES

In studies on diet and cancer, diet assessment should address long-term intake. Therefore, the authors determined the 5-year reproducibility of a self-administered 150-item food frequency questionnaire (FFQ) used to assess dietary habits in the Netherlands Cohort Study on diet and cancer. Since the FFQ was repeated more than once, the pure test-retest reliability of the FFQ could be distinguished from the intra-individual change in nutrient intake over time. These results were furthermore used to investigate the measurement error structure of the FFQ.

DESIGN

After baseline administration in 1986, the FFQ was annually repeated from 1987 to 1991 in independent random samples of the cohort (n = 400). Pearson correlation coefficients (r) between baseline and repeated measurements of nutrient intake, calculated for each time interval, were regressed on time interval to provide separate estimates of the test-retest correlation (intercept of regression line) and of the decline in correlation over time (slope). The proportion of correlated measurement error was derived from combining the test-retest results with those from a validation study, in which the FFQ was validated against three 3-day diet records.

RESULTS AND CONCLUSIONS

Response was stable at 82%. The test-retest r ranged from 0.42 for selenium intake to 0.90 for alcohol intake. The slopes of the regression lines were relatively flat, but negative for most nutrients; on average, the decline in r amounted to 0.07 after 5 years, indicating that the potential of a single FFQ measurement to rank subjects according to nutrient intake dropped only slightly over time. This is important for studies on cancer since a long induction period may be involved. It was furthermore shown that the proportion of within-subject (error) variance of the FFQ method that could be attributed to correlated error ranged from 0 to 50%. This finding confirms that a reliability study may underestimate the measurement error of a method.

SPONSORSHIP

Dutch Cancer Society (grants CIVO 86-1 and CIVO 90-3).