Correction of diet-outcome association for day-to-day variance in dietaray intake: performance evaluation by...

The use of one or a few 24-hour recalls (24hR) to measure dietary exposure in models to estimate diet-outcome association leads to attenuation of the measure of association and a decrease in the test's power. This is due to daily variation in food intake. The measure of association can be corrected using regression calibration and requires at least one repetition of the 24hR in a subsample of the study population. However, the calibration's performance and the precision of the corrected coefficients can vary according to the characteristics of the study to which it is applied. The aim of this study is to evaluate the performance of correction in different research scenarios in relation to the estimated measure of association and its precision. A population (n = 1,000) was simulated with information on food intake for 200 days and an outcome with an association defined with usual intake (mean for the 200 intake days). The scenarios evaluated were: (a) 100%, 60%, 40%, and 20% of the sample with 2 intake days; (b) individuals with 2, 3, 4, and 5 24hR; and (c) populations with 1,000, 600, and 300 individuals. The coefficients were estimated for 300 random combinations of intake days; mean corrected coefficients were similar to the true coefficient. Precision was lower in all the scenarios: the probability of finding a significant association (when true) varied from 0.47 to 0.29 (100% to 20% with repetition, respectively); 0.47 to 0.78 (2 to 5 days); and 0.47 to 0.15 (1,000 to 300 individuals).