Effects of intraindividual and interindividual variation in repeated dietary records

In this study 5,115 daily diet records were collected from 151 women on two randomly selected days per sampling month (29 days) over a two-year period. Yearly estimates of the ratios of intraindividual (sigma w2) and interindividual (sigma b2) variance components were calculated for dietary intake of 15 nutrients and for dietary intake + vitamin/mineral supplements. From one year to the next, corresponding ratios of sigma w2/sigma b2 were nearly identical. Intraindividual variation in dietary intake of all 15 nutrients was greater than interindividual variation (sigma w2/sigma b2 greater than 1). Variance component ratios for highly supplemented nutrients such as iron, vitamin C and B vitamins were less than one (sigma w2/sigma b2 less than 1). Using the ratios of sigma w2/sigma b2, it was calculated that between six and 99 repeated records measuring dietary intake and between two and 31 records measuring total intake may be needed per individual to ensure that the estimate of the population correlation (rho DF) between an individual's "usual" dietary intake or total intake of a dietary risk factor and an individual's mean or usual level of a physiologic risk factor was within 10% of the true population correlation coefficient (rho xy). It was also found that twice as many dietary records per individual were required to estimate the population slope (beta xy) within 10%. These results have serious implications for the design and analysis of prospective nutritional studies.