Method issues in dietary data analyses in the Multiple Risk Factor Intervention Trial.

Abstract

The selection process in the Multiple Risk Factor Intervention Trial caused relations between risk factors to differ between participants who were randomly assigned into the study and the screening population. Cigarette smoking, blood pressure, and serum cholesterol were moderately inversely related to each other in the randomly assigned population whereas these relations in the unselected population were direct and small in magnitude. This problem was addressed by covariate adjustment in analyses. The selection process also created an artificially high initial mean concentration of serum cholesterol; the mean plasma concentration at the second screening was 15 mg/dL lower than at the first screening. Most of this difference is attributable to regression to the mean. To account for this problem, emphasis was placed on change in plasma cholesterol over time, calculated from the second-screening measurement. Examination of the reliability of nutrition data based on one 24-h dietary recall showed that nutrient-biochemical relations are subject to considerable regression-dilution bias. The ratios of "within" to "between" components of variability were typically between one and four. Analyses in which multiple follow-up measures were averaged are emphasized in this monograph. Men assigned to the special intervention group reported considerable reductions in total energy intake, which was not consistent with observed weight loss. The most likely explanation for this is underreporting or underconsumption the day before the recall. To partially adjust for this, nutrient data are often expressed both in absolute units and as nutrient densities.