Daily dietary intake data derived from self-reported dietary recall surveys are widely considered inaccurate. In this study, methods were developed for adjusting these dietary recalls to more plausible values. In a simulation model of two National Health and Nutrition Examination Surveys (NHANES), NHANES I and NHANES 2007-2008, a predicted one-third of raw data fell outside a range of physiologically plausible bounds for dietary intake (designated a 33% failure rate baseline). To explore the nature and magnitude of this bias, primary data obtained from an observational study were used to derive models that predicted more plausible dietary intake. Two models were then applied for correcting dietary recall bias in the NHANES datasets: (a) a linear regression to model percent under-reporting as a function of subject characteristics and (b) a shift of dietary intake reports to align with experimental data on energy expenditure. After adjustment, the failure rates improved to <2% with the regression model and 4-9% with the intake shift model - both substantial improvements over the raw data. Both methods gave more reliable estimates of plausible dietary intake based on dietary recall and have the potential for more far-reaching application in correction of self-reported exposures.
Keywords: NHANES; bias (epidemiology); computer simulation; diet surveys; energy intake; questionnaires.