**Background:** Self-reported dietary intake is widely used in nutrition epidemiology but is known to be biased compared with biomarker measurements. The Goldberg cutoffs are commonly applied to exclude individuals whose reported energy intake is implausible relative to predicted energy expenditure, under the assumption that this also reduces bias in other self-reported nutrients and in nutrition-health associations. However, whether this approach actually reduces bias in associations has not been rigorously tested.
**Methods:** The authors analyzed data from 303 participants in the Interactive Diet and Activity Tracking in the American Association of Retired Persons (IDATA) study. Self-reported intakes (energy, sodium, potassium, protein) were measured using the Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24). Biomarker-based intakes were derived from doubly labeled water (energy) and 24-hour urine samples (sodium, potassium, protein). The Goldberg cutoffs were applied using predicted basal metabolic rate (Cunningham equation: BMR = 370 + 21.6 × fat-free mass) and a physical activity level of 1.75, with within-subject coefficients of variation of 23% for energy intake, 8.5% for BMR, and 15% for PAL, over 7 days of assessment. Percent bias (b) and percent remaining bias (r) were calculated for regression coefficients between each nutrition intake and six health outcomes (body weight, waist circumference, heart rate after fitness test, resting systolic and diastolic blood pressure, VO2 max). Jackknife estimation was used for 95% confidence intervals. A simulation study (1000 replicates, n=100 each) preserved the characteristics of IDATA by resampling biomarker intakes and fat-free mass, generating health outcomes from linear models, and adding reporting errors modeled as polynomial functions of biomarker intake percentiles with normally distributed errors.
**Key Results:** In the empirical data, mean underreporting was significant for all four nutrients: energy intake (EI_SR − EI_BIO = −352.3 ± 811.0 kcal/day, p<0.001), sodium (−557.0 ± 2075.2 mg/day, p<0.001), potassium (−279.5 ± 1404.8 mg/day, p<0.001), and protein (−11.5 ± 41.6 g/day, p<0.001). The Goldberg cutoffs excluded 120 of 303 participants (40%). In accepted cases, mean bias was not significantly different from zero for any nutrient (e.g., energy: −26 ± 33 kcal/day; potassium: 3 ± 98 mg/day), while rejected cases showed significant underreporting (energy: −850 ± 89 kcal/day, p<0.001). For nutrition-health associations, significant underestimation (b<0) was observed only for energy intake with body weight and waist circumference. After applying Goldberg cutoffs, bias was reduced but remained significant for these two associations. In the simulation, bias in regression coefficients was reduced in 14 of 24 nutrition-outcome combinations after applying Goldberg cutoffs, with the largest reductions for energy intake. However, bias was not eliminated: coverage probabilities improved but remained below 95% in most cases. Mean squared error also decreased but did not reach biomarker-based levels. Sensitivity analyses varying sample size (n=50, 100, 200, 300) and error parameters showed consistent results; bias magnitude increased with larger reporting error standard deviation.
**Clinical Implications:** These findings have important implications for nutrition epidemiology. While the Goldberg cutoffs effectively correct mean intake estimates—useful for population surveillance—they are insufficient for removing bias in diet-disease association studies. Researchers should not assume that excluding extreme reporters adequately addresses confounding by reporting bias in studies of nutrition and health outcomes. More sophisticated statistical methods, such as regression calibration using biomarker data in validation subsamples, may be needed. The study is limited by its predominantly white, older sample (mean age 63, 93.4% non-Hispanic white), use of only bivariate associations, and reliance on a single self-report tool (ASA24). Generalizability to other populations and dietary assessment methods requires further investigation.