Evaluation of the 24-hour recall as a reference instrument for calibrating other self-report instruments in nutritional cohort studies: Evidence from the Validation Studies Pooling Project

Authors: FREEDMAN, LAURENCE - Information Management Services, Inc; COMMINS, JOHN - Information Management Services, Inc; WILLETT, WALTER - Harvard School Of Public Health; TINKER, LESLEY - Fred Hutchinson Cancer Research Center; SPIEGELMAN, DONNA - Harvard School Of Public Health; Rhodes, Donna; POTISCHMAN, NANCY - National Cancer Institute (NCI, NIH); NEUHOUSER, MARIAN - Fred Hutchinson Cancer Research Center; Moshfegh, Alanna; Baer, David; KIPNIS, VICTOR - National Cancer Institute (NCI, NIH); ARAB, LENORE - Geffen School Of Medicine; PRENTICE, ROSS - Fred Hutchinson Cancer Research Center; SUBAR, AMY - National Cancer Institute (NCI, NIH)

Submitted to: American Journal of Epidemiology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/2/2016
Publication Date: 7/1/2017
Citation: Freedman, L.S., Commins, J.M., Willett, W., Tinker, L.F., Spiegelman, D., Rhodes, D.G., Potischman, N., Neuhouser, M.L., Moshfegh, A.J., Kipnis, V., Baer, D.J., Arab, L., Prentice, R.L., Subar, A. 2017. Evaluation of the 24-hour recall as a reference instrument for calibrating other self-report instruments in nutritional cohort studies: Evidence from the Validation Studies Pooling Project. American Journal of Epidemiology. 186(1):73-82. https://doi.org/10.1093/aje/kwx039.
DOI: https://doi.org/10.1093/aje/kwx039

Interpretive Summary: Measurement error is inherent in self-reported dietary assessment instruments and leads to biased estimates of diet-disease associations in nutritional cohort studies. Currently, the most commonly used main instrument in dietary cohort studies is a food frequency questionnaire (FFQ), and the most commonly used reference instruments are multiple 24-hour recalls. Results from five large validation studies with biomarkers were pooled for common analysis in order to evaluate the 24-hour recall as a reference instrument for calibrating a FFQ. In nutritional cohort studies based on FFQ reports, 24-hour recalls can be used as reference instruments for adjusting estimates of associations of health outcome with continuous dietary intakes for measurement error. Such use reduces but does not completely eliminate the bias in the risk estimates for continuous dietary intakes and provides over-optimistic estimates of correlations between FFQ and true intake.

Technical Abstract: Dietary self-report instruments involve errors and require calibration for proper interpretation of results. However, only a limited set of biomarkers can be used for calibration, so most investigators use self-report instruments, such as 24-hour recalls. We evaluate the performance of 24-hour recalls as reference instruments. We used data from the Validation Studies Pooling Project, a set of five large dietary reporting validation studies using recovery biomarkers. For food frequency questionnaire (FFQ) reported intakes of energy, protein, potassium and sodium and their densities, we compared estimates of the following FFQ parameters using biomarkers with those using 24-hour recalls: attenuation factors, correlations with truth, and calibration equations. We assessed the bias in estimated relative risk resulting from using 24HRs as reference instruments compared to the bias resulting from making no adjustment for measurement error. Using 24HRs as reference instruments, attenuation factors were substantially overestimated for absolute energy and sodium intakes, but less for protein and potassium and hardly at all for densities. Correlations with truth were substantially overestimated for all dietary components. Calibration equations did not capture dependencies of personal characteristics, especially body mass index. In a disease model with energy and one or more nutrient intakes, the bias in the estimated nutrient relative risks was on average reduced, but the bias in the risk coefficient for energy was unchanged. Using 24HRs as reference instruments for adjusting relative risk estimates for intakes on a continuous scale improves results in comparison to making no adjustment for measurement error.