Learn More about Regression Calibration
[glossary term:] Regression calibration is the most popular method in nutritional epidemiology to adjust estimates of associations between diet and health outcomes for [glossary term:] measurement error (see Key Concepts about Measurement Error). In a standard analysis, these diet-health associations are estimated from risk regression models (often logistic regression or Cox regression) relating a health [glossary term:] outcome (often the occurrence, or not, of a disease) to dietary intake. The coefficient of the reported dietary intake represents the estimated diet-health association (often the log relative risk).
In the regression calibration approach, the reported dietary intakes used as explanatory variables in the risk model are replaced by the expected values of the true usual intake predicted from the reported intakes and other [glossary term:] covariates (such as confounders) that are included in the risk model. This expected value of true usual intake is usually obtained from a “calibration equation” (see below). Regression calibration produces an approximately unbiased estimate of the true relative risk for a dietary intake under the condition that measurement errors in the reported dietary intakes are non-differential, that is, they are independent of the disease outcome. This condition of non-differential measurement error is most reliably fulfilled in prospective studies where the main dietary assessment instrument, usually a food frequency questionnaire (FFQ), is assessed before the occurrence of the health outcomes (see the Food Frequency Questionnaire Profile). It is in these studies that regression calibration is most commonly applied.
The predicted values of true usual intake are estimated using “calibration equations” derived from data gathered in a [glossary term:] validation study that includes (often repeated) reference measurements (see Key Concepts about Validation). Ideal reference measurements for usual intake are [glossary term:] recovery biomarkers, because they are unbiased at the level of the individual for true usual intake (Learn More about Biomarkers). However, these are available for only a few nutrients. Therefore, most validation studies use as [glossary term:] reference instruments more detailed and less biased dietary assessment instruments, such as [glossary term:] 24-hour dietary recalls (24HR) or multiple-day [glossary term:] food records (see 24-hour Dietary Recall Profile and Food Record Profile). The working assumption is that such reference instruments are unbiased for true usual intake, even though they fall short of this ideal. Although using these imperfect reference instruments does not completely adjust estimated diet-outcome associations for the [glossary term:] bias caused by dietary measurement error, on average, it may produce less biased results than an unadjusted standard analysis based on the reported intakes.
Typically, in large cohorts, an FFQ is administered to all study participants as the [glossary term:] main dietary assessment instrument, and an internal validation study is conducted in a subset of the participants to which repeat administrations of a reference instrument are also administered. The regression calibration method can also be applied using an [glossary term:] external calibration study, in which the reference data are collected in a different but preferably similar population to the study population, and the participants report their dietary intakes with the same instrument as used in the main study.
For More Information
Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu C. Measurement Error in Nonlinear Models: A Modern Perspective. Chapman & Hall/CRC Monographs on Statistics & Applied Probability (2 ed.). CRC Press. ISBN 9781584886334.
Freedman LS, Schatzkin A, Midthune D, Kipnis V. Dealing with dietary measurement error in nutritional cohort studies. J Natl Cancer Inst 2011 Jul 20;103(14):1086-92. [View Abstract]
Kipnis V, Subar AF, Midthune D, Freedman LS, Ballard-Barbash R, Troiano RP, Bingham S, Schoeller DA, Schatzkin A, Carroll RJ. Structure of dietary measurement error: results of the OPEN biomarker study. Am J Epidemiol 2003 Jul 1;158(1):14-21; discussion 22-6. [View Abstract]