Differential Measurement Error across Time and Treatment Groups in Self-Reported Nutrition Data
Embargo until
Date
2019-08-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
In dietary observational studies and randomized controlled trials, researchers often try to ascertain the relationship between nutrient intake and biological outcomes, or the effect of interventions on intake and other outcomes. From a study design perspective, this requires capturing true participant intake, but properly doing so is (nearly) impossible. Intake measurements often rely heavily on self-reported nutrition measurements which are an easy and cost-effective proxy to implement in studies, but may not accurately reflect true nutrient intake. Less frequently, intake measurement relies on measured biological components (such as blood or urine) known as biomarkers. Biomarkers, considered to be the “gold standard”, are more resource and financially intensive, but better represent true intake.
When studies contain both self-reported and biomarker nutrient values (which happens sparingly), researchers can model the measurement error structure for self-reporting errors and attempt to produce less biased results for calculating true but unobservable nutrient intake. Previous measurement error work that investigates the relationship between biomarker and self-reported levels has typically been at a single time point, in a single treatment group, or with respect to basic patient demographics. Few studies have examined the measurement error structure in longitudinal studies, where nutrient intake and self-reported values may change over the course of a study, and by treatment exposure.
Using two longitudinal randomized controlled trials with internal validation data (urine biomarkers and self-reported values), we examine how self-reported sodium error changes as a function of time and/or treatment assignment by comparing it to measured urine sodium.
We find that although true sodium consumption changes across time and treatment group, there is essentially no evidence that the measurement error varies across time or treatment groups. While researchers should consider the effects of time and treatment status when designing longitudinal studies, more evidence is needed on how measurement error changes with regards to time and treatment.
Description
Keywords
Measurement Error