Exposure misclassification of self-reported vegetarian diet and its impact on health outcomes

Literature around the health impact of a vegetarian diet is expanding, yet exposure measurement and its effect on the study results receive far less attention. The aim of our study was to show the effect of bias from exposure misclassification on the direction, amount, and uncertainty of study results.

We included 11 180 participants from the National Health and Nutrition Examination Survey (2007-10). The exposure was self-reported diet type (being vegetarian). Using sensitivity, specificity, and positive and negative predictive values as bias parameters, we quantified the effect of exposure misclassification in non-differential and differential scenarios. We compared conventional analysis with bias-adjusted results.

Among self-reported vegetarians (2.2%), more than half (68%) reported consuming meat, poultry, or fish in the 24-hour dietary recalls. The sensitivity of self-reported diet type was 32% (95% CI, 0.26-0.38) and the specificity was 98% (95% CI, 0.98-0.99). The exposure misclassification was differential. The quantitative bias analysis of differential exposure misclassification was more informative. Using positive and negative predictive values as bias parameters was helpful to avoid negative bias-adjusted cell counts.

Assessing the impact of exposure misclassification in the self-reported vegetarian diet on four health outcomes, our study demonstrates that quantitative bias analysis can help to make more plausible assumptions about the possible direction, strength, and uncertainty of bias and its impact on the effect estimates of conventional analysis. However, comparing different scenarios showed that the model assumptions play a key role in the informative value of quantitative bias analysis.