The impacts of measurement errors on a dietary pattern analyses:a simulation study based on dietary data from the China Multi-Ethnic Cohort (CMEC) study

Abstract

Background: Measurement error is a significant challenge in nutritional epidemiology research. Compared with traditional, isolated-nutrient research, dietary-pattern studies provide a more comprehensive approach to chronic disease prevention and have become popular in recent years. However, few studies have examined the impacts of measurement errors on dietary pattern analyses.

Objectives: We investigated the impacts of measurement errors on the 2 most commonly used dietary pattern derivation methods: principal component factor analysis (PCFA) and K-means cluster analysis (KCA).

Methods: We conducted a simulation study by taking the dietary data collected in the China Multi-Ethnic Cohort study as the "true values" and adding linear measurement errors for each food group to consider both systematic and random errors. We investigated the impacts of measurement errors from 2 aspects: distortion of the derived dietary patterns and estimated associations between the dietary patterns and disease.

Results: For both systematic and random errors, larger measurement errors caused more serious distortion of dietary patterns derived by PCFA and KCA, with consistency rates ranging from 67.5% to 100% and from 13.4% to 88.4%, respectively. In addition, for both systematic and random errors, larger measurement errors caused greater attenuation effects on the association coefficients. For a beneficial association (coefficient, -0.5), the estimated coefficients ranged from -0.287 to -0.450 and from -0.231 to -0.394 in the PCFA and KCA, respectively. For a harmful association (coefficient, 0.5), the estimated coefficients ranged from 0.295 to 0.449 and from -0.003 to 0.373 in the PCFA and KCA, respectively. Dietary patterns derived by PCFA with factor loadings of low discrepancies and dietary patterns derived by KCA with small cluster sample sizes are more vulnerable to measurement error.

Conclusions: Measurement errors could distort dietary patterns and attenuate the dietary pattern-disease association. The stability of dietary patterns under measurement errors differs when using PCFA and KCA.

Keywords: attenuation effect; dietary pattern; measurement error; nutritional epidemiology; simulation study.