Abstract
AbstractIn nutritional epidemiology, self-reported assessments of dietary exposure are prone to measurement errors, which is responsible for bias in the association between dietary factors and risk of disease. In this study, self-reported dietary assessments were complemented by biomarkers of dietary intake. Dietary and serum measurements of folate and vitamin-B6 from two nested case-control studies within the European Prospective Investigation into Cancer and Nutrition (EPIC) study were integrated in a Bayesian model to explore the measurement error structure of the data, and relate dietary exposures to risk of site-specific cancer. A Bayesian hierarchical model was developed, which included: 1) an exposure model, to define the distribution of unknown true exposure (X); 2) a measurement model, to relate observed assessments, in turn, dietary questionnaires (Q), 24-hour recalls (R) and biomarkers (M) to X measurements; 3) a disease model, to estimate exposures/cancer relationships. The marginal posterior distribution of model parameters was obtained from the joint posterior distribution, using Markov Chain Monte Carlo (MCMC) sampling techniques in JAGS. The study included 554 and 882 case/control pairs for kidney and lung cancer, respectively. In the measurement error component, the error correlation between Q measurements of vitamin-B6 and folate was estimated to be equal to 0.82 (95% CI: 0.76, 0.87). After adjustment for age, center, sex, BMI and smoking status, the kidney cancer odds ratios (OR) were 0.55 (0.16, 1.31) and 1.07 (0.33, 3.44) for one standard deviation increase of vitamin-B6 and folate, respectively. For lung cancer ORs were 0.85 (0.27, 2.42) for vitamin-B6 and 0.55 (0.14, 1.39) for folate. Bayesian models offer powerful solutions to handle complex data structures. After accounting for the role of measurement error, folate and vitamin-B6 were not associated to the risk of kidney and lung cancer.
Publisher
Cold Spring Harbor Laboratory