Plenary session - Mediation analyses for environmental epigenetic studies; Linda Valeri, Harvard Medical School
Does DNA-methylation mediate the effect of maternal smoking on birth weight? - The impact of exposure measurement error in mediation analyses for environmental epigenetic studies
Linda Valeri, Instructor at Harvard Medical School at the Psychiatry Biostatistics Laboratory, USA
Mediation analyses that investigate whether epigenetic alterations mediate toxicity from environmental chemicals are popular in the field of environmental epidemiology. Such analyses often ignore that the environmental exposure is measured with error. We assess the impact of exposure measurement error on the validity of estimation and hypothesis testing of the indirect effect of maternal smoking on birth weight, potentially mediated by DNA methylation in a Norwegian cohort.
We study bias due measurement error on a continuous or binary exposure in the estimation of direct and indirect causal effects. Further, we evaluate type I error for tests of mediation in this setting. SIMEX and regression calibration approaches to mitigate bias due to measurement error are proposed and examined. The approaches are applicable when a gold standard is available, or are coupled with sensitivity analyses in the absence of auxiliary data. We then evaluate recent evidence of methylation of cytosine-phosphate-guanine sites in the GFI1 gene as potential mediators of the smoking-birth weight relationship in a prospective birth-cohort study accounting for measurement error.
Our analytic results and simulation study demonstrate that ignoring measurement error leads to conservative estimates of direct effects, while the indirect effects could be biased either away or towards the null. Further, exposure measurement error can lead to inflated Type I errors of tests for mediation. Analysis of the cohort study indicates that the indirect effects previously reported, that ignore measurement error in self-reported smoking behavior, are potentially over-estimated.
When the environmental exposure is measured with error the validity of mediation analysis can be severely undermined. Measurement error correction approaches can mitigate this bias and should become routine practice in environmental epigenetic studies.
Contact person: Lars Alfredsson