Methods: We adopted two statistical approaches. First, we carried out an additive nonlinear analysis assuming homogeneous prenatal MeHg-outcome relationships to explore overall associations. Second, we applied the regression tree to the Woodcock-Johnson Calculation subtest (it was significantly associated in earlier analyses) and identified 4 clusters based on covariates. Then we used additive models to assess the prenatal MeHg association in each of the four clusters for all seven outcomes. This approach assumes nonlinear associations in each cluster and non-homogeneous associations between clusters.
Results: The additive nonlinear analysis yielded prenatal MeHg curves similar to the linear analysis. For the regression tree analysis, the curves relating prenatal MeHg to outcomes between the 4 clusters differed and some crossed at higher prenatal MeHg levels, suggesting non-homogeneity in the upper range of exposure. Additionally, some of the curves suggested a possible non-linear relationship within the range of exposure we studied.
Conclusion: This non-linear analysis supports the findings from the linear analysis. It shows little evidence to support an adverse association of prenatal MeHg exposure through maternal consumption of fish contaminated with natural background levels. However, the tree analysis suggests that the prenatal exposure/outcome relationship may not be homogeneous across all individuals and that some subpopulations may have an adverse association in the upper range of the exposures studied. More robust data in the higher levels of exposure in this cohort are needed to confirm this finding.
|Journal||Stochastic Environmental Research and Risk Assessment|
|Early online date||6 Sep 2017|
|Publication status||E-pub ahead of print - 6 Sep 2017|
- Child development
- Prenatal exposure
- Regression tree