Association between methylmercury exposure from fish consumption and child development at five and a half years of age in the Seychelles Child Development Study: an evaluation of nonlinear relationships.

Studies to date of the developmental effects of pre- and postnatal methylmercury exposure from fish consumption in the Seychelles Islands, using linear regression models for analysis, have not shown adverse effects on neurodevelopmental test scores. In this study we evaluated whether nonlinear effects of methylmercury exposure were present, using scores on six tests administered to cohort children in the Seychelles Child Development Study at 66 months of age. Prenatal exposure was determined by measuring mercury in a segment of maternal scalp hair representing growth during pregnancy. Postnatal exposure was measured in a segment of the child's hair taken at 66-months of age. Generalized additive models (GAMs), which make no assumptions about the functional form of the relationship between exposure and test score, were used in the analysis. GAMs similar to the original linear regression models were used to reanalyze the six primary developmental endpoints from the 66-month test battery. Small nonlinearities were identified in the relationships between prenatal exposure and the Preschool Language Scale (PLS) Total score and Child Behavior Check List (CBCL) and between postnatal exposure and the McCarthy General Cognitive Index (GCI) test scores. The effects are best described graphically but can be summarized by computing the change in the predicted test score from 0 to either 10 or 15 ppm and then above this point. For the PLS the trend involved a decline of 0.8 points between 0 and 10 ppm followed by an increase (representing improvement) of 1.3 points above 10 ppm. For the CBCL there was an increase of 1 point from 0 to 15 ppm, and then a decline (improvement) of 4 points above 15 ppm. The GCI increased by 1.8 points through 10 ppm and then declined 3.2 points (representing worse performance) above 10 ppm. These results are not entirely consistent. Two of the trends involve what appear to be beneficial effects of prenatal exposure. The one possibly adverse trend involves postnatal exposure. In every case the trend changes direction, so that an effect in one direction is followed by an effect in the opposite direction. Because of the descriptive nature of GAMs it is difficult to provide a precise level of statistical significance for the estimated trends. Certainly above 10 ppm there is less data and trends above this level are estimated less precisely. Overall there was no clear evidence for consistent (across the entire range of exposure levels) adverse effects of exposure on the six developmental outcomes. Further nonlinear modeling of these data may be appropriate, but there is also the risk of fitting complex models without a clear biological rationale.