A pooled analysis of the Iraqi and Seychelles methylmercury studies
Title:
A pooled analysis of the Iraqi and Seychelles methylmercury studies
Author:
Carrington, Clark D. Bolger, P. Michael
Appeared in:
Human and ecological risk assessment
Paging:
Volume 6 (2000) nr. 2 pages 323-340
Year:
2000-04
Contents:
Several epidemiology studies have investigated the impact of maternal exposure to methylmercury (MeHg) on childhood development of the central nervous system (CNS). In the present report, data from the Iraqi episode that occurred in 1970 from contaminated grain are integrated with those from a more recent study of a population with a high fish intake in the Seychelles Islands. The latter study had many more subjects whose mercury hair levels that were much lower and more representative of levels typically found in consumers whose MeHg exposure is from fish. The age of onset of talking (AOT), the age of onset of walking (AOW) and a combined measure (CM) that integrated the two were used as common scales of MeHg effect for the two studies. The first step of the analyses involved the construction of separate two-dimensional cumulative frequency tables for each study for different groups spanning the range of hair levels and observed effect for each measure. Models were then fit to the values in the tables that were constructed from four components: (1) A dose-effect function that related hair MeHg to the effect measure; (2) a frequency distribution describing population variability; (3) parameters to represent dose-independent influences on effect; and (4) parameters to represent study dependent influences on effect. When the four submodels were assembled, a series of 1092 candidate models resulted which contained 3 to 7 parameters (e.g., slope, standard-deviation, dose-independent age of talking) whose value could be adjusted to improve the fit. After optimizing the fit of each model, a weighting algorithm that rewards for fit and penalizes for the number of parameters in the model was used to identify the best 200 models. The same algorithm was then used to assign a probability to each model in a probability tree. A two-dimensional Monte-Carlo simulation using the resulting function in combination with exposure values typical of U.S. consumers yielded predicted delays in AOT, AOW, and CM attributable to fish consumption in a variable and uncertain range of 0.000 to 1 day.
Journal description