Council on Environmental Health
Article Review

Prepared by: Helen J. Binns, MD, MPH
January 2008

Dose-response relationship of prenatal mercury exposure and IQ: an integrative analysis of epidemiologic data

Axelrad DA, Bellinger DC, Ryan LM, Woodruff TJ

US EPA and Harvard

Environ Health Perspect 2007;114:609-15


This paper applies complex statistical techniques to estimate the relationship between maternal mercury burden and childhood IQ decrement.  Bring along your biostatistician when you read this paper!

Mercury is a ubiquitous environmental pollutant, which is released naturally from geologic actions and has additional widespread dispersal through anthropogenic sources.  The potential for toxicity in humans is greatest during developmental stages.  Mercury accumulates within organisms (bioaccumulation) and its concentration increases as it progresses up the food chain (biomagnification).  Additionally, the potential toxicity of mercury increases as it cycles through the marine environment and is transformed, commonly to monomethyl mercury.  Fetal exposure to mercury, principally through maternal fish consumption, has been associated with reductions in measures of child development and behavior.

Three prospective, longitudinal studies informed the development of the US reference dose for mercury (i.e., estimate of a daily exposure to the human population that is likely to be without appreciable risk of deleterious effects during a lifetime).  These studies included samples of mothers and children in areas with high fish intake: 1) the Faroe Islands in the North Atlantic Ocean, 2) New Zealand, and 3) the Republic of Seychelles in the Indian Ocean.  The Faroe Islands study measured cord blood (parts per billion), while the other two measured maternal hair (parts per million).  Two studies (Faroe Islands and New Zealand) found significant adverse effects of methylmercury, while the Seychelles study did not.  Within the Faroe and New Zealand study populations there was no evidence of a threshold effect. 

In 2001, the US EPA to establish the benchmark dose for mercury (i.e., the lowest dose, estimated from the modeled data, expected to be associated with a small increase in the incidence of an adverse outcome).  Based on expert recommendations, they used the Boston Naming Test (a test of memory retrieval of a word that expresses a concept) results from the Faroe Island study for the modeling process.  The benchmark dose level reflects the lower 95% confidence limit of the cord blood mercury level that is estimated to result in a 5% increase in the incidence of abnormal scores in the Boston Naming Test.  The cord blood mercury level that fit this analysis was 58 PPB (parts per billion), which equates to an ingested dose of 1.081 mg/kg body weight/day.  They then applied a total uncertainty factor of 10 to account for various aspects of the modeling process to achieve a reference dose of 0.1 mg/kg body weight/day.  At this intake the whole blood mercury should be <5.0 PPB and hair <1.0 PPM (parts per million).

Bayesian statistics are used in this analysis.  Bayesian statistics examine the probability of a model given the data (versus “traditional” frequentist statistics that examine the probability of the data given a model [hypothesis]).  Frequentist statistics allow one to compute confidence intervals for estimated parameters, which are assumed to be fixed (nonrandom).  In the Bayesian approach, model parameters are treated as random variables, which allow estimation of the most probable range of the parameter given the observed data.    (Here is a link to a primer on Bayesian Statistics designed to be “easily” understood by high school students --

They applied a Bayesian approach to estimate a dose-response relationship between maternal mercury burden and childhood IQ decrements using published summary statistics from the 3 cohorts.  They focused their outcomes on cognitive endpoints—most specifically using the various study measures (which differed by study and were done at ages 6, 7, or 9 years, also depending on study).  Coefficients from the 3 studies were rescaled so data from all 3 studies could be expressed in similar terms.   Cord blood data from the Faroe Islands study was rescaled to maternal hair using ~200:1 relationship, as previously reported in Faroe Island analyses.  They used a hierarchical random-effects model, which includes study and end point variability as random effects and assumed that variability between studies was 3 times greater than variability within a study.

Best estimate for the change in IQ for each 1 PPM increase in maternal hair mercury is -0.18 (probable range of -0.378 to -0.009). 

Over 50% of US women have mercury concentrations below the lowest measured in these studies; dose-response relationships at such lower levels may not be linear.  Neurodevelopment deficits other than IQ are not considered.

These data can be used to estimate benefits of reducing mercury in the population.

Return to COEH Page