The x-axis represents the residuals of mercury regressed on age, race
and ethnicity, sex, technician, educational achievement, assets, body mass
index, alcohol use, and diabetes. The y-axis represents the residuals of Rey
complex figure delayed recall and finger tapping scores regressed on the same
Weil M, Bressler J, Parsons P, Bolla K, Glass T, Schwartz B. Blood Mercury Levels and Neurobehavioral Function. JAMA. 2005;293(15):1875–1882. doi:10.1001/jama.293.15.1875
Author Affiliations: Departments of Environmental
Health Sciences (Ms Weil and Drs Bressler, Bolla, and Schwartz) and Epidemiology
(Drs Glass and Schwartz), Johns Hopkins Bloomberg School of Public Health,
and Departments of Neurology (Dr Bolla), Medicine (Dr Schwartz), and Psychiatry
and Behavioral Sciences (Dr Bolla), Johns Hopkins Medical Institutions, and
Kennedy Krieger Institute (Dr Bressler), Baltimore, Md; and Wadsworth Center,
Trace Metals Laboratory, New York State Department of Health, Albany (Dr Parsons).
Context Due to its cardiovascular benefits, fish consumption is widely encouraged
among older Americans. However, this fast-growing population is at increased
risk of cognitive impairment and may be particularly sensitive to methylmercury,
a neurotoxicant found in fish.
Objective To describe associations of blood mercury levels with neurobehavioral
test scores in an urban adult population.
Design, Setting, and Participants Cross-sectional analysis to determine the effect of mercury levels on
neurobehavior in 474 randomly selected participants in the Baltimore Memory
Study, a longitudinal study of cognitive decline involving 1140 Baltimore
residents aged 50 to 70 years. We measured total mercury in whole blood samples
and used multiple linear regression to examine its associations with neurobehavioral
test scores. First-visit data were obtained in 2001-2002.
Main Outcome Measures Twenty scores from 12 neurobehavioral tests.
Results The median blood mercury level was 2.1 μg/L (range, 0-16 μg/L).
After adjustment for covariates, increasing blood mercury was associated with
worse performance on Rey complex figure delayed recall, a test of visual memory
(β, −0.224; 95% confidence interval, −0.402 to −0.047).
However, increasing blood mercury levels were associated with better performance
on finger tapping, a test of manual dexterity (β for dominant hand, 0.351;
95% confidence interval, 0.017-0.686).
Conclusion Overall, the data do not provide strong evidence that blood mercury
levels are associated with worse neurobehavioral performance in this population
of older urban adults.
Mercury is ubiquitous in the environment and enters the air during fossil-fuel
combustion, mining, smelting, solid-waste incineration, and natural degassing
of the earth.1 It is converted to methylmercury
by microorganisms, enters the food chain, and bioaccumulates in predatory
fish. Consumption of certain fish and crustaceans (hereafter referred to as
fish) is the primary source of methylmercury exposure in the general population.1,2
Methylmercury distributes rapidly throughout the body and easily crosses
the blood-brain barrier into the brain, where it may become trapped after
demethylation.1 Generally, changes in nervous
system function are considered the most sensitive health end point1,3; however, recent evidence indicates
that adverse cardiovascular effects may occur at even lower levels,3 possibly leading to further cognitive effects.4- 9 Total
blood mercury is considered the most valid biomarker of recent methylmercury
Recent regulations for mercury emissions, the increasing trend in fish-consumption
advisories, clinical studies, and heightened media attention have led to the
emergence of mercury as a leading public health concern.11- 16 The
US Environmental Protection Agency, the US Food and Drug Administration, and
the National Research Council all recently addressed the risks associated
with eating mercury-contaminated fish, focusing on children and women of child-bearing
age.3,17- 20 Fish
consumption, however, is frequently recommended for older adults due to its
high omega-3 fatty acid content, well-documented cardiovascular benefits,
and, more recently, its possible protective association with Alzheimer disease.21- 24 Since
the aging nervous system is more sensitive to neurotoxicants, there is reason
for concern about mercury contamination in fish, especially now that baby
boomers are approaching that point when age-related cognitive decline becomes
apparent.25- 27 Given
the longer life expectancy of that generation, a dramatic increase in the
prevalence of cognitive dysfunction is anticipated.28 For
this reason, investigating mercury exposure in the older population is considered
a public health priority.
We analyzed blood mercury levels and neurobehavioral test scores in
474 participants from the Baltimore Memory Study, which involved 1140 randomly
selected, 50- to 70-year-old Baltimore, Md, residents. To our knowledge, this
is the first study to examine associations between mercury exposure and neurobehavioral
outcomes in a representative sample of older adults in the United States.
The population, design, sampling, and recruitment methods for the Baltimore
Memory Study have been described.29 In brief,
residents were sampled by neighborhood to ensure variability by socioeconomic
status, race, and ethnicity. A total of 18 826 households with telephone
numbers were randomly selected and recruited. Eligibility requirements included
living in a targeted neighborhood for at least 5 years and being between 50
and 70 years old. Among the 2351 eligible residents, 1430 (60.8%) were scheduled
for a clinic visit and 1140 were enrolled. The first of 3 study visits occurred
between May 30, 2001, and September 20, 2002. The Committee for Human Research
at the Johns Hopkins Bloomberg School of Public Health approved the study.
All participants provided written informed consent before testing and were
paid $50 for their time. The current study involved cross-sectional analysis
of first-visit data from 474 randomly selected participants of the Baltimore
Memory Study with complete first-visit data and adequate blood specimens for
mercury measurement. Sample size was based on power calculations30 (2-tailed α
= .05; power = 0.89; effect size = 0.03) and budget available for mercury
Data collection methods have been described.29 In
brief, trained technicians administered 20 neurobehavioral tests in the following
7 domains: nonverbal reasoning and intelligence, Ravens coloured progressive
matrices31,32; language, Boston
naming test,33,34 letter fluency,35 category fluency35;
verbal memory, Rey auditory verbal learning test36;
visual memory, Rey complex figure–delayed recall,37 symbol-digit
paired associate learning38; visuoconstruction
and visuoperception, Rey complex figure–copy37;
motor and manual dexterity, Purdue pegboard,39 finger
tapping,40 simple reaction time41;
and executive function, Purdue pegboard–assembly,39 Stroop
test,42 trail-making tests.40 A
structured interview obtained self-reported information on race or ethnicity,
sex, age, medications, medical history, alcohol and tobacco use, educational
achievement, and household income and household assets. Race and ethnicity
was ascertained to ensure representativeness of the population and because
it is associated with both mercury level and cognitive function. All testing
was performed without knowledge of blood mercury level or dietary history.
Technicians weighed and measured the height of the participants, and a phlebotomist
obtained a blood specimen. Specimens were stored at −20°C (mean
7.3 days) and later transferred to −70°C (mean 252 days) until analysis.
Total mercury was measured in whole blood using a flow-injection mercury
system with on-line microwave digestion and cold-vapor, atomic-absorption
spectrometry in the Trace Elements Laboratory of the New York State Department
of Health’s Wadsworth Center. The methods were based on the comparison
method described in Barbosa et al43 and required
a 0.2-mL sample. Collection tubes and storage containers were screened for
mercury contamination. Samples were analyzed in duplicate, and all quality-control
specifications were met. The intraday and interday coefficient of variation
(CV) for the 1.1-μg/L mercury control was 17.6% and 13.9%, respectively.
The intraday and interday CV for the 5.4-μg/L mercury control was 8.7%
and 8.8%, respectively. The detection limit was 0.1 μg/L. For the statistical
analysis, results below the detection limit (n=7) were assigned a value equal
to the detection limit divided by the square root of 2.
A commercial laboratory measured serum homocysteine levels using fluorescence
polarization immunoassay (Abbott AxSYM, Abbott Park, Ill); the CV ranged from
2.2% to 3.6%. The metals laboratory of the Kennedy Krieger Institute, Baltimore,
Md, measured blood lead using anodic stripping voltammetry.44 The
intraday CV was 11% and the interday CV was 7% (for 5.9 μg/dL of lead).
Another commercial laboratory measured serum cholesterol levels using an Olympus
AU5200 or AU600(Olympus America, Melville, NJ), with the CV ranging from 2.15%
to 2.28%. Serum triglycerides were measured on an AU5200 (CV from 2.88% to
3.32%). Apolipoprotein E (APOE) genotyping was performed
by the Malaria Institute laboratory at the Johns Hopkins Bloomberg School
of Public Health using previously published methods.45
Participants completed the Block 98.2 Food Frequency Questionnaire (Berkeley
Nutrition Services, Berkeley, Calif) before their second study visit. Completed
forms were optically scanned, and data were returned electronically. The questionnaire
assessed the participant’s “usual eating habits in the past year
or so” for the following foods: oysters, shellfish, tuna, fried fish,
and other fish. Participants estimated average serving sizes by choosing 1
of 4 pictures that looked like the portion size they normally eat, ranging
from a quarter cup to 2 cups. Frequency information was divided into 9 categories
ranging from “never consumed” to “one serving per day.”
Berkeley Nutrition Services also provided an estimate of average daily intake
of omega-3-fatty acids (grams) using US Department of Agriculture data46 and the following formula: (portion size ×
nutrient content × daily food frequency × seasonality factor)/100.47,48
The main objectives were to (1) explore associations between blood mercury
concentration and neurobehavioral test scores, adjusting for age, race and
ethnicity, sex, educational achievement, neurobehavioral testing technician,
fish consumption, and other potential confounding variables and (2) evaluate
whether these associations were influenced by potential effect modifiers,
such as APOE genotype; race and ethnicity; sex; age;
homocysteine, cholesterol, and triglyceride levels; blood lead; body mass
index, calculated as weight in kilograms divided by the square of height in
meters; antihypertensive medication use; diabetes; and tobacco use. Intercooled
Stata 7.0 (Stata Corp, College Station, Tex) software was used.
Treatment methods for the outcome variables have been reported.29 In brief, some of the measures were natural-log transformed
because of departures from normality, were negated to standardize the signs
of the β coefficients so that a negative coefficient always indicates
that test performance worsens with increasing blood mercury levels, or both.
Multiple linear regression was used to evaluate associations of blood
mercury levels with neurobehavioral test scores, adjusting for confounders;
only associations that achieved statistical significance (P<.05) are discussed. In the base model, mercury was regressed on
neurobehavioral score, adjusting for age, race and ethnicity, sex, educational
achievement, and testing technician. Race and ethnicity was categorized as
white (reference group), black, black-mixed, or other.29 Educational
achievement was divided into 9 categories, based on years of education and
possession of degrees or trade certificates, or both. The reference group
possessed a high school diploma and a trade certificate. Finally, the testing-technician
variable was modeled as 3 dummy variables, using the technician who tested
the largest number of participants as the reference.
To arrive at a final model, other covariates were added to the base
model using a biologically driven, forward, stepwise technique. These variables
were chosen a priori and added to the model individually: time of day of the
interview (morning, afternoon, or evening), household income and assets (both
natural-log transformed to minimize the influence of very large values), blood
lead level, APOE genotype (presence of the ε4
allele vs none), body mass index, smoking status (current, previous, or never),
alcohol consumption in the past month (yes vs no), history of diabetes (yes
vs no), history of myocardial infarction (yes vs no), use of antihypertensive
medications in the past 2 weeks (yes vs no), history of stroke (yes vs no),
use of antidepressant medications in past 2 weeks (yes vs no), use of antianxiety
medications in past 2 weeks (yes vs no), homocysteine level, total cholesterol
level, and triglycerides. Variables were retained if they fulfilled at least
1 of the following: (1) they were significant predictors of neurobehavioral
test scores or (2) their inclusion changed the mercury coefficient by 25%
or more. In addition to the covariates included in the base model, the final
model included assets, body mass index, alcohol consumption, and diabetes.
Because 58 participants did not complete the food questionnaire, a third
model (base-for-food model) served as a base with which to compare 2 models
containing food variables: a model that controlled for fish consumption and
a model that controlled for omega-3 fatty acid intake. All 3 models were based
on the final model. For each fish type, consumption frequency and portion
size were multiplied to estimate annual consumption. These estimates were
then added to yield an estimate of total annual fish consumption. This was
divided into quartiles and entered into models as 3 dummy variables.
The final model was used for evaluation of effect modification by the
variables listed previously. For these analyses, we evaluated the significance
of the cross-product term that resulted from multiplying mercury by each variable,
one at a time. For continuous variables,we used quartiles and tested the significance
of all 3 cross-product terms at once.
Adequacy of the final models was evaluated by (1) examining added variable
plots showing adjusted regression lines,49 (2)
comparing these lines with lowess regression lines,50 and
(3) plotting residuals against predicted values. To evaluate the magnitude
of the associations, test scores were Z transformed
and then multiplied by mercury’s interquartile range (2.4 μg/L).
The 474 study participants consisted of 325 women (68.57%), 185 blacks
(39.03%), and 263 whites (55.49%). These individuals did not differ by age,
race and ethnicity, or sex from the 666 participants who were not selected.
They were, however, more likely to have a postbaccalaureate education, greater
assets, and higher fish consumption than those not selected for the study
(Table 1). Blood mercury levels were
consistent with those found in populations that do not have high fish consumption.1- 3,51
In the base model, higher blood mercury was associated with worse performance
on Rey complex figure delayed recall and better performance on finger tapping
and Purdue pegboard (Table 2). Comparing
the base model with the final model, we observed an increase in the magnitude
of the association between mercury and the Rey complex figure delayed recall,
a decrease in the magnitude of the associations between mercury and finger
tapping), and a loss of significance on Purdue pegboard (Figure).
In the base-for-food model (Table 3),
the association of blood mercury with the Rey complex figure delayed recall
was of larger magnitude compared with the original base model (Table 2). The coefficients for Purdue pegboard and finger tapping
declined in both significance (except for nondominant finger tapping) and
magnitude. There were only small differences in the magnitude and significance
of associations comparing the base-for-food models, the fish model, and the
omega-3 model (Table 3). Exploratory
analysis did not reveal any consistent evidence of effect modification by
the variables examined.
Because results across models were similar, we only present magnitude
analysis for the final model. For Rey complex figure delayed recall, on average,
an increase of blood mercury from the 25th to the 75th percentile was associated
with a 0.12 SD decline in performance. Four SD units encompass approximately
95% of a normal distribution; therefore, a decline of 0.12 SD units is approximately
equivalent to a 3% decline in performance. For finger tapping, an increase
of blood mercury from the 25th to the 75th percentile was associated with
approximately a 2% improvement.
To our knowledge, this is the first study to investigate whether mercury
is associated with adverse neurobehavioral outcomes in older adults from the
general US population. This study was important given the levels of mercury
found in fish,52 the growing population of
older adults at risk of cognitive impairment, the well-known benefits of fish
consumption, and evidence that such benefits may counteract the negative effects
of consuming mercury-contaminated fish.23
In summary, the study provided no compelling evidence that blood mercury
levels were adversely associated with neurobehavioral test scores. There were
some consistent associations across models but because of the large number
of comparisons and the observation that statistically significant associations
were in different directions (ie, worse performance on a test of visual memory
and better performance on tests of manual dexterity), we cannot exclude the
possibility that associations were due to chance.
This study had many design strengths including the random selection
of participants with diversity by race and ethnicity, extensive neurobehavioral
battery in a broad set of cognitive domains, assessment of and control for
a large number of potential confounders and effect modifiers, and a relatively
large sample size. Previous epidemiological studies have documented overt
neurological outcomes following mercury-poisoning incidents including dysarthria,
ataxia, constriction of visual fields, distal paresthesias, hearing loss,
muscle weakness, and tremor.53- 55 However,
effects of long-term exposure to lower levels of methylmercury are likely
to be subclinical, similar to effects associated with lead and other neurotoxicants.3,27,56 Several recent studies
have investigated the neurobehavioral effects of such exposures in adults;
the majority concluded that higher mercury levels were associated with poorer
performance on neurobehavioral tests. Of these studies, not one looked at
the general US population57- 64 and
most focused on frequent fish consumers,57,59,61- 64 populations
with mercury levels higher than that found in the general US population, or
many of the studies focused on populations living in highly contaminated areas
(eg, the Amazon)57- 60,62,63 and
with little racial or ethnic diversity (such as that typically seen in the
United States).57- 59,61- 64 Many
had a small sample size with insufficient power,57,59,60 lack
of appropriate statistical techniques (such as only looking at correlation
and not using regression modeling),58,60,64 possibly
biased sampling of study participants,57,60,64 and
inadequate neurobehavioral assessment.64 It
is difficult to draw strong conclusions from these studies or to determine
whether the findings have relevance to the general adult US population.
In evaluating whether toxicants have adverse effects on central nervous
system function, it is important to consider whether exposure was recent or
cumulative, whether effects are acute or chronic, and whether the biomarker
is adequate to assess differing dose patterns. Clearance half-time of mercury
in blood is approximately 50 days, so blood mercury likely represents integrated
dose over the past 5 to 6 months. In frequent regular fish consumers, blood
mercury levels reach a steady state and may provide a better picture of cumulative
dose.3 If patterns of fish consumption vary
dramatically over a lifetime, then a single blood-mercury level may not be
adequate to assess longer latency effects or effects related to cumulative
dose, particularly if individuals were exposed in utero. Hair mercury is thought
to provide a longer-term estimate of dose, but average concentration of mercury
in hair is highly correlated with the concentration of mercury in blood.1,3,53,65,66 Two
additional factors favor use of blood mercury. First, the concentration of
methylmercury in blood is considered to be the best indicator of not only
total body burden but also dose to the brain.65 Second,
blood mercury is the most relevant clinical measure and the one with which
patients are most likely to be familiar.
Our study has some relative limitations. First, cross-sectional assessment
precluded evaluation of the temporality or causality of any associations.
Second, although self-reported fish consumption was associated with blood
mercury (evidence of the validity of the food questionnaire), the questionnaire
may not accurately measure omega-3 fatty acid dose. Third, fish consumption
was assessed at the second study visit while blood mercury was determined
during the first; however, the questionnaire did use an intake period of 1
year. A final limitation is that our subsample had individuals with more graduate
degrees, higher assets, and higher fish intake than the Baltimore Memory Study
participants not selected, possibly reducing the external validity of the
sample. Otherwise, the results may be expected to be generalizable to other
urban-dwelling, 50- to 70-year-old US residents.
Current fish consumption recommendations are based on risk assessments
for children and women of child-bearing age; according to the Environmental
Protection Agency and the National Research Council an “acceptable”
blood mercury level for this group is 5.8 μg/L or less.3,19 Since
the aging population may be particularly vulnerable to neurotoxicants, this
study was an attempt to examine whether this rapidly growing group is sensitive
to even lower levels of exposure. Since the blood mercury levels in our study
did not appear to be associated with adverse neurobehavioral effects, our
results suggest that these levels of exposure may not present a concern for
older adults. Studies with more detailed dose assessment are necessary to
confirm this conclusion since a single blood-mercury level may not be an optimal
estimate of cumulative dose.
Corresponding Author: Megan Weil, MHS, Association
of State and Territorial Health Officials, 1275 K Street NW, Suite 800, Washington,
DC 20005-4006 (email@example.com).
Author Contributions: Ms Weil had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Weil, Schwartz.
Acquisition of data: Weil, Bressler, Parsons,
Bolla, Glass, Schwartz.
Analysis and interpretation of data: Weil,
Bolla, Glass, Schwartz.
Drafting of the manuscript: Weil, Bolla, Schwartz.
Critical revision of the manuscript for important
intellectual content: Bressler, Parsons, Glass, Schwartz.
Statistical analysis: Weil, Glass, Schwartz.
Obtained funding: Weil, Glass, Schwartz.
Administrative, technical, or material support:
Weil, Bressler, Parsons, Schwartz.
Study supervision: Bressler, Bolla, Glass,
Financial Disclosures: None reported.
Funding/Support: This work was supported by
training grants ES-07141 from the National Institute of Environmental Health
Sciences and T42/CCT310419 from the National Institute of Occupational Safety
and Health and research grant R01 AG19604 from the National Institutes of
Role of the Sponsors: None of the funding organizations
had a role in the design and conduct of the study; collection, management,
analysis, and interpretation of the data; or preparation, review, or approval
of the manuscript.
Acknowledgment: We thank Margaret Mintz, MS,
and Anne E. Jedlicka, MS, both from the Johns Hopkins University Bloomberg
School of Public Health, W. Harry Feinstone Department of Molecular Microbiology
and Immunology, for the APOE genotyping analysis.
We also thank Fernando Barbosa, Jr, PhD, for assistance with the blood mercury