A Meta-Analysis of the Neuropsychological Effects of Occupational Exposure to Mercury

ABSTRACT

This paper reports a meta-analysis of 36 peer-reviewed published studies of the neuropsychological effects of occupational exposure to mercury, which yielded 43 independent samples. These studies included 2,512 exposed participants and 1,846 controls, for a total sample size of 4,358. Because the independent variables defining mercury exposure varied across studies, effect sizes were calculated for exposed versus non-exposed workers. Dose–response relations were considered for measures of mercury levels in urine (81% of studies reported), blood (42% of studies reported), and air samples (33% of studies reported). Level of exposure was also estimated by reported years of exposure (M = 11.3, SD = 5.6). Cohen's d statistic yielded a statistically significant weighted study-mean effect size of −.23, p < .0001 for occupational mercury exposure. However, an effect this small is typically undetectable when evaluating individuals because it is smaller than the typical 95% confidence interval used for most neuropsychological measures. None of the exposure variables analyzed reached statistical significance. The magnitude of self-reported symptoms (−.30) was slightly larger than that obtained from objective test scores (−.22), though the difference was not statistically significant. Also, the weighted mean effect size for psychomotor skills (−.34) was the largest in magnitude, whereas the weighted mean effect size for verbal comprehension measures had the smallest (−.06). However, an analysis of the differential effects of mercury exposure across cognitive domains found significant differences between verbal comprehension measures and all other domains. None of the other domains were significantly different from one another. The weighted study-mean effect size suggests that the prevalence of neuropsychological deficits due to occupational exposure to mercury is small and difficult to detect on an individual case-by-case basis.

Notes

^a Type of dependent variable.

^b Cohen's d for weighted effect size.

^c Standard error of estimate of Cohen's d.

^d Number of studies used to calculate effect size.

^e Lower limit of the 95% confidence interval for the effect size.

^f Upper limit of the 95% confidence interval for the effect size.

^g One sample z statistic (weighted effect size/standard error of the mean).

^h Probability that z is significantly different than zero.

ⁱ Q statistic—a test of homogeneity of variance, distributed aboutthe χ² distribution.

^j Probability of heterogeneity in variances.

^a Type of dependent variable.

^b Cohen's d for weighted effect size.

^c Standard error of estimate of Cohen's d.

^d Number of studies used to calculate effect size.

^e Lower limit of the 95% confidence interval for the effect size.

^f Upper limit of the 95% confidence interval for the effect size.

^g One sample z statistic (weighted effect size/standard error of the mean).

^h Probability that z is significantly different than zero.

ⁱ Q statistic—a test of homogeneity of variance, distributed about the χ² distribution.

^j Probability of heterogeneity in variances.

^a Number of exposed participants in study.

^b Number of control participants in study.

^c Number of dependent variables in study.

^d Cohen's d for unweighted effect size. The statistics listed are based on Cohen's (1988) recommendation that the beta-to-alpha ratio equals 4.0, which results in a beta of .20 when alpha is set at .05. We set beta at .20, used each study's actual sample size and mean-study effect size, and generated the t and p values listed in the Table 2.

^e Probability of difference in means based on t test. Also see footnote d above.

^f Note: A negative effect size means the exposed group scored lower than controls or “worse” than controls.

[Note: Articles included in the meta-analysis are marked with an asterik.]