Skip Navigation LinksHome > July 2004 - Volume 15 - Issue 4 > Assessing Environmental Neurotoxicant Exposures and Child Ne...
doi: 10.1097/01.ede.0000129525.15064.a4

Assessing Environmental Neurotoxicant Exposures and Child Neurobehavior: Confounded by Confounding?

Bellinger, David C.

Free Access
Article Outline
Collapse Box

Author Information

From the Department of Neurology Harvard Medical School, Children's Hospital, Boston, Massachusetts.

Correspondence: David C. Bellinger, Professor of Neurology, Harvard Medical School, Children's Hospital, Boston, MA 02115. E-mail: David.Bellinger@childrens.harvard.edu

Increasing numbers of studies seek to identify effects of environmental exposures on neurobehavior, especially among children. One recurring worry in such studies has been that an observed association might be the result, either wholly or in part, of confounding. In this issue, Pamela Mink and her colleagues1 show that a failure to measure known confounders could easily produce effects on the order of those often attributed to environmental neurotoxins. This is certainly important but only one aspect of the problem.

The major potential confounders in studies of child neurobehavior are well-known, and most studies at least try to assess them. The concern is generally not that such factors are unmeasured, but that they are not measured well. The impact of this poor measurement quality is harder to quantify but might ultimately be more critical. What can investigators do, in terms of adjustment for confounding, to increase the validity of inferences drawn in neurobehavioral studies?

This problem has conceptual aspects for which the prospects of purely statistical solutions seem rather bleak. The usual adjustment strategy assumes, as does Mink, that all the variance in neurobehavior shared by the exposure of interest and the confounder “belongs” to the confounder. In some settings, this is likely to be excessively conservative, because confounders can, to some extent, also be proxies for exposure.

One example of this is blood lead and social class (the key confounder identified by Mink and colleagues). Social class is certainly associated with child neurobehavior (independent of lead), and higher blood lead levels are usually associated with lower social class. Social class, however, can convey information about a child's lead exposure opportunities. For example, lower social class might be associated with more lead paint in poor repair, higher soil lead levels, and so on. Because lead levels in blood can fluctuate over the short-term, a child's social class might convey even more information than a current blood lead level about the child's past lead exposure at some critical age. If lead does affect neurobehavior, and if part of this effect is carried by the association between social class and neurobehavior, then adjusting for social class in the usual manner will underestimate the association between lead and neurobehavior.2

To remedy this problem, we need more differentiated measures of complex constructs such as social class—measures that would permit control for only those aspects that do not reflect exposure opportunities. Such an aspect might still be correlated, in a noncausal manner, with an exposure biomarker such as blood lead, making it a confounder for which adjustment must be made. It is easier to identify the need for such measures than it is to develop them. Perhaps social epidemiologists can provide some help.

Investigators can also choose study settings in which the exposure of interest and known potential confounders are dissociated as much as possible. For lead, this would be settings in which the primary route of exposure tends to provide “equal opportunity for all,” regardless of social class, such as the public water supply3 or smelter emissions.4 A hospital with a catchment area in which social class is reasonably homogeneous could also provide a birth cohort in which lead and social class are relatively disassociated.5 Fetal exposure to methylmercury, which occurs primarily through maternal fish consumption during pregnancy, is usually highly confounded with nutrients in fish that promote neurodevelopment. (Daniels and her colleagues6 illustrate this point in a paper also in this issue.) In one study,7 this dilemma was addressed by matching high fish-consuming women who had hair mercury levels above 6 μg/g with high fish-consuming women who had hair mercury levels below 6 μg/g, thereby dissociating, at least in part, methylmercury intake and “fish-eating” behavior and its correlates. This solution is not perfect because the 2 groups of women chose to consume fish species with different levels of methylmercury, and, probably, different levels of micronutrients. Still, it seems a step in the right direction.

For any particular observational study, we can, of course, never determine with certainty whether the association observed between an exposure and neurobehavior is entirely the result of residual confounding. Adjusting for social class and other factors does often reduce the association with the exposure of interest, leading to a suspicion that it would recede even further toward the null if only we could adjust for all pertinent factors. However, this line of argument is unsatisfying because it has no logical end point. One could propose an endless succession of unmeasured (or poorly measured) factors that might result in residual confounding, so such a discussion can never be resolved to everyone's satisfaction.

However, why try to draw inferences about causality on the basis of an association observed in a single study? Besides being imprudent, it is often not necessary. Ideally, we have recourse to several studies evaluating essentially the same hypothesis. Consistency across studies provides a compelling basis for causal inference, especially when the studies were conducted in diverse sociocultural settings that vary in the details, perhaps even in the direction, of confounding of the exposure–outcome relationship. Such consistency in findings amid a diversity of settings helped to build the case, for example, that the association between low-level lead exposure and neurobehavioral deficits is not likely to be entirely spurious.8

Although consistency across studies supports the case for causality, it cannot make the case. The discussion about causality need not, however, end mired in the uncertainties associated with even a large number of observational studies with consistent findings. Often evidence from experimental animal studies is available—evidence not compromised by the possibility of confounding by, for instance, poorly measured social factors. It seems unnecessary to say that such evidence should be accorded considerable weight in any discussion about causality. It is remarkable, however, how frequently discussions about lead have ended with the uncertainties of the epidemiologic evidence, failing to refer to the enormous experimental animal literature that proves that lead at low levels causes neurobehavioral deficits and even provides insights into mechanisms. To fail to place epidemiologic evidence on environmental exposures in this larger context when drawing causal inferences is to ignore an essential piece of the puzzle.

Back to Top | Article Outline


David Bellinger is Professor of Neurology at Harvard Medical School and Professor of Environmental Health at the Harvard School of Public Health. For the past 25 years, he has conducted studies of child neurobehavior and environmental exposures to metals such as lead, mercury, and manganese. He has served on committees for the National Research Council, WHO, EPA, and ATSDR on issues of childhood environmental health. He is currently conducting a randomized trial on the neurotoxicity of dental amalgams in children.

Back to Top | Article Outline


1. Mink PJ, Goodman M, Barraj LM, et al. Evaluation of uncontrolled confounding in studies of environmental exposures and neurobehavioral testing in children. Epidemiology. 2004;15:385–393.

2. Bellinger D, Leviton A, Waternaux C. Lead, IQ, and social class. Int J Epidemiol. 1989;18:180–185.

3. Fulton M, Thomson G, Hunter R, et al. Influence of blood lead on the ability and attainment of children in Edinburgh. Lancet. 1987;1:1221–1226.

4. Wasserman GA, Liu X, Lolacono NJ, et al. Lead exposure and intelligence in 7-year-old children: the Yugoslavia Prospective Study. Environ Health Perspect. 1997;105:956–962.

5. Bellinger D, Leviton A, Waternaux C, et al. Longitudinal analyses of prenatal and postnatal lead exposure and early cognitive development. N Engl J Med. 1987;316:1037–1043.

6. Daniels J, Longnecker MP, Rowlane AS, et al., for the ALSPAC Study Team. Fish intake during pregnancy and early cognitive development of offspring. Epidemiology. 2004;15:394–402.

7. Kjellstrom T, Kenndy P, Wallis S, et al. Physical and Mental Development of Children with Prenatal Exposure to Mercury From Fish. National Swedish Environmental Protection Board Report No. 3642, 1989.

8. Lanphear BP, Hornung R, Khoury J, et al. Low-level environmental lead exposure and children's intellectual function: an international pooled analysis. Presented at the XXIst International Neurotoxicology Conference, Honolulu, HI, February 2004.

Cited By:

This article has been cited 2 time(s).

Journal of Occupational and Environmental Medicine
Neurocognitive Impairment in Lead-Exposed Children of Andean Lead-Glazing Workers
Counter, SA; Buchanan, LH; Ortega, F
Journal of Occupational and Environmental Medicine, 47(3): 306-312.

PDF (551)
Measuring Mercury Concentration
Daniels, JL; Longnecker, MP; Rowland, AS; Golding, J
Epidemiology, 16(1): 134.
PDF (231) | CrossRef
Back to Top | Article Outline

© 2004 Lippincott Williams & Wilkins, Inc.

Twitter  Facebook 


Article Tools


Article Level Metrics