Epidemiologic studies are subject to two types of error: systematic and random. Systematic errors (or bias) are by far the most problematic as they are generally not measured and they do not decrease as the sample size increases. Key sources of bias include those associated with aspects of selection and the distortion of the cause-effect relation by confounding. Reasons for the controversy over the lead-IQ link include a) the large number of confounders that must be considered when measuring an effect on children's intelligence; and b) the frequent finding that the more covariates included in regression models, the smaller the effect of blood lead on IQ becomes, although it remains in the same direction (WHO 1995). The most important confounders are SES, parental IQ, and the quality of the home environment. Other factors associated with both IQ and blood lead levels include sex, nutritional status, and parental smoking behavior.
Three studies shed light on the area of confounding (Needleman and Bellinger 2001; Tong and Lu 2001;Wasserman and Factor-Litvak 2001). Blood lead levels have been negatively and positively associated with SES. Because of the sociodemographics and geography of Boston, Massachusetts, USA, increased prenatal lead levels were found in children of higher SES, and after adjustment for covariates the association of lead with IQ loss increased (Needleman and Bellinger 2001). This effect was also seen in the Yugoslavia prospective lead study in which children living near a smelter were from higher SES backgrounds than those living in a nearby control town with lower lead exposure (Wasserman and Factor-Litvak 2001). A study on the identification of confounders in the Port Pirie cohort study (Tong and Lu 2001) found that the size of the relationship between blood lead levels and mean IQ scores decreased by up to 40% when adjustment was made for 4 confounders but by less than 10% when a further 10 confounders were added to the regression models. The four main covariates were SES, quality of home environment, maternal IQ, and parental smoking behavior. The 10 extra confounders, which had little effect individually, included age, sex, birth weight, birth rank, maternal age, number of siblings, and duration of breastfeeding (Tong and Lu 2001).
Bellinger (2000) has argued that factors such as SES and sex should not be viewed solely as confounders but as effect modifiers as well. Unlike confounding, effect modification is a true characteristic of the association between an exposure and its end point. An example is the association between alcohol consumption and blood pressure, which varies in size depending on the modifying effects of the age, sex, and smoking status of the individual. Using data from the Boston prospective lead study, Bellinger showed that children from the lower half of the social class distribution demonstrated a decrease in performance at lower blood lead levels than children in the upper half of the distribution. However, this protective effect of higher SES was lost in the group of children with the highest cord blood lead levels. The author's hypothesis is that at a given exposure level the magnitude of the estimated effect varies depending on the individual's location on the social class continuum (Bellinger 2000). This idea is not new. In 1984 Winneke and Krämer (1984) showed the protective effects of higher SES on visual-motor performance deficits in lead-exposed children and concluded,
the common practice of merely removing the effects of confounding factors, such as SES, appears doubtful. . . . In addition, some of the inconsistencies in this area of research might be due to differential sampling of subgroups of lead-exposed children characterized by different levels of psychosocial adversity.
Intuitively, Bellinger's hypothesis is very attractive and provides a possible explanation for the variability between ostensibly similar studies. In the lead field in the past, study results have been deemed right or wrong, usually on the basis of how the issue of confounding was handled. If dose-effect relationships are not independent of other host characteristics, it will be necessary to model three (or more)-way interactions. However, most prospective studies are designed with only enough statistical power to detect main effects and do not have the power to detect effect modification in subgroups of the main cohort. Bellinger urges a move away from broad, population-based cohorts toward a greater use of focused sampling frames, which should include adequate numbers within specific subgroups (Bellinger 2000). The report on the Port Pirie cohort at 11-13 years of age supports the hypothesis that children from socially disadvantaged backgrounds are apparently more sensitive to the effects of lead than children from higher SES families (Tong et al. 2000).
The powerful influence of SES on developmental outcome has been elegantly demonstrated in a report on school-age children born to mothers with heroin dependency (Ornoy et al. 2001). The study followed children born to mothers with heroin dependency raised at home or adopted at a very young age. These children were compared with groups of control children with average SES, children raised in families with a heroin-dependent father, or children born in families with low SES and high environmental deprivation. The children with environmental deprivation or raised at home by parents with heroin dependency had the lowest intellectual achievements. The adopted children had normal scores on the verbal WISC-R and on the Bender test, as well as normal reading and arithmetic skills, although they had a higher rate of attention deficit hyperactivity disorder than control children. Ornoy later extended this work to include two other high-risk cohorts: children born to mothers with diabetes and children born prematurely with low birth weights. Again he was able to demonstrate that a good home environment had a strong influence on subsequent intellectual abilities but not on motor skills or attention span (Ornoy 2003).
Animal models using spatial learning in rats have shown the protective effect of an enriched environment on lead-induced neurotoxicity (Schneider et al. 2001). Of particular relevance is a recent report on rats exposed to low levels of lead during early development, that is, from birth to weaning at day 21. This exposure produced a lasting deficit in spatial learning that could be completely reversed by raising the rats in an enriched environment after weaning. This reversal was accompanied by nerve growth factor gene induction and recovery of deficits in hippocampal glutamate receptor gene expression (Guilarte et al. 2003).
Genetic predisposition can also affect vulnerability to lead-induced neurotoxicity; this area of research has recently been reviewed by Lidsky and Schneider (2003). Three genes are currently believed to play a role in lead neurotoxicity: the ALAD gene, which codes for δ-aminolevulinic acid dehydratase; the vitamin D receptor (VDR) gene; and the hemochromatosis gene coding for a defective protein known as HFE. There are two forms of the ALAD protein, ALAD1 and ALAD2; lead has a higher affinity for ALAD2. Preliminary evidence has shown adolescents with the ALAD1 phenotype are more resistant to the effects of lead on behavior and attention than ALAD2 individuals. There are at least two alleles (b and B) and three variants of the VDR genotype, and among adults occupationally exposed to lead, b individuals have higher lead levels in blood and bone. Mutated HFE protein is known to cause hemochromatosis, in which large quantities of iron are deposited in internal organs. Because lead can be incorporated into processes requiring iron, polymorphisms in HFE might be expected to influence lead absorption. It is likely that future epidemiologic studies will include analysis of ALAD status and possibly other biomarkers.
Generally, no single epidemiologic study should be treated as the sole source of convincing evidence. The weight of evidence for any causal link comes when a number of studies using similar or preferably different methodologies in different populations reveal the same finding. In the low-level lead-IQ link, the balance has come down in favor of an association, with the methodologically sound study by Canfield et al. (2003) indicating that these effects are seen at peak blood lead levels below 10 µg/dL. Having established a valid association, the use of a number of the Bradford Hill criteria can assist in making causal inferences: temporal relationship (Does lead exposure precede the effect on cognition?), biological plausibility (Are there neurotoxic mechanisms to explain the effect of lead on cognition?), and biological gradient and strength (Is there a dose-response relationship, and if so, how strong is it?).
Evidence is increasing for a temporal relationship. The finding that 4-5 years of age is the critical period for manifestation of earlier (postnatal) lead exposure (Schnaas et al. 2000) might explain the wide variability in effects reported in cross-sectional studies that only looked at children 6 years of age or older. Further support for the critical period comes from the finding by Rogan et al. (2001) that chelation therapy in lead-poisoned children has no beneficial effect when given at 4-6 years of age.
Mechanistically, no unifying theory explains the neurotoxicity of lead or how lead might affect cognition. The ability of lead to substitute for calcium is a common factor underlying many of its toxic actions, including apoptosis and influences on neurotransmitter storage and release, second messengers, cerebrovascular endothelial cells, and glial cells. A variety of mechanisms may be important, and these are summarized in recent reviews by De Gennaro (2002) and Lidsky and Schneider (2003). Lead activates calmodulin, calcineurin, and protein kinase C at very low doses (Deng and Poretz 2002; Kern and Audesirk 2000). Glutamate receptors are thought to be involved in mediation of learning and memory, and changes in N-methyl-d-aspartate glutamate receptor subunits are observed in animals that show cognitive deficits induced by exposure to lead (Lau et al. 2002; Nihei and Guilarte 2001). Lead-induced decreases in hippocampal neurotrophic factor gene expression in rats can be reversed by raising the animals in an enriched environment (Schneider et al. 2001).
Concerning dose-response relationships, IQ tests are blunt measures of neurologic status, and blood lead is at best only a crude index of lead-induced neurotoxicity. However, a negative association has been found across groups of children from a range of populations around the world. Visual-motor tests and tests of attention are designed to assess more limited cognitive domains than IQ tests, and it is of interest that more consistent decreases have been reported for these measures in cross-sectional studies (Al-Saleh et al. 2001; Calderón et al. 2001; Walkowiak et al. 1998) and prospective studies (Dietrich et al. 1993; Tong et al. 1996; Wasserman et al. 1997). The bluntness of IQ tests to measure cognitive function is underlined by a study on five children who underwent left temporal lobectomy for epilepsy. Each patient experienced significant language-related cognitive loss after surgery, and these losses were clinically evident in four of the five patients. However, IQ testing alone did not reliably identify these deficits. Only one child showed a loss of verbal IQ; the other four children showed increases in verbal IQ (Dlugos et al. 1999).
It is clear that blood lead levels have fallen significantly over the last 40 years. During the 1970s, childhood blood lead concentrations of 40 µg/dL were not unusual. The available evidence suggests that mean blood lead levels are now in the range 2-4 µg/dL in the United States and much of Europe. Despite this reduction in lead exposure, it could be argued that current baseline blood lead levels continue to constitute a global public health risk, as preindustrial humans are estimated to have had 100- to 1,000-fold lower blood lead levels than the population of today (Owen and Flegal 1998). With the recent evidence demonstrating an inverse association between blood lead levels and cognitive function in children exposed to low levels of lead, there is no safety margin at existing exposures. Clearly, efforts must continue to minimize childhood exposure. However, we would urge that these efforts be seen in perspective. The magnitude of the lead-IQ dose-response relationship is small on a population basis and should be set against the far greater combined effect of SES status and quality of the caregiving environment. It has been argued that, instead of "chasing after an ever-receding lead threshold," attention and funds should be focused on "the more complex social ills that are associated with continued lead exposure in a small segment of the population" (Gee and McKay 2002). Current lead exposure accounts for a very small amount of variance in cognitive ability (1-4%), whereas covariates such as social and parenting factors account for 40% or more (Weiss 2000).
The authors are indebted to R. Canfield, Cornell University, Ithaca, New York, USA, for providing further details of his study.
We gratefully acknowledge the support of the U.K. Department for Environment, Food and Rural Affairs.Reprint Address
Address correspondence to K. Koller, MRC Institute for Environment and Health, University of Leicester, 94 Regent Rd., Leicester LE1 7DD, UK. Telephone: 44 0 116 223 1629. Fax: 44 0 116 223 1601. E-mail: email@example.com
Environ Health Perspect. 2004;112(9) © 2004 National Institute of Environmental Health Sciences
Cite this: Recent Developments in Low-Level Lead Exposure and Intellectual Impairment in Children - Medscape - Jun 01, 2004.