Occupational versus environmental studies
While studies of occupationally exposed populations provided initial data about the harmful effects of lead at high levels of exposure, environmental studies are much less troubled by the healthy worker effect, survivor cohorts, and other sources of bias inherent to occupational studies. Environmental studies also have the capacity to encompass much larger sample sizes with more socioeconomic and racial/ethnic diversity than occupational studies.
These are important differences. In some cases, researchers conducting studies in subjects with high chronic occupational lead exposure have failed to observe the adverse impacts associated with lower levels of lead in environmental settings. For example, several studies of renal function in smelter workers have found no evidence of clinical renal dysfunction or changes in markers of tubular dysfunction compared with controls (Gerhardsson et al. 1992
; Omae et al. 1990
; Roels et al. 1994
; Wang et al. 2002
). Similarly, some occupational studies have not found significant differences in cognitive symptoms or psychosocial disturbances between occupationally exposed workers and controls (Parkinson et al. 1986
There are many potential explanations for null associations in occupational studies. Occupational studies tend to be based on small sample sizes, making them vulnerable to type II error. Some studies had nonexposed, or control, groups of workers with BLLs well above current background-exposure levels (< 5 μg/dL), which is likely to underestimate the effect being studied. Perhaps the most important problem is the vulnerability of occupational studies to the healthy worker effect, that is, the bias inherent in studying populations of workers who remain after the departure of sicker and/or more susceptible workers, especially problematic in cross-sectional studies of current workers. In studies that compare health effects in workers with general population controls, the healthy worker effect could explain why associations may not be observed, because workers are, especially for symptomatic conditions, more likely to be healthier than the general population. This problem is often manifested as an attenuation of exposure–response curves in occupational studies at high exposure levels (Stayner et al. 2003
). If lead causes most of its health effects at low levels, occupational studies that examine only the high end of the dose–response range could miss associations at the lower end or across the entire range (Nuyts et al. 1993
). It has also been reported that there may be selection bias by δ-aminolevulinic acid dehydratase genotype, as lead exposure and exposure duration increase and symptomatic workers leave the workplace (Schwartz et al. 1995
Such selection bias can be mitigated in two ways. First, although more difficult, this bias could be avoided in occupational studies that assembled complete cohorts and randomly selected workers for study, including those who left the workplace early or late in their careers. Of interest is that studies that have assembled cohorts of all workers ever employed in a given plant or industry have reported many significant findings not previously observed (Schwartz et al. 2000
; Stewart et al. 1999
). Second, longitudinal studies, in comparing subjects with themselves in changes in health status over time, are somewhat less susceptible to this kind of bias, whether occupational or environment. Community-based studies of the general population avoid the healthy worker effect entirely, making them an especially important study design, although a similar form of bias can be encountered in studies limited to just older subjects in the general population.
In this mini-monograph, both occupational and environmental studies were chosen for inclusion in the systematic reviews (Navas-Acien et al. 2007
; Shih et al. 2007
). When evaluating the associations of cumulative lead dose with health outcomes, investigators need to acknowledge that nonoccupational sources of lead exposure were present for all members of the general population in the United States, including lead workers, throughout most of the 20th century until public health interventions progressively removed lead from gasoline and many consumer products during the 1970s and 1980s (Agency for Toxic Substances and Disease Registry 1999
; Annest et al. 1983
; Pirkle et al. 1998
). Lead remains a low-level and ubiquitous neurotoxicant in the environment and is found in measurable levels in all individuals (Hoppin et al. 1995
). Thus, current tibia lead levels may represent a mix of environmental exposures and potential occupational exposures.
Race/ethnicity, socioeconomic status, and other factors/covariates
Environmental lead exposure differs by race/ethnicity and socioeconomic status. Persons with low socioeconomic status (e.g., educational attainment, income, assets) have been known to have higher blood lead levels throughout at least the period of the recurrent National Health and Nutrition Examination Survey (NHANES) blood lead surveys (Annest et al. 1983
; Elreedy et al. 1999
; Pirkle et al. 1998
). Several investigators have also reported higher bone lead levels in minorities compared with whites. For example, Martin et al. (2006)
recently reported that tibia lead levels among a population-based sample of individuals 50–70 years of age in Baltimore, Maryland, were 30% higher in African Americans than in whites (Martin et al. 2006
). Lin et al. (2004)
reported higher tibia and patella lead levels in minorities compared with predominantly white subjects who were older than 55 years and living in Boston, Massachusetts (Lin et al. 2004
). One study reported higher tibia and patella lead levels in blue-collar workers compared with white-collar workers and this association was modified by race/ethnicity (Elmarsafawy et al. 2002
The strong associations of cumulative lead dose with race/ethnicity and socioeconomic status raises methodologic concerns. Factors that in the past were simply termed “confounding variables” are now more carefully evaluated as potential mediators (i.e., in the biological causal pathway), moderators (i.e., risk modifiers), direct causes, or otherwise parts of complex causal pathways (Kraemer et al. 2001
). It is now understood that such complex causal pathways also apply to lead exposure and chronic disease, including cognitive dysfunction, hypertension, and renal dysfunction. These pathways can include connections between individual-level indicators (e.g., age, sex, race/ethnicity, socioeconomic status); behavioral risk factors; biological factors (e.g., genetics); social factors (e.g., social capital, social cohesion); lead dose (i.e., both recent and cumulative); health conditions (e.g., diabetes, heart disease, hypertension); and other biological markers predictive of disease (e.g., homocysteine levels) that may be thought of as either outcomes by themselves or as intermediate pathological states that result in other conditions (e.g., renal dysfunction, cognitive declines).
What are the implications of the fact that race/ethnicity and socioeconomic status may be causally related to cumulative lead dose? Although low socioeconomic status is associated with higher BLLs in population-based surveys, early life lead exposure has been shown to cause intellectual impairment and worse educational outcomes (Banks et al. 1997
; Canfield et al. 2003
; Needleman et al. 1979
; Pocock et al. 1994
), which in turn may influence socioeconomic status attainment in later life. Thus, although early studies of lead and cognitive function argued that controlling for education was an important necessity, when including the contribution of early life lead exposures, the potential reciprocal causation may lead to an underestimation of the association between lead dose and cognitive function if education is included in regression models. This has also been recently discussed in the context of lead and blood pressure (Martin et al. 2006
A similar issue has been raised concerning race/ethnicity. To the extent that race/ethnicity serves as a proxy for other factors influenced by early lead exposure and also adversely affect cognitive function or cardiovascular outcomes (Navas-Acien et al. 2007
; Shih et al. 2007
), adjusting for race/ethnicity could underestimate the overall effect of that early lead exposure. Some authors have thus argued that there should not be adjustment for race/ethnicity (Martin et al. 2006
; Shih et al. 2006
). If later life lead exposure also affects the measured cognitive or cardiovascular outcome, then to the extent that race/ethnicity also serves as a proxy for factors that influence later life lead exposure and these outcomes, not adjusting for race/ethnicity may introduce bias.
Thus, it can be concluded that inclusion of race/ethnicity in models evaluating relations of cumulative lead dose and cognitive function or cardiovascular outcomes could lead to an underestimation of the direct effect of lead. Given these complex causal pathways, we believe relations of tibia lead and these outcomes are likely to be best estimated by parsimonious regression models that control for such variables as age, sex, and testing technician, for example, but not necessarily those that include race/ethnicity and socioeconomic status, which is at odds with what has been concluded by other authors (Goodman et al. 2002
; Lindgren et al. 1996
) and with our earlier thinking on this issue (Balbus-Kornfeld et al. 1995
). It may be most informative in the future if analyses were explicitly reported (and appropriately interpreted) that both included and excluded race/ethnicity and measures of socioeconomic status.
The ideal solution to such a conundrum, although not often possible, would be to have separate measures of early-life and late-life lead exposures and/or direct measures of the underlying factors for which race/ethnicity is serving as a proxy (LaVeist 1994
) and possibly applying statistical methods such as marginal structural models to account for variables that are both mediators in the pathway to exposures and confounders of later exposures (Robins et al. 2000
). Another issue of concern when considering race/ethnicity is that there might be plausible physiologic differences by race/ethnicity that affect the association between lead and cognitive or cardiovascular outcomes. In such a case, one must consider effect modification by race/ethnicity in analyses, although not without the same considerations raised above regarding the precise factors for which race/ethnicity is serving as a proxy (LaVeist 1994
). Although stratified analysis by race/ethnicity is one approach to these concerns, very few epidemiologic studies are designed to be adequately powered for stratified analysis, and thus loss of statistical significance is often the result. Stratified analysis should not be necessary if evaluation of effect modification in a single model (i.e., by inclusion of an interaction term between lead dose and race/ethnicity) reveals no evidence that the association with lead dose differs by race/ethnicity.
A number of important individual-level factors, in addition to race/ethnicity and socioeconomic status, are frequently considered in the body of literature on the health effects of lead. When comparing associations of health outcomes with blood and bone biomarkers, it is essential to recognize that factors such as age, sex, and elevated bone turnover accompanying osteoporosis (Silbergeld et al. 1988
; Webber et al. 1995
) may modify lead toxicokinetics. These factors may co-vary with age, race/ethnicity, and sex, and genetic polymorphisms, complicating consideration of these issues.
Two important health behaviors—tobacco and alcohol consumption—have been linked with risk of cardiovascular and cognitive outcomes. This begs the question of whether it is critical to adjust for tobacco and alcohol consumption in evaluating relations of lead dose with cardiovascular and cognitive outcomes. The numerous studies of tobacco and alcohol consumption and its relations with cognitive function are conflicting (Carmelli et al. 1999
; Cervilla et al. 2000
; Crawford et al. 2001
; Elias et al. 1999
; Elwood et al. 1999
; Hendrie et al. 1996
; Schinka et al. 2002
). We do not believe that control for these variables is absolutely necessary and can be guided by modeling and appropriate interpretation of causal pathways.
As another example, smoking is often included in models of hypertension, and its impact on blood pressure remains an important potential mechanism for its status as a risk factor for end-organ dysfunction (Orth 2004
). Tobacco is well known to have been contaminated by lead arsenate pesticide in the past and smoking has been identified as a risk factor for increased cumulative lead burden (Hu et al. 1996b
). Thus, it is possible that epidemiologic studies of hypertension that adjust for smoking are “over-controlling” for a risk factor (smoking) that is associated with the exposure of interest (lead), underestimating the direct effect of lead. Similar concerns pertain to alcohol, which has been linked to the pathogenesis of hypertension [with moderate to heavy consumption (Beilin and Puddey 2006
)] and is also well-known risk factor for elevated lead exposure (Lee et al. 2005