Click on the links below to access all the ArticlePlus for this article.
Please note that ArticlePlus files may launch a viewer application outside of your web browser.
The central nervous system is a critical target organ of lead toxicity. Lead encephalopathy was first described in case reports after high-intensity doses resulting in rapid elevations in blood lead levels.1–3 Nervous system symptoms such as irritability, poor attention and concentration, forgetfulness, depressed affect, and sleep disturbance are common after lower doses. During the past 3 decades, the more subtle cognitive deficits associated with lead have been evaluated in epidemiologic studies using neurobehavioral test batteries and measures of lead dose.4–6
Most of these studies were cross-sectional in design and relied on blood lead levels to estimate dose. Because blood lead has a clearance half-time of approximately 30 days, blood lead is more reflective of recent exposures, although bone lead stores also can contribute to blood levels. In contrast, lead has a clearance half-time of approximately 25–30 years in tibia; therefore, tibia lead is a measure of retained cumulative dose. Depending on the pattern of exposure over time and the relation between lead dose history and nervous system effects, either recent dose or cumulative dose (or some combination of the 2) might be the more relevant parameter. Given the complexities of this situation, it is not surprising that different authors have reached conflicting conclusions regarding the relation of lead exposure to neurobehavioral and neurologic outcomes.4–6
Here, we report on a longitudinal study of the neurobehavioral and peripheral nervous system effects of lead by using both blood lead and tibia lead as measures of dose. In a cross-sectional analysis of first-visit data, we previously observed that lead levels consistently were associated with decrements in neurobehavioral test scores, mainly in the cognitive domains of executive abilities and manual dexterity, and that blood lead was a better predictor of test scores than tibia lead.7 We concluded that in cross-sectional analysis of current workers, the acute effects of recent dose, as measured by blood lead, may be easier to detect than the chronic effects of cumulative dose, as measured by tibia lead.
In the present report, we consider the longitudinal effects of blood lead and tibia lead on cognitive decline. Evaluation of longitudinal change in cognitive function in relation to lead is of particular interest because such analysis allows assessment of the transience, persistence, or progression of effects.8 Furthermore, lead biomarkers allow, to some degree, dose to be partitioned into time-dependent pools. This permits the development of longitudinal models to separate recent dose (using blood lead) from cumulative dose (using tibia lead) and to separate acute effects (possibly reversible) from chronic effects (probably irreversible).
Study Population, Study Design, and Follow-Up
A description of the study population has been previously reported.7 In brief, 803 lead workers were enrolled at visit 1 (V1), 625 (78%) were retested at V2, and 652 (81%) were retested at V3. Enrollment began in October 1997, and study visits were completed in June 2001. The mean time (± standard deviation [SD]) from V1 to V2 was 1.0 ± 0.20 years (range, 0.6–2.3 years) and from V1 to V3 was 2.2 ± 0.5 years (range, 1.6–3.6 years). Among lead workers, 576 (72%) completed all 3 study visits, 127 (16%) had 2 study visits, and 100 (12%) had only 1 study visit. During the time of the study, several plants closed. Of the 576 subjects with complete visit data, 71, 97, and 150 were no longer currently working in the lead industry at V1, V2, and V3, respectively.
Data collection methods have been previously described.7,9,10 Two technicians performed all neurobehavioral testing sessions throughout the study, ensuring consistency of testing across study visits. Information was collected on demographic variables, medical history and occupational history, blood pressure, and height and weight. Participants provided a blood specimen and a spot urine sample.11 X-ray fluorescence was used to measure tibia lead concentration (in units of micrograms lead per gram bone mineral [μg/g]) at V1 and V2.9
The neurobehavioral and peripheral nervous system testing has been previously described.7 Neurobehavioral function was assessed with a modified version of the World Health Organization Neurobehavioral Core Test Battery12; Purdue pegboard (model 32020, Lafayette Instrument Company, Inc., Lafayette, IL) was substituted for the Santa Ana Dexterity Test, and the Center for Epidemiologic Studies Depression scale was substituted for the Profile of Mood States. Trail-making tests A and B and Raven's Colored Progressive Matrices (Psychologic Corporation, San Antonio, TX) also were conducted. Peripheral vibration thresholds were measured with the Vibratron II tester (Physitemp Instruments, Inc., Clifton, NJ) and grip and pinch strength were assessed with the Jamar Hydraulic Hand and Pinch Gauge dynamometers (Sammons Preston, Bolingbrook, IL). We modeled 15 neurobehavioral tests and 4 peripheral nervous system outcomes.7
The goal of the analysis was to assess the relation of lead dose to changes in cognitive functioning over the course of time. We developed regression models that separated recent from cumulative dose, acute from chronic effects, and cross-sectional from longitudinal relations. Conceptually, there are 3 main associations of lead biomarkers with health outcomes: cross-sectional, historical, and longitudinal. Cross-sectional associations are those between baseline lead biomarkers and baseline health measures. Historical associations are those between lead biomarkers at time t (or earlier) and change in health measures from time t to time t′; these represent chronic effects and thus were evaluated only for cumulative dose (ie, tibia lead). Historical associations with blood lead were not examined because a blood lead level predicting change in health over the following year was not deemed to be biologically sensible. Longitudinal associations are those between change in lead biomarkers from time t to time t′ and change in health measures from time t to time t′. These were evaluated only for blood lead because annual change in tibia lead levels could not be validly measured with the equipment that was then in use; the error in tibia lead measurement was larger than the change in tibia lead levels during the course of 1 year. Both cross-sectional and longitudinal associations potentially represent acute effects.
Employing this framework for defining associations, we constructed regression equations of neurobehavioral test scores for each visit and for the change in test scores from t to time t′. The following equations define 4 primary models (Table 1). In the equations, we use blood lead (PbBi, j, where i = ith individual, j = jth visit) as the lead dose variable, but other lead dose biomarkers could be entered.
where 1 indicates V1, εi,1 is residual error, and C designates “cross-sectional.”
where H designates “historical” and L designates “longitudinal.”
Equations 1 to 3 can be combined to derive equations for change in test score from V1 to V2 (Eq 4), and change from V2 to V3 (Eq 5), as follows. For average change from V1 to V2 (subtracting Eq 1 from Eq 2 [intercepts not shown]):
For average change from V2 to V3 (subtracting Eq 2 from Eq 3 [intercepts not shown]):
We used 4 models of change to distinguish various effects of lead on neurobehavioral test scores (Table 1): short-term change associated with recent dose (model 1); longer-term change associated with cumulative dose, controlling for the cross-sectional influence of recent (model 2) or cumulative (model 3) dose; or a combination of these effects (model 4). These models were formulated before examination of the data to comport with hypotheses about the kinetics and toxicity of lead and to limit the possible lead dose terms that could otherwise be used in such models so as to simplify biologic interpretation and minimize chance associations.
Our models assume that blood lead and tibia lead are biologically independent (as opposed to statistically independent), with blood lead reflecting recent exposure and tibia lead reflecting cumulative exposure. However, blood lead and tibia lead are biologically dependent in the cross-sectional, historical, and longitudinal time frames, which suggests that it may be difficult to distinguish between these 2 effects given the change caused by lead that is likely to have already occurred at baseline assessment and the short time frame for further change. Furthermore, some of the historical association of tibia lead with change in test scores could be captured by the cross-sectional blood lead term if baseline blood lead is correlated with tibia lead and the test scores at baseline are correlated with change in test scores over the course of time. Another consideration is that, unless serial correlations are powerful, cross-sectional measures are likely to have less error than change measures. This does not mean that the cross-sectional terms are necessarily more precisely estimable than the historical or longitudinal terms, because the relative precision with which the terms can be estimated also depends on the proportion of covariation between lead biomarkers and health outcomes that is accounted for by between-person versus within-person variation in lead levels. Finally, the error in blood lead measurement is less than the error in tibia lead measurement, which may make associations more easily observed with blood lead than tibia lead.
Because of the departures from normality, 6 outcomes were natural-log-transformed (simple reaction time, mean and SD, Trails A, Trails B, and vibration threshold, index finger and great toe). To standardize the direction of associations, values for 8 outcomes were made negative for the analysis (the 6 that were transformed, plus pursuit aiming test [only the metric for number incorrect] and the Center for Epidemiologic Studies depression scale). All analyses of neurobavioral tests were performed using multivariate linear regression models of workers’ repeated measures on a given test in terms of lead and other predictor variables, using maximum likelihood fitting for multivariate normally distributed data (generalized estimating equations).13 Standard errors and confidence intervals were thus estimated while accounting for correlations among repeated measures within subjects. Models for the repeated measures correlations were unstructured (ie, did not constrain the correlations to be the same across all of a worker's visits).14,15
All regression models controlled for baseline age, visit number, education, and sex, plus height (for vibration measures), and body mass index (for pinch and grip strength). Other potential confounding variables were assessed but did not merit inclusion in the final models (eg, tobacco and alcohol consumption, job duration) because they were not associated with test scores and did not appreciably alter associations of the lead measures with test scores. Models 1–4 (Table 1) were fitted to each of 15 neurobehavioral and 4 peripheral nervous system outcomes separately. From these models, β coefficients and 95% confidence intervals are presented.
To interpret association magnitudes, we describe the mean proportional difference in the test interquartile range (IQR) that would be associated with an increase from the 25th to the 75th percentile in lead level at baseline. Specifically, the blood lead or tibia lead β coefficients from model 4 were multiplied by the IQR of blood lead or tibia lead at V1 and then divided by the IQR of the neurobehavioral test at V1. We have performed these normalizations to permit direct comparison of magnitudes across test scores. Dividing by the IQR of a given test is similar in its interpretation to a Z-transformation, in which one divides by the standard deviation; this procedure eliminates test score units. We calculated the magnitude of the lead associations in terms of age (model 4) by deriving the number of years of increased age at baseline that was equivalent to an increase of lead from the 25th to the 75th percentile.
For the final models we examined added variable plots and partial residual plots16 to evaluate linearity, influential points, and homoscedasticity. Nonlinearity also was evaluated by including quadratic terms for age and the lead biomarkers. Because of concerns that temporal changes in lead levels caused by dropout could result in spurious associations, our primary longitudinal models included only lead workers who completed all 3 study visits (n = 576). Results were unchanged, and so only the results from the primary models are reported.
Description of Study Subjects
Baseline information has previously been reported.7,9,10 There was a wide range of blood and tibia lead levels among the 576 current and former lead workers. Mean blood lead level was 31.4 μg/dL (SD, 14.2; range, 4–76), and mean tibia lead was 38.4 μg/g (SD, 43; range, −7 to 338). Among the lead workers with complete visit data, 76% were men; mean age at the first study visit was 41.4 years (SD 9.5) and mean job duration was 8.5 years (SD, 6.3). Lead workers who completed the 3 study visits were older (3.3 years), had lower baseline blood lead levels (2.0 μg/dL), had longer job durations (1.6 years), were more likely to be women (24% of completers vs. 10% of noncompleters), and were more likely to have less than high school education compared with all lead workers at enrollment. These covariates were included in our regression models, thereby adjusting for differential dropout by observable characteristics of subjects.17
There was no difference in baseline tibia lead levels between those with complete visit data and other lead workers. Blood lead levels at V1 were correlated with levels at V2 and V3 (Pearson's r correlation coefficients = 0.86 and 0.79, respectively). The Pearson's r for the correlation of tibia lead at V1 and V2 was 0.90.
Modeling of Longitudinal Data
The 4 models in Table 1 were compared to evaluate transient versus longer-term effects. The results of model 1 (Table 2), which evaluated short-term change, show consistent cross-sectional associations of blood lead with lower test scores, mainly in the domains of executive abilities and manual dexterity. In this model, longitudinal blood lead was consistently associated only with declines in Purdue pegboard scores, thus providing limited evidence for a longitudinal association between changes in blood lead and neurobehavioral decline. In addition, there were 9 positive β coefficients, which is not strongly deviant from expectation under sampling variation. These findings indicate that higher recent doses, as assessed by blood lead, are cross-sectionally associated with worse neurobehavioral performance, mainly on tests of executive abilities and manual dexterity.
The results of model 2 (Table 2) show associations of cross-sectional blood lead with lower test scores similar to those in model 1, although historical tibia lead was more consistently associated with declines in test scores (manual dexterity, executive abilities, neuropsychiatric, and peripheral nervous system sensory function) than was longitudinal blood lead. In model 3 (Table 2), associations of cross-sectional tibia with lower test scores were less consistent than those with cross-sectional blood lead, suggesting that lower test scores are more associated with recent dose than with cumulative dose. Finally, in the combined model assessing both short-term and longer-term change (model 4, Table 2), there were associations of declines in test scores (executive abilities, manual dexterity, neuropsychiatric, and peripheral nervous system sensory function) with both historical tibia lead and longitudinal blood lead, similar to those already described for models 1 (blood lead) and 2 (tibia lead). Thus, the models suggest cross-sectional associations of recent lead exposure with poorer executive function and manual dexterity, and also of longer-term lead exposure with worse maintenance of functioning in peripheral nervous system sensory perception, executive abilities, and neuropsychiatric symptoms.
The lead associations were of a magnitude that could be important from both the clinical and public health perspectives (Table 3). Consider, for example, the Purdue pegboard dominant hand (a test with which all lead measures in model 4 were adversely associated). An increase of blood lead from 21 μg/dL to 40 μg/dL (the IQR for blood lead) at V1 was associated with test scores at baseline that were lower by 11% of the test IQR at baseline (cross-sectional blood lead column in Table 3) and lower by 6% of the test IQR per year (longitudinal blood lead column). An increase of tibia lead from 14 μg/g to 47 μg/g (the IQR for tibia lead) was associated with test scores that were lower by 2% of the test IQR per year (historical tibia lead column). On average, for the tests that were adversely associated with the lead measures and whose confidence interval did not include zero, for an increase of the lead measures from the 25th to the 75th percentile, blood lead (cross-sectional) was associated with test scores that were lower by 11% of the IQR at baseline; tibia lead (historical) was associated with test scores that were lower by 2% of the IQR per year; and blood lead (longitudinal) was associated with test scores that were lower by 9% of the IQR per year. These associations were equivalent to 3.8 years of age at baseline for cross-sectional blood lead, 0.9 years of age at baseline for historical tibia lead, and 4.8 years of age at baseline for longitudinal blood lead.
It was important to understand the behavior of the cross-sectional, historical, and longitudinal lead terms in our models; therefore, we evaluated how associations with cross-sectional and historical age terms (βC and βH for age) compared. We evaluated age associations from model 4, comparing regressions with cross-sectional (baseline) age alone to those with baseline age and historical age. Baseline age alone was associated strongly with all 19 outcomes. The addition of historical age had little effect on these estimates, whereas the association of historical age with the outcomes was much weaker. For the 5 strongest associations with historical age, the magnitudes were, on average, only 12% of those of the associations with baseline age. This result is important and suggests that the different time periods covering the cross-sectional and historical associations (ie, change over the period of exposure duration minus some induction period, compared with that over the past year) influence the consistency and magnitude of the associations.
Evaluation of Assumptions and Robustness of Findings
Cross-sectional adjustment for job duration and inclusion of a quadratic term for age did not alter the associations and, thus, these terms were not included in the final models. Nonlinearity of associations with the lead biomarkers was evaluated by inclusion of quadratic terms for the lead terms, one at a time, in each of the 4 models. There was no evidence of nonlinearity and, thus, the final models did not include quadratic terms. Examination of added variable plots and partial residual plots did not suggest nonlinearity, influential points, or heteroscedasticity. Finally, to evaluate induction period, models were repeated in lead workers with 1 year or more (n = 527) and then 3 years or more (n = 450) of employment in the lead industry. In general, the associations in models 1–4 increased slightly in magnitude and consistency when going from no employment criterion to 3 or more years.
There have been 4 longitudinal studies of the neurobehavioral effects of adult inorganic lead exposure that generally were conducted in the occupational setting.18–21 These have been limited by poor follow-up rates, follow-up periods of short duration, small sample sizes, no measures of cumulative dose, and no attempts to separate recent and cumulative dose or acute and chronic effects. In short, there are no adequate previous studies to which we can compare our results. In a recently published study of 535 former organolead manufacturing workers with past mixed exposures to organic and inorganic lead, peak tibia lead levels predicted longitudinal declines for 6 tests of verbal memory and learning, visual memory, executive ability, and manual dexterity.22 Blood lead levels were not associated with test scores at cross-section or over time.
In a previous cross-sectional analysis of neurobehavioral test scores from the first year of our study, blood lead was a consistent predictor of worse performance on 8 of 19 tests of central and peripheral nervous system function.7 After adjustment for covariates, tibia lead was not associated with neurobehavorial test scores at cross-section. In the present analysis of the longitudinal data, we attempted to separate the possibly transient effects from the longer-term effects of lead. Our analyses show consistent associations of both lead biomarkers with current neurobehavioral test scores and also with declines in test scores over time, with a stronger association with blood lead than tibia lead. The magnitude of these associations across an interquartile range of exposure was equivalent to 1 to 5 years of aging.
This study is the largest published longitudinal analysis of neurobehavioral function in current and former lead workers. Compared with previous studies, it includes measurement of a broader panel of lead biomarkers (some repeatedly), a high follow-up rate (with 72% of lead workers who had complete visit data), and longer average duration of follow-up (2.2 years). The repeated-measures design allowed us to identify biases that would select more highly exposed workers on their neurobehavioral characteristics at some point in time but not on the robustness of their neurobehavioral characteristics to dose-induced changes.
We previously reviewed several caveats of our approach to blood and tibia lead modeling, namely lack of biologic independence, cross-sectional terms taking explained variance from the historical terms, smaller errors in cross-sectional compared with change measures, and smaller error in blood lead measurement compared with tibia lead measurement. Taken together, these caveats suggest that, in longitudinal models of this type, stronger associations are likely to be observed with cross-sectional rather than historical or longitudinal terms and with blood lead rather than tibia lead. The likely significant measurement error, especially for tibia lead and change measures, and the relatively short follow-up interval, could obscure relations between lead dose and changes in test scores. This knowledge must be factored into our inferences about the likely health effects. We conclude that lead likely has an acute effect on cognitive test scores as a function of recent dose, and a chronic, possibly progressive, effect on cognitive decline as a function of cumulative dose.
We thank Patrick J. Parsons for performing urine lead measurement and Yong-Bae Kim, Kyu-Yoon Hwang, Sung-Soo Lee, and Kyu-Dong Ahn for assisting with data collection in Korea.
1.Agency for Toxic Substances and Disease Registry. Toxicological Profile for Lead. Update, 1999
. Atlanta, GA: US Department of Health and Human Services, Public Health Services; 1999.
2.Gordon JN, Taylor A, Bennett PN. Lead poisoning: case studies. J Clin Pharmacol
3.Tüzün M, Tüzün D, Salan A, et al. Lead encephalopathy: CT and MR findings. J Comput Assist Tomogr
4.Meyer-Baron M, Seeber A. A meta-analysis for neurobehavioral results due to occupational lead exposure with blood lead concentrations < 70 μg/100 ml. Arch Toxicol
5.Goodman M, LaVerda N, Clarke C, et al. Neurobehavioral testing in workers occupationally exposed to lead: systematic review and meta-analysis of publications. Occup Environ Med
6.Balbus-Kornfeld, Stewart W, Bolla KI, et al. Cumulative exposure to inorganic lead and neurobehavioral test performance in adults: an epidemiologic review. Occup Environ Med
7.Schwartz BS, Lee B-K, Lee G-S, et al. Associations of blood lead, dimercaptosuccinic acid-chelatable lead, and tibia lead with neurobehavioral test scores in South Korean lead workers. Am J Epidemiol
8.Rothman KJ, Greenland S. Modern Epidemiology
edition). Philadelphia: Lippincott–Raven Publishers, 1998.
9.Todd AC, Lee B-K, Lee G-S, et al. Predictors of DMSA chelatable lead, tibial lead, and blood lead in 802 Korean lead workers. Occup Environ Med
10.Schwartz BS, Lee B-K, Lee G-S, et al. Associations of blood lead, dimercaptosuccinic acid-chelatable lead, and tibia lead with polymorphisms in the vitamin D receptor and δ-aminolevulinic acid dehydratase genes. Environ Health Perspect
11.Weaver VM, Lee B-K, Lee G-S, et al. Associations of lead biomarkers with renal outcomes in Korean lead workers. Occup Environ Med
12.Johnson BL, Baker EL, Batawi M, et al. Prevention of Neurotoxic Illness in Working Populations
. New York: John Wiley and Sons; 1987.
13.Diggle PJ, Liang K-Y, Zeger SL. Analysis of Longitudinal Data
. Oxford: Clarendon Press; 1994;131–145.
14.Zeger SL, Liang K-Y. Longitudinal data analysis for discrete and continuous outcomes. Biometrics
15.Liang K-Y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika
16.Cook RD, Weisberg S. Residuals and Influence in Regression
. New York: Chapman and Hall; 1982.
17.Fitzmaurice GM, Laird NM. Analysing incomplete longitudinal binary responses: a likelihood-based approach. Biometrics
18.Baker EL, White RA, Pothier LJ, et al. Occupational lead neurotoxicity: improvement in behavioural effects after reduction of exposure. Br J Ind Med
19.Mantere P, Hänninen H, Hernberg S, et al. A prospective follow-up study on psychological effects in workers exposed to low levels of lead. Scand J Work Environ Health
20.Stollery BT, Broadbent DE, Banks HA, et al. Short term prospective study of cognitive functioning in lead workers. Br J Ind Med
21.Yokoyama K, Araki S, Aono H. Reversibility of psychological performance in subclinical lead absorption. NeuroToxicol
22.Schwartz BS, Stewart WF, Bolla KI, et al. Past adult lead exposure is associated with longitudinal decline in cognitive function. Neurology