Past adult lead exposure is associated with longitudinal decline in cognitive function

Study design and subject recruitment.

A 4-year prospective evaluation of cognitive function was conducted in former employees of a chemical manufacturing facility in the eastern United States that produced tetraethyl and tetramethyl lead.13-15⇓⇓ At the time the study was initiated, all subjects had ceased occupational exposure to lead, an average of 16.0 years before the first study visit. The study was approved by the Committee for Human Research at the Johns Hopkins School of Hygiene and Public Health, and all subjects provided informed, written consent.

Details of the selection and recruitment of study subjects have been reported elsewhere.14,15⇓ Former lead workers were eligible for recruitment if they were ever employed in the lead facility on or after January 1, 1950, were men, and were between 40 and 70 years of age in 1995. Of the estimated 968 eligible former workers who we attempted to contact, 73% enrolled into the study.14,15⇓ Limited data available on refusals suggested that the mean age and work duration of refusals did not differ from subjects who were enrolled. For comparison, community-based nonexposed control subjects were also recruited for the study. A database of telephone numbers in the residential areas of former lead workers was purchased from a commercial vendor. Numbers in the exchanges of former lead workers were randomly selected and telephoned to identify households with eligible men who were then frequency-matched to former lead workers for age, education, and race.

Subjects were enrolled from June 1994 to October 1997 and completed two to four annual visits in which cognitive function was assessed. A total of 703 former lead workers and 130 nonexposed controls were enrolled in the study. For former lead workers, the current report is based on 543 subjects on whom tibia lead levels were measured during the third year of the study. Tibia lead measurements were completed on 84% of eligible former lead workers (i.e., the denominator did not include those who were too ill, deceased, or had moved out of the area by the third year of the study).14,15⇓ Tibia lead measurements were not performed in nonexposed controls.

Follow-up status.

The 543 former lead workers were enrolled over a 3-year period, with 403 (74.2%) enrolled in the first year, 101 (18.6%) in the second year, and 39 (7.2%) in the third year. All but one former lead worker had at least one follow-up visit (99.8%); 393 (72.4%) subjects had four visits, 99 (18.2%) had three visits, and 51 (9.4%) had two visits. The nonexposed controls were also enrolled over a 3-year period, with 119 (91.5%) in the first year, 7 (5.4%) in the second year, and 4 (3.1%) in the third year. A total of 91.6% of controls had at least one follow-up visit; 92 (70.2%) had four visits, 13 (9.9%) had three visits, and 15 (11.5%) had two visits. The mean ± SD (range) time intervals in years between visit one to visit two, visit two to visit three, and visit three to visit four were 1.05 ± 0.34 (0.60 to 2.78), 0.89 ± 0.21 (0.54 to 2.19), and 0.92 ± 0.17 (0.33 to 1.81).

Data collection.

A well-constructed neurobehavioral battery consists of tests assessing a wide range of cognitive domains, including intelligence, verbal and visual learning and memory, visuo-constructive and organizational skills, executive functioning, reaction time, and motor performance. The vocabulary subtest from the Wechsler Adult Intelligence Scale–Revised (WAIS-R)16 was used as a measure of verbal intellectual functioning. Verbal learning and memory were assessed with the Rey Auditory Verbal Learning Test (RAVLT)17,18⇓ and the serial digit learning task,19 whereas visual learning and memory skills were determined using the symbol digit paired associate learning test20 and the Rey-Osterrieth complex figure, delayed score (hereafter referred to as Rey complex figure).21 The Rey complex figure, copy21 was used to measure visual planning and organization, whereas the block design subtest of the WAIS-R16 assessed both visual organization and visuo-construction. Trail making tests A and B (Trails A and Trails B),22 as well as the digit symbol subtest from the WAIS-R16 and the Stroop test (A form minus C form),23 assessed mental flexibility, executive functioning, and visual scanning. The choice reaction time task evaluated reaction time and response discrimination (computer administered test with a “1” or “2” appearing at random intervals ranging from 1 to 10 seconds for a total of 100 trials, the first 17 of which were practice).24 Olfactory function was measured with the University of Pennsylvania Smell Identification Test (UPSIT).25 The finger tapping task22 (five trials each hand alternating from one to the other) was used to measure motor performance, whereas the Purdue pegboard26 was used to assess manual dexterity. Subjects also completed the Center for Epidemiologic Studies–Depression Scale (CES-D),27 a measure of depressive symptomatology, and the Symptom CheckList-90 (SCL-90), a measure of the frequency and magnitude of psychological symptoms.28 Herein, “neurobehavioral tests” refers to the entire test battery and “cognitive tests” refers to all tests except the neuropsychiatric symptom scales and the smell identification test.

After obtaining consent, data were collected in the following order: SCL-90, blood pressure, height, weight, an initial interview, neurobehavioral testing, and detailed occupational history. Information on physician diagnosis of CNS disease, including neurodegenerative disease and head injury, and psychiatric disorders was obtained at all study visits.

Biologic measurements.

Blood lead levels were measured by a commercial laboratory during the first year of the study on former lead workers by the standard addition method of the National Institute of Occupational Safety and Health.29 During the third year, current tibia lead levels were measured (in units of μg lead per gram of bone mineral, hereafter referred to as μg/g) by ¹⁰⁹Cd-induced X-ray fluorescence (XRF) to fluoresce K-shell X-rays from lead.8-10,30⇓⇓⇓ This was performed at the left mid-tibia shaft via a 30-minute measurement with a back scatter geometry.

Because lead is eliminated from bone over time, we were concerned that the large differences among individuals in time since last occupational exposure to lead could obscure true differences in cumulative lead absorption as assessed by current tibia lead levels. Moreover, preliminary analysis revealed that years since last exposure was a strong predictor of decline in neurobehavioral test scores (6 of 19 cognitive tests had p values for years since last exposure ≤0.05), and the coefficients for interaction terms between current tibia lead levels and years since last exposure were consistently negative (9 of 19 cognitive tests had p values for the interaction term <0.05). Years since last exposure was not a biologically plausible predictor of declines in neurobehavioral test scores, and years since last exposure to lead and age were only moderately correlated (Pearson’s r = 0.33, p < 0.01), so age did not explain these observations. These theoretical concerns and empirical observations thus motivated us to use years since last exposure to estimate tibia lead levels at the time of last occupational exposure to lead, termed peak tibia lead. Current tibia lead levels were extrapolated back using a clearance half-time of lead in tibia of 27 years,9 assuming first-order (mono-exponential) clearance from tibia.15

Statistical analyses.

Before data analysis, all neurobehavioral test results were reviewed by a neuropsychologist. The complete study records of all subjects with outlying values on the neurobehavioral measures were evaluated for possible explanations of unusually poor performance. One former lead worker was missing information on years since last exposure, and nine subjects (seven former lead workers and two controls) were excluded because they developed medical conditions not considered to be related to past lead exposure but which could adversely affect test performance (e.g., carpal tunnel syndrome, stroke, demyelinating or neurodegenerative disease). Thus, a total of 535 former lead workers and 118 controls were included in the analysis.

Neurobehavioral test change scores were calculated by subtracting the neurobehavioral test score at the first study visit from all subsequent scores (Y_i − Y₁, where i = 2, 3, or 4 and Y₁ = score at the first study visit). For each test, subjects could thus contribute one, two, or three change scores to the analysis. Change scores were then divided by the time between measurements to derive measures of change per year. This method assumes that the annual changes for each measure were equivalent from year to year. Tests were standardized for direction so that a negative coefficient indicated a relative decline in performance with increasing peak tibia lead.

All analyses of neurobehavioral test performance were made using linear regression models of annual change scores on a given test in terms of lead and other predictor variables. Because each regression model involved repeated measures on individuals over time, the analyses needed to account for the intrasubject correlation among repeated measures. Generalized estimating equations fitting was used for this purpose, for all the regression models described below.31,32⇓ The analysis was performed under an exchangeable correlation assumption. Thus, regression coefficients had the interpretation of predictor effects on mean neurobehavioral change per year, and standard errors and confidence intervals correctly accounted for correlations among repeated measures within subjects.

First, using linear regression models as just described, annual change scores were compared in former lead workers and controls, adjusting for age at the first study visit, education (less than high school, high school graduate, some college, college graduate), testing technician (four different technicians), number of previous visits (to control for practice effects), and neurobehavioral test score at the first study visit (to control for regression toward the mean). Next, to determine if age-related declines in cognitive function differed between former lead workers and nonexposed controls, cross-product terms of age at the first study visit and former worker versus control status were added to the regression models.

The influence of peak tibia lead levels on annual change for each of the 22 neurobehavioral tests was investigated in 22 separate regression models. Each regression model controlled for confounding by age at the first study visit, educational level, testing technician, number of previous visits, and neurobehavioral test score at the first study visit. Associations of blood lead levels obtained at the first study visit with annual change scores were also evaluated in separate regression models.

The influence of self-reported health conditions and tobacco and alcohol use were also assessed in the final models but were not included because such use neither predicted annual change scores nor confounded relations between lead exposure and annual change scores. To evaluate nonlinear (curvilinear) relations between peak tibia lead and neurobehavioral test scores, both linear and quadratic terms for peak tibia lead were evaluated.33 Plots of the relations were examined to be certain that the models described the data accurately and the associations were not driven by outlying points. All significance testing was two-sided.

The XRF system provides a continuous unbiased estimate of the concentration of lead in bone that oscillates around the true concentration, so estimates below zero can be obtained when the true concentration is close to zero.34 Seven subjects had negative current tibia lead estimates, which were retained in the analysis because this method minimizes bias and does not require censoring of data.34 One subject had an extreme value of peak tibia lead (117.9 μg/g) and was excluded on statistical grounds before modeling neurobehavioral outcomes; exclusion of this subject did not significantly influence the regression results.