Dear Editor

We thank Drs Schwartz and Seeber, as well as their colleagues, for the commentaries on our paper “Neurobehavioral Testing in Workers Occupationally Exposed to Lead: Systematic Reviewand Meta-Analysis of Publications”[1] and would like to respond to their criticisms.

Response to specific points made by Schwartz et al.:

(i) “No evaluation of the quality of the evidence from available studies based on study design and analytic method.”
Our article first reviews the individual studies to assess their quality, summarises findings in Table 1 and only then presents the results of a meta-analysis. In our methods section, we specifically discuss the criteria for assessing the quality of the individual studies. These included:
a) evaluation of pre-exposure status;
b) adjustment for age;
c) adjustment for other occupational exposures;
d) adjustment for alcohol use;
e) adjustment for socio-economic confounding factors such as income level, education, etc.; and
f) use of blinding procedures.

(ii) “Data were combined from poorly done studies with data from welldone studies.”
Table 1 clearly shows that no study satisfied all of the above criteria for quality. As Schwartz et al. did not provide criteria to distinguish a “poorly done” from a “well done” study, we re-analysed the data based on five studies that adjusted results for age, education, and alcohol use and, therefore, in our opinion can be considered relatively well-designed. These five studies (Baker et al.,[2] Campara et al.,[3] Chia et al.,[4] Maizlish et al.,[5] and Williamson and Teo[6]) contained sufficient information to conduct a meta-analysis for only three tests: Santa Ana preferred hand, Santa Ana non-preferred hand, and Digit Symbol. The results were as follows:

For the Santa Ana preferred hand test, the effect size changed from non-significant negative to non-significant positive. For the Santa Ana non-preferred hand the result changed slightly towards the null and remained statistically non-significant. For the Digit Symbol test, the result changed away from the null and remained statistically significant in the fixed effects model, but changed slightly towards the null and was no longer statistically significant in the two random effects models. (See table)

(iii) “Inclusion of studies that did not control for age and education.”
Although adjustment for age and education is undoubtedly important, Schwartz et al. do not provide evidence that these are “the two most important predictors.” One could certainly argue that alcohol use, or the presence of pre-existing neuropsychiatric conditions would also act as powerful confounders. Our review indicated that the 22 studies had non-overlapping strengths and limitations and further inclusion or exclusion based on quality appeared to be a matter of judgment. Nevertheless, we accepted the Schwartz et al. argument and re-analysed the data based on the 13 studies that adjusted for age and education. The results indicated that, as opposed to our original findings (based on all 22 studies), none of the tests showed a statistically significant difference in all three models. (See table)

(iv) “No adjustment for age, education, or lead dose differences across studies.”
This criticism appears to be somewhat theoretical, as the data did not allow such adjustment. Inclusion of only those studies that adjusted for age and education in part addresses this issue with no apparent impact on the overall conclusions.

(v) “Reliance on exposed v. control comparisons” rather than “only including studies that reported beta coefficients for the blood lead v. test score relation, or adjusting for mean blood lead levels in exposed and non-exposed groups.”
As noted in the abstract and in the methods section, we defined low to moderately exposed populations as those whose central tendency for blood lead concentration was less than 70mg/dl. We used the same definition of exposure as the previously published meta-analysis by Meyer-Baron and Seeber[7] to find out if the results of our two studies were reproducible.The direct comparison of the two analyses in the discussion section was important in explaining our position with regards to meta-analysis as a research technique. We do agree that other approaches such as those proposed by Schwartz and colleagues could be informative. We were surprised to read the following statement:
“The authors conclude that blood lead levels, that are described as “moderate” in one location in the manuscript and “low” in another, are not associated with neurobehavioral test scores.” We believe this statement misrepresents our conclusions listed on page 222.[1]

(vi) “Reliance on a small number of unspecified studies for effect estimates. Table 2 of the study reports the number of studies that were combined to derive effect estimates, but does not specify which studies were combined.”
The originally submitted version of the article included tables with information on each individual study. However, following the reviewers’ comments, and at the request of the editor, we had to shorten the manuscript substantially. We will make this information available upon request.

With respect to the omission from our meta-analysis of the recent article by Schwartz et al.,[8] we have to point out that our manuscript was submitted for publication in December 2000, while the Schwartz et al. study appeared in press in May 2001. The other two studies cited in their letter did not meet our inclusion criteria.

We have not had an opportunity to evaluate the association between cumulative exposure to lead and neurobehavioral testing results. However, we did note that the 2001 article by Schwartz et al.,8 found no association between tibia lead levels and test scores.

Response to Seeber:
Although Dr Seeber agrees with our conclusion about the need for prospective studies he states that “the repeated information on cross-sectional studies should also be accepted as source for conclusions on (neurobehavioral) effects due to exposures” and that meta-analyses are one approach to search such summarizing information.”

After having reviewed the results of five meta-analyses on the subject (two presented in the recent Seeber and Meyer Baron article,[9] our paper,[1] and the two additional re-analyses presented here) we found five different sets of results with no evidence of consistency to qualify these results as “repeated”. Therefore, we have to adhere to our original conclusions.

References

(1) Goodman M, LaVerda N, Clarke C, Foster ED, Iannuzzi J, Mandel J. Neurobehavioural testing in workers occupationally exposed to lead: systematic review and meta-analysis of publications. Occup Env Med 2002;59:217-23.

(2) Baker EL, Feldman RG, White RA, Harley JP, Niles CA, Dinse GE, Berkey CS. Occupational lead neurotoxicity: a behavioral and electrophysiological evaluation. Study design and year one results. Br J Ind Med 1984;41:352-61.

(3) Campara P, D’Andrea F, Micciolo R, Savonitto C, Tansella M, Zimmerman-Tansella C. Psychological performance of workers with blood-lead concentration below the current threshold limit value. Int Arch Occup Environ Health 1984;53:233-46.

(4) Chia S-E, Chia H-P, Ong C-N, Jeyaratnam J. Cumulative blood lead levels and neurobehavioral test performance. Neurotoxicology 1997;18:793-804.

(5) Maizlish N, Parra G, Feo O. Neurobehavioral evaluation of Venezuelan workers exposed to inorganic lead. Occup Environ Med 1995;52:408-14.

(6) Williamson A, Teo R. Neurobehavioral effects of occupational exposure to lead. Br J Ind Med 1986;43:374-80.

(7) Meyer-Baron M, Seeber A. A meta-analysis for neurobehavioral results due to occupational lead exposure with blood lead concentrations <70 µg/ml. Arch Toxicol 2000;73:510-8.

(8) Schwartz BS, Lee BK, Lee GS, Stewart WF, Lee SS, Hwang KY, Ahn KD, Kim YB, Bolla KI, Simon D, Parsons PJ, Todd AC. Associations of blood lead, dimercaptosuccinic acid-chelatable lead, and tibia lead with neurobehavioral test scores in South Korean lead workers. Am J Epidemiol 2001 Mar 1;153(5):453-64.

(9) Seeber A, Meyer-Baron M, Schäper M. A summary of two meta- analyses on neurobehavioural effects due to lead exposure. Arch Toxicol2002;76:137-45.

Dear Editor

Whether or not low-to-modest levels of exposure to lead have a detrimental effect on cognition is an important issue given the growing attention, for example, in the United States, that has recently been paid towards potentially revising downward the levels of lead exposure allowed in the workplace. Thus, we read with interest the meta-analysis of studies on this topic that appeared in this journal by Goodman et al. [1] Unfortunately, we believe that the authors’ conclusions are not valid. Specifically, the authors state that “the data available to date are inconsistent and are unable to provide adequate information on the neurobehavioral effects of exposure to moderate blood concentrations of lead.” We found no direct support for this conclusion in the publication. Moreover, numerous flaws in their method limit any specific inferences that can be made. In general, we found that the meta-analysis combined evidence from studies of widely varying quality and did not account for significant confounding within and between studies. Given these and other flaws, it was predictable that the authors did not find an association between blood lead levels and neurobehavioral test scores.

Specific concerns that we had with the methods include:
(i) The authors offer no evaluation of the quality of the evidence from available studies based on study design and analytic method.
(ii) The authors combine data from poorly done studies with data from well done studies, clouding any effects that is observable from the better conducted studies.
(iii) Although age and education adjustment within studies is assessed, six studies were included that did not adjust for age and another three studies did not adjust for education. These are the two most well established predictors of neurobehavioral test scores and the most important potentially confounding variables.
(iv) Even among the remaining studies that did adjust for age and education, the authors do not address the confounding in the meta-analysis that is caused by variation in age and education across study populations.
(v) The authors’ main effect measure is an exposed v. control comparison. Among the options that could have been pursued, this is the effect measure with the lowest power. It is unable to assess a dose-effect relation, and it is also the one most prone to selection bias.
(vi) Relatively few of the 22 studies listed in Table 2 contribute to the estimate of the effect size for each neurobehavioral outcome. Moreover, the authors do not state which studies contributed to the effect estimate.

It is important to note that several recent studies, all published before this article was accepted for publication, reported that blood lead was associated with neurobehavioral test scores in multiple cognitive domains. One study of 803 Korean lead workers is the largest study reported to date and observed consistent associations of blood lead with test scores in the domains of executive abilities, manual dexterity, and peripheral motor strength at blood lead levels as low as 18 µg/dL. [2] In another study of former organolead manufacturing workers, tibia lead was associated with test scores at cross-section [3] and with longitudinal declines in test scores.[4] These findings suggest that lead may have both short-term and progressive influences on neurobehavioral performance.

We elaborate on our main concerns, below.

(i) No evaluation of the quality of the evidence available from studies, and (ii) data were combined from poorly done studies with data from well done studies. It is traditional in meta-analysis to establish a priori criteria for what defines acceptable evidence from studies. The authors only had three inclusion criteria, none of which refer to the quality of the study designs, analytic method, adjustment for confounding, evaluation of bias in selection of exposed and non-exposed subjects, and other such methodologic factors. There is apparently no consideration for this arguably single most important step in meta-analysis. The meta-analytic results could simply reflect wide heterogeneity in the quality of the evidence that was combined. This factor alone could account for the overall conclusion of no association.

(iii) Inclusion of studies that did not control for age and education. Age and education are the two most important predictors of neurobehavioral test scores in working populations. In the absence of adjustment for these confounders there should be convincing evidence that the two groups being compared were equivalent in age and education. Eight of the included studies did not adjust for age and/or education. The authors offer no explanation for why these studies should be included in the meta- analysis.

(iv) No adjustment for age, education, or lead dose differences across studies. By not adjusting for age and education differences across studies, the authors make an implicit assumption that age and education do not modify the relation between blood lead and neurobehavioral test scores. This may or may not be true. In the meta-analysis, the authors also implicitly assume a fixed difference in blood lead levels between exposed and non-exposed groups. Table 1 clearly indicates that this assumption does not hold.

(v) Reliance on exposed v. control comparisons. This is a weak test and a test that is not germane to the conclusions that the authors make. The authors conclude that blood lead levels, that are described as “moderate” in one location in the manuscript and “low” in another, are not associated with neurobehavioral test scores. All studies included exposed workers with a range of blood lead levels, from very low to high. More appropriate approaches could have been considered, for example, only including studies that reported beta coefficients for the blood lead v. test score relation, or adjusting for mean blood lead levels in exposed and non-exposed groups.

(vi) Reliance on a small number of unspecified studies for effect estimates. Table 2 of the study reports the number of studies that were combined to derive effect estimates, but does not specify which studies were combined. This omission does not allow the reader to determine whether solid evidence was combined with more questionable evidence, or to evaluate whether any of the issues described above were germane to the effect estimates reported.

Two more concerns exist regarding the author’s treatment of the issue of cumulative v. on-going lead exposure as well as the identification of the source of funding for this study. In their introduction, the authors quote the review by Balbus-Kornfeld et al. [5] which noted that “the current scientific evidence is flawed because of inadequate estimation of cumulative exposure to or absorption of lead…” but fail to acknowledge this issue in the interpretation of their own meta-analysis. In fact, as has been widely reported in the literature, methods are now available to non-invasively measure bone lead levels as a reliable and accurate measure of cumulative lead dose. Several studies (such as Payton et al.[6], Stewart et al.[3], and Schwartz et al.[4]) suggest that cumulative lead exposure is a very important biological marker that may be related to cognitive decrements not predicted by blood lead levels. With regards to funding, the authors note that they are mainly from Exponent Health Group in Alexandria, Virginia and Menlo Park, California; however, they fail to describe what motivated the study or sources of funding for the study. We believe this information would be of interest to scientists and policy-makers engaged in work on this topic.

References

(1) M Goodman, N LaVerda, C Clarke, E D Foster, J Iannuzzi, J Mandel. Neurobehavioural testing in workers occupationally exposed to lead: systematic review and meta-analysis of publications. Occup Environ Med 2002;59: 217-23.

(2) Schwartz BS, Lee BK, Lee GS, Stewart WF, Lee SS, Hwang KY, Ahn KD, Kim YB, Bolla KI, Simon D, Parsons PJ, Todd AC. Associations of blood lead, dimercaptosuccinic acid-chelatable lead, and tibia lead with neurobehavioral test scores in South Korean lead workers. Am J Epidemiol 2001;153:453-64.

(3) Stewart WF, Schwartz BS, Simon D, Bolla KI, Todd AC, Links J. Neurobehavioral function and tibial and chelatable lead levels in 543 former organolead workers. Neurology 1999;52:1610-7.

(4) Schwartz BS, Stewart WF, Bolla KI, Simon PD, Bandeen-Roche K, Gordon PB, Links JM, Todd AC. Past adult lead exposure is associated with longitudinal decline in cognitive function. Neurology2000;55:1144-50.

(5) Balbus-Kornfeld JM, Stewart W, Bolla KI, et al. Cumulative exposure to inorganic lead and neurobehavioral test performance in adults: an epidemiological review. Occup Environ Med 1995;52:2–12.

(6) Payton M, Riggs KM, Spiro A 3rd, Weiss ST, Hu H. Relations of bone and blood lead to cognitive function: the VA Normative Aging Study. Neurotoxicol Teratol 1998 Jan-Feb;20(1):19-27

Brian S. Schwartz, MD, MS
Professor and Director
Division of Occupational and Environmental Health
Johns Hopkins Bloomberg School of Public Health

Walter Stewart, PhD, MPH
Adjunct Professor
Department of Epidemiology
Johns Hopkins Bloomberg School of Public Health

Howard Hu, MD, ScD, MPH, MS
Associate Professor and Director
Occupational/Environmental Medicine
Department of Environmental Health
Harvard School of Public Health

Dear Editor,

We appreciate dr Wijngaarden´s interest in our report and his suggestion for understanding the differences in risk. Dr Wijngaarden suggests that difference in unemployment rate between electricians and glassworker and wood workers could be an explanation.

We have no data on employment status at time of death and can therefore not test this hypothesis. However, if employment status is an important predictor, this could explain some of the difference as the wood workers had a different employment than the other groups. Electricians and glass workers have had permanent positions since a long time, while wood workers had an employment for a certain object, e.g. a house before 1990s. When the object was finished he had to find another employer. Today, most construction workers have permanent positions in Sweden.

In our search of literature to understand differences in suicide rates between occupations we found little knowledge. This might be an important area of research in the future.

Dear Editor,

Järvholm and Stenberg [1] evaluated suicide mortality rates among electricians ("exposed to electromagnetic fields (EMFs)") and glass and woodworkers ("unexposed to EMFs") in the Swedish construction industry. Standard mortality rates were lower for the two job groups as compared to the Swedish general population. This is likely due to the healthy worker effect. The internal cohort analysis showed that electricians had a lower suicide mortality rate than glass and woodworkers.

As the authors rightfully point out, these results should not be seen as evidence against the association between exposure to EMFs and suicide, in particular because no quantitative estimates of exposure were obtained to directly evaluate this association. Jarvholm and Stenberg cited a small measurement survey in the Swedish construction industry, which indicated that exposure levels were low and comparable between the two occupational groups. Therefore, one would not expect to see an EMF-mediated increase in suicide risk among electricians as compared to glass and woodworkers, if an association between EMF exposure and suicide truly exists.

Järvholm and Stenberg suggested that the difference in suicide rate between the two job groups was unlikely to be due to differences in socioeconomic factors, but they did not provide an alternative explanation. One possible explanation may be a healthy survivor worker effect related to employment status (e.g., at time of death) within this cohort. That is, active workers may be more physically and mentally fit than those who left the industry or are unemployed, and may therefore be at lower risk of committing suicide [2]. A large body of literature suggests that unemployment and suicide are positively related [3,4], and being out of work was positively associated with suicide in the electric utility industry [2]. Since cessation of work also leads to cessation of work-related exposures, employment status may be an important potential confounder (or perhaps effect modifier) for the association between work-related exposures and suicide. The lower suicide rate among electricians as compared to glass and wood workers may be explained by a larger proportion of glass and wood workers with an inactive employment status at the time of death.

Although it is unlikely that consideration of employment status, if possible, would greatly alter the conclusions reached by Järvholm and Stenberg [1], it would be informative to see its influence on the rate ratio.

References:

(1) Jarvholm B, Stenberg A. Suicide mortality among electricians in the Swedish construction industry Occup Environ Med 2002; 59: 199-200.

(2) van Wijngaarden E, Savitz DA, Kleckner RC, Cai J, Loomis D. Exposure to electromagnetic fields and suicide among electric utility workers: A nested case-control study. Occup Environ Med 2000;57:258-263.

(3)Dooley D, Fielding J, Levi L. Health and unemployment. Annu Rev Public Health 1996;17:449-465.

(4)Kposowa AJ. Suicide mortality in the United States: differentials by industrial and occupational groups. Am J Ind Med 1999;36:645-652.