The Authors Reply: Thank you for giving us the opportunity to reply to the letters by Bove et al and Frumkin and Oris commenting on our mortality study of PCB-exposed capacitor workers.1 We disagree with the statement by Bove et al that ". . . limitations in the study approach tend to dilute any excesses in cancer mortality resulting from PCB exposure . . . ." These assertions are speculative and not supported by the data. Although some degree of misclassification in observational studies is unavoidable, it is usually not possible to determine whether this misclassification is differential or non-differential. Furthermore, non-differential misclassification does not always result in bias toward the null hypothesis. Neither the type nor the effect of the misclassification can be determined by Bove et al. In our article, we do, however, discuss at length the measures taken to limit misclassification, and we feel strongly that we were successful in doing so.
Bove et al assert that the healthy worker effect (HWE) results are an underestimate of the SMRs for all-causes mortality and cancer mortality. This is partially true. The HWE is most pronounced for cardiovascular deaths and thus affects all-causes mortality.2 It has much less of an effect on cancer deaths.3
The presentation by Bove et al of the all-cancers SMRs and selected cancer-specific SMRs without confidence intervals (CIs) gives incomplete information and is misleading. Had the confidence intervals been reported, the lack of significance for these SMRs would have been immediately obvious to the reader. Bove et al selected the female hourly employees' all-cancers SMR of 110 (95% CI, 93 to 129), intestinal cancer (SMR = 157; 95% CI, 96 to 242), rectal cancer (SMR = 169; 95% CI, 46 to 434), melanomas (SMR = 144; 95% CI, 30 to 421), and melanomas in male hourly employees (SMR = 130; 95% CI, 42 to 303). Notably absent from this list of SMRs considered by Bove et al are the male hourly SMRs for intestinal and rectal cancer (SMR = 57; 95% CI, 25 to 112; and SMR = 87; 95% CI, 18 to 255, respectively).
Bove et al suggest that the male all-cancers SMRs of 81 (hourly employees; 95% CI, 68 to 97) and 69 (salaried employees, 95% CI, 52 to 90) are largely due to the HWE. A careful examination of Table 4 in our article suggests that the statistically significantly low all-cancers SMRs in both the hourly and salaried males result primarily from the lower than expected lung cancer SMR (for hourly workers: 42 observed/54.5 expected; SMR = 77; 95% CI, 56 to 104; and for salaried workers: 12 observed/29.6 expected; SMR = 41; 95% CI, 21 to 71).
The statement by Bove et al that these elevations were consistent with elevations found in other studies of PCB-exposed workers is not correct.4-6 In addition to the three studies cited by Bove et al, there is the Bertazzi cohort and its update by Bertazzi et al7 and Tironi et al.8 The results of the Brown4 and Sinks et al5 studies are inconsistent with each other. The Loomis et al6 study of utility workers, not capacitor workers, did report an elevation in melanomas in some subsets of the cohort that were presumed to have had exposure to PCBs while working outdoors. Exposure to sunlight was not adequately accounted for by Loomis et al.6 Brown and Jones9 and Brown4 found an excess of liver and rectal cancers. Neither Sinks et al5 nor Loomis et al6 reported such increases. Sinks et al5 reported a non-significant elevation in brain and nervous system cancers. Neither Brown and Jones,9 Brown,4 Bertazzi et al,7 or Tironi et al8 found an elevation in brain cancer. These inconsistencies were discussed in our article.
Bove et al state that we only included a latency-period analysis for all cancers and for intestinal cancer. This was done primarily because of space limitations. Cumulative exposure and latency tables were computed and evaluated for many other causes of death, including all of the cancers of interest. The interpretation by Bove et al that the intestinal cancer SMR increases to a significant level for women with ≥20 years of latency ignores the importance of examining the trend associated with latency and length of employment. Furthermore, it might be worth noting that for women employed for 10 years or longer with a latency period ≥20 years, the SMR was 100. The individual category-specific SMRs cannot be interpreted as meaningful without examination of the trend across cumulative exposure categories. Although the intestinal cancer SMR for latency ≥20 years was significantly elevated, there was no significant trend indicating an increase in risk with cumulative exposure or latency, as discussed in our article. Furthermore, comparison with the regional population resulted in a much-reduced SMR (SMR = 120; 95% CI, 74 to 186) for intestinal cancer in female hourly workers. The regional comparison is more representative because higher rates of intestinal cancer are observed among the white population of the northeastern part of the United States.
Bove et al raise concerns about our exposure assessment. Several factors need to be recognized when assessing the propriety of our exposure assessment and our use of length of employment as a surrogate of exposure. Workers accumulate PCB body burdens over time, which persist for many years even after their occupational PCB exposure is discontinued. To suggest that PCB body burdens among capacitor workers were comparable to those found in the general population is unjustified and is not supported by previously published data.10-13 The fact that workers in capacitor plants had significantly higher body burdens than the general population has been demonstrated in other capacitor plants.14 As reported in our article, average serum PCB levels in the general population between 1976 and 1979 were 5 to 7 parts per billion (ppb; µg/L).14 Geometric mean serum PCB levels in GE workers in 1979 (2 years after PCBs were no longer used) were 277 ppb (µg/L) reported as Aroclor 1242 and 55 ppb (µg/L) reported as Aroclor 1254. In 1983, 5 years after termination of the use of PCBs, geometric mean serum levels were 116 ppb (µg/L) for Aroclor 1242 and 34 ppb (µg/L) for Aroclor 1254. In 1988, the geometric mean serum PCB levels were 90 ppb (µg/L) quantitated as Aroclor 1242 and 32 ppb (µg/L) quantitated as Aroclor 1254.15 Workers preferentially retained the more persistent congeners so that the gas chromatographic pattern of their body burden gradually approached that observed in the general population, with primary retention of the more highly chlorinated, poorly metabolized congeners.12 The half-lives of the major PCB congeners retained in these workers were as follows: for 2,4,4′ trichlorobiphenyl, 1.4 years; for 2,4,4′5 tetrachlorobiphenyl, 3.2 years; for 2,3′,4,4′,5 pentachlorobiphenyl, 5.8 years; and for 2,2′,4,4′,5,5′ hexachlorobiphenyl, 12.4 years.16 Even though different commercial mixtures of PCBs were used in the capacitor plants, the congeneric composition on a qualitative basis is similar.17 Production began in 1946 with the highly chlorinated Aroclor 1254, and small amounts of Aroclor 1254 were used in the plant at least through 1971.
The statement that length of employment alone was an inadequate surrogate for exposure and a likely source of exposure misclassification bias leading to an underestimation of the effect and a distortion of the exposure-response relationship is not supported by the toxicokinetics of PCBs, nor is it an accurate representation of the data analyses conducted on our cohort and reported in the article.
Bove et al report that the majority of hourly workers never worked in a high-exposure job, when in fact 1268 of the 2984 male hourly employees (42.4%) did work in a high-exposure job. Only 13.8% of the female hourly employees worked in a high-exposure job, not an uncommon occurrence in an industrial setting. To suggest that the remaining portion of the cohort experienced PCB exposure similar to that of the general population is not an accurate representation of the facts. This is presented in the exposure-assessment section of our article.
Bove et al state in the opening sentence that although the goal of the study was to evaluate six specific cancers, we focused almost entirely on all-cancers mortality. Table 4 in the article presents SMRs and 95% CIs not only for the six cancers of interest but for 32 other causes of death, including 15 additional cancers. The issue of statistical power is raised by Bove et al and two tables were provided. These tables were not properly referenced nor was the methodology used to generate these calculations explained. It is unclear why an SMR of 150 should be considered the "highest expected" for these cancers, when previous publications on smaller cohorts reported statistically significant SMRs well above 150. Our study was an attempt to evaluate these earlier observations in a larger study with a longer follow-up period.
Bove et al question the decision to limit the latency by length of employment calculations to cancers with more than two observed cases and a lower boundary of the 95% CI of 90 or above. This decision was made by the investigators to limit the multiple comparison problem and to provide more meaningful data, rather than to obscure data. Additionally, the lack of presentation of data should not be interpreted as the data not having been analyzed. All six a priori cancers of concern were examined carefully; however, publication space is limited and presenting a table of latency by cumulative exposure for liver cancer, for instance, with two deaths was deemed unwarranted.
In their summary statement, Bove et al dismiss our study findings because of the HWE effect, failure to account for latency, exposure misclassification, potentially insufficient dosage differences between exposed and comparison groups, and poor statistical power, yet they still insist that we did find excess cancer risk for three of the six a priori cancers of interest and give credence to those findings. It is inconceivable to the investigators of this study how Bove et al, given this litany of problems, were able to differentiate the impact and direction of these biases with such certainty and specificity.
The authors take exception to the tone of the letter by Frumkin and Orris and find statements such as "conspicuously silent" and "willful effort to avoid a positive finding" inflammatory and suggest that such statements do little to advance the understanding of PCBs and cancer risk.
Most of the issues raised by Frumkin and Orris have been addressed earlier. Their suggestion to include more capacitor plants to increase power has merit, however. The General Electric Company had only the two facilities in upstate New York (Hudson Falls and Fort Edward) where capacitors were made using PCBs.
Frumkin and Orris question whether high-exposure jobs actually entailed high exposure and raise concerns about misclassification. The exposure misclassification suggested by Frumkin and Orris is highly improbable, given the distinction between jobs with direct dermal and inhalation exposure and those with only inhalation exposure to PCB air levels in the plant, as explained and referenced in our article. Additionally, the characterization of this bias as substantial is unwarranted and is an overstatement of the potential effect. Assignment of exposure for specific job categories was done before determination of vital status. At both plants, workers were located in the same building, and the same air-ventilating system served the entire building. We verified the physical layout by conducting a walk through the building and by talking to present and former employees. Many workers had different jobs in the different exposure categories (high, undefinable, and low). All workers, including those in low-exposure jobs, had significantly higher exposures than the general population, on the basis of PCB serum levels reported by Lawton et al,11 Brown et al15,16 and Brown.18
The PCB blood levels (from 194 and 290 workers) mentioned by Frumkin and Orris were of limited value in validating an exposure job matrix for 7075 workers. Although the job histories and the exposure assignment did confirm that workers in high-exposure jobs had high PCB blood levels, these workers were selected either because of their known high-exposure job11 or they were self-selected.10 The high-exposure jobs were readily identified by plant personnel and were confirmed by PCB air-level readings and PCB blood levels. Misclassification of jobs into the high-exposure category or misclassifying high-exposure jobs as lower-level exposure jobs was extremely unlikely.
Frumkin and Orris suggested that PCBs are serious health hazards, irrespective of carcinogenicity, with effects that include decreased birth weight, neurodevelopmental effects, and interference with thyroid and estrogen hormone function. It has not been shown that PCBs interfere with estrogen-hormone function in humans. Studies conducted to examine the effects of PCBs in infants and children have been critically reviewed19-25 or could not be supported.26 Results from thyroid function tests performed in infants were within the normal range. Furthermore, Koopman-Esseboom et al27 stated, "The mean dioxin-like PCB toxic equivalent levels and the mean total PCB and dioxin toxic equivalent levels of the neurological normal infants were significantly higher (p = 0.04 for both) compared with the levels of the neurologically (mildly or definitely) abnormal infants. There was no relationship between the TT3 (serum total triiodothyroxine), FT4 (free thyroxine), and TSH (thyroid stimulating hormone) levels in maternal, umbilical, or infant plasma (collected in the second week after birth) and the results of the neonatal neurological examinations. We conclude that overt abnormalities found in the neonatal period are not caused by either direct effects of PCB or dioxin exposure or lowered thyroid hormone levels." According to the National Center for Health Statistics,28 birth weight is affected by education of the mother, mother's age, birth order, interval between births, gender, inadequate prenatal nutrition, alcohol consumption, smoking, lack of prenatal care, incidence of elective induction, contraceptive utilization, out-of-wedlock births, metropolitan areas (lower), and race. The body size of the parents and maternal illnesses such as diabetes also play a role. These many variables exemplify the difficulties of appropriately designing studies to examine a single factor affecting birth weight. Given these uncertainties and the published criticisms of studies reporting "other health effects of PCBs," it has not been conclusively shown that PCBs cause other "serious" health problems in humans.
We disagree with the final comment by Frumkin and Orris that this study was a great leap sideways on the path to a more complete understanding of the health effects of PCBs. The issue of PCBs and potential health effects has been a significant public health concern for more than 30 years. The lack of consistent findings in the previous cohort studies was assumed to have resulted from small cohort sizes and short follow-up periods. Given the disparate findings in these smaller capacitor cohorts, the appropriate next step was to assemble a larger cohort of PCB-exposed workers and examine them throughout a longer follow-up period. The fact that we were unable to confirm any of the previously reported findings is important and adds to the knowledge about PCBs and health effects. The assumption-that a negative study does not provide valuable information imposes significant restrictions on the scientific process and the ability to adequately and objectively assess all data.
Errata: The correct number of female salaried workers with a length of employment of 10 to < 15 years in Table 2 is 27; 5.8% is the correct percentage. In Table 6, line 2, last column, total SMR for ≥20 years of latency should be 117. The total number of workers in the upper panel of Table 2 should be 7075.
Renate D. Kimbrough, MD
Martha L. Doemland, PhD
Maurice E. LeVois, PhD
Institute for Evaluating Health Risks; Washington, DC
1. Kimbrough RD, Doemland ML, LeVois ME. Mortality in male and female capacitor workers exposed to polychlorinated biphenyls. J Occup Environ Med.
2. McMichael AJ. Standardized mortality ratios and the "healthy worker effect': scratching beneath the surface. J Occup Med.
3. Checkoway H, Pearce NE, Crawford-Brown DJ. Issues of study design and analysis. In: Research Methods in Occupational Epidemiology.
New York: Oxford University Press; 1989:78-79.
4. Brown DP. Mortality of workers exposed to polychlorinated biphenyls: an update. Arch Environ Health.
5. Sinks T, Steele G, Smith AB, et al. Mortality among workers exposed to polychlorinated biphenyls. Am J Epidemiol.
6. Loomis D, Browning SR, Schenk AP, et al. Cancer mortality among electrical workers exposed to polychlorinated biphenyls. Occup Environ Med.
7. Bertazzi PA, Riboldi L Pesatori A, et al. Cancer mortality of capacitor manufacturing workers. Am J Ind Med.
8. Tironi A, Presatori A, Consonei D, et al. The mortality of female workers exposed to PCBs. Epidemiol Prev.
9. Brown DP, Jones J. Mortality and industrial hygiene study of workers exposed to polychlorinated biphenyls. Arch Environ Health.
10. Wolff MS, Fischbein A, Thornton J, Rise C, Lilis R, Selikoff IJ. Body burden of polychlorinated biphenyls among persons employed in capacitor manufacturing. Int Arch Occup Environ Health.
11. Lawton RW, Ross MR, Feingold J, Brown JF Jr. Effects of PCB exposure on biochemical and hematological findings in capacitor workers. Environ Health Perspect.
12. Lawton RW, Brown JF, Ross MR, Feingold J. Comparability and precision of serum PCB measurements. Arch Environ Health.
13. Taylor PR, Reilly AA, Stelma J, Lawrence CE. Estimating serum polychlorinated biphenyl levels in highly exposed workers: an empirical model. J Toxicol Environ Health.
14. Kimbrough RD. Polychlorinated biphenyls (PCBs) and human health: an update. Crit Rev Toxicol.
15. Brown JF Jr, Lawton RW, Ross MR, Feingold J. Assessing the human health effects of PCBs. Chemosphere.
16. Brown JF Jr, Lawton RW, Ross MR, et al. Persistence of PCB congeners in capacitor workers and Yusho patients. Chemosphere.
17. Mayes BA, McConnell EE, Neal BH. Comparative carcinogenicity in Sprague-Dawley rats of the polychlorinated biphenyl mixtures Aroclors 1016, 1242, 1254, and 1260. Toxicol Sci.
18. Brown JF Jr. Determination of PCB metabolic, excretion, and accumulation rates for use as indicators of biological response and relative risk. Environ Sci Technol.
19. Paneth N. Human reproduction after eating PCB-contaminated fish. Health Environ Digest.
20. Paneth N. Adopting a public health approach to developmental neurotoxicity. Neurotoxicol Teratol.
21. Buck GM. Epidemiologic perspective of the developmental neurotoxicol in PCBs in humans. Neurotoxicol Teratol.
22. Guo YL, Yu M-LM, Ryan JJ. Different congeners of PCBs/PCDFs may have contributed to different health outcomes in the Yu-Cheng cohort. Neurotoxicol Teratol.
23. Schantz SL. Developmental neurotoxicity of PCBs in humans: what do we know and where do we go from here? Neurotoxicol Teratol.
24. Schantz SL. Response to commentaries. Neurotoxicol Teratol.
25. Borak J, Israel L. Does in utero
exposure to PCBs cause developmental toxicity? The occupational and environmental medicine report. J Occup Environ Med.
26. Grandjean P, Weihe P, White RF, et al. Cognitive deficit in 7-year-old children with prenatal exposure to methylmercury. Neurotoxicol Teratol.
27. Koopman-Esseboom C, Huisman M, Touwen BC, et al. Newborn infants diagnosed as neurologically abnormal with relation to PCB and dioxin exposure and their thyroid hormone status. Dev Med Child Neurol.
28. NCHS. Trends in Low Birth Weight in the United States 1975-1985.
Washington, DC: Department of Health and Human Services, CDC National Center for Health Statistics; October 1989. National Center for Health Statistics Series 21, No. 48.
Readers are invited to submit letters for publication in this department. Submit them to: The Editor, Journal of Occupational and Environmental Medicine, PO Box 370, Bryn Mawr, PA 19010. Letters should be typewritten and double spaced and should be designated “For Publication.”