Data Loading...
Formaldehyde and Leukemia: An Updated Meta-Analysis and ... Flipbook PDF
ORIGINAL ARTICLE Formaldehyde and Leukemia: An Updated Meta-Analysis and Evaluation of Bias Erika Schwilk, MD, MPH, Luop
103 Views
25 Downloads
FLIP PDF 232.2KB
ORIGINAL ARTICLE
Formaldehyde and Leukemia: An Updated Meta-Analysis and Evaluation of Bias Erika Schwilk, MD, MPH, Luoping Zhang, PhD, Martyn T. Smith, PhD, Allan H. Smith, MD, PhD, and Craig Steinmaus, MD, MPH Objective: Formaldehyde exposures are common, and data linking these exposures to leukemia have been mixed and controversial. The objective of this analysis is to review the current epidemiologic literature on formaldehyde and leukemia. Methods: We performed an updated meta-analysis focusing on high-exposure groups and myeloid leukemia and included two large recent studies: one involving ⬎25,000 workers in US formaldehyde industries and the other involving a cohort of ⬎13,000 funeral directors and embalmers. Results: Formaldehyde was associated with increased risks of leukemia (relative risk ⫽ 1.53; 95% confidence interval ⫽ 1.11 to 2.21; P ⫽ 0.005; 14 studies), specifically myeloid leukemia (relative risk ⫽ 2.47; 95% confidence interval ⫽ 1.42 to 4.27; P ⫽ 0.001; 4 studies). Conclusion: These findings provide evidence of increased myeloid leukemia risk with exposure to formaldehyde.
M
illions of people in the United States and worldwide are exposed to formaldehyde in the workplace,1 and environmental exposures may be even more common. In a recent study, 38% of Federal Emergency Management Agency supplied trailers used after Hurricane Katrina had formaldehyde levels ⬎0.1 ppm,2 a level over six times higher than the recommended exposure limit of 0.016 ppm.3 The International Agency for Research on Cancer has classified formaldehyde as a group 1 human carcinogen based on sufficient evidence in humans that formaldehyde causes both nasopharyngeal cancer and leukemia.4 Nevertheless, these conclusions have been controversial. For example, some authors have noted that the human epidemiologic evidence on nasopharyngeal cancer is heavily reliant on a cluster of cases from a single US factory.5,6 And, although a recent meta-analysis by our research group found evidence of an association between formaldehyde and leukemia,7 other meta-analyses have not reported clear associations.5,6,8 There may be several reasons for these discrepant results for leukemia. One is that, in some studies, all subtypes of leukemia are combined, despite the possibility that not all subtypes may be related to formaldehyde. The second reason is that, in some studies, all cohort members were combined into a single “exposed” group, despite the fact that some cohort members (eg, administrative or management staff) may have had very low or no formaldehyde From the Division of Occupational and Environmental Medicine (Dr Schwilk, Dr Steinmaus), University of California, San Francisco, Calif; Division of Environmental Health Sciences (Dr Zhang, Dr Smith, Dr Smith, Dr Steinmaus), School of Public Health, University of California, Berkeley, Calif; and Office of Environmental Health Hazard Assessment (Dr Steinmaus), California Environmental Protection Agency, Oakland, Calif. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.joem.org). Address correspondence to: Craig Steinmaus, MD, MPH, Office of Environmental Health Hazard Assessment, 1515 Clay Street, 16th Floor, Oakland, CA 94612; E-mail: [email protected]. Copyright © 2010 by American College of Occupational and Environmental Medicine DOI: 10.1097/JOM.0b013e3181ef7e31
878
exposure. The third issue is the healthy worker effect.9 This may occur in studies that compare workers with the general population, which contains many people who cannot work because of health conditions including cancer. If true associations exist, failure to account for each of these three factors could drive relative risk (RR) estimates toward 1.0 and could limit the ability of studies and meta-analyses to identify real effects. This article presents an updated meta-analysis on formaldehyde and leukemia, which includes two large recently published studies not used in our previous meta-analyses. The first new study is a case-control investigation nested in a large cohort of ⬎13,000 US funeral directors and embalmers, occupations with known high formaldehyde exposure.10 The second is the National Cancer Institute’s formaldehyde cohort study that involves ⬎40 years of follow-up and includes ⬎25,000 workers from 10 US formaldehydeproducing or formaldehyde-using plants.11 Because of their large sizes, these new studies add substantially to the precision with which formaldehyde-leukemia associations can be assessed. The major goals of this meta-analysis are to summarize the current epidemiologic literature and to explore the impacts of various factors that may affect the interpretation of these data. Several issues are evaluated including 1) the impact of combining workers with high and low exposure; 2) the effect of including subtypes of leukemia that may not be associated with formaldehyde; 3) the potential role of the healthy worker effect; and 4) the possible differences between industry workers and professional workers such as funeral directors and embalmers. This is the first meta-analysis to objectively explore all of these issues and the first to include both of the new large studies mentioned above. Other major tenets of causal inference are also assessed including biologic plausibility, the possibility that elevated RRs in professionals may be because of confounding, and an evaluation of possible doseresponse relationships.
METHODS Multiple sources including PubMed were searched for all epidemiologic studies on leukemia and formaldehyde exposure or formaldehyde-using industries and occupations. Searches included the keywords formaldehyde, leukemia, lymphohematopoietic, cancer, myeloid, and others. Bibliographies of all articles included in the meta-analysis, all relevant review articles, and previous metaanalyses were also searched. Only data published in peer reviewed scientific journals or edited books were used in our primary analyses. Government reports were excluded but evaluated in sensitivity analyses in which these studies are included.12,13 Other studies were excluded for the following reasons: 1) RRs or estimates of variance were not provided or could not be estimated14,15; 2) study subjects were the same as those used in an another included study16 –21; 3) there was no clear formaldehyde-exposed group22,23; 4) RRs were reported only as standardized proportionate cancer incidence ratios, which can be biased if formaldehyde or a correlated exposure (eg, asbestos and silica) increases the risks of other cancer types.24 Sensitivity analyses were done to evaluate the impact of these exclusions. JOEM
•
Volume 52, Number 9, September 2010
JOEM
•
Volume 52, Number 9, September 2010
Searches included both case-control and cohort studies, although only three case-control studies met our initial inclusion criteria.10,25,26 The case-control study by Hauptmann et al10 was included in the meta-analysis because it was nested in a large cohort of US embalmers. Another nested case-control study, Partanen et al,26 was excluded because exposure was assessed much differently in cases (personal interviews and company records) than in controls (only company records). This left only the population-based casecontrol study by Blair et al,25 which had only one myeloid leukemia case in the high-exposure group. This study was excluded from our main analyses because significant formaldehyde exposure is much less likely in population-based studies than in studies that take place in cohorts of highly exposed workers. Nevertheless, a sensitivity analysis was done to assess the impact of this exclusion. Several publications reported results on the same cohort. In these instances, only the most recent publication was included, unless 1) only the less recent publication reported data specifically on myeloid leukemia or 2) only the less recent study reported data for a highly exposed group (defined later). We focused on myeloid leukemia because previous evidence suggests that associations with formaldehyde may be strongest for this subtype.7 If data were given for myeloid leukemia, those data were used. Otherwise, RRs for all leukemia subtypes combined were used in the main analysis. If formaldehyde is associated with a particular subtype of leukemia (eg, myeloid), but not other subtypes, combining all subtypes together could bias summary RRs toward the null. Because of this, a separate analysis was done that only included data on myeloid leukemia. And, to evaluate the magnitude of the potential bias caused by including unrelated leukemia subtypes, the myeloid-only analysis was compared with a separate analysis that included only RRs in which all leukemia subtypes were combined. The major goal of this meta-analysis was to evaluate whether formaldehyde is associated with leukemia and not to define exact dose-response relationships or to evaluate risks in people with low exposures. The simple type of cause and effect relationship we sought to evaluate is best assessed in groups with high exposures. This is because, if a true relationship exists, RRs are likely to be higher in groups with high exposure than in groups with low exposure. All else being equal, higher RRs have greater statistical power than lower RRs (ie, those just ⬎1.0). In addition, higher RRs are less likely to be solely because of confounding or other bias.27–29 Because of these factors, when studies provided RR estimates for different levels of exposure (eg, low, medium, and high), the RR for the highest level was used in the main meta-analysis. Nevertheless, in addition, a separate analysis was also done in which we evaluated each study for evidence of dose-response relationships. Several studies did not specifically report data on highly exposed workers but only provided RRs for their entire cohort as a whole. Nevertheless, several of these cohorts included workers with high exposure (eg, ⬎0.75 ppm, the permissible exposure limit [PEL]) and workers with low or no exposure. If a true association exists, including workers with very low or no exposure in the exposed group can dilute RR estimates toward 1.0. To evaluate this issue, we performed a subgroup meta-analysis that only included studies reporting RRs in highly exposed groups. For this analysis, high exposure was defined as an average or peak exposure above the PEL (0.75 ppm) or short-term exposure limit (STEL; 2 ppm), respectively; an exposure duration of ⱖ10 years; or work in a formaldehyde-producing facility before 1961. For comparison, we also performed an analysis that only included studies in which the exposed group included all cohort members, regardless of whether their exposures were high, low, or none. In many of the studies included in this meta-analysis, different types of exposure metrics were used. For example, Hauptmann et al10 reported separate RRs for peak exposure, average intensity,
Formaldehyde and Leukemia: An Updated Meta-Analysis
cumulative exposure, and duration. Other studies only gave data for one metric. In observational epidemiology, it is uncommon for all, or even most, studies on a given topic to report data using the exact same exposure metric. As a consequence, meta-analyses of epidemiologic data frequently combine RRs based on different metrics and different exposure categorizations. Our meta-analysis is no different. When RRs for different exposure metrics were given, we selected one in the following order: peak exposure, average intensity, cumulative exposure, exposure duration, and earlier date of hire. Peak exposure was chosen first a priori because metrics such as average intensity and cumulative exposure may be less accurate measures of true exposure if workers with very high exposure also have long intervening periods with little or no exposure.30,31 Microsoft Excel 2008 for Mac version 12.2.3 and STATA version 8.0 (College Station, TX) were used for all calculations. Summary RR estimates were calculated using both the fixed effects inverse variance weighting method32 and the random effects method.33 Heterogeneity was evaluated using the general variance-based method.34 If heterogeneity is present, the random effects model incorporates between-study variation into the summary variance estimate and confidence intervals (CI)s. Some authors have suggested that the random effects model may be more conservative.34 Nevertheless, unlike the fixed effects model, where weights are directly proportional to study precision, the random effects model weights studies are based on a highly complex and nonintuitive mix of study precision, RR, and meta-analysis size (ie, the number of studies included).33 As a consequence, this model assigns greater weight to smaller studies than the fixed effects model and, therefore, may actually be less conservative.35 To avoid this problem, we used the method presented by Shore et al36 and used in several subsequent meta-analyses.37– 41 In this method, the summary RR estimate is calculated by directly weighing individual studies by their precision, whereas between-study heterogeneity is only incorporated into the summary RR’s variance (ie, the 95% CI). Occupational studies reporting standardized mortality ratios (SMRs) can be biased by the healthy worker effect.9 To objectively evaluate the impact of this, separate analyses were performed, in which SMRs were adjusted for this potential bias using the SMRs for all-cause mortality (or all-cancer mortality when all-cause data were not available) and the methods described by Miettinen and Wang42 and Smith et al.43 All-cancer mortality was not selected first because formaldehyde or some other possibly correlated exposure (eg, silica and asbestos) may increase the risks of some cancers and mask an otherwise true healthy worker effect. In Hauptmann et al,10 RRs were provided using two different comparison groups: those who conducted ⬍500 embalmings and those who conducted no embalmings.10 The former was chosen because the later included very few subjects and led to unstable RR estimates. This resulted in a more conservative estimate because the RR using the latter group was higher (RR ⫽ 2.9 vs 13.0). Funnel plots and Egger’s and Begg’s tests were used to evaluate publication bias.44,45 Missing CIs in cohort studies were calculated using Byar’s approximation.46 All P values are one-sided (unless otherwise reported) given our clear a priori hypothesis that formaldehyde increases, not decreases, leukemia risk.
RESULTS Table 1 shows the results and weights of the studies used in this meta-analysis. In total, 13 cohort studies and 1 nested casecontrol study were included. Six studies were from professional groups, and eight studies were from industry groups. No study received ⬎20% of the weight in the fixed effects model. Table 2 summarizes the results of the meta-analysis. The summary RR for all studies combined was 1.53 (95% CI ⫽ 1.11 to 2.11; P ⫽ 0.005; Fig. 1). Summary RRs in industry workers and
© 2010 American College of Occupational and Environmental Medicine
879
Schwilk et al
TABLE 1.
JOEM
•
Volume 52, Number 9, September 2010
Description of Studies in the Current Meta-Analysis
Author
RR
95% CI
%W
Type
N
Location
Outcome (ICD)
Exposure Category
Andjelkovich et al51 Beane Freeman et al11 Coggon et al49 Dell and Teta47 Hall et al69 Harrington and Shannon50 Harrington and Shannon50 Hauptmann et al10 Levine et al70 Pinkerton et al71 Stellman et al52 Stern et al72 Stroup et al48 Wong73
0.43 1.78 0.71 2.65 1.52 0.63
0.05–1.57 0.87–3.64 0.31–1.39 1.15–5.24 0.41–3.89 0.01–3.48
2.2 12.8 11.7 11.4 5.2 0.8
Coh/SMR Coh/RR Coh/SMR Coh/SMR Coh/SMR Coh/SMR
2 19 8 8 4 1
Leukemia 204–7 Myeloid 205 Leukemia 204–8 Leukemia Leukemia Leukemia 204–7
Exposed Peak 4⫹ ppm Average 2⫹ ppm R and D workers Total cohort Pathologists
0.45
0.01–2.53
0.9
Coh/SMR
1
Leukemia 204–7
2.90 1.60 2.19 0.96 1.70 8.80 1.35
0.90–9.50 0.43–4.10 0.94–4.32 0.54–1.71 0.63–3.73 1.80–25.50 0.15–4.87
4.7 5.2 11.3 19.7 8.3 3.7 2.2
NCC Coh/SMR Coh/SMR Coh/RR Coh/SMR Coh/SMR Coh/SMR
11 4 8 12 6 3 2
Iron foundry 10 US formaldehyde industries 6 United Kingdom chemical plants Plastics manufacturing United Kingdom pathologists (1974–1987) United Kingdom pathologists/laboratory technicians (1955–1973) United Kingdom pathologists/laboratory technicians (1955–1973) US embalmers and funeral directors Ontario undertakers US garment facilities ACS US population cohort Tannery workers US anatomists Formaldehyde chemical plant
Laboratory technicians Peak 9.3⫹ ppm Total cohort Duration 10⫹ yr Exposed Duration 10⫹ yr Total cohort Employed before 1961
Myeloid 205 Leukemia 204–7 Myeloid 205 Leukemia Leukemia 204 Myeloid* Leukemia 204–7
*Chronic myeloid leukemia only (data not given for acute myeloid leukemia) for year in which specific reference subtype data were available (1969 –1979). ACS, American Cancer Society Cancer Prevention Study II; Coh, cohort study; N, number of exposed cases; NCC, nested case-control study; R and D, research and development; %W, percent weight given in the fixed effects model.
TABLE 2.
Results of Meta-Analysis of Formaldehyde and Leukemia Fixed Effects
All studies Leukemia type Myeloid only All leukemias combined† High exposure vs all exposed High exposure All exposed‡ Healthy worker effect adjusted All studies Myeloid only Sensitivity analyses Exclude Dell and Teta47 Add excluded studies§ Add Blair et al25 Professional workers Industry workers Industry: high exposure
Random Effects
Heterogeneity
Studies
Cases*
RR
95% CI
Shore-Adjusted 95% CI
14
89
1.53
1.18–1.98
1.11–2.11
1.58
1.12–2.24
20.66
0.08
4 16
41 193
2.47 1.42
1.57–3.86 1.21–1.67
1.42–4.27 —
2.64 —
1.49–4.69 —
4.50 11.42
0.21 0.72
6 16
54 185
1.55 1.07
1.08–2.22 0.90–1.26
1.04–2.31 0.86–1.32
1.57 1.16
1.04–2.36 0.90–1.51
6.24 24.93
0.28 0.05
14 4
89 41
1.72 2.77
1.34–2.21 1.79–4.30
1.18–2.51 1.39–5.52
1.86 3.21
1.23–2.82 1.55–6.67
29.67 7.41
0.01 0.06
13 18 15 6 8 5
81 124 90 24 65 43
1.42 1.45 1.51 2.27 1.38 1.45
1.08–1.87 1.18–1.78 1.17–1.95 1.29–3.99 1.04–1.84 1.00–2.12
1.02–1.99 1.15–1.83 1.10–2.06 1.15–4.45 0.96–1.99 0.95–2.22
1.49 1.41 1.56 2.21 1.39 1.45
1.03–2.14 1.08–1.84 1.11–2.19 1.09–4.49 0.95–2.04 0.94–2.24
18.38 21.87 21.26 7.08 11.25 5.05
0.10 0.19 0.10 0.21 0.13 0.28
RR
95% CI
2
P
*Number of exposed cases. †Hauptmann et al10 was replaced with Hayes et al,19 Walrath and Fraumeni,20 and Walrath and Fraumeni21 in this analysis because Hauptmann et al did not provide data for all leukemias combined. ‡Includes RRs for all exposure categories combined for each study. Stern et al72 provided separate data for two tanneries and Dell and Teta47 provided separate data for salary vs hourly workers. §Adds the four studies in Bachand et al6 that were not included in our meta-analysis. 㛳Unadjusted data for AML and CML combined were used because adjusted data for both combined were not provided. 2, 2 test statistic.
professional workers were 1.38 (95% CI ⫽ 0.96 to 1.99; P ⫽ 0.04) and 2.27 (95% CI ⫽ 1.15 to 4.45; P ⫽ 0.009), respectively. The summary RR of the four studies on myeloid leukemia was 2.47 (95% CI ⫽ 1.42 to 4.27; P ⫽ 0.001). In contrast, in the analysis using only data in which all leukemia subtypes are combined, the 880
summary RR was lower (RR ⫽ 1.42; 95% CI ⫽ 1.21 to 1.67; P ⬍ 0.001). In analyses comparing myeloid and lymphatic subtypes using studies and exposure categories where data on both were provided, the summary RR was elevated for myeloid leukemia (RR ⫽ 1.55, P ⬍ 0.001) but not for lymphatic leukemia (RR ⫽
© 2010 American College of Occupational and Environmental Medicine
JOEM
•
Volume 52, Number 9, September 2010
Formaldehyde and Leukemia: An Updated Meta-Analysis
Andjelkovich et al., 1995 Coggon et al., 2003 Dell and Teta, 1995 Hall et al., 1991 Harrington & Shannon, 1975 Harrington & Shannon, 1975 Beane Freeman et al., 2009 Hauptmann et al., 2009 Levine et al., 1984 Pinkerton et al., 2004 Stellman et al., 1998 Stern et al., 1987 Stroup et al., 1986 Wong, 1983 Summary RR .1
1 odds ratio
FIGURE 1. Forrest plot of studies used in the meta-analysis of formaldehyde and leukemia.
10
4
logor
2
0
-2
-4 0
.5
1
1.5
s.e. of: logor
0.95, P ⫽ 0.42; data and methods provided in Supplemental Table 1, http://links.lww.com/JOM/A39). The RR for the six studies that provided data on a highexposure group was 1.55 (95% CI ⫽ 1.04 to 2.31; P ⫽ 0.016). In the analysis in which all cohort members are combined into one exposed group (regardless of whether their exposures are high, medium, or low), the summary RR is close to 1.0 (RR ⫽ 1.07; 95% CI ⫽ 0.86 to 1.32). Adjusting for the healthy worker effect increased the summary RR for all studies and for myeloid studies to 1.72 (95% CI ⫽ 1.18 to 2.51) and 2.77 (95% CI ⫽ 1.39 to 5.52), respectively. Sensitivity analyses were conducted to determine the impacts of excluding certain studies. Excluding the Dell and Teta47 data on research and development workers decreased the summary RR from 1.53 to 1.42 (P ⫽ 0.02). Excluding the study with the highest RR, Stroup et al,48 lowered the summary RR slightly, from 1.53 to 1.43 (95% CI ⫽ 1.08 to 1.89; data not shown). Neither the Stroup et al nor Dell and Teta studies were included in the high-exposure or myeloid analyses. Including all four studies used in the recent meta-analysis by Bachand et al6 but excluded here lowered the summary RR from 1.53 to 1.45 (P ⫽ 0.001).12,13,25,26
FIGURE 2. Funnel plot of the log odds ratio (logor) versus the standard error (SE) of the log odds ratio of the studies used in the meta-analysis.
The funnel plot showed no evidence of asymmetry consistent with publication bias (Fig. 2). Egger’s (Kendall’s score ⫽ ⫺7; P ⫽ 0.74) and Begg’s (bias coefficient ⫽ 0.04; P ⫽ 0.96) tests also showed no evidence of publication bias. Because dose-response relationships can be an important element of assessing causal inference, we evaluated whether these relationships were present in the individual studies included in this meta-analysis. Table 3 shows the RRs by exposure category for each study that presented dose-response data. In every study except for Coggon et al,49 there is some evidence that RRs increase as the potential exposure increases. Wong et al had few leukemia cases, although evidence of dose response was seen for all lymphatic and hematopoietic cancers combined.
DISCUSSION
The increased summary RRs of 1.53 (P ⫽ 0.005) and 2.47 (P ⫽ 0.001), respectively, provide evidence that formaldehyde is associated with leukemia, specifically myeloid leukemia. The low P values show that these findings are unlikely due to chance. Given the major differences across studies in design, populations, leuke-
© 2010 American College of Occupational and Environmental Medicine
881
Schwilk et al
JOEM
•
Volume 52, Number 9, September 2010
TABLE 3. Evaluation of Dose-Response Relationships in Studies of Leukemia and High Exposure to Formaldehyde Study
Results by Exposure Category
P Trend*
Dose Response†
49
Coggon et al, 2003 Average exposure (ppm) N SMR 95% CI Beane Freeman et al, 200911 Peak exposure (ppm) N RR 95% CI Hauptmann et al, 200910 Peak exposure (ppm) N RR 95% CI Pinkerton et al, 200471 Duration exposed (yr) N SMR 95% CI Stern et al, 198772 Duration employed (yr) N SMR 95% CI Wong, 1983‡73 Year of hire N SMR 95% CI
Total cohort 31 0.91 0.62–1.29
⬎2 ppm 8 0.71 0.31–1.39
0 4 0.82 0.25–2.67
0.1–1.9 14 1.00 Reference
Reference 5 1.0 Reference
No
2.0–3.9 11 1.30 0.58–2.92
ⱕ7.0 9 2.9 0.9–9.8
⬎7.0–9.3 9 2.0 0.6–6.6
⬍3 3 0.83 0.17–2.43
3–9 4 1.26 0.34–3.23
ⱖ10 8 2.19 0.94–4.32
⬍1 2 0.45 0.05–1.68
1–9 2 1.00 0.11–3.61
ⱖ10 6 1.70 0.63–3.73
After 1960 0 0 0–NA
Before 1961 2 1.35 0.15–4.87
ⱖ4 19 1.78 0.87–3.64 ⬎9.3 11 2.9 0.9–9.5
0.035
Yes
0.018
Yes
Yes
Yes
—‡
*One-sided P values. P trends were only provided in these two studies. †Classified as “Yes” if some evidence of a dose-response relationship was present. ‡In Wong,73 there were no leukemia cases after 1960, although the expected number was low. SMRs for all lymphatic and hematopoietic cancers by length of employment (⬍5, 5–9, 10 –14, 15–19, 20⫹ yr) were 0.77, 1.53, 1.64, 1.75, and 2.10, respectively (data by length of employment was not provided for leukemia alone). N, number of exposed cases.
mia classifications, time periods, and exposure scenarios, some heterogeneity in results are expected. Nevertheless, the relatively low heterogeneity statistics (and the heterogeneity P ⬎ 0.05) highlight the overall consistency in these data. In the analysis of all studies combined, 9 of the 14 studies (64%) reported RRs ⬎1.0. Three of the five studies with RRs ⬍1.0 involved two or fewer exposed cases each and received little weight (⬍4%).50,51 Interestingly, in one of the other remaining two studies, Stellman et al,52 although formaldehyde exposure outside the wood industry was not associated with leukemia, the combination of work in the wood industry and formaldehyde exposure was associated (RR ⫽ 5.79; 95% CI ⫽ 1.44 to 23.2).52 Five of the six studies (83%) reporting data on high-exposure groups had RRs of 1.7 or higher, and all four of the myeloid leukemia studies (100%) had RRs ⬎1.0. Most of the high-exposure studies showed evidence of a dose-response relationship (Table 3). Summary RRs were elevated in high-exposure studies (RR ⫽ 1.55, P ⫽ 0.02) but were ⬃1.0 when subjects with high and low exposure were combined into a single exposed group (RR ⫽ 1.07, P ⫽ 0.28). This difference shows the diluting effect that can occur when people with relatively low or no formaldehyde exposure 882
are included in exposed groups. The lower summary RR for lymphatic leukemia (RR ⫽ 0.95) when compared with myeloid leukemia (see Supplemental Table 1, http://links.lww.com/JOM/A39) highlights the additional diluting effect that can occur when related and unrelated leukemia subtypes are combined. This meta-analysis included two recently published large studies. Beane Freeman et al11 involved 25,619 workers in 10 US formaldehyde-producing or formaldehyde-using plants employed before 1966. Vital status (13,951 deaths) was assessed through 2004 using the National Death Index Plus and other sources, with a median follow-up of 42 years. Exposure for each job held by each worker was evaluated through 1980 using expert assessments involving job titles, tasks, and current and past formaldehyde measurements. For myeloid leukemia, a dose-response trend was seen with increasing peak exposure (Table 3). Similar trends were less clear or not present for average exposure and cumulative exposure, although long intervening periods of low or no exposure can the dilute effects of short-term high exposures when these metrics are used. Myeloid leukemia RRs decreased as the follow-up period increased, which could indicate a relatively short latency
© 2010 American College of Occupational and Environmental Medicine
JOEM
•
Volume 52, Number 9, September 2010
Formaldehyde and Leukemia: An Updated Meta-Analysis
period. The other new study, Hauptmann et al,10 was a case-control study nested in a large cohort of US funeral workers. Workers were identified through industry associations and licensing boards, and all deaths from lymphohematopoietic malignancies (contributing or underlying) from 1960 to 1986 were collected based on death certificate information from state vital statistics offices. Control subjects (n ⫽ 265) were randomly selected from funeral workers who died of other causes and matched by sex, dates of birth and death, and data source. Structured interviews with next of kin and coworkers were used to collect information on job history, number of embalmings, and workplace ventilation. Exposure estimates were obtained by linking this information to monitoring data from a previous exposure-assessment study. Despite the small number of myeloid leukemia cases (n ⫽ 34), statistically significant trends were seen with increasing job duration, average exposure (onesided P ⫽ 0.029), and peak exposure (Table 3). Trends were seen for cumulative exposure and number of embalmings, but these were not statistically significant. Increased risks were not seen for other cancer types, suggesting no diagnostic bias. According to the authors, exposure to benzene and radiation were likely to be low in this cohort, and adjustment for smoking had little effect on results. Although some exposure misclassification was likely in this study and in Beane Freeman et al, the misclassification was likely nondifferential with a resulting bias toward the null.53 Overall, both this study and Beane Freeman et al provided new evidence that formaldehyde is associated with leukemia. In this meta-analysis, as in almost all meta-analyses of observational epidemiologic data, studies using different exposure metrics (eg, peak exposure, average exposure, and duration) and different exposure categorizations were combined. Because expo-
TABLE 4.
sure was assessed independently of disease in all studies used in this meta-analysis, misclassification of exposure is likely to be nondifferential, and bias from this is likely to be toward the null and not toward a false association.53 Similarly, nondifferential misclassification of leukemia status or subtype (eg, misdiagnoses or use of different International Classification of Diseases codes) would also likely cause bias toward finding no effect. Importantly, the priority order used to choose the exposure metric selected from each study was determined a priori and was not based on selecting the highest RR from each study. Combining different exposure metrics can impact RR estimates, but the likely direction of the impact would be to reduce the ability to determine true associations. This is because, if formaldehyde does cause leukemia, some exposure metrics are likely to be more strongly associated with leukemia risks than others.30,31 In this meta-analysis, we may have included some metrics that are less strongly associated or unassociated with leukemia than others. Including less relevant metrics will dilute summary RR estimates toward 1.0. If every study in this meta-analysis had reported data on the same single metric that was most strongly associated with leukemia risk, it is possible that the true RRs in people with high formaldehyde exposure are even greater than the ones reported here. The recent meta-analysis by Bachand et al6 did not find evidence of an association between formaldehyde and leukemia (RR ⫽ 1.05 for all leukemia and 1.09 for myeloid leukemia). As seen in Table 4, the major difference with the meta-analysis reported here is our emphasis on high-exposure groups and myeloid leukemia. This resulted in our selecting higher RRs than Bachand et al for several studies. For example, for Beane Freeman et al,11 we used a RR of 1.78 for myeloid leukemia and peak exposures ⱖ4
Comparison of the Current Meta-analysis With the Meta-analysis of Bachand et al6 Current Meta-analysis*
Author
Cohort 51
Andjelkovich et al, 1995 Hall et al, 199169 Harrington and Shannon, 197550 Harrington and Shannon, 197550 Levine et al, 198470 Stellman et al, 198852 Beane Freeman et al, 200911 Coggon et al, 200349 Dell and Teta, 199547 Hauptmann et al, 200910 Pinkerton et al, 200471 Stern et al, 198772 Stroup et al, 198648 Wong, 198373 Blair et al, 2001§25 Partanen et al, 199326 Marsh et al, 200417 Matanoski, 199112 Robinson et al, 198713
Exposed group
Bachand et al
Leukemia
RR
N
Exposed group
Leukemia
RR
N
Iron foundry Pathologists Pathologists
Exposed Total cohort Total cohort
All All All
0.43 1.52 0.63
2 4 1
Same* Same* Same*
Same* Same* Same*
0.43 1.52 0.63
2 4 1
Laboratory technicians
Total cohort
All
0.45
1
Same*
Same*
0.45
1
Undertakers US population Formaldehyde industry Chemical plants Plastics workers Embalmers Garment workers Tannery workers Anatomists Chemical workers Iowa and Minnesota Wood industry Ten US industries Pathologists Plywood mill
Total cohort Exposed Peak 4⫹ ppm Average 2⫹ ppm R and D Peak 9.3⫹ ppm Duration 10⫹ yr Duration 10⫹ yr Total cohort Employed before 1961 High exposure Not used Not used Not used Not used
All All Myeloid All All Myeloid Myeloid All Myeloid‡ All Myeloid
1.60 0.96 1.78 0.71 2.65 2.90 2.19 1.70 8.80 1.35 0.68
4 12 19 8 8 11 8 6 3 2 1
Same* Same* All exposed Total cohort Not used Not used Total cohort Total cohort Same* Total cohort All exposed Same* All exposed Total cohort Total cohort
Same* Same* All Same*
1.60 0.96 1.02 0.91
4 12 116 31
All Same* All Same* All Same* All All All
1.09 0.75† 1.50 1.18 0.98 1.40 0.79 1.35 0.59
24 10 10 2 64 2 69 31 1
*Same data as used in the current meta-analysis. †Two relative risks were used: 0.77 from Tannery A and 0.75 from Tannery B. ‡Includes only chronic myeloid leukemia (CML). Data for acute myeloid leukemia (AML) were not provided. §Not used in the main analysis, but included in a sensitivity analysis. Unadjusted data for AML and CML combined were used because adjusted data for both combined were not provided. N, number of exposed cases; R and D, research and development; RR, relative risk.
© 2010 American College of Occupational and Environmental Medicine
883
Schwilk et al
JOEM
ppm, whereas Bachand et al used a RR of 1.02 for all leukemias combined and any formaldehyde exposure. Importantly, this cohort includes some people who were not highly exposed (eg, not above the current PEL or STEL). In fact, average exposure in this study ranged from 0.01 to 4.3 ppm, cumulative exposure ranged from 0.0 to 107.4 ppm-years, and 71% did not have peak exposures ⬎4 ppm. As discussed, combining people with high and low exposure into a single exposure group and including unrelated leukemia subtypes can both mask true associations. Other data support the biologic plausibility of the findings linking formaldehyde to leukemia, and these are reviewed in greater detail elsewhere.7 Briefly, several case-control studies and at least one cohort study have identified links between formaldehyde and nasopharyngeal cancer, providing evidence that formaldehyde is a human carcinogen.54 A variety of animal studies have also shown increases in cancers, although findings are not consistent across all studies.54 Cell culture experiments and in vivo studies in human cells and experimental animals have shown that formaldehyde is genotoxic and induces both DNA damage and chromosome changes. Some studies have shown that these effects occur in the lymphocytes of exposed people, although the results of these studies are variable.55– 60 Several mechanisms have been proposed for formaldehyde-caused cancer including direct damage to stem cells in the bone marrow, damage to hematopoietic stem/progenitor cells circulating in the peripheral blood, and damage to the primitive pluripotent stem cells in the nasal turbinates or olfactory mucosa.7 In a recent study, formaldehyde-exposed workers had
TABLE 5.
•
Volume 52, Number 9, September 2010
reduced white blood cell counts, including decreases in all major myeloid cell types, and increases in leukemia-specific chromosome changes in myeloid blood progenitor cells.61 In this meta-analysis, the summary RR was higher for professional workers than industry workers, although the findings for industry workers are not negative. That is, the summary RR in highly exposed industry workers is increased (RR ⫽ 1.45) and is unlikely due to chance (P ⫽ 0.04). In addition, as seen in Table 3, each of the industry studies with dose-response data and sufficient statistical power except for one49 shows evidence of a dose-response relationship. The lower summary RR in industry workers is mostly related to a single study, Coggon et al,49 the only industry study with data on a high-exposure group that reported a RR ⬍1.0. Without this study, the summary RR for high-exposure industry workers (RR ⫽ 1.85, 1.20 to 2.86, P ⫽ 0.003) was closer to that of professionals. The reason why the results of the study by Coggon et al differ is unknown. One factor could be this study’s use of average intensity as the exposure metric. It is possible that this metric may be less strongly associated with leukemia risk than peak exposure. This would be true if high-intermittent peak exposures overwhelm detoxification mechanisms to a greater extent than more constant, but lower, average exposures. Indices of average exposure may also underestimate truly relevant exposure in workers with high peak exposures but long intervening periods of little or no exposure. High peak exposures (eg, ⬎STEL of 2 ppm) seem to be more common in professionals than in industry workers.10,62,63 For example, in Hauptmann et al,10 estimated peak exposures exceeded 7 ppm in ⬃70% of the control embalmers and funeral directors. In
Relative Risks for Other Lymphohematopoietic Cancers in Professionals*
Study Harrington and Shannon, 197550 Walrath and Fraumeni, 198421 Walrath and Fraumeni, 198421 Stroup et al, 198648 Harrington and Oakes, 198416 Walrath and Fraumeni, 198320 Logue et al, 198674 Harrington and Oakes, 198416 Hauptmann et al, 200910 Walrath and Fraumeni, 198320 Stroup et al, 198648 Hayes et al, 199019 Hayes et al, 199019 Walrath and Fraumeni, 198320 Walrath and Fraumeni, 198421 Walrath and Fraumeni, 198320 Hall et al, 199169 Walrath and Fraumeni, 198320 Hayes et al, 199019 Walrath and Fraumeni, 198421 Hayes et al, 199019 Harrington and Shannon, 197550 Walrath and Fraumeni, 198320 Stroup et al, 198648 Hayes et al, 199019 Walrath and Fraumeni, 198421 Walrath and Fraumeni, 198421
Outcome Hodgkin lymphoma (laboratory technicians) Lymphatic leukemia Hodgkin lymphoma Hodgkin lymphoma Other LHPM (women) Monocytic leukemia Other LHPM Other LHPM (men) LHPM-Lymphoid Non-specific leukemia Lymphosarcoma† Hodgkin lymphoma Lymphatic leukemia Hodgkin lymphoma Lymphosarcoma† Lymphosarcoma† Hodgkin lymphoma Other lymphatic Non-Hodgkin lymphoma Other lymphatic Multiple myeloma Hodgkin lymphoma (pathologists) Lymphatic leukemia Other lymphatic Other leukemia Other leukemia Monocytic leukemia
RR
95% CI
N (E)
0 0 0 0 0 0.33 0.48 0.54 0.6 0.67 0.7 0.72 0.74 0.87 0.97 1.08 1.21 1.23 1.26 1.33 1.37 1.43 1.54 2.0 2.28 2.86 6.67
0–NA 0–NA 0–NA 0.0–2.0 0–18.7 0–1.85 NA 0.05–4.29 0.2–1.3 0.01–3.71 0.1–2.5 0.15–2.10 0.29–1.53 0.10–3.14 0.19–2.83 0.35–2.52 0.03–6.71 0.45–2.68 0.87–1.76 0.36–3.41 0.84–2.12 0.02–7.95 0.41–3.94 0.7–4.4 1.39–3.52 0.77–7.32 0.75–24.1
0 (1.6) 0 (2.2) 0 (2.5) 0 (1.9) 0 (0.16) 1 NA 1 15 1 2 3 7 2 3 5 1 6 34 4 20 1 4 6 20 4 2
E, expected number of cases; LHPM, lymphohematopoietic malignancy; N, observed number of cases; NA, not available. *Includes RRs for all reported types lymphohematopoietic cancers that do not include myeloid leukemia. Levine et al70 did not report on other subtypes. †Includes lymphosarcoma and reticulosarcoma (ICD-8 codes 200).
884
© 2010 American College of Occupational and Environmental Medicine
JOEM
•
Volume 52, Number 9, September 2010
contrast, peak exposures exceeding 4 ppm were seen in only 24% of all subjects in the large National Cancer Institute cohort study of formaldehyde industries.18 Given these data, the higher summary RR seen in professionals compared with industry workers is consistent with the hypothesis that peak exposure is an important metric in determining leukemia risk. Another reason why the results of the study by Coggon et al are different from other studies may be because that they combined all leukemia subtypes, including leukemias of unspecified cell type (International Classification of Diseases-9 code 208). These unspecified leukemias can account for 25% of all leukemias and were not included in other studies used in our meta-analysis.64 As discussed, the inclusion of unrelated leukemia subtypes can dilute summary RRs toward 1.0. There are other possible reasons why summary RRs are lower in industry workers, but these seem unlikely. One possibility is confounding by some other agent (eg, benzene, radiation, or viruses) in professionals. Nevertheless, funeral directors, embalmers, and anatomists are not known to be exposed to benzene or radiation levels associated with ⬎2-fold increased risks of myeloid leukemia.65,66 In addition, a recent International Agency for Research on Cancer working group concluded that there was little evidence that professionals have more viral infections or that viruses cause myeloid leukemia.54 Other factors including pesticides, chemotherapy agents, electromagnetic fields, or inherited disorders have also been linked to leukemia67,68 but are unlikely to be prevalent enough or strongly enough related to formaldehyde exposure to cause substantial confounding. Another reason why summary RRs are higher in professionals than in industry workers may be a diagnosis bias; that is, myeloid leukemia could be more likely to be diagnosed in professionals who may have higher quality medical care than industry workers.6 If this were true, it would be expected that RRs for other lymphopoietic cancers would also be increased in professionals. Table 5 shows RRs for all lymphopoietic cancer types not including myeloid leukemia in all the studies of professionals reviewed for this meta-analysis. Most of these RRs are ⬍1.0 (15 of 27) and only three (11%) are ⬎2.0. This suggests that diagnostic bias is unlikely to have caused the summary RR of 2.27 identified for formaldehyde-exposed professionals in this meta-analysis. In this meta-analysis, excess summary RRs increased by ⬃25% after adjusting for the healthy worker effect. Although the impact of this bias has been well described for all-cause mortality, its role in cancer mortality is less well documented. To evaluate this, all-cause and all-cancer SMRs were collected from all studies on formaldehyde and lymphohematopoietic cancer reviewed for this meta-analysis (see Supplemental Table 2, http://links.lww.com/JOM/A40). Twenty-nine of the 34 all-cause SMRs (85.3%; average SMR ⫽ 0.86) and 26 of 36 all-cancer SMRs (72.2%; average SMR ⫽ 0.89) were ⬍1.0. Allcancer SMRs would probably be even lower if occupationally caused cancers, including some lung cancers and some leukemias, were excluded. Overall, the general trend for all-cancer SMRs being ⬍1.0 provides some evidence that the healthy worker effect is affecting cancer rates in many of these studies.
CONCLUSIONS In summary, the findings of this meta-analysis suggest that formaldehyde exposure is associated with increased risks of leukemia, particularly myeloid leukemia. These findings also highlight the importance of focusing on high-exposure groups and myeloid leukemia when evaluating the human carcinogenicity of formaldehyde. Although confounding, publication bias, diagnostic bias, or substantial exposure or outcome misclassification cannot be completely ruled out, our evaluations suggest that these biases are unlikely causes of the associations identified. Despite the overall
Formaldehyde and Leukemia: An Updated Meta-Analysis
evidence of an association, some inconsistencies remain, including the RR ⬍1.0 reported in the large Coggon et al49 study. Future research is needed to explain these inconsistencies and to elucidate the possible mechanisms of formaldehyde carcinogenicity, the role of short-term peak versus long-term average exposures, and the possible impacts of low environmental exposures. ACKNOWLEDGMENT We thank the support of the Center for Occupational and Environmental Health (COEH), University of California, Berkeley, CA.
REFERENCES 1. OSHA. Occupational Safety and Health Administration: Occupational Exposure to Formaldehyde Fact Sheet: US Department of Labor, 1995. Available at: http://www.medgasexperts.com/docs/01-01-1995%20-%20Occupational %20Exposure%20to%20Formaldehyde.htm. Accessed January 4, 2010. 2. Centers for Disease Control and Prevention (CDC). Final Report on Formaldehyde Levels in FEMA-Supplied Travel Trailers, Park Models, and Mobile Homes, July 2, 2008. Available at: www.cdc.gov/nceh/ehhe/trailer study/pdfs/FEMAFinalReport.pdf. Accessed February 15, 2010. 3. NIOSH. Pocket Guide to Chemical Hazards. Formaldehyde. National Institute for Occupational Safety and Health, 2005. Available at: http://cdc.gov/ NIOSH/npg/npgd0293.html. Accessed February 15, 2010. 4. Baan R, Grosse Y, Straif K, et al; WHO International Agency for Research on Cancer Monograph Working Group. A review of human carcinogens— part F: chemical agents and related occupations. Lancet Oncol. 2009;10: 1143–1144. 5. Bosetti C, McLaughlin JK, Tarone RE, Pira E, La Vecchia C. Formaldehyde and cancer risk: a quantitative review of cohort studies through 2006. Ann Oncol. 2008;19:29 – 43. 6. Bachand A, Mundt K, Mundt D, Montgomery R. Epidemiological studies of formaldehyde exposure and risk of leukemia and nasopharyngeal cancer: a meta-analysis. Crit Rev Toxicol. 2010;40:85–100. 7. Zhang L, Steinmaus C, Eastmond DA, Xin XK, Smith MT. Formaldehyde exposure and leukemia: a new meta-analysis and potential mechanisms. Mutat Res. 2009;681:150 –168. 8. Collins JJ, Lineker GA. A review and meta-analysis of formaldehyde exposure and leukemia. Regul Toxicol Pharmacol. 2004;40:81–91. 9. Rothman K. Epidemiology. An Introduction. New York, NY: Oxford University Press; 2002:97–98. 10. Hauptmann M, Stewart PA, Lubin JH, et al. Mortality from lymphohematopoietic malignancies and brain cancer among embalmers exposed to formaldehyde. J Natl Cancer Inst. 2009;101:1696 –1708. 11. Beane Freeman LE, Blair A, Lubin JH, et al. Mortality from lymphohematopoietic malignancies among workers in formaldehyde industries: the National Cancer Institute Cohort. J Natl Cancer Inst. 2009;101:751–761. 12. Matanoski G. Risk of Pathologists Exposed to Formaldehyde. Springfield, VA: National Technical Information Service; 1991. 13. Robinson C, Fowler D, Brown D, Lemen R. Plywood Mill Workers’ Mortality Patterns 1945: National Toxicology Information Service. Cincinnati, OH: National Institute of Occupational Health and Safety; 1977. 14. Edling C, Jarvholm B, Andersson L, Axelson O. Mortality and cancer incidence among workers in an abrasive manufacturing industry. Br J Ind Med. 1987;44:57–59. 15. Ott MG, Teta MJ, Greenberg HL. Lymphatic and hematopoietic tissue cancer in a chemical manufacturing environment. Am J Ind Med. 1989;16: 631– 643. 16. Harrington JM, Oakes D. Mortality study of British pathologists 1974 – 80. Br J Ind Med. 1984;41:188 –191. 17. Marsh GM, Youk AO. Reevaluation of mortality risks from leukemia in the formaldehyde cohort study of the National Cancer Institute. Regul Toxicol Pharmacol. 2004;40:113–124. 18. Hauptmann M, Lubin JH, Stewart PA, Hayes RB, Blair A. Mortality from lymphohematopoietic malignancies among workers in formaldehyde industries. J Natl Cancer Inst. 2003;95:1615–1623. 19. Hayes RB, Blair A, Stewart PA, Herrick RF, Mahar H. Mortality of U.S. embalmers and funeral directors. Am J Ind Med. 1990;18:641– 652. 20. Walrath J, Fraumeni JF Jr. Mortality patterns among embalmers. Int J Cancer. 1983;31:407– 411. 21. Walrath J, Fraumeni JF Jr. Cancer and other causes of death among embalmers. Cancer Res. 1984;44:4638 – 4641.
© 2010 American College of Occupational and Environmental Medicine
885
Schwilk et al
JOEM
22. Linos A, Kyle RA, O’Fallon WM, Kurland LT. A case-control study of occupational exposures and leukaemia. Int J Epidemiol. 1980;9:131–135. 23. Band PR, Le ND, Fang R, et al. Cohort mortality study of pulp and paper mill workers in British Columbia, Canada. Am J Epidemiol. 1997;146:186–194. 24. Hansen J, Olsen JH. Formaldehyde and cancer morbidity among male employees in Denmark. Cancer Causes Control. 1995;6:354 –360. 25. Blair A, Zheng T, Linos A, Stewart PA, Zhang YW, Cantor KP. Occupation and leukemia: a population-based case-control study in Iowa and Minnesota. Am J Ind Med. 2001;40:3–14. 26. Partanen T, Kauppinen T, Luukkonen R, Hakulinen T, Pukkala E. Malignant lymphomas and leukemias, and exposures in the wood industry: an industrybased case-referent study. Int Arch Occup Environ Health. 1993;64:593–596. 27. Greenland S. Causation and causal inference. In: Rothman K, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: Lippincott Raven; 1998:7–28. 28. Schlesselman J. Assessing effects of confounding variables. Am J Epidemiol. 1978;108:3– 8. 29. Axelson O. Aspects on confounding in occupational health. Scand J Work Environ Health. 1978;4:85– 89. 30. Lubin JH, Caporaso N, Wichmann HE, Schaffrath-Rosario A, Alavanja MC. Cigarette smoking and lung cancer: modeling effect modification of total exposure and intensity. Epidemiology. 2007;18:639 – 648. 31. Lubin JH, Moore LE, Fraumeni JF Jr, Cantor KP. Respiratory cancer and inhaled inorganic arsenic in copper smelters workers: a linear relationship with cumulative exposure that increases with concentration. Environ Health Perspect. 2008;116:1661–1665. 32. Greenland S. Meta-analysis. In: Rothman K, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: Lippincott Raven; 1998:643– 673. 33. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177–188. 34. Petitti D. Statistical Methods in Meta-analysis. In: Meta-Analysis, Decision Analysis, and Cost-Effectiveness Analysis: Methods for Quantitative Synthesis in Medicine. New York, NY: Oxford University Press, 1994:94 –118. 35. Poole C, Greenland S. Random-effects meta-analyses are not always conservative. Am J Epidemiol. 1999;150:469 – 475. 36. Shore RE, Gardner MJ, Pannett B. Ethylene oxide: an assessment of the epidemiological evidence on carcinogenicity. Br J Ind Med. 1993;50:971–997. 37. Bhatia R, Lopipero P, Smith AH. Diesel exhaust exposure and lung cancer. Epidemiology. 1998;9:84 –91. 38. Zhuo H, Smith AH, Steinmaus C. Selenium and lung cancer: a quantitative analysis of heterogeneity in the current epidemiological literature. Cancer Epidemiol Biomarkers Prev. 2004;13:771–778. 39. Steinmaus C, Smith AH, Jones RM, Smith MT. Meta-analysis of benzene exposure and non-Hodgkin lymphoma: biases could mask an important association. Occup Environ Med. 2008;65:371–378. 40. Steinmaus CM, Nunez S, Smith AH. Diet and bladder cancer: a metaanalysis of six dietary variables. Am J Epidemiol. 2000;151:693–702. 41. Setia MS, Steinmaus C, Ho CS, Rutherford GW. The role of BCG in prevention of leprosy: a meta-analysis. Lancet Infect Dis. 2006;6:162–170. 42. Miettinen OS, Wang JD. An alternative to the proportionate mortality ratio. Am J Epidemiol. 1981;114:144 –148. 43. Smith MT, Jones RM, Smith AH. Benzene exposure and risk of nonHodgkin lymphoma. Cancer Epidemiol Biomarkers Prev. 2007;16:385–391. 44. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics. 1994;50:1088 –1101. 45. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315:629 – 634. 46. Breslow N, Day N. Statistical Methods in Cancer Research Vol II: The Design and Analysis of Cohort Studies. Lyon, France: International Agency for Research on Cancer; 1987. 47. Dell L, Teta MJ. Mortality among workers at a plastics manufacturing and research and development facility: 1946–1988. Am J Ind Med. 1995;28:373–384. 48. Stroup NE, Blair A, Erikson GE. Brain cancer and other causes of death in anatomists. J Natl Cancer Inst. 1986;77:1217–1224. 49. Coggon D, Harris EC, Poole J, Palmer KT. Extended follow-up of a cohort of British chemical workers exposed to formaldehyde. J Natl Cancer Inst. 2003;95:1608 –1615. 50. Harrington JM, Shannon HS. Mortality study of pathologists and medical laboratory technicians. BMJ. 1975;4:329 –332. 51. Andjelkovich DA, Janszen DB, Brown MH, Richardson RB, Miller FJ. Mortality of iron foundry workers: IV. Analysis of a subcohort exposed to formaldehyde. J Occup Environ Med. 1995;37:826 – 837.
886
•
Volume 52, Number 9, September 2010
52. Stellman SD, Demers PA, Colin D, Boffetta P. Cancer mortality and wood dust exposure among participants in the American Cancer Society Cancer Prevention Study-II (CPS-II). Am J Ind Med. 1998;34:229 –237. 53. Greenland S. Basic methods of sensitivity analysis and external adjustment. In: Rothman K, Greenland S, eds. Modern Epidemiology. 2nd ed. Philadelphia, PA: Lippincott Raven; 1998:343–357. 54. Cogliano VJ, Grosse Y, Baan RA, Straif K, Secretan MB, El Ghissassi F. Meeting report: summary of IARC monographs on formaldehyde, 2-butoxyethanol, and 1-tert-butoxy-2-propanol. Environ Health Perspect. 2005; 113:1205–1208. 55. ATSDR. Toxicological Profile for Formaldehyde. Agency for Toxic Substances and Disease Registry 1999. Available at: http://www.atsdr.cdc.gov/ toxprofiles/tp111.pdf. Accessed April 23, 2010. 56. IARC. Formaldehyde, 2-butoxyethanol, and 1-tert-butoxy-2propanol. International Agency for Research on Cancer, 2006. Available at: http://www. who.int/bookorders/anglais/detart1.jsp?sesslan⫽1&codlan⫽1&codcol⫽ 72&codcch⫽88. Accessed February 15, 2010. 57. Ye X, Yan W, Xie H, Zhao M, Ying C. Cytogenetic analysis of nasal mucosa cells and lymphocytes from high-level long-term formaldehyde exposed workers and low-level short-term exposed waiters. Mutat Res. 2005;588:22–27. 58. Iarmarcovai G, Bonassi S, Sari-Minodier I, Baciuchka-Palmaro M, Botta A, Orsiere T. Exposure to genotoxic agents, host factors, and lifestyle influence the number of centromeric signals in micronuclei: a pooled re-analysis. Mutat Res. 2007;615:18 –27. 59. Orsiere T, Sari-Minodier I, Iarmarcovai G, Botta A. Genotoxic risk assessment of pathology and anatomy laboratory workers exposed to formaldehyde by use of personal air sampling and analysis of DNA damage in peripheral lymphocytes. Mutat Res. 2006;605:30 – 41. 60. Yu LQ, Jiang SF, Leng SG, He FS, Zheng YX. [Early genetic effects on workers occupationally exposed to formaldehyde]. Zhonghua Yu Fang Yi Xue Za Zhi. 2005;39:392–395. 61. Zhang L, Tang X, Rothman N, et al. Occupational exposure to formaldehyde, hematotoxicity, and leukemia-specific chromosome changes in cultured myeloid progenitor cells. Cancer Epidemiol Biomarkers Prev. 2010;19:80–88. 62. Kerfoot EJ, Mooney TF. Formaldehyde and paraformaldehyde study in funeral homes. Am Ind Hyg Assoc J. 1975;36:533–537. 63. NIOSH. Formaldehyde. Evidence of Carcinogenicity. Current Intelligence Bulletin 34. National Institute of Occupational Safety and Health, 1981. Available at: http://www.cdc.gov/niosh/81111_34.html#Appendix%20II. Accessed January 12, 2010. 64. CDC. CDC Wonder. Compressed Mortality, 1979 –1998 Results Centers for Disease Control and Prevention, 2010. Available at: http://wonder.cdc.gov/ controller/datarequest/D16. Accessed March 16, 2010. 65. EPA. Integrated Risk Information System. Benzene. US Environmental Protection Agency, 2003. Available at: http://www.epa.gov/iris/subst/0276.htm. Accessed February 25, 2010. 66. Romanenko AY, Finch SC, Hatch M, et al. The Ukrainian-American study of leukemia and related disorders among Chornobyl cleanup workers from Ukraine: III. Radiation risks. Radiat Res. 2008;170:711–720. 67. Kheifets L, Monroe J, Vergara X, Mezei G, Afifi AA. Occupational electromagnetic fields and leukemia and brain cancer: an update to two metaanalyses. J Occup Environ Med. 2008;50:677– 688. 68. Van Maele-Fabry G, Duhayon S, Mertens C, Lison D. Risk of leukaemia among pesticide manufacturing workers: a review and meta-analysis of cohort studies. Environ Res. 2008;106:121–137. 69. Hall A, Harrington JM, Aw TC. Mortality study of British pathologists. Am J Ind Med. 1991;20:83– 89. 70. Levine RJ, Andjelkovich DA, Shaw LK. The mortality of Ontario undertakers and a review of formaldehyde-related mortality studies. J Occup Med. 1984;26:740 –746. 71. Pinkerton LE, Hein MJ, Stayner LT. Mortality among a cohort of garment workers exposed to formaldehyde: an update. Occup Environ Med. 2004; 61:193–200. 72. Stern FB, Beaumont JJ, Halperin WE, Murthy LI, Hills BW, Fajen JM. Mortality of chrome leather tannery workers and chemical exposures in tanneries. Scand J Work Environ Health. 1987;13:108 –117. 73. Wong O. An epidemiologic mortality study of a cohort of chemical workers potentially exposed to formaldehyde, with a discussion on SMR and PMR. In: Gibson J, ed. Formaldehyde Toxicity. New York, NY: Hemisphere Publishing; 1983:256 –272. 74. Logue JN, Barrick MK, Jessup GL Jr. Mortality of radiologists and pathologists in the Radiation Registry of Physicians. J Occup Med. 1986;28:91–99.
© 2010 American College of Occupational and Environmental Medicine