Pooled Analysis and Meta-Analysis of GSTM1 and Bladder Cancer
by Lawrence S. Engel¹, Emanuela Taioli², Ruth Pfeiffer¹, Montserrat Garcia-Closas¹, Pamela M. Marcus³, Qing Lan¹, Paolo Boffetta⁴, Paolo Vineis⁵, Herman Autrup⁶, Douglas A. Bell⁷, Robert A. Branch⁸, Jürgen Brockmöller⁹, Ann K. Daly¹⁰, Susan R. Heckbert¹¹, Ivan Kalina¹², Daehee Kang¹³, Takahiko Katoh¹⁴, Amalia Lafuente¹⁵, Henry J. Lin¹⁶, Marjorie Romkes⁸, Jack A. Taylor⁶, Nathaniel Rothman¹

¹Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, USA
²IRCCS University of Milan, Milan, Italy
³Division of Cancer Prevention, National Cancer Institute, Bethesda, MD, USA
⁴International Agency for Research on Cancer, Lyon, France
⁵University of Torino, Turin, Italy
⁶University of Aarhus, Aarhus C., Denmark
⁷National Institute of Environmental Health Sciences, Research Triangle Park, NC, USA
⁸University of Pittsburgh, Pittsburgh, PA, USA
⁹Department of Clinical Pharmacology, University Goettingen, Germany
¹⁰University of Newcastle upon Tyne, Newcastle upon Tyne, England
¹¹University of Washington, Seattle, WA, USA
¹²Safárik University, Kosice, Slovak Republic
¹³Seoul National University, Seoul, South Korea
¹⁴Miyazaki Medical College, Miyazaki, Japan
¹⁵Universitat de Barcelona, Barcelona, Spain
¹⁶Harbor-UCLA Medical Center, Torrance, CA, USA

Correspondence and Reprint requests to: Lawrence Engel, Occupational Epidemiology Branch, National Cancer Institute, 6120 Executive Blvd., Room 8113, Bethesda, MD 20892-7240; Phone number: (301) 402-7825; Fax number: (301) 402-1819; E-mail: engell@mail.nih.gov.

Smoking is a known risk factor for bladder cancer. The product of the GSTM1 gene, glutathione S-transferase M1, is involved in the detoxification of polycyclic aromatic hydrocarbons found in tobacco smoke; a homozygous deletion of this gene in approximately 50% of Caucasians and Asians results in a lack of GSTM1 enzyme activity. Most studies examining the relationship between bladder cancer and GSTM1 have reported increased risk associated with a lack of GSTM1 activity. We performed meta- and pooled analyses of published and unpublished case-control genotype-based studies that examined this association (17 studies, 2,149 cases, 3,646 controls). We excluded studies conducted in populations with a high prevalence of exposure to known bladder cancer risk factors other than tobacco smoke. Using random effects models in the meta-analysis, we obtained a summary odds ratio (OR) of 1.44 (95% confidence interval (CI): 1.23, 1.68) for GSTM1 null status with all studies included. Results from studies with at least 100 cases and 100 controls produced a summary OR of 1.42 (95% CI: 1.26, 1.60). Pooled analyses using original datasets from ten studies (1,496 cases and 1,444 controls) and adjusting for age, sex, and race produced similar results. There was no evidence of multiplicative interaction between GSTM1 null genotype and ever smoking in relation to bladder cancer, although there was a suggestion of additive interaction (additive interaction=0.45; 95 percent CI: -0.03, 0.93). These results indicate that, among populations studied to date, GSTM1 null status is associated with a modest increase in risk of bladder cancer.

The glutathione S-transferases (GSTs) are a family of enzymes that are important in the metabolism of a wide variety of xenobiotics, including environmental carcinogens, reactive oxygen species, and chemotherapeutic agents (1). They act as phase II metabolizing enzymes, catalyzing reactions between glutathione and various electrophilic compounds. Five GST enzyme classes (a, g, q, m, and p) have been identified in humans. Because of their detoxification role, these enzymes and the genes encoding them may play an important role in cancer susceptibility. Although the vast majority of reactions catalyzed by the GSTs result in detoxification products, there are a few cases in which the reaction is reversible or in which the product, or a metabolite of the product, is more reactive than the parent compound (1).

The GSTM1 gene codes for the cytosolic enzyme GST-m. This enzyme has received considerable attention in relation to bladder cancer and other smoking-related cancers because of its role in the detoxification of benzo[a]pyrene and other polycyclic aromatic hydrocarbons (PAHs) found in tobacco smoke. Three polymorphisms of the GSTM1 gene, which is located on chromosome 1p13.3, have been identified. One polymorphism is a deletion that results in a lack of functional gene product (GSTM1-0). The other two (GSTM1a and GSTM1b) differ by a CàG substitution at base position 534, resulting in LysàAsn substitution at amino acid 172 (2). Because there is no evidence of functional difference between GSTM1a and GSTM1b, the two are typically categorized together as a single functional phenotype.

Individuals with homozygous deletion of GSTM1 (GSTM1 null) show no GST-m activity. Several studies have shown high agreement (i.e., > 94 percent) between GSTM1 genotype and GST-m phenotype (3-7). Most studies of GSTM1 and cancer have compared the homozygous deletion genotype with the genotypes containing at least one functional allele.

The frequency of the GSTM1 null genotype varies across ethnic groups. This frequency distribution has been extensively reviewed in previous HuGE reviews by Cotton et al (8) and Geisler and Olshan (9) and will be described only briefly here. Most of the studies from which these estimates were obtained were case-control studies. Variation in study results may reflect differences in the source of controls (e.g., hospital-based vs. population-based), sample size, or criteria for defining ethnicity.

In the United States, the frequency of the GSTM1 null genotype varies between ethnic groups (8). The mean reported frequency is 29 percent (95 percent confidence interval [CI]: 26, 32) among persons of African descent, 50 percent (95 percent CI: 45, 55) among those of Asian descent, 46 percent (95 percent CI: 41, 52) for individuals of Hispanic descent, and 51 percent (95 percent CI: 49, 52) for those of European descent.

The mean frequencies of the GSTM1 null genotype among the ethnic groups listed above are very similar to the frequencies observed in those groups' places of origin (2, 8, 9). The highest frequencies - 64-100 percent - have been observed among small samples of Pacific Islanders; these estimates were obtained using Southern blot while most others were obtained via polymerase chain reaction (PCR). The estimates presented here are very similar to estimates based on control distributions in a database of results from studies of metabolic gene polymorphisms maintained by the International Collaborative Study on Genetic Susceptibility to Environmental Carcinogens (GSEC), which also includes unpublished data (10).

Incidence and mortality rates of bladder cancer vary about 10-fold worldwide (11-13). The highest rates are found in North America and Europe, although the rates in eastern Europe tend to be lower than elsewhere in Europe. Rates are low in many parts of Asia. Differences in pathologic classification of bladder tumors, especially the registration of "benign" tumors or "papillomas" as cancer in some countries, may explain some of the geographic variation in rates; however, evidence suggest that this plays a relatively small role in the observed patterns (14-17).

Internationally, transitional cell carcinomas account for about 95 percent of bladder neoplasms (14, 15).  The remaining 5 percent consist of squamous cell carcinomas and adenocarcinomas, with squamous cell carcinomas being the more common. Most registries and national statistics do not distinguish between different bladder tumor types, making comparisons of rates difficult. However, limited evidence suggests that the proportion of bladder cancers that are squamous cell is higher in countries with endemic schistosomiasis (i.e., infection by Schistosoma hematobium) (18-20). Bladder cancer also tends to occur at a younger age in these countries, with many cases being diagnosed in their 40s or 50s (14). Information is lacking on incidence and mortality of bladder adenocarcinoma.

In the United States, bladder cancer incidence varies markedly by race (21). African Americans and Hispanic whites have incidence rates approximately half those of non-Hispanic whites. Rates are even lower among Asian Americans; among persons of Filipino descent, they are about one-quarter that of non-Hispanic whites.

Males consistently show a higher incidence than females throughout the world, with male to female ratios varying from approximately 2.5 to 5 (12, 15, 17). In the United States, the ratio of males to females is about 2.6 among African Americans, 3.3-3.7 among Hispanic whites and most Asian groups, and 4.0 among non-Hispanic whites and Filipinos (22). This gender disparity may be partially due to historical differences between males and females in occupational exposures and cigarette smoking behaviors.

The incidence and mortality of bladder cancer rises exponentially with age, with very few cases under the age of 30 and about two-thirds of cases occurring over the age of 65 (12). Both gender differences and histologically high grade tumors are strongly associated with increasing age (15).

Age-adjusted incidence rates have been rising in many parts of Europe and North America over the last few decades (12, 14).   In contrast, rates have remained relatively constant in Asia and Australia. Increases have generally been greater among males than among females. These trends may reflect changing geographic and gender patterns of smoking.

Cigarette smoking is a well-documented risk factor for bladder cancer (12, 14, 23). In fact, while certain occupational exposures such as aromatic amines present greater relative risks for bladder cancer, cigarette smoking has a greater population attributable risk because of its higher prevalence (23). An estimated 66 percent of bladder cancers in Western countries is attributable to tobacco smoking (23). This increased risk is likely due to the presence in cigarette smoke of polycyclic aromatic hydrocarbons, nitrosamines, and aromatic amines. Smokers have 2.5 to 3.5 times the risk of bladder cancer compared to nonsmokers. Dose-response relations have been observed for both intensity (i.e., number of cigarettes smoked per day) and duration of smoking, although the relation for intensity appears to level off at higher exposures (24). Quitting smoking reduces bladder cancer risk, although ex-smokers continue to be at increased risk relative to never smokers (25, 26). Smokers of black tobacco have two to three times the risk of bladder cancer than do smokers of blond tobacco (27-33). This is probably due to the higher levels of aromatic amines found in black tobacco. The evidence is inconsistent for increased risk of bladder cancer related to other forms of tobacco use, although there is suggestion of an association with pipe smoking (12, 14, 26).

Bladder cancer is associated with a number of occupations or occupational exposures. The first such association was observed in 1895 by Rehn (34), who reported high rates of bladder cancer among men employed in the aniline dye industry. Subsequent research among dyestuffs workers identified the aromatic amines benzidine and 2-naphthylamine, and possibly 1-naphthylamine, as bladder carcinogens (35). Since then, these and several other aromatic amines and related compounds have been identified as either likely or suspected human bladder carcinogens (36-46). These include 4-aminobiphenyl, 4-chloro-o-toluidine, o-toluidine, 4,4'-methylenedianiline, and 4,4'-methylene-bis(2-chloroaniline). Excess risk of bladder cancer has also been observed among rubber workers (with possible exposure to 2-naphthylamine or a precursor) (47-49); painters (potentially exposed to dyes, PCBs, formaldehyde, asbestos, and solvents) (50-59); truck, bus, and taxi drivers (probably exposed to polycyclic aromatic hydrocarbons (PAHs) and nitro-PAHs in automobile exhaust) (30, 55, 60-70); aluminum workers (possibly from exposure to coal-tar pitch volatiles) (71-76); and leather workers (with possible exposure to benzidine-based azo dyes, aromatic solvents, formaldehyde, chromium/chromate, and leather dust) (53, 61, 77-81).

Other known or suspected risk factors for bladder cancer include chlorination by-products in drinking water (82-84) and arsenic (often through drinking water) (85, 86), ionizing radiation (87, 88), bladder infection (89-91), high consumption of phenacetin-containing analgesics (92, 93), and hair dyes (12). Most studies suggest a causal relationship between Schistosoma hematobium and bladder cancer (94-96).

Several genetic susceptibility factors have been examined in relation to bladder cancer. Polymorphisms in the N-acetyltransferase 2 (NAT2) gene, which is involved in the Phase II detoxification of aromatic amines, appear to affect bladder cancer risk, with slow acetylators having modestly increased risk (97-100). There is also evidence for a multiplicative interaction between smoking and NAT2 (98). Glutathione S-transferase (GSTM1), involved in the Phase II detoxification of PAHs (FIGURE 1), has been studied in relation to bladder cancer in two previous meta-analyses (100, 101), and is examined in greater detail below. There is a suggestion of increased risk of bladder cancer among subjects with high activity of CYP1A2, which is a cytochrome P450 enzyme involved in the metabolic activation of aromatic amines (102). Inconsistent results have been observed for CYP2D6 (16), while no association has been observed for CYP2E1 (12, 103, 104).

click to view

Figure 1: Role of glutathione S-transferase M1 (GSTM1) in the detoxification of benzo[a]pyrene. CYP, cytochrome P450.

Materials and Methods
To examine the association between GSTM1 and bladder cancer, meta- and pooled analyses of all identified studies were undertaken. To identify articles or abstracts containing information on GSTM1 status and bladder cancer, we used review papers (2, 100, 105, 106), searched MEDLINE (using keywords "glutathione S-transferase," "gstm1," and "bladder cancer") without restriction on language, and searched abstracts in the American Association for Cancer Research (AACR) Proceedings. Eligible studies were those that used a case-control design, were population- or hospital-based, and that assessed GSTM1 status via genotype (TABLE 1). We excluded studies conducted in populations with a high prevalence of exposure to known bladder cancer risk factors other than tobacco smoke, including schistosomiasis infection (107-109) and occupational exposure to aromatic amines and polycyclic aromatic hydrocarbons (PAHs) (110, 111). We also excluded phenotype-only studies (112) to reduce possible misclassification of GSTM1 status (TABLE 2). Although cigarette smoking likely contributed to the elevated risks of bladder cancer reported in the excluded studies, our goal was to focus on tobacco smoke as the primary known risk factor and, more specifically, on a particular class of carcinogen found in tobacco smoke, polycyclic aromatic hydrocarbons, which are detoxified by GSTM1. We then attempted to obtain the original data from all eligible studies published in manuscript or abstract form through May 2000 in one of three ways. First, we included all relevant data that had been previously contributed to a database maintained for such purposes by the International Collaborative Study on Genetic Susceptibility to Environmental Carcinogens (GSEC) (113). These data included, for each subject, case-control status, GSTM1 genotype, method used to ascertain genotype, smoking status and history, age, gender, race, alcohol consumption, and bladder tumor stage and/or grade. Second, the principal investigators of all other eligible studies were sent letters inviting them to participate in a pooled analysis by sharing the data from their relevant study (or studies). The data requested were similar to those collected by the GSEC. Investigators who did not respond to the letter within eight weeks were sent a second letter further explaining the project and again inviting them to participate. Lastly, in order to reduce the risk of publication bias, we contacted all participating investigators, including those who had contributed their data to the GSEC, and asked them to identify any analyzed but unpublished data on this topic. One study (114) was excluded because the published paper provided case data only aggregated with other smoking-related cancers and we were unable to obtain the original data. Some study datasets differed slightly from the data contained in the published studies. For both the meta- and pooled analyses, when there was more than one publication based on related datasets, we included only the most recent results.

click to view

Table 1  Source and description of data used in the pooled and meta-analyses of  glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk.

click to view

Table 2  Source and description of data excluded from this study

For the meta-analysis, we included all eligible studies published in manuscript or abstract form through August 2000 that either provided a cross-tabulation of GSTM1 genotype status by case status or for which we had the original data. We used the unpublished genotype results from one study published with only phenotype data (115) and also included results from one unpublished study (Romkes M). When possible, all data were extracted from the published manuscript or abstract; otherwise, the necessary data were obtained from the dataset. A total of seventeen studies (2,149 cases and 3,646 controls) were identified and included (115-130; Romkes M, unpublished) (TABLE 1). We fitted random effects models (131-133), which estimate summary measures by weighting each individual-study result by a factor of within- and between-study variance. Homogeneity of study results in different groupings was assessed via the Q-statistic, with p-values < 0.05 indicating lack of homogeneity. Meta-analysis can produce biased results if the included studies are a biased sample of studies in general, so we assessed publication bias via funnel plots (134), Begg's test (134), and Egger's test (135).

For the pooled analyses, we included all eligible studies for which we had the original data, including two that were unpublished (TABLE 1). This comprised a total of thirteen studies (1,935 cases and 1,444 controls, including twelve studies and 74 percent of cases included in the meta-analysis): four in the United States (115-117; Romkes M, unpublished), seven in Europe (118-120, 123-125; Daly AK, unpublished), and two in Asia (128, 129). Because some of the datasets we were provided contained only case data (117, 125; Daly AK, unpublished), these datasets were included only in case-only analyses; the remaining ten datasets were included in case-control analyses. We used fixed effects models with study-specific intercepts to estimate odds ratios (OR) and 95 percent confidence intervals (CI) with SAS (136). Estimates were adjusted for age (< 60, 61-74, ³ 75 years), gender, race (Caucasian, other), and study site. Random effects models produced very similar results to fixed effects models and are not presented here.

Summary odds ratios for both the meta-analysis and pooled analysis were calculated for all the studies combined as well as for subgroups of studies. Subgroups were defined by geographic region (USA, Europe, and Asia) and by explicitly-defined incident vs. prevalent cases (incident: 115, 118, 119, 121, 123, 124 and prevalent: 118, 122, 123; studies that provided data separately for both incident and prevalent cases had the data included in both of the corresponding analyses). Subgroup analyses were also performed for i) studies whose participants were known or presumed to be Caucasian or that contained identifiable Caucasian subgroups (115-127; Romkes M, unpublished), ii) studies or subgroups within studies of transitional cell carcinoma (115-123; Romkes M, unpublished), and iii) studies published in manuscript form (i.e., excluding 119 and 128). We also attempted to examine multiplicative and additive interactions between GSTM1 status, various smoking measures, and risk of bladder cancer. Additive interactions were assessed using a model in which the confounders were loglinear and the exposures of interest (GSTM1, smoking, and their interaction) were linear (137). We did not estimate absolute risks because most of the studies included in this analysis were clinic- or hospital-based and it would not be appropriate to estimate baseline rates from these study controls. In addition, reliable estimates of baseline rates do not exist for some of the study populations included. Interaction analyses were performed with Epicure for Windows (138).

Most of the studies included in our analyses used clinic- or hospital-based controls. Use of hospital-based controls can introduce bias in the assessment of gene-environment interactions when disease resulting in hospitalization among the controls is related to the exposure or gene being studied (Wacholder et al., Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, unpublished manuscript). Our risk estimate for the main effect of cigarette smoking on bladder cancer (OR=1.86; 95 percent CI: 1.54, 2.26) was lower than the estimates of 3.02 (95 percent CI: 1.56, 5.82) obtained for the three population-based studies in our dataset and the 2.5-3.5 reported in other population-based case-control studies of bladder cancer (23, 139-142), suggesting a biased control sample. We examined this bias by performing stratified analyses of hospital-based and population-based studies. We also assessed the magnitude of this bias using parameters estimated from external and internal information, especially the two main effects and interaction for being hospitalized. Because case-only analyses of GSTM1 and ever smoking indicated no multiplicative interaction and there was no observed association between GSTM1 and control disease, we examined bias only in the additive interaction parameter. Details of this method are presented in the Appendix.

Because appreciable variability in study results can arise from differences in control selection, we further examined gene-environment interactions via case-only analyses involving GSTM1 status and smoking. In such analyses, an OR is calculated from cross-classification of exposure and genetic information among cases only (143-146). A case-only OR > 1 in the present study would indicate that the relationship between smoking and bladder cancer is stronger among GSTM1-null subjects than among GSTM1-active subjects. The same study subgroups and smoking measures used for case-control analyses were used for the case-only analyses, some case-only subgroups contained more cases than their case-control counterparts because some of the datasets we were provided contained only case data. Because valid interpretation of the case-only OR depends upon independence of the exposure and genetic factor in the underlying population, we assessed this independence via the c ² statistic among the associated controls.

Results
Meta-analysis
Although the studies included in the meta-analysis had odds ratios for the main effect of GSTM1 null and bladder cancer ranging from 0.76 (95 percent CI: 0.40, 1.44) to 3.81 (95 percent CI: 1.50, 9.70), all but two reported odds ratios above 1 (TABLE 1, FIGURE 2). Seven of the studies observed significantly elevated risk estimates. The larger studies tended to have similar risk estimates. There was appreciable overlap of confidence intervals.

click to view

Figure 2 : Odds ratios and 95 percent confidence intervals for glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk. Solid circles are proportional in area to number of cases. Vertical axis is on a log scale.  Romkes, unpubl: unpublished data of Romkes M, University of Pittsburgh, Pittsburgh, PA.

Including all the study results produced a significant summary OR for GSTM1 null of 1.44 (95 percent CI: 1.23, 1.68). Statistical tests indicated no significant heterogeneity of individual study results (Q-statistic with p-value=0.08) (TABLE 3). Grouping study results by geographic region produced similar summary ORs for Asia, Europe, and the US of 1.73 (95 percent CI: 1.66, 1.81), 1.39 (95 percent CI: 1.09, 1.77), and 1.44 (95 percent CI: 1.38, 1.50), respectively, although there was significant heterogeneity among European study results (Q statistic p-value=0.02). Results from studies with at least 100 cases and 100 controls (8 studies, 1,506 cases, 2,592 controls) produced a summary OR of 1.42 (95 percent CI: 1.26, 1.60), with Q-statistic p value=0.60. Statistical tests indicated significant heterogeneity of results from the six studies that either restricted to incident cases or presented data on this subgroup (Q statistic p-value=0.03). Grouping the published results from the eleven studies for which we had the datasets gave a summary OR of 1.40 (95 percent CI: 1.20, 1.64) (Q-statistic p=0.29). A funnel plot (FIGURE 3) and Begg's and Egger's statistical tests (p=0.43 and p=0.34, respectively) indicated no substantial publication bias. disease.

click to view

Table 3  Odds ratios and 95 percent confidence intervals for the association of GSTM1 null status and bladder cancer – meta-analyses

click to view

Figure 3 : Begg's funnel plot for assessing publication bias in relation to glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk.

Pooled-analysis
Summary ORs for GSTM1 null based on the pooled data were very similar to the summary ORs obtained from the meta-analysis (TABLE 4). The adjusted summary OR for all studies combined (10 studies) was 1.40 (95 percent CI: 1.20, 1.64). Examination by geographic region produced adjusted summary ORs of 1.70 (95 percent CI: 1.19, 2.42) for Asia, 1.29 (95 percent CI: 1.05, 1.58) for Europe, and 1.49 (95 percent CI: 1.06, 2.08) for the US. Grouping results of the five studies having at least 100 cases and 100 controls gave an adjusted summary OR (OR=1.37; 95 percent CI: 1.16, 1.63) similar to that for all studies combined. Most subgroup summary odds ratios ranged between approximately 1.3 and 1.5. Crude summary ORs were similar to adjusted ORs for all subgroups.

click to view

Table 4  Odds ratios and 95 percent confidence intervals for the association of GSTM1 null status and bladder cancer – pooled analyses

Nine studies provided sufficient information to classify subjects by smoking status (TABLE 5). Pooling these nine studies gave a significant adjusted summary OR for ever smoking of 1.91 (95 percent CI: 1.57, 2.32). The adjusted summary OR for European studies was 1.73 (95 percent CI: 1.35, 2.20) and was a substantially higher 3.34 (95 percent CI: 2.18, 5.11) for US studies. We did not estimate a summary OR among Asian studies because only one Asian study had smoking information on both cases and controls. The pooled estimate for Caucasians (OR=2.04; 95 percent CI: 1.65, 2.53) was similar to the all-studies pooled estimate, as was the pooled estimate for studies with at least 100 cases and 100 controls (OR=1.83; 95 percent CI: 1.49, 2.25). Summary risk estimates for dose-response relations involving pack-years of smoking and cigarettes smoked per day are not presented due to the appreciable heterogeneity of individual study estimates.

click to view

Table 5  Odds ratios and 95 percent confidence intervals for the association of ever smoking and bladder cancer – pooled analyses

There was no statistical evidence of multiplicative interaction, although there was a suggestion of additive interaction, between GSTM1 and ever smoking for bladder cancer risk from the pooled case-control analyses (TABLE 6). In stratified analyses of all studies combined, the odds ratio for GSTM1 null status was 1.21 (95 percent CI: 0.86, 1.72) among never smokers and a significant 1.36 (95 percent CI: 1.13-1.63) among ever smokers. The multiplicative interaction odds ratio was only slightly, and non-significantly, elevated (OR=1.15; 95 percent CI: 0.79, 1.67). In contrast, the additive interaction term, which was elevated, approached statistical significance (additive interaction=0.45; 95 percent CI: -0.03, 0.93). Estimates were similar across subgroups of studies, although for studies in the US and Europe combined, the multiplicative interaction odds ratio increased slightly to 1.23 (95 percent CI: 0.82, 1.85) and the additive interaction term became a significantly elevated 0.53 (95 percent CI: 0.03, 1.04). When we performed analyses that also included all available data extracted from published manuscripts for which we lacked individual-level data (an additional 2 studies with 22% more subjects, for a total of 11 studies with 1,628 cases and 1,552 controls) and adjusted only for study, the multiplicative and additive interaction terms changed little, although the additive term became significant (OR=1.14; 95 percent CI: 0.82, 1.59; additive interaction=0.46; 95 percent CI: 0.04, 0.88, respectively). We were unable to examine pooled estimates of potential interaction by pack-years or intensity of smoking because individual study results were heterogeneous and some analyses had low statistical power.

click to view

Table 6  Adjusted odds ratios and 95 percent confidence intervals for the main and interaction effects of GSTM1, ever smoking, and bladder cancer – pooled analyses*

We performed stratified analysis of hospital-based and population-based studies to assess bias in the estimate of additive interaction resulting from the use of hospital-based controls. Analysis of the six hospital-based studies in our dataset (containing a total of 1,222 cases and 1,003 controls) produced an additive interaction of 0.36 (95 percent CI: -0.17, 0.89); the three population-based studies (containing a total of 135 cases and 257 controls) produced an additive interaction of 1.09 (95 percent CI: -0.57, 2.74). This difference, although modest, suggests that the observed additive interaction value of 0.45 (95 percent CI: -0.03, 0.93), based predominantly on hospital-based controls, could be an underestimate. We further examined this bias via adjustment described by Wacholder et al. (Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, unpublished manuscript). Based on estimates from population-based studies (23, 139-142), we considered a range of 2.5 to 3.5 for the risk of bladder cancer, using population-based controls, related to ever smoking among GSTM1 active subjects (see Appendix). The adjusted additive interactions were higher than observed additive interactions across the range of ever smoking ORs, varying from 0.68 assuming a true ever smoking OR of 2.5, to 0.86 with an ever smoking OR of 3.0, to 1.03 with an ever smoking OR of 3.5.

Case-only analyses of ever smoking and bladder cancer also suggested a lack of multiplicative interaction. The case-only odds ratio for the twelve studies with GSTM1 and smoking data (1,836 cases) was 1.08 (95 percent CI: 0.84, 1.38). The Q-statistic indicated no significant heterogeneity of individual study results (p=0.34). Subgroups of studies produced similar, non-significant, case-only odds ratios. Case-only analyses involving pack-years (10 studies, 1,325 cases) and cigarettes per day (8 studies, 851 cases) also provided no evidence of interaction (data not shown). The c² statistics showed no association among controls between GSTM1 and either ever smoking (p=0.97), pack-years (p=0.90), or cigarettes per day (p=0.41), indicating independence of the exposure and genetic factor in the underlying population, which is necessary for valid interpretation of these results.

Discussion
The results of this meta-analysis suggest that persons with the GSTM1 null genotype are at increased risk of bladder cancer, with a summary OR for all studies combined of 1.44 (95 percent CI: 1.23, 1.68). The pooled analysis produced a very similar risk estimate. The similar summary risk estimates obtained using published data, pooled original data (including unpublished data), and homogeneous subgroups of published and pooled data, strengthen these findings.

Results of our analyses are consistent with those observed in other related meta-analyses (100, 101). D'Errico et al (100), in a 1996 meta-analysis of 10 studies, reported summary ORs of 1.54 (95 percent CI: 1.28, 1.85) for Caucasians and 2.40 (95 percent CI: 1.30, 4.45) for Asians. The higher summary OR among Asians in that meta-analysis was based on only two studies. One of those studies reported a highly elevated but unstable OR (117). The other included OR, obtained from an earlier report (147) of a study included in our meta-analysis (129), was unadjusted for potential confounders and was based on less-well matched cases and controls. A recent meta-analysis of 15 studies by Johns and Houlston (101), which included 9 studies that were part of the current meta-analysis, reported a summary OR of 1.53 (95 percent CI: 1.28, 1.84). In contrast to our analyses, which were restricted to studies of subjects whose primary risk factor for bladder cancer was likely to be tobacco smoke, the meta-analysis of Johns and Houlston included studies of subjects with known or likely occupational exposure to bladder carcinogens (110, 111) as well as subjects with a high prevalence of schistosomiasis (107-109).

We found a suggestion of additive interaction, but no statistical evidence of multiplicative interaction, between GSTM1 null and smoking, on bladder cancer risk. Stratified analysis and adjustment of additive interaction estimates for possible bias resulting from the use of hospital-based controls suggested that the observed additive interaction could be an underestimate of the true value. Heterogeneity of dose-response estimates across studies limited our ability to examine gene-environment interactions. Differences in the methods of obtaining detailed smoking histories may account for the variation observed when considering pack-years or cigarettes per day. The crude exposure classification represented by the ever smoking measure may have masked an interaction between level of smoking, GSTM1, and bladder cancer. Larger studies with consistent methods for ascertaining smoking history should help clarify this issue in the future.

The pooling of original study data provided us with several unique opportunities. By combining individual efforts, we obtained a large overall sample size. We were able to adjust uniformly for confounding factors across studies. In addition, we were able to examine some gene-environment interactions that most of the original studies were too small to adequately consider.

Publication bias, if present, could bias the results of a meta-analysis or pooled analysis away from the null (148). Such bias can occur when studies with null or unexpected results are not published, leading, in most instances, to their not being included in meta-analyses. We attempted to address this issue by including in our analyses all unpublished data on this topic that we were able to identify. In addition, statistical tests indicated no substantial publication bias.

These results indicate that, among populations studied to date, the GSTM1 null genotype is associated with a modest increase in risk of bladder cancer. There is suggestion of an additive interaction between GSTM1, smoking, and bladder cancer; however, neither the additive nor the multiplicative interaction parameters were statistically significant. Future studies would benefit from looking at the interaction of different levels of smoking with adequate statistical power. (> 25 repeats) 57.

Methods used for determining GSTM1 status have been described previously and will not be discussed further here. All studies included in the present analysis used genomic DNA extracted from blood. Most used polymerase chain reaction (PCR) with internal control primers (116-130). Some used ELISA in addition to PCR (118, 119) or used Southern blot hybridization  (115).

To date, there has been no population testing of GSTM1 in relation to bladder cancer. GSTM1 deficiency confers only a modestly increased relative risk and a low absolute risk. The lack of a reliable non-invasive bladder cancer screening test and the low absolute risk associated with GSTM1 deficiency limit the feasibility and usefulness of identifying GSTM1-deficient persons for bladder cancer screening purposes.

The authors wish to thank Dr. Jay Lubin, Dr. Sholom Wacholder, and Dr. Nilanjan Chatterjee for their statistical assistance.

Calculations used to assess possible bias in the estimation of gene-environment additive interaction resulting from use of hospital-based controls. 

Wacholder et al. (Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, unpublished manuscript) have shown how the magnitude of this bias depends on parameters that can be estimated from external and internal information, in particular the two main effects and interaction for being hospitalized. Adjusted interaction terms, presented in the text, were calculated using the following estimates:

• Odds ratio for bladder cancer, using population-based controls, from GSTM1 null among never-smokers of 1.20. Obtained from our pooled analysis of all studies. Based on the fact that the prevalence of GSTM1 null among controls in this study (52%) was similar to the prevalence among population controls of similar ethnicity reported in other published and unpublished studies of GSTM1 (2, 8-10)

• Odds ratio for study controls (i.e., for hospitalization for other diseases) from GSTM1 null among never-smokers of 1.00. Also based on the similar prevalence of GSTM1 null among controls in this study and comparable population controls in other studies.

• Odds ratio ranging from 2.5 to 3.5 for bladder cancer, using population-based controls, from ever smoking among GSTM1 active persons; obtained from other population-based studies (23, 139-142)

• Odds ratio ranging from 1.45 to 2.02 for study controls from ever smoking among GSTM1 active persons; estimated by dividing the odds ratio for bladder cancer, using population-based controls, from ever smoking among GSTM1 active persons (range: 2.5-3.5) by the corresponding odds ratio estimate in our pooled analysis (OR=1.73).

• Odds ratio for the association of GSTM1 null and ever smoking among study controls of 0.98.

• Multiplicative interaction of GSTM1 null and ever smoking for bladder cancer, using population-based controls, of 1.13. Calculated as the product of the observed multiplicative interaction, 1.15, and the association of GSTM1 null and ever smoking among study controls, 0.98.

• Table 1 - Source and description of data used in the pooled and meta-analyses of glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk
• Table 2 - Source and description of data excluded from this study
• Table 3 - Odds ratios and 95 percent confidence intervals for the association of GSTM1 null status and bladder cancer – meta-analyses
• Table 4 - Odds ratios and 95 percent confidence intervals for the association of GSTM1 null status and bladder cancer – pooled analyses
• Table 5 - Odds ratios and 95 percent confidence intervals for the association of ever smoking and bladder cancer – pooled analyses
• Table 6 - Adjusted odds ratios and 95 percent confidence intervals for the main and interaction effects of GSTM1, ever smoking, and bladder cancer – pooled analyses

Figures:

• Figure 1 - Role of glutathione S-transferase M1 (GSTM1) in the detoxification of benzo[a]pyrene.
  CYP, cytochrome P450
• Figure 2 - Odds ratios and 95 percent confidence intervals for glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk. Solid circles are proportional in area to number of cases. Vertical axis is on a log scale.
• Figure 3 - Begg's funnel plot for assessing publication bias in relation to glutathione S-transferase M1 (GSTM1) null status and bladder cancer risk.

List of References

Bladder Cancer (all links last accessed 2/2008)

• American Cancer Society
• National Cancer Institute: Bladder Cancer Home Page
• University of Pennsylvania Cancer Center OncoLink

Data on Bladder Cancer Frequency (all links last accessed 2/2008)

• IARC - Cancer Mondial
• SEER
• American Cancer Society- Statistics

GSTM1 (all links last accessed 2/2008)

• Human Gene Mutation Database
• OMIM
• UniGene
• GeneCards
• GenAtlas