Registration Dossier

Administrative data

Description of key information

A study on "Hazardous air pollutants and breast cancer risk in California teachers: a cohort study" (Garcia et al Environmental Health, 2015; 14:14) is described.

Separate commentary on this study prepared by E. Delzell is presented below.

Additional information

E. Delzell COMMENTS ON

Garcia et al.Environmental Health, 2015; 14:14.

Hazardous air pollutants and breast cancer risk in California teachers: a cohort study.

Overview

The paper by Garcia et al. reports the results of a prospective cohort study that evaluated the relationship between environmental exposure to air pollutants and the incidence of invasive breast cancer among 112,378 women enrolled in the California Teacher Study (CTS). These women were active or retired teachers or administrators and lived in California at the time of their completing a baseline questionnaire in 1995-1996. To be included in the analyses of Garcia et al., they had to have a home address at the time of study enrolment that could be geocoded, and they had to meet several additional eligibility requirements. The cohort was followed annually using questionnaires and record linkages through 2011 to determine cancer incidence, dates and causes of death and changes of address.

Each subject’s exposure to hazardous outdoor air pollutants (HAPs) was estimated by linking the subject’s baseline home address, geocoded to a 2000 Census tract, to 2002 emissions data from the US EPA National-Scale Air Toxics Assessment (NATA), which has been conducted every three years since 1996. The HAPs analyzed by Garcia et al. were 24 compounds identified as potential mammary gland carcinogens (MGCs) on the basis of toxicological studies in laboratory animals. Cox proportional hazards models examined associations between each of two types of exposure variable and breast cancer. The two types of exposure variable were: (a) each compound, categorized by quintile of concentration estimated in 2002 for the census tract of residence at baseline and (b) a summary exposure variable computed as the sum of the standardized log base 10 value of each compound. The outcomes considered were all breast cancers combined and the separate outcomes of: (a) estrogen or progesterone receptor-positive breast cancers and (b) estrogen and progesterone receptor negative breast cancer. In addition, analyses examined associations in the overall cohort and in cohort subgroups specified on the basis of endogenous estrogen status at baseline, as indicated by baseline menopausal status, body mass index and history of hormone therapy. Analyses included Cox models that adjusted hazard ratios (HRs) for race/ethnicity only and models that adjusted HRs for race/ethnicity plus a number of important personal risk factors for breast cancer (obtained from baseline questionnaire data). Analyses also were performed without and with adjustment for multiple comparisons. Sensitivity analyses were performed on cohort members who were nonsmokers at baseline and on cohort members who did not move during follow-up.

The cohort experienced 5,676 incident invasive breast cancers during follow-up, 76.7% of which were estrogen or progesterone receptor-positive. Analyses of all forms of breast cancer combined, without adjustment for multiple personal risk factors for breast cancer or for multiple comparisons, found statistically significant HRs or trend p-values for residence in areas with high estimated ambient concentrations of acrylamide, carbon tetrachloride, chloroprene, 4,4’-methyl bis(2-chloro aniline), propylene oxide or vinyl chloride. However, results for these (and other) compounds were not statistically significant after adjustment for multiple potential confounders and multiple comparisons. For propylene oxide, the largest HR (1.11; 95% confidence interval, 1.02-1.20) occurred in the third quintile of exposure concentration and remained statistically significant after adjustment for multiple comparisons but not after adjustment for multiple confounders; and the p-value for trend was 0.18 (not statistically significant).

Further analyses, apparently done without adjusting for multiple personal risk factors, indicated that estrogen or progesterone receptor-positive breast cancer was associated with acrylamide, carbon tetrachloride, vinyl chloride, benzidine and ethylidene dichloride, while estrogen and progesterone receptor negative breast cancer was associated with benzene. For propylene oxide, HRs of about 1.2 were reported for each of the four upper quintiles of exposure; the HR for the third quintile remained statistically significant after adjustment for multiple comparisons; and the p-value for trend was 0.012, although the HR did not increase consistently with increasing exposure. The summary exposure variable was not associated with all breast cancer or with breast cancer subtypes. Analyses restricted to nonsmokers or to women who did not move during follow-up yielded results described by the authors as similar to results for the total cohort.

The investigators concluded that the study found “little evidence of risk associated with ambient exposure to MGC HAPs” and that “elevations in risk associated with some compounds for certain subpopulations of women and/or types of breast cancer tumors” should be interpreted cautiously but were “interesting” and may warrant further investigation. They particularly discussed results for carbon tetrachloride, ethylidene dichloride or vinyl chloride (ambient exposures to these two compounds were highly correlated in the study) and benzene.

Strengths and limitations

The study has notable strengths. It is a large, prospective cohort study with high-quality data on personal risk factors for breast cancer and on incident invasive breast cancer cases occurring during follow-up. It used census tract level data on ambient HAPs. It had the ability to adjust for multiple potential confounders. Limitations stem mainly from the approaches used to estimate exposure to MGC HAPs and to control for confounding and from the presentation of results.

Exposure estimation. Although the investigators mention several limitations of their approach (e.g., they assumed that 2002 air pollution data were representative of the entire follow-up time and that the available data provided good estimates of inhalation exposure; they did not consider exposure from indoor air pollutants or exposure routes other than inhalation), they do not mention their lack of data on exposure before the start of the study in 1995. If earlier life exposures were relevant to the risk of developing breast cancer, then misclassification of relevant exposures could be massive. It is not possible to predict the extent of exposure misclassification and direction of any resulting bias in this study.

Confounding. Some analyses controlled for multiple personal risk factors for breast cancer, while other analyses did not. Analyses of a particular HAP that adjusted for exposure to other HAPs apparently were not done.

Presentation of results. The paper does not display results of any of the analyses that adjusted for multiple personal risk factors for breast cancer. Instead, the authors simply state, with regard to data pertaining to overall breast cancer (Table 2), that results did not remain statistically significant after such adjustment. Therefore, it is not possible to discern whether such adjustment had an impact on the direction and strength of the point estimate of the HR or merely affected p-values and confidence intervals. In addition, results for subtypes of breast cancer and for cohort subgroups (Table 3 in the publication) do not appear to have been adjusted for multiple breast cancer risk factors. Thus, it is not possible to determine if confounding occurred in this study.

Interpretation

This study provides, at most, weak evidence of an association between exposure to outdoor MGC HAPs, including propylene oxide, and invasive breast cancer. HRs by quintile of exposure concentration were not markedly elevated for any compound (most “elevated” HRs were in the range of 1.1 to 1.2 and none was above 1.5); trends of increasing HR with increasing quintile of exposure were, for the most part, irregular; quintile-specific HRs and trend tests were not statistically significant after adjustment for multiple comparisons and for multiple confounders; subgroup analyses were not adjusted for confounding by personal breast cancer risk factors; and no analyses were adjusted for multiple HAP exposures. Moreover, there is little external support for the possible associations observed in the study. Thus, chance, exposure misclassification and confounding cannot be ruled out as alternative explanations of the weak associations reported by Garcia et al.