Assessing Long-Term Exposure in the California Teachers Study

In an article published in Environmental Health Perspectives (Ostro et al. 2010), we analyzed the relationships of long-term exposure to fine particulate matter (≤ 2.5 μm in aerodynamic diameter; PM2.5) and its components with mortality in a cohort of > 100,000 active and retired female professionals participating in the California Teachers Study (CTS) cohort. We used a Cox proportional hazards model in which pollution exposure was measured as a continuous variable over the study period. Monthly average pollutant concentrations were obtained for each participant from measurements at the nearest PM2.5 monitor within either 8 or 30 km of her geocoded residential address. Each participant was assigned a single exposure value over the follow-up period, defined as the average pollutant concentration from the beginning of the observation period (1 June 2002) to the woman’s date of death, loss to follow-up, or study termination (31 July 2007). Thus, exposure assignment was dependent on the duration of follow-up for each participant. In our article (Ostro et al. 2010), we reported associations of mortality from all causes, cardiopulmonary disease, and ischemic heart disease (IHD) with PM2.5 mass and several of its components. However, the estimated hazard ratios (HRs) were generally higher than those reported from previous cohort studies (Dockery et al. 1993; Eftim et al. 2008; Krewski et al. 2009; Laden et al. 2006; Pope et al. 1995). Part of this difference was likely due to the nature of the exposure assignment. Most previous cohort studies have assigned the same exposure period to all study subjects, regardless of when deaths occurred. Thus, estimated exposures for some study participants in several studies occurred after their deaths. In addition, exposures have usually been assigned to participants based on their residential address at enrollment only, without taking into account exposure changes that may have occurred throughout the study period or when participants relocated. Finally, many previous studies measured exposure for only a subset of the years during which the cohort was followed. In an effort to reduce these aspects of exposure misclassification, we estimated exposures beginning prior to the cohort follow-up period, continuing to the end of the study or until the participant died or relocated out of state, incorporating updated exposure assignments when the subjects moved. Importantly, measured concentrations of several pollutants in California declined substantially from 2002 through 2007; annual average PM2.5, organic carbon (OC), and nitrates decreased by around 30% each. These marked decreases in ambient PM2.5 concentrations resulted in lower average exposure estimates for cohort members who survived to the end of our study. Thus, the exposure assigned to a participant who died at time t would tend to be greater for events occurring early in the observation period, compared with the long-term average exposures of the participants who comprised the remainder of the risk set (i.e., those who were still part of the cohort study at time t and who subsequently experienced lower ambient pollution levels). We have reanalyzed the CTS data using time-dependent pollution metrics—in which the exposure estimates for everyone remaining alive in the risk set were recalculated at the time of each death—in order to compare their average exposures up to that time with that of the individual who had died. In this way, decedents and survivors comprising the risk set had similar periods of pollution exposure, without subsequent pollution trends influencing the surviving women’s exposure estimates. As in our previous study (Ostro et al. 2010), we restricted the sample in this reanalysis to women living within 30 km of one of eight fixed-site monitors in the U.S. Environmental Protection Agency’s Speciation Trend Network (STN), resulting in a study population of almost 44,000 women. Residential addresses from study enrollment forward were geocoded and linked with monthly pollutant averages at the nearest STN monitor to generate estimates of long-term exposure. We also used the same set of individual and ecological covariates in a Cox proportional hazards model as was used in the original study. Pollutants entered separately into the model included PM2.5 mass, elemental carbon (EC), OC, sulfate, nitrate, iron, potassium, silicon, and zinc. We used data on primary cause of death from August 2002 through July 2007 to examine the relationships between pollutants and mortality from all causes and cardiopulmonary, pulmonary, and IHDs. The results are summarized in Erratum Table 1, scaled to the interquartile range (IQR) for each pollutant. HRs were significantly attenuated from our previous results. No associations were observed between all-cause mortality and PM2.5 or its components. For cardiopulmonary mortality, we observed significant associations for PM2.5 mass, nitrate, sulfate, and silicon, with more modest associations for zinc. PM2.5 mass and all of its components were associated with mortality from IHD, whereas none of the pollutants was associated with pulmonary mortality. This Erratum Table 1 should replace Table 5 in our previous article (Ostro et al. 2010). Erratum Table 1 Association between mortality outcomes and PM2.5 and its components using a 30-km buffer (n = 43,220). Compared with our previous results (Ostro et al. 2010), these updated PM2.5 HRs are more consistent with several other published estimates of mortality risks, which are scaled to an increment of 10 μg/m3 of long-term average PM2.5 and summarized in Erratum Table 2. For example, relative to our revised HR of 1.19 for cardiopulmonary disease, analogous HRs from previous studies include 1.09 (95% CI, 1.03–1.16) from the American Cancer Society–Cancer Prevention II (ACS) cohort (cardiopulmonary disease; Pope et al. 2004), 1.28 (95% CI, 1.13–1.44) from the Harvard Six Cities study (cardiovascular disease; Laden et al. 2006), and 1.10 (95% CI, 0.94–1.28) from the Los Angeles subcohort of the ACS study (cardiopulmonary disease; Jerrett et al. 2005). Much higher HRs were observed in the observational study of the Women’s Health Initiative cohort for cardiovascular and IHD mortality (Miller et al. 2007). Erratum Table 2 Comparative HRs (95% CIs) associated with a 10-μg/m3 change in long-term exposure to PM2.5 in several cohort studies conducted in the United States. These revised results still support the existence of elevated risks of PM2.5-associated cardiopulmonary disease and IHD, and illustrate the importance of considering the impact of long-term pollution trends in modeling estimates of exposure.