You are here

Assessing adverse health effects of long-term exposure to low levels of ambient pollution

Principal Investigators: 

Harvard T.H. Chan School of Public Health


Harvard T.H. Chan School of Public Health

This study will examine health effects of low levels of air pollution in the US using data from about ~56 million people enrolled in Medicare and Medicaid. In addition, they will develop new causal modeling methods to characterize the shape of the exposure-response function. See also this Program Summary of HEI's research program on low levels of air pollution. 

Funded under
In review

Effects of Confounder Control and Causal Modelling in MAPLE, ELAPSE, and Medicare Cohorts

Danielle Braun1,2 on behalf of the MAPLE, ELAPSE, and Harvard (Medicare) teams; Bert Brunekreef3; Michael Brauer4,5Francesca Dominici1

1Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
2Dana-Farber Cancer Institute, Boston, Massachusetts, USA
3Institute for Risk Assessment Sciences, Utrecht University, Utrecht, Netherlands
4University of British Columbia, Vancouver, British Columbia, Canada
5Institute for Health Metrics and Evaluation, University of Washington, Seattle, Washington, USA

Background. The MAPLE, ELAPSE, and Harvard Medicare studies, are three Health Effects Institute funded grants assessing health effects of long-term exposure to low levels of ambient air pollution, PM2.5. Ideally when studying the health effects of air pollution one would randomize patients to low/high air pollution, but this is not feasible. Observational studies, such as these three studies, have limitations due to lack of randomization. Factors that are associated both with both exposures and health outcomes (e.g. socioeconomic status (SES)-related factors) may confound exposure comparisons. 

Methods. All three studies address the challenges of covariate adjustment and confounding differently when assessing the association between PM2.5 exposure and mortality. Both the MAPLE and ELAPSE studies fit three main models with increasing levels of covariate adjustment; Model 1 was an unadjusted model with only strata specific information, Model 2 was Model 1 with additional individual level covariates, Model 3 was Models 1 and 2 with additional contextual level covariates. In addition, both the MAPLE and ELAPSE studies conducted an indirect adjustment to adjust for unmeasured behavioral covariates in the main study. The Harvard Medicare study conducted sensitivity analysis using a subset of the Medicare cohort which contains information on more than 150 potential covariates. In addition, the Harvard Medicare study applied various causal inference models that are more robust to model misspecification. 

Results. The MAPLE study showed that the relationship between PM2.5 and non-accidental mortality was robust to the addition of behavioral covariates.  Results from the indirect adjustment suggests that behavioral covariates (e.g., smoking and diet) only slightly confounded the PM2.5–mortality association. The ELAPSE study showed that across all seven cohorts, hazard ratios (HRs) were sensitive to more complete adjustment for potential confounders. Adjustment for individual and especially area-level confounders increased the HRs in some cohorts (e.g. Rome, Swiss, Norwegian) and decreased HRs in other cohorts (e.g. Dutch, English). Results from the indirect adjustment showed that in the Dutch, Swiss and Norwegian cohorts, HRs were attenuated but remained (borderline) significant.  In the Rome and Belgian cohorts, HRs increased after indirect adjustment. The Harvard Medicare study showed that HRs were robust to the inclusion of additional individual level covariates. Furthermore, all five approaches (traditional and casual) lead to similar results. 

Conclusions. In all three studies missing data on covariates were unlikely to significantly confound the PM2.5–mortality relationship. The Harvard Medicare study showed robust and reproducible evidence on causal link between PM and mortality by applying five different approaches to analyze the data. Future work includes harmonization of covariates and confounding adjustment approaches across the three studies.