A hierarchical model for estimating the exposure-response curve by combining multiple studies of acute lower respiratory infections in children and household fine particulate matter air pollution

Joshua P Keller; Joanne Katz; Amod K Pokhrel; Michael N Bates; James Tielsch; Scott L Zeger

doi:10.1097/EE9.0000000000000119

A hierarchical model for estimating the exposure-response curve by combining multiple studies of acute lower respiratory infections in children and household fine particulate matter air pollution

Environ Epidemiol. 2020 Nov 18;4(6):e119. doi: 10.1097/EE9.0000000000000119. eCollection 2020 Dec.

Authors

Joshua P Keller¹, Joanne Katz², Amod K Pokhrel³, Michael N Bates^{3

4}, James Tielsch⁵, Scott L Zeger⁶

Affiliations

¹ Department of Statistics, Colorado State University, Fort Collins, Colorado.
² Department of International Health, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland.
³ Division of Environmental Health Sciences, School of Public Health, University of California, Berkeley, California.
⁴ Division of Epidemiology, School of Public Health, University of California, Berkeley, California.
⁵ Department of Global Health, Milken Institute School of Public Health, George Washington University, Washington, District of Columbia.
⁶ Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland.

Abstract

Adverse health effects of household air pollution, including acute lower respiratory infections (ALRIs), pose a major health burden around the world, particularly in settings where indoor combustion stoves are used for cooking. Individual studies have limited exposure ranges and sample sizes, while pooling studies together can improve statistical power.

Methods: We present hierarchical models for estimating long-term exposure concentrations and estimating a common exposure-response curve. The exposure concentration model combines temporally sparse, clustered longitudinal observations to estimate household-specific long-term average concentrations. The exposure-response model provides a flexible, semiparametric estimate of the exposure-response relationship while accommodating heterogeneous clustered data from multiple studies. We apply these models to three studies of fine particulate matter (PM_2.5) and ALRIs in children in Nepal: a case-control study in Bhaktapur, a stepped-wedge trial in Sarlahi, and a parallel trial in Sarlahi. For each study, we estimate household-level long-term PM_2.5 concentrations. We apply the exposure-response model separately to each study and jointly to the pooled data.

Results: The estimated long-term PM_2.5 concentrations were lower for households using electric and gas fuel sources compared with households using biomass fuel. The exposure-response curve shows an estimated ALRI odds ratio of 3.39 (95% credible interval = 1.89, 6.10) comparing PM_2.5 concentrations of 50 and 150 μg/m³ and a flattening of the curve for higher concentrations.

Conclusions: These flexible models can accommodate additional studies and be applied to other exposures and outcomes. The studies from Nepal provides evidence of a nonlinear exposure-response curve that flattens at higher concentrations.

Keywords: Bayesian; Exposure-response curve; Hierarchical models; Household air pollution; Measurement error; Particulate matter.