Estimating flexible exposure-response curves is critical for quantifying the impacts of environmental exposures on health. In intervention trials targeting household air pollution, the limited range of concentrations and small sample sizes of individual studies are important factors that may be limiting their statistical power. Pooling data from multiple studies provides an opportunity to increase power for estimating the exposure-response relationship.We present hierarchical approaches to modeling exposure concentrations and combining data from multiple studies to estimate a common exposure-response curve. The models are designed to accommodate features relevant to many environmental epidemiology studies: exposure measurements and outcome ascertainments that are irregularly spaced across time, clustered by individual and groups of individuals, and impacted by seasonal trends. The exposure concentration model additionally incorporates shrinkage to reduce error from high-variance observations. The exposure-response curve model provides a flexible, semi-parametric estimate of the joint exposure-response relationship across multiple studies from different times and contexts.We demonstrate this modeling approach using data from three studies of cookstoves and acute lower respiratory infections in children in Nepal.
Read full abstract