2198

Background/Aim: Global quantitative estimation of household air pollution (HAP) exposure is critical for assessing associated health impacts. Previous global HAP modeling studies have aggregated published measurements, which can introduce bias by combining data from various study protocols.
Methods: The Prospective Urban and Rural Epidemiology (PURE)-AIR study, one of the largest HAP measurement studies to-date, included 48-hour fine particulate matter (PM_2.5) kitchen concentrations in a stratified-sample (2,541 households) proportional to primary cooking fuel type across 120 rural communities within eight countries (Bangladesh, Chile, China, Colombia, India, Pakistan, Tanzania, Zimbabwe). Additionally, survey data captured detailed information on household cooking characteristics/behaviors. Random forest modeling was used to rank predictors of measured PM_2.5 concentrations. Selected factors and weakly informative priors from a previous global model using PM_2.5 measurements available in the WHO Global HAP database were included in a Bayesian hierarchical predictive model of PM_2.5 concentrations. Model performance was assessed via leave-one-out cross-validation. The chosen model was then applied to 26,197 households in the eight countries, with available survey data but no PM_2.5 monitoring.
Results: Primary cooking fuel type, heating fuel type, roofing material, primary drinking water source, household size, household income and kitchen ventilation (windows) were the most important predictors of household PM_2.5 concentrations; an R² of 0.49 and mean absolute error of 49 μg/m³ resulted from the Bayesian model. Modeled global average 48-hour PM_2.5 concentrations among households using gas as a primary cooking fuel (46 μg/m³; 95%CI:[32,64]) were lower than those using coal (68 μg/m³; 95%CI:[37,124]), wood (72 μg/m³; 95%CI:[65,80]), grass/shrubs (81 μg/m³; 95%CI:[45,147]) and animal dung (98 μg/m³; 95%CI:[50,194]).
Conclusions: HAP monitoring in a strategic sub-sample of households alongside detailed survey data collection provides a feasible, multinational quantitative assessment of HAP exposure. Improved global estimates of PM_2.5 concentrations can be used to improve risk assessment models and epidemiological analyses of HAP.