Multinational modelling of PM_2.5 and CO exposures from household air pollution in peri-urban Cameroon, Ghana and Kenya

In sub-Saharan Africa, approximately 85% of the population uses polluting cooking fuels (e.g. wood, charcoal). Incomplete combustion of these fuels generates household air pollution (HAP), containing fine particulate matter (PM_2.5) and carbon monoxide (CO). Due to large spatial variability, increased quantification of HAP levels is needed to improve exposure assessment in sub-Saharan Africa. The CLEAN-Air(Africa) study included 24-h monitoring of PM_2.5 and CO kitchen concentrations (n_pm2.5 = 248/n_CO = 207) and female primary cook exposures (n_pm2.5 = 245/n_CO = 222) in peri-urban households in Obuasi (Ghana), Mbalmayo (Cameroon) and Eldoret (Kenya). HAP measurements were combined with survey data on cooking patterns, socioeconomic characteristics and ambient exposure proxies (e.g. walking time to nearest road) in separate PM_2.5 and CO mixed-effect log-linear regression models. Model coefficients were applied to a larger study population (n = 937) with only survey data to quantitatively scale up PM_2.5 and CO exposures. The final models moderately explained variation in mean 24-h PM_2.5 (R² = 0.40) and CO (R² = 0.26) kitchen concentration measurements, and PM_2.5 (R² = 0.27) and CO (R² = 0.14) female cook exposures. Primary/secondary cooking fuel type was the only significant predictor in all four models. Other significant predictors of PM_2.5 and CO kitchen concentrations were cooking location and household size; household financial security and rental status were only predictive of PM_2.5 concentrations. Cooking location, household financial security and proxies of ambient air pollution exposure were significant predictors of PM_2.5 cook exposures. Including objective cooking time measurements (from temperature sensors) from (n = 143) households substantially improved (by 52%) the explained variability of the CO kitchen concentration model, but not the PM_2.5 model. Socioeconomic characteristics and markers of ambient air pollution exposure were strongly associated with mean PM_2.5 measurements, while cooking environment variables were more predictive of mean CO levels.