Household air pollution and personal CO:PM2.5 relationships during cooking in the GRAPHS cohort: important covariates include wearing compliance

Abstract

BACKGROUND AND AIM: Prior household air pollution studies have seen conflicting results using carbon monoxide (CO) measurements as predictors of personal PM₂.₅ exposures. We evaluated CO:PM₂.₅ relationships in the large Ghana Randomized Air Pollution and Health Study (GRAPHS). METHODS: 1,414 expecting mothers were enrolled into the study in rural communities in central Ghana. Participants tended to cook outdoors in dry weather and in semi-enclosed kitchens in wet weather. At baseline all cooking was done on traditional biomass and charcoal cookstoves. Personal exposure was assessed for seven 72-hour periods on each adult participant. CO was measured using Lascar EL-CO-USB devices. PM₂.₅ was also assessed during 2 of the 7 sessions in ~60% of the mothers using RTI MicroPEMs, which measured both real time and integrated PM, and included a motion sensor. We developed regression models to predict PM₂.₅ concentrations based on CO and other covariates. We built linear prediction models using a dataset of 2,118 validated exposure sessions of 24-hour, estimating average exposures in two ways: 1) during cooking hours (6-10 am and 4-8 pm); 2) for 24-hour periods. We conducted a forward stepwise variable selection process, successively adding covariates to the base model. RESULTS:The base model for the cooking period had an R² of 0.17, which was higher than the model based on 24-hour averages (R² =0.05). Adding four covariates, month, day, compliance and community, to both models moderately improved statistics (R² = 0.34 to 0.31). When we restricted data to sessions with higher wearing compliance (based on the motion sensor), predictive ability improved (R² = 0.54 - 0.51). CONCLUSIONS:In general we found limited ability to predict PM₂.₅ from CO and covariates in both 24-hour and 8-hour cooking periods. Poor wearing compliance can weaken associations between personal CO and PM₂.₅. KEYWORDS: Air pollution, Particulate matter, Modeling, Exposures