Factors Determining Black Carbon Exposures among Pregnant Women Enrolled in the HAPIN Trial

Abstract

Residential biomass burning is an important source of black carbon (BC) exposure among rural communities in low- and middle-income countries. We collected 7165 personal BC samples and individual/household level information from 3103 pregnant women enrolled in the Household Air Pollution Intervention Network trial. Women in the intervention arm received free liquefied petroleum gas stoves and fuel throughout pregnancy; women in the control arm continued the use of biomass stoves. Median (IQR) postintervention BC exposures were 9.6 μg/m³ (5.2-14.0) for controls and 2.8 μg/m³ (1.6-4.8) for the intervention group. Using mixed models, we characterized predictors of BC exposure and assessed how exposure contrasts differed between arms by select predictors. Primary stove type was the strongest predictor (R² = 0.42); the models including kerosene use, kitchen location, education, occupation, or stove use hours also provided additional explanatory power from the base model adjusted only for the study site. Our full, trial-wide, model explained 48% of the variation in BC exposures. We found evidence that the BC exposure contrast between arms differed by study site, adherence to the assigned study stove, and whether the participant cooked. Our findings highlight factors that may be addressed before and during studies to implement more impactful cookstove intervention trials.

Keywords: Guatemala; India; Peru; Rwanda; biomass fuel stoves; black carbon (BC); exposure assessment; exposure models; household air pollution; intervention; liquefied petroleum gas.

Figure 1

HAPIN-wide multivariable mixed-effects regression coefficients (with 95% confidence intervals). Numeric coefficients represent the mean percentage change of the geometric mean in BC exposure compared to the reference category based on the final multivariable linear regression models. Coefficients for relative humidity and kitchen volume represent a 5 unit increase in percentage and 10 unit increase in volume, respectively.

Figure 2

Hapin-wide postrandomization BC exposure contrasts (with 95% confidence intervals) between treatment arms (control vs intervention) by selected factors. Effect estimates outside of the confidence intervals reported in Johnson et al. 2022 (triangle) show how select factors potentially modified the effectiveness of the intervention in reducing personal exposures to BC. The percent differences in personal BC exposure between treatment arms were calculated within each subvariable using the emmeans package in R, which computes and compares marginal means.