Household Air Pollution and Child Lung Function: The Ghana Randomized Air Pollution and Health Study

Abstract

Rationale: The impact of a household air pollution (HAP) stove intervention on child lung function has been poorly described. Objectives: To assess the effect of a HAP stove intervention for infants prenatally to age 1 on, and exposure-response associations with, lung function at child age 4. Methods: The Ghana Randomized Air Pollution and Health Study randomized pregnant women to liquefied petroleum gas (LPG), improved biomass, or open-fire (control) stove conditions through child age 1. We quantified HAP exposure by repeated maternal and child personal carbon monoxide (CO) exposure measurements. Children performed oscillometry, an effort-independent lung function measurement, at age 4. We examined associations between Ghana Randomized Air Pollution and Health Study stove assignment and prenatal and infant CO measurements and oscillometry using generalized linear regression models. We used reverse distributed lag models to examine time-varying associations between prenatal CO and oscillometry. Measurements and Main Results: The primary oscillometry measure was reactance at 5 Hz, X5, a measure of elastic and inertial lung properties. Secondary measures included total, large airway, and small airway resistance at 5 Hz, 20 Hz, and the difference in resistance at 5 Hz and 20 Hz (R5, R20, and R5-20, respectively); area of reactance (AX); and resonant frequency. Of the 683 children who attended the lung function visit, 567 (83%) performed acceptable oscillometry. A total of 221, 106, and 240 children were from the LPG, improved biomass, and control arms, respectively. Compared with control, the improved biomass stove condition was associated with lower reactance at 5 Hz (X5 z-score: β = -0.25; 95% confidence interval [CI] = -0.39, -0.11), higher large airway resistance (R20 z-score: β = 0.34; 95% CI = 0.23, 0.44), and higher AX (AX z-score: β = 0.16; 95% CI = 0.06, 0.26), which is suggestive of overall worse lung function. The LPG stove condition was associated with higher X5 (X5 score: β = 0.16; 95% CI = 0.01, 0.31) and lower small airway resistance (R5-20 z-score: β = -0.15; 95% CI = -0.30, 0.0), which is suggestive of better small airway function. Higher average prenatal CO exposure was associated with higher R5 and R20, and distributed lag models identified sensitive windows of exposure between CO and X5, R5, R20, and R5-20. Conclusions: These data support the importance of prenatal HAP exposure on child lung function. Clinical trial registered with www.clinicaltrials.gov (NCT01335490).

Keywords: household air pollution; lung function; oscillometry; prenatal; sensitive windows.

Figure 1.

Associations between the Ghana Randomized Air Pollution and Health Study (GRAPHS) arm and z-scores for oscillometry at age 4 in (A) intention-to-treat and (B) sensitivity models. This figure demonstrates the associations between the GRAPHS study arms—control (traditional open-fire stove), improved biomass stove, and LPG stove—and oscillometry z-scores, including for resistance at 5 and 20 Hz (R5 and R20, respectively), difference in resistance at 5 Hz and 20 Hz (R5–20), reactance at 5 Hz (X5), Fres, and AX. In (A), intention-to-treat analyses are unadjusted, whereas (B) sensitivity analyses are adjusted for asset index (relative household wealth), and secondhand smoke exposure. All models include cluster robust standard errors. AX = area of reactance; Fres = resonant frequency; IOS = oscillometry; LPG = liquefied petroleum gas.

Figure 2.

Associations between prenatal household air pollution exposure, as indexed by personal carbon monoxide (CO), and z-scores for (A) resistance at 5 Hz, (R5) (B) resistance at 20 Hz (R20), and (C) area of reactance (AX), measured by oscillometry at age 4 (Model 2). This figure demonstrates the associations between personal CO exposure measurements over pregnancy and oscillometry at age 4 for (A) R5, (B) R20, and (C) AX z-scores assuming week-specific effects for the overall sample. Models were adjusted for child sex, child body mass index, average infant CO exposure, asset index (relative household wealth), maternal education, secondhand smoke exposure, and child exposure to particulate matter ⩽2.5 μm in aerodynamic diameter (PM_2.5) at age 4, and they included inverse probability weights to account for the PM_2.5 exposure at age 4 measurement subset. The y-axis represents the time-varying association between oscillometry measure z-score and CO exposure; the x-axis depicts gestational age in weeks. The solid line indicates the predicted correlation, and the gray shaded areas represent the 95% confidence intervals. A sensitive window is identified when the confidence intervals do not include zero.

Figure 3.

Associations between prenatal household air pollution exposure, as indexed by personal carbon monoxide (CO), and z-scores for (A) reactance at 5 Hz (X5) and (B) difference in resistance at 5 Hz and 20 Hz (R5–20) by oscillometry at age 4 (Model 2). This figure demonstrates the associations between personal CO exposure measurements over pregnancy and z-scores for oscillometry at age 4 for (A) X5 and (B) R5–20, assuming week-specific effects for the overall sample. Models were adjusted for child sex, child body mass index, average infant CO exposure, asset index (relative household wealth), maternal education, secondhand smoke exposure, and child exposure to particulate matter ⩽2.5 μm in aerodynamic diameter (PM_2.5) at age 4, and they included inverse probability weights to account for the PM_2.5 exposure at age 4 measurement subset. The y-axis represents the time-varying association between oscillometry measure z-score and CO exposure; the x-axis depicts gestational age in weeks. The solid line indicates the predicted correlation, and the gray shaded areas represent the 95% confidence intervals. A sensitive window is identified when the confidence intervals do not include zero.