Cooking with liquefied petroleum gas or biomass and fetal growth outcomes: a multi-country randomised controlled trial

Abstract

Background: Household air pollution might lead to fetal growth restriction during pregnancy. We aimed to investigate whether a liquefied petroleum gas (LPG) intervention to reduce personal exposures to household air pollution during pregnancy would alter fetal growth.

Methods: The Household Air Pollution Intervention Network (HAPIN) trial was an open-label randomised controlled trial conducted in ten resource-limited settings across Guatemala, India, Peru, and Rwanda. Pregnant women aged 18-34 years (9-19 weeks of gestation) were randomly assigned in a 1:1 ratio to receive an LPG stove, continuous fuel delivery, and behavioural messaging or to continue usual cooking with biomass for 18 months. We conducted ultrasound assessments at baseline, 24-28 weeks of gestation (the first pregnancy visit), and 32-36 weeks of gestation (the second pregnancy visit), to measure fetal size; we monitored 24 h personal exposures to household air pollutants during these visits; and we weighed children at birth. We conducted intention-to-treat analyses to estimate differences in fetal size between the intervention and control group, and exposure-response analyses to identify associations between household air pollutants and fetal size. This trial is registered with ClinicalTrials.gov (NCT02944682).

Findings: Between May 7, 2018, and Feb 29, 2020, we randomly assigned 3200 pregnant women (1593 to the intervention group and 1607 to the control group). The mean gestational age was 14·5 (SD 3·0) weeks and mean maternal age was 25·6 (4·5) years. We obtained ultrasound assessments in 3147 (98·3%) women at baseline, 3052 (95·4%) women at the first pregnancy visit, and 2962 (92·6%) at the second pregnancy visit, through to Aug 25, 2020. Intervention adherence was high (the median proportion of days with biomass stove use was 0·0%, IQR 0·0-1·6) and pregnant women in the intervention group had lower mean exposures to particulate matter with a diameter less than 2·5 μm (PM_2·5; 35·0 [SD 37·2] μg/m³vs 103·3 [97·9] μg/m³) than did women in the control group. We did not find differences in averaged post-randomisation Z scores for head circumference (0·30 vs 0·39; p=0·04), abdominal circumference (0·38 vs 0·39; p=0·99), femur length (0·44 vs 0·45; p=0·73), and estimated fetal weight or birthweight (-0·13 vs -0·12; p=0·70) between the intervention and control groups. Personal exposures to household air pollutants were not associated with fetal size.

Interpretation: Although an LPG cooking intervention successfully reduced personal exposure to air pollution during pregnancy, it did not affect fetal size. Our findings do not support the use of unvented liquefied petroleum gas stoves as a strategy to increase fetal growth in settings were biomass fuels are used predominantly for cooking.

Funding: US National Institutes of Health and Bill & Melinda Gates Foundation.

Translations: For the Kinyarwanda, Spanish and Tamil translations of the abstract see Supplementary Materials section.

Figure 1

Trial profile Participants could have more than one reason for exclusion (ie, categories are not mutually exclusive).

Figure 2

Violin plots for head circumference (A), abdominal circumference (B), femur length (C), and estimated fetal weight or birthweight (D) by categories of gestational age and intervention group Violin plots are hybrids of boxplots and density plots, and are used here to visualise the distribution of fetal growth outcomes. The thick band represents the 50th percentile, and the thinner bands represent the 25th and 75th percentiles.

Figure 3

Results of intention-to-treat analyses for the differences in Z scores of head circumference (A), abdominal circumference (B), femur length (C), and estimated fetal weight or birthweight (D) by visit, displayed as a forest plot Stacked panels represent one of the four fetal growth outcomes. The difference was estimated from multiple linear regressions of fetal growth Z scores as a function of intervention group adjusted for randomisation strata at each visit, and multiple linear regressions of fetal growth Z score averages across visits as a function of intervention group adjusted for randomisation strata. Values lower than 0 indicate benefit to the control group, whereas values higher than 0 indicate benefit to the intervention group. The estimated averaged post-randomised means by trial group and the mean differences between groups (intervention minus control) are displayed with a diamond and the corresponding 98·75% CIs as horizontal lines. The first pregnancy visit was at 24–28 weeks of gestation; and the second pregnancy visit was at 32–36 weeks of gestation. We did not find differences in averaged post-randomisation Z scores between intervention participants and controls for head circumference (0·30 vs 0·39; p=0·04), abdominal circumference (0·38 vs 0·39; p=0·99), femur length (0·44 vs 0·45; p=0·73), and estimated fetal weight or birthweight (–0·13 vs –0·12; p=0·70).

Figure 4

Results of exposure–response analyses of head circumference, abdominal circumference, femur length, and estimated fetal weight or birthweight Z scores scaled to the interquartile difference of PM_2·5, black carbon, and carbon monoxide and estimated at 25, 33, and 40 weeks (for the estimated fetal weight Z score only) Differences were estimated from generalised additive mixed models (by pollutant) of the fetal growth Z scores as a function of a smooth surface comprising gestational age and pollutant concentrations and adjusted for potentially confounding variables (socioeconomic status, maternal age, nulliparity, diet diversity, maternal education, maternal haemoglobin, and second-hand smoke). Each of the 12 panels corresponds to the combination of fetal growth Z score and personal exposures to each pollutant. Values lower than 0 indicate a lower Z score for the interquartile difference of each pollutant (ie, household air pollution worsens fetal growth Z scores), whereas values higher than 0 indicate higher Z score for the interquartile difference of each pollutant. The estimated mean differences are displayed with a diamond and corresponding 98·75% CIs as horizontal lines. PM_2·5=fine particulate matter.