Perinatal & Pediatric
Linking Intervention Effectiveness in Reducing Exposure to Household Air Pollution and Adverse Health Outcomes: An Instrumental Variable Analysis in the Household Air Pollution Intervention Network (HAPIN) Trial Mohamed Alhassan* Mohamed Alhassan Alhassan Alhassan Alhassan Department of Environmental and Radiological Health Sciences, Colorado State University, Fort Collins, Colorado, USA
Rationale: Prenatal exposure to household air pollution is associated with adverse birth outcomes. Randomized trials of clean cooking interventions often yield null intention-to-treat estimates, possibly due to heterogeneity in achieving exposure reduction. Estimating causal effects among individuals whose exposure would be lowered by the intervention may provide insight regarding whether reduced prenatal exposure to fine particulate matter (PM₂.₅) improves birth outcomes.
Methods: We evaluated the causal effect of achieving sustained prenatal PM₂.₅ exposure below 35 µg/m³, the World Health Organization (WHO) interim target, on birthweight and gestational age using data from 3,200 individuals in the Household Air Pollution Intervention Network (HAPIN) trial. Random assignment to a liquefied petroleum gas cooking intervention was used as an instrumental variable to estimate the average causal effect. Two-stage least squares models were fit for birthweight and gestational age, adjusting for maternal, household, and study-site covariates. Instrument strength was assessed using first-stage F-statistics.
Results: The adjusted estimated average causal effect of achieving PM₂.₅ exposure below 35 µg/m³ was a 65.5 g increase in birthweight (95% CI: -35.8, 166.7). No meaningful effect was observed for gestational age (adjusted mean difference: 0.13 weeks; 95% CI: -0.27, 0.52). Instruments were strong (first-stage F-statistics > 10).
Conclusions: In HAPIN, achieving prenatal PM₂.₅ exposure below the WHO interim target resulted in higher birthweight and gestational age, though confidence intervals included the null. Effect estimates were larger than prior intention-to-treat estimates, reflecting a different target population defined by achieved exposure reduction. Although confidence intervals were wider, the estimated effect exceeded the original HAPIN minimum detectable effect, suggesting limited statistical power in this analysis.