Prenatal exposure to airborne particles is a potential risk factor for infant neuropsychological development. This issue is usually explored by regression analysis under the implicit assumption that all relevant confounders are accounted for. Our aim is to estimate the causal effect of prenatal exposure to high concentrations of airborne particles with a diameter < 2.5 µm (PM2.5) on children's psychomotor and mental scores in a birth cohort from Gipuzkoa (Spain), and investigate the robustness of the results to possible unobserved confounding. We adopted the propensity score matching approach and performed sensitivity analyses comparing the actual effect estimates with those obtained after adjusting for unobserved confounders simulated to have different strengths. On average, mental and psychomotor scores decreased of -2.47 (90% CI: -7.22; 2.28) and -3.18 (90% CI: -7.61; 1.25) points when the prenatal exposure was ≥17 μg/m3 (median). These estimates were robust to the presence of unmeasured confounders having strength similar to that of the observed ones. The plausibility of having omitted a confounder strong enough to drive the estimates to zero was poor. The sensitivity analyses conferred solidity to our findings, despite the large sampling variability. This kind of sensitivity analysis should be routinely implemented in observational studies, especially in exploring new relationships.
Keywords: Monte Carlo simulations; airborne particles; bias analysis; child development; propensity score matching; sensitivity analysis.