Decomposing the variance: The unique and shared associations of fine and ultrafine particulate matter exposed during pregnancy with child autism spectrum disorder

While fine particulate matter (PM_2.5) has been associated with autism spectrum disorder (ASD), few studies focused on ultrafine particles (PM_0.1). Given that fine and ultrafine particles can be highly correlated due to shared emission sources, challenges remain to distinguish their health effects. In a retrospective cohort of 318,371 mother-child pairs (4549 ASD cases before age 5) in Southern California, pregnancy average PM_2.5 and PM_0.1 were estimated using a California-based chemical transport model and assigned to residential addresses. The correlation between PM_2.5 and PM_0.1 was 0.87. We applied a two-step variance decomposition approach: first, decomposing PM_2.5 and PM_0.1 into the shared and unique variances using ordinary least squares linear regression (OLS) and Deming regression considering errors in both exposures; then assessing associations between decomposed PM_2.5 and PM_0.1 and ASD using Cox proportional hazard models adjusted for covariates. Prenatal PM_2.5 and PM_0.1 each was associated with increased ASD risk. OLS decomposition showed that associations were driven mainly by their shared variance, not by their unique variance. Results from Deming regression considering assumptions of measurement errors were consistent with those from OLS. This decomposition approach has potential to disentangle health effects of correlated exposures, such as PM_2.5 and PM_0.1 from common emissions sources.