Live Birth Bias May Affect Associations between Air Pollution and Autism

摘要

In a recent publication of ours, we implemented a distributed lag model with autism spectrum disorders (ASD) data and N02 models from Israel that are highly resolved in time and space. When mutually adjusted, postnatal exposures had positive associations, while prenatal exposures had negative peak around the end of the 1st trimester. We suggest two possibilities in which the apparently protective associations could be the result of live-birth bias. The first possibility results from selection that affects the distribution of exposures in the analyzed sample. Selecting on live births may result in a group less likely to have co-exposure to both high air pollution and other causes of ASD that also affect the likelihood of live birth. In the analyzed sample, those with high exposure are less likely to have other causes of ASD (and thus a lower ASD prevalence) compared with those with lower exposure. This scenario will be described in a directed acyclic graph demonstrating conditioning over a collider. In order for the bias to act, only two assumptions need to hold: (a) air pollution during pregnancy affects the chance of a live birth; and (b) another variable (U) exists that affects both the chance of a live birth and ASD. The second possibility is that air pollution selects into the analyzed sample (live-born children) fetuses that are less susceptible to developing ASD later on. For a paradoxical protective association to arise with an exposure like air pollution, the key factor is that air pollution must lead to fetal loss preferentially among fetuses who are more susceptible to ASD (because of any other factor, e.g. genetic factors). In this case, the proportion of susceptible fetuses selected into the analysis sample (live-born children that survive to the age of ASD diagnosis) will vary across air pollution levels, with lower relative proportion of susceptible fetuses at higher exposures.