Primary Submission Category: Applications in Health and Biology
A Protocol for Comparing the Causal Impact of Pre- and Perinatal Factors on Autism
Authors: Rachel Hanger, Amy Cochran, Olivia Pokoski, Sarah Furnier, Maureen Durkin, Camara Gregory, Dadit Hidayat,
Presenting Author: Rachel Hanger*
Over the past several decades, autism has been pushed to the forefront of the medical zeitgeist. Recent government policy emphasizes identification of contributing factors to autism through efforts such as the Autism CARES Act of 2024. However, it is both hard to identify such factors and measure their impact due to limitations of observational studies. An autistic child cannot have a perinatal factor retroactively changed and then be reevaluated for autism. In this work, we present a protocol for analyzing case-control data from a multi-site study of autism. We use Bayesian Additive Regression Trees (BARTs) to calculate two important metrics for candidate pre- and perinatal factors: the individual causal relative risk (the relative change in an individual’s chance of having autism if one factor’s value is changed) and the population causal relative risk (the relative change in the prevalence of autism in general if the entire population had the same factor value changed). We carefully justify the assumptions required to identify these causal effects from case-control data. In addition, we describe how to build on these analyses so that we can attribute specific percentages of the likelihood of having autism to each factor in both individualized and generalized public health contexts, and present a guideline for future applications and analyses, medical or not.