Risk Factors for Autism Among California Births 1992-2002: a within-Family Framework

Background:  Male sex and increased parental ages are widely established risk factors for autism. Associations between autism and pregnancy-related factors such as maternal parity and birth complications have also been reported, and indicate that sub-optimal prenatal environment may play a role in risk for the disorder. Variation in risk over subsequent pregnancies may implicate biological factors or environmental exposures that vary from one pregnancy to the next, and could help inform the search for modifiable risk factors for autism. However, the weight of evidence for these associations stem from study designs drawing comparisons between rather than within families, while the processes of interest are of relevance within women or families.

Objectives:  To examine associations between sociodemographic and perinatal factors and autism while controlling for potential confounding by family-level characteristics, by conducting within-family analyses.

Methods:  We identified sets of full singleton siblings from among all births in California between 1992 and 2002. Cases of autism were identified using California's Department of Developmental Services client files.  The association between autism and birth order and other covariates within families was investigated using conditional logistic and multi-level random effects models.

Results:  Second and later born children were at increased risk of autism relative to their first-born siblings. The association between birth order and autism diagnosis was time dependent, such that the increased risk occurred when interval between pregnancies was short (OR and 95% CI for second versus first born siblings= 2.15 (1.87, 2.47) for a 6 month versus 0.99 (0.87, 1.13) for a 24 month pregnancy interval, based on conditional logistic models). The first multi-level analysis to address this question, the results also suggest that increasing parental age plays a significant role in influencing risk between, rather than within, families; and that being male is associated with a higher manitude of association within families than between.

Conclusions:  Confounding by family-level factors influences both the magnitude and direction of observed associations for risk factors for autism, which may be relevant to causal inference about the mechanisms behind these associations. Epidemiologic studies of risk factors for autism should include within-family comparisons when feasible.