ESTIMATED AUTISM RISK, OLDER REPRODUCTIVE AGE, AND PARAMETERIZATION

In their recent article, King et al.¹ seemed to dismiss the conclusion supported by previous epidemiologic studies (including 1 authored by us)^2,3 that increasing maternal age and paternal age are both independently associated with autism risk. Based on their analysis of California Department of Disabilities Services (DDS) data, King et al. argued that the previously observed paternal age effect is an artifact of pooling data across successive birth cohorts during a period in which both the prevalence of children receiving services for autism and the proportion of births to older parents have increased, essentially asserting that observed paternal age effects are confounded by birth cohort effects. Although King et al. acknowledge that our study,² based on a single birth year showing significant independent effects of both parents' ages on autism risk, is not subject to this artifact and, although the other study they reference found a paternal age effect after controlling for birth year,³ they still conclude that “analyses that do not suffer from the statistical fallacies produced by pooling demonstrate that advanced maternal age, rather than paternal age, may pose a greater risk.”^1(p1678)

Moreover, much of King et al.'s own analysis of DDS data does not support a conclusion that confounding by birth cohort explains the paternal age effect. The relative risks that they calculated show similar-magnitude effects for fathers aged 40 years and older in each birth cohort (falling between 1.29 and 1.71). These do not differ drastically from the pooled estimate, suggesting no confounding. It is only when King et al. move to estimating age effects on a continuous scale that they show birth cohort-specific age effects that are consistently stronger for mothers than they are for fathers.¹ Consequently, any shift in thinking about which parental age effect might be stronger in these data appears driven by the choice of parameterization of the parental age terms, and not confounding by birth cohort. Using a continuous parameterization assumes that a fixed increment change in parental age beginning anywhere in the age range leads to the same percent change in autism risk. This approach diminishes the ability to detect threshold effects and will force the estimated percent change in risk for, say, a parent aged 35 years compared with a parent aged 25 years to be the same as that for a parent aged 45 years compared with a parent aged 35 years. The question of what parameterization—continuous or categorical—is more appropriate for understanding the true role of parental age in autism is legitimate and, by all indications, seems more important in these DDS data than does the issue of confounding by birth cohort, or reversal paradox, that King et al. make much of in their article.