Another idea for solving the puzzle of autism: put DOGE on it

Big data and AI may make progress where traditional science has failed

Image: Bret Baier of Fox News interviews members of the DOGE team in March 2025.

The vast epidemic of autism began in the late 1980s/early 1990s and has escalated without interruption — and without answers — through today.

We have poured more than a billion dollars into causation research, mainly in the realms of genetics and fetal exposures, but have precious little to show for it. Sure, robust genetics studies on tens of thousands of cases have revealed that about 10% or so of cases have a genetic origin. And on the environmental side, we see that some exposures and stressors raise risk for autism, such as premature birth and maternal drugs such as valproic acid (an anticonvulsant). But the great majority of cases, probably about 80%, remain “idiopathic,” meaning cause unknown.

Bottom line: autism rates keep growing and people have not a clue what to do to limit risks for their future children.

Given the glaring failure of research to date, the last thing we need is more of the same. We don’t need more repetitive genomics studies looks for more (non-existent) genes or more epidemiological studies finding that air pollution raises autism risk by a tiny and very questionable fraction.

So let’s try another approach. Think DOGE. Think massive computing. Think outside of the traditional laboratory box. Now, I’m not suggesting that the actual DOGE guys pursue this, but I am suggesting that people with those sorts of computational talents take a stab at this intractable mystery.

But how? Many systems and countries have huge medical and insurance databases and population health registries. Israel is a good example of a country where most all citizens have their key medical information in huge insurance databases. Denmark and Sweden are examples of countries with population-wide health registries.

Now this exercise would be limited to one input and one endpoint, consistent with the findings that autism is highly heritable and that fetal exposures generally pose a low risk for offspring autism. The input would be the lifetime medical procedures and medications given to parents prior to conception of their last child. The endpoint would be the neurodevelopment outcomes of the offspring, including autism, ADHD and others. It would be a way of probing exposures to the parents that could influence the molecular program of their germline, relative to the brain development of their children.

For example, using big data we could probe some of the following questions:
—Does a parent’s early life history of surgery affect risk for neurodevelopmental disorders in the offspring?
—Does anesthesia used in IVF egg retrieval affect risk for autism in the offspring?
—Does paternal use of certain drugs or medications influence risk for autism in subsequent offspring?

There have already been some studies trying big data techniques in autism. For example, Rhzetsky et al. in “Environmental and State-Level Regulatory Factors Affect the Incidence of Autism and Intellectual Disability” (2014) used an insurance claims dataset covering nearly one-third of the US population. They used environmental, phenotypic, socioeconomic and state policy factors to analyze spatial incidence patterns of ASD and intellectual disability (ID). They found ASD incidence rates were strongly linked congenital malformations of the reproductive system in males but that such malformations were barely significant for ID, suggesting that endocrine disruption in the parents or fetus was associated with ASD risk.

Similarly, Ejlskov et al, in “Prediction of Autism Risk From Family Medical History Data Using Machine Learning: A National Cohort Study From Denmark” (2021) found that parents’ mental and non-mental medical conditions increased autism risk in offspring, suggesting either genetic or environmentally-induced impacts on the heritability of autism.

Another study using a massive dataset was published recently by Grosvenor et al, “Autism Diagnosis Among US Children and Adults, 2011-2022” (2024). Using data from across 12 U.S. hospital systems and about 12 million patients it found rapidly rising prevalence of autism from across more than five decades of birth cohorts. It did not look for associations between past exposures, eg, parent exposures, and offspring outcomes, but there’s no reason a massive medical records resource like this could not be used for such purposes. All we need is the data-mining technical expertise.

While big data has not yet found any smoking gun of autism causation, the studies I just discussed were not calculated to do so. But they do demonstrate some feasibility of asking a variety of questions about factors that may correlate with autism with a big data approach. Beyond knowing a bit about heritability and basics of biological plausibility the data researchers need not be experts in biology. Rather, they need the skills to seek out data patterns and relationships, across time and across two generations, that could highlight risks that are now only poorly understood.

Those findings could be followed up with more meticulously conducted studies.

Why not? Anything’s better than the dead ends we’ve been staring down for decades. Let’s DOGE autism.

Jill Escher is an autism research philanthropist and advocate. Read more the topic of investigating the roots of autism at JillEscher.com.