Time-to-event estimation of birth prevalence trends
In my book, No Place for Autism, I employ Reser’s Solitary Forager Hypothesis of Autism, the work being done on the current human genome, and the study of the humanity’s genetic past to posit that what we now call autism has been with humanity since at least the time when homo sapiens and neanderthal were slugging it out. Reser’s work suggests an even longer timeline for the presence of autism in our species.
A new paper on the subject both challenges the assertion, in a way, and offers a model to potentially prove it out - at least in the living population of autistics. In Time-to-event estimation of birth prevalence trends: A method to enable investigating the etiology of childhood disorders including autism (source), author Alexander G. MacInnis presents a novel model that transforms the idea of prevalence.
If scientists or other stakeholders want to know if more autistic kids are being born than in the past, how would they check? It's hard to tell if there are really more cases, if doctors are just getting better at noticing autism, or if there are market forces at work. The paper talks about a new method to figure this out called "time-to-event birth prevalence estimation" or TTEPE.
TTEPE uses math and statistics to separate the number of autism cases at birth from changes in how doctors diagnose it. It looks at the age when people are first diagnosed. It also accounts for updated definitions of autism over time.
The method treats diagnosing autism like a survival problem. The autism cases that haven't been found yet are like the people still “surviving.” TTEPE uses the age when people are diagnosed to estimate how many cases are “surviving” undiagnosed at each age.
By modeling diagnosis age and changes in diagnosing, TTEPE can estimate the actual number of autism cases being born each year. Computer simulations showed it gives accurate estimates adjusted for how doctors diagnose autism.
TTEPE may be a new way to finally answer the big question - are more autistic people being born than before?
What about looking backwards?
Consider my case. I was diagnosed in 2012. I’m in my fifties. Could the method be used to estimate the birth prevalence in 1965 … or 1945 for that matter? Based on the details in the paper, it seems TTEPE could potentially estimate autism birth prevalence backwards, with some caveats:
The method requires data on age of initial autism diagnosis over a range of birth cohorts. Such data may not exist going back that far.
Diagnostic criteria and factors likely changed dramatically between 1965 and today. The model may need to be adjusted to account for major historical changes.
Limited data points from a small number of older birth cohorts could make it difficult to accurately estimate eligibility and probability of diagnosis.
Data quality and population coverage further back in time may be lower.
So whilst TTEPE does not have an inherent limitation precluding estimates for older cohorts like 1965, the availability and quality of data may make it challenging to apply the method accurately that far back.
The paper shows through simulation that TTEPE works well with 20 recent cohorts. Applying it to older cohorts may require some modifications and assumptions to account for changing conditions. But if good data on initial diagnoses by age exist going backwards, TTEPE could potentially provide autism birth prevalence estimates, though with greater uncertainty than more recent cohorts.