Sunday, April 22, 2012

Rapid online diagnosis of ASD

The diagnosis of ASD can be complex, subjective and time-consuming. Spending large amounts of time at specialised clinics, often far from home, can be anxious. The current resources available for a diagnosis of ASD mean can cause a considerable delay between the onset of symptoms and receiving a diagnosis. A group at Harvard has addressed this problem by creating a computer program that seeks a diagnosis using the least number of interview questions. Taking the 93 typical ASD diagnostic questions, the computer program could match the traditional ASD diagnosis in 94% of cases using only 7 questions.
Such a means of diagnosis has been proposed as a screening tool, and Harvard continues to test this approach online (video and survey). If administered by first tier health care service providers such as GPs, referrals could then be made to specialists. It is unlikely that the 7 questions will be enough for a full diagnosis. A full examination by specialists will be required in any case before intervention can commence.
The full paper can be found here
The group's facebook page is here

It remains to be seen if the software can distinguish ASD from ADHD, or another developmental disorder or challenging behaviour. It would also be nice to think that international roll-out of the software, or online international access is not impeded by patent rights or prohibitive costs.


1 comment:

  1. It's an interesting idea. But a few things to bear in mind:

    The press release talks about the ADI but the paper is only about the ADOS.

    The claims based on the ADOS data have been massively oversold. They reduced the number of items on the ADOS by only looked at the most clear-cut cases (those with an ADOS diagnosis of "autism", excluding anyone with an "autism spectrum" diagnosis). We have to assume that many more items are required to identify individuals on the milder end of the spectrum. Even then, it doesn't do much for shortening the ADOS, because they still have to do 8 out of 10 activities to get all the items they need.

    Given the above, I'm waiting for the ADI-R data to be published before I get too excited. That said, the ADI-R is massively redundant. It takes about 3 hours and most of the items don't go into the diagnostic algorithm. We don't need a fancy machine learning algorithm to tell us this, but if it's what it takes to get people to change the ADI-R, I'm all in favour!