Hacker News new | ask | show | jobs
by robrenaud 2924 days ago
Statistical reporting is often pretty frustrating.

> “We are able to predict with 88 percent accuracy whether children have autism,” said Juergen Hahn, lead author

> It is estimated that approximately 1.7 percent of all children are diagnosed with ASD

    def has_autism(child):
      return False
apparently has 98% accuracy.
2 comments
icegreentea2 2924 days ago
It's 88% correct classification on a validation data set consisting of 154 patients diagnosed with ASD. Their validation dataset did not contain a "typically developing" population since there's only one dataset for that data and it was consumed in their model generation. Their training data resulted in a 5% misclassification rate of their controls - that is 5% of patients that were not diagnosed with ASD in current tests would be classified as having ASD in this scheme.

The full paper is here: https://onlinelibrary.wiley.com/doi/10.1002/btm2.10095

link
perl4ever 2924 days ago
If you give the test to everyone, your results are:

~4.9% incorrectly diagnosed with autism (0.05 x .983)

~1.5% correctly diagnosed with autism (0.88 x 0.017)

Thus, 76% of those diagnosed would be diagnosed incorrectly.

link
alexandercrohde 2924 days ago
That's true in a statistical sense. Or it's possible that the human-observers are less accurate about autism than this brain scan.

It's unclear if those who have the odd brain structure but aren't diagnosed are entirely asymptomatic.

Also they may be non-autistic in spite of this brain pattern, but nonetheless have a genetic predisposition to pass it on. Big step.

link
perl4ever 2924 days ago
It's not a genetic test, and it isn't a test for something that defines autism, so you're begging the question IMO. Sure, human observers are imperfect, but that doesn't mean you can choose an arbitrary test that is precise and make it the standard just because. "We need something better than human judgement, and this is something, thus it is better"
link
vorotato 2924 days ago
I would hope that there's more rigor than that. I mean if you're willing to be that dishonest with your sample selection you could select only people who don't have autism and get 100% accuracy.
link