a small intervention study in australia reports that 10 sessions of infant therapy can reduce the likelihood of developing autism by age three. in fact, all the statistics are weak and multiple uncorrected tests were done. ignore.  

original article: Whitehouse et al., 2021 (JAMA Pediatrics), reported in: The Week on 2nd October 2021

this story was in episode 19 #autism #treatment #therapy #cure

the error bar says

The Week, a magazine that I read every, er, week, reported on the 2nd October that infants who display signs of autism are 'far less likely to later be diagnosed as autistic' several years later if they receive ten sessions of social interaction therapy over five months.

the effect of this remarkable therapy was that, at age 3 years, 'only 6.7 percent' of the children receiving therapy were diagnosed with autism, while 'one in five' of the control group were.

does therapy prevent autism?

it might do, but the numbers of children involved in this study does not inspire confidence. autism refers to a range of developmental conditions that may affect around 1% of people.

when media stories report figures like '6.7 percent' of a population, we expect that there are at least 1000 people being studied - why else would they give us the percentage figures to the nearest tenth of a percent?

in fact, this '6.7 percent' is three children. & the 'one in five' figure is nine children. why doesn't The Week just say three and nine?

is three very different from nine? well that depends on how many children were treated. in the paper, published in JAMA Paediatrics, they started with 104 children aged 9 to 14 months who displayed potential signs of autism. the children were divided into two groups of 52, half were given 10 sessions of social interaction therapy plus the usual care; the other half received just the usual care.

assessments were carried out at 12, 18, 24 & 36 months of age. the authors' main measure was the severity of autism symptoms, but they also looked at other factors, including whether children were classified as autistic or not.

some children dropped out of the study, so the final sample was 45 children who had therapy & 44 [sic] who had not.

as a nerdy stats fan, three out of 45 & nine out of 44 look to me like a weak effect of treatment. in the paper, the statistical tests they used were all 'one-tailed' - that is, they were only looking for improvements in the treated group & were happy to ignore any negative effects of therapy. not a good idea where children are concerned. indeed, if you do the more justifiable 'two-tailed' tests, then there are no statistically significant improvements in the treated group.

i also ran my own statistical test, dear listener. the humble chi-squared test, which sounds like a form of group meditation, is a simple way to test whether there is an association between a particular group of people & a particular outcome. putting the numbers 42, 3, 35, & 9 into the test gave me the following result...

... not significant.

that's right - the 'far less likely' reported by The Week was, in fact, 'not less likely' when analysed over a cocktail at the error bar. looking across the paper as a whole, there were three assessment times, multiple outcome measures, no corrections for multiple comparisons & a parade of unusually weak 1-tailed p-values which error bar nerds will enjoy: .04, .04, .02, .02, .02, .02.


autism is a collection of conditions that neuroscience & psychology is, as yet, unable to explain or effectively intervene on. the importance of the topic makes clear that small, weak studies like this one should be ignored, for now.


The Week: fiction - no scientific story here