Development and Validation of a Short Version Eye-Tracking Paradigm for the Screening and Diagnosis of Autism Spectrum Disorder in Qatar.

The children were compared after diagnosis rather than screened before it: one group already had autism and the controls were mostly typically developing, which is the design most likely to make a test look more accurate than it will be in a real clinic. The authors themselves say the estimates probably represent upper-bound performance. Children who could not complete the task were excluded, so children with more marked behavioural or attentional difficulties are not represented. Only 16 children with other developmental delays were included, and they were pooled with the typically developing children, so the tool's ability to tell autism apart from other developmental conditions is thinly tested. Some controls were brothers and sisters of autistic children, which the authors list as a methodological limitation. The social and non-social clips differed in basic visual properties such as colour and contrast, so attention differences could partly reflect those rather than social content. The test-retest check rested on only 19 children, and the authors call it exploratory. The age range was 1 to 16 years, deliberately broad, and the authors say this developmental heterogeneity needs age-specific confirmation. The comparison with the older 10-minute version is descriptive only, because the two versions were tested in different groups of children. Children with autism had lower-quality eye-tracking recordings than controls, so the two groups were not compared on identical data.

Funded by the Qatar Biomedical Research Institute under grant SP001172; the study was reviewed and approved by the institutional review board of Hamad Bin Khalifa University. The authors declare no conflicts of interest.

Do not read the high accuracy as proof this test can screen children in the community. The autistic children already had a diagnosis and the controls were mostly typically developing, which is the setup most likely to make a test look accurate, and the authors say the estimates are probably an upper bound. No cut-off score was fixed in advance either, so there is no agreed threshold for calling a result positive.