Advantages of combining multiple eye-tracking paradigms for distinguishing young autistic from non-autistic children.
Dan Xu, Lan Zhang, Xiangqin Wang and 10 others
Combining five eye-tracking paradigms identified autistic from non-autistic children with 90% accuracy, outperforming any single paradigm.
Each paradigm captured distinct attentional patterns, and the combined model showed net clinical benefit at a 5% threshold.
Key findings
1The combined five-paradigm model reached 90.0% accuracy and 95.3% AUC, significantly outperforming every single-paradigm model (best single: SI at 0.87 AUC).
2Across all five paradigms, autistic children showed reduced attention to social stimuli (15.0% vs. 28.5% TFD% on eyes; 35.5% vs. 61.7% on interacting parents) and increased attention to non-social stimuli (61.1% vs. 43.7% on spinning objects; 42.5% vs. 67.1% on dancing children).
3Cross-paradigm dissimilarity was significant (p < 0.001) and remained significant after removing group mean differences, so each paradigm captures information the others do not.
Still to come
How it was doneWhat they found
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Sex was imbalanced between groups (64 of 77 autistic children were male vs. 41 of 63 non-autistic), and developmental quotient scores differed significantly across all five Gesell subdomains, so some observed effects may reflect developmental level rather than autism-specific attention. Validation was internal only (cross-validation within the same sample). The authors state the model requires testing in an independent, larger cohort before any clinical use. The decision-curve analysis used assumed clinic prevalence rates (10%, 15%, 20%) rather than actual referral data, so the net-benefit figures are projections, not measured outcomes. The three eye-tracking subgroups are explicitly described as exploratory and hypothesis-generating; their stability and clinical utility have not been tested in any independent or longitudinal sample. Single-site study; the authors note that portability across laboratories and specificity relative to other neurodevelopmental conditions remain untested.
Declared interests
Funded by the National Natural Science Foundation of China, the Yingcai Scheme at Chengdu Women's and Children's Central Hospital, a Guangdong Province key technological project, and a Sichuan Province key R&D project. No industry sponsor is named and no author declares a competing interest in the supplied text.
The easy way to misread this
Do not read the 90% accuracy as a ready-to-use clinical test. The model was validated only within this single sample using cross-validation, the three subgroups are explicitly hypothesis-generating, and the decision-curve analysis relies on assumed clinic prevalence rather than real-world deployment.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →