Profiles of autism characteristics in thirteen genetic syndromes: a machine learning approach.
Natali Bozhilova, Alice Welham, Dawn Adams and 14 others
PMID 36639821WHAT IT FOUND
Autism profiles on the Social Communication Questionnaire differed across thirteen genetic syndromes, but many overlapped enough that a model assigned 55% of people to the correct syndrome group.
It is not a diagnostic test.
Key findings
01Using lifetime Social Communication Questionnaire responses, a machine learning model classified 55% of individuals with genetic syndromes into the correct syndrome group.
02When an autistic group without a genetic syndrome was added, individuals with fragile X syndrome were more likely to be misclassified into that group, and vice versa, despite high classification accuracy for both groups.
03Items about social interaction quality, imaginative play and complicated body movements contributed most to classification across a large proportion of syndromes, while communication items contributed substantially for fewer groups.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
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
The Social Communication Questionnaire is a screening tool, not a full diagnostic assessment. The data were retrospective parent and carer reports, and recall for behaviour at ages four to five may be biased for older participants. Genetic syndrome diagnoses were reported by professionals, but genetic confirmation was not always available. Autism diagnoses were made by different professionals, and some may not have had specialist autism training. Group sizes varied, and smaller groups may have lower accuracy and less generalisability. The machine learning model does not show which specific features caused each classification. Language ability affected item scoring, because no language use was coded separately from absence of autistic characteristics. The study did not evaluate cognitive impairment directly, only self-help skills as an estimate. The study reports profile classification, not treatment effects or clinical outcomes.
Declared interests
The work was supported by Cerebra, the Cornelia de Lange Syndrome Foundation UK and Ireland, Research Autism, Newlife the Charity for Disabled Children and the Baily Thomas Charitable Fund.
The easy way to misread this
Do not use the 55% classification accuracy as evidence that autism can be diagnosed from a Social Communication Questionnaire profile. The model was built from research questionnaire data, many syndromes were misclassified into specific other syndromes, and the Social Communication Questionnaire is a screening tool rather than a full diagnostic assessment.