SLPOtherJournal of autism and developmental disorders2024

Machine Learning Differentiation of Autism Spectrum Sub-Classifications.

R Thapa, A Garikipati, M Ciobanu and 7 others

PMID 37751097

WHAT IT FOUND

A machine learning model using routine questionnaire and demographic data classified autism spectrum subtypes with high accuracy.

It correctly identified 80.5% of cases in the main dataset, though positive predictive value for PDD-NOS was low at 0.221.

Key findings

01The model achieved an AUROC of 0.980 for identifying non-spectrum individuals and 0.932 for autistic disorder in the primary dataset.

02Positive predictive value for PDD-NOS was low (0.221) in the primary testing dataset, indicating many false positives for this specific subtype.

03The model correctly classified 5,951 individuals (80.5%) in the primary testing dataset, with most errors occurring between autism subtypes rather than misclassifying autistic individuals as non-spectrum.

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.

Already have one?

What it does not show

The study used retrospective data, so real-world clinical performance is unknown. The model was trained on DSM-IV criteria, which are no longer the standard of care (DSM-5 is used), limiting direct applicability to current diagnoses. Positive predictive value for PDD-NOS was very low (0.221), meaning the tool generates many false positives for this group. The datasets did not allow for cross-validation between SPARK and ABIDE due to differing feature sets. SPARK data did not include DSM-5 diagnoses, so concordance with current standards could not be fully assessed.

Declared interests

No specific funding sources or conflicts of interest were declared in the provided text.

The easy way to misread this

Do not use this model for diagnosis. It was trained on outdated DSM-IV criteria and has a high false-positive rate for PDD-NOS (PPV 0.221). Furthermore, it has never been tested in a prospective clinical setting.

Read it on PubMed →