An analysis of the real world performance of an artificial intelligence based autism diagnostic
Abstract Rapidly rising demand for pediatric autism evaluations has outpaced specialist capacity and created a crisis of delayed diagnoses and treatment. Streamlining the diagnostic process could reduce wait times and optimize use of limited specialist resources. Following strong clinical trial resu...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-08-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-15575-8 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Rapidly rising demand for pediatric autism evaluations has outpaced specialist capacity and created a crisis of delayed diagnoses and treatment. Streamlining the diagnostic process could reduce wait times and optimize use of limited specialist resources. Following strong clinical trial results, Canvas Dx, an AI-based diagnostic, was FDA authorized to support accurate diagnosis or rule-out of autism in children 18–72 months with caregiver or healthcare provider concern for developmental delay. To gain insight into real-world device performance, a de-identified aggregate data analysis of the initial 254 Canvas Dx prescriptions fulfilled post-market authorization was conducted to determine: accuracy of autism predictions compared to clinical reference standard diagnosis and prior clinical trial data, key real-world prescriber and patient characteristics, proportion of determinate device outputs (positive or negative for autism) and impact of decision threshold settings on device performance. In this sample of 254 children with a 54.7% autism prevalence rate (29.1% female, average age 39.99 months), Canvas Dx had a NPV of 97.6% (CI- 92.8% -100.0%) and a PPV of 92.4% (CI-87.7%-97.2%). A majority of cases (63.0%) received a determinate result. Sensitivity and specificity of determinate results were 99.1% (CI-97.3%-100.0%) and 81.6% (CI-70.8%-92.5%) respectively. The median age of children who received a positive for autism output was 37.2 months, which is more than 2 years earlier than the current median age of autism diagnosis. No performance differences were noted based on patients’ sex. Compared to clinical trial results, real world performance was equivalent for all key metrics, with the exception of the determinate rate and the PPV which were significantly improved in real world performance. Analysis of real-world Canvas Dx data highlights its feasibility and utility in supporting accurate, equitable and early diagnosis or rule out of autism. With medical coverage and broader clinical adoption, innovative solutions such as Canvas Dx can play an important role in helping to address the growing specialist waitlist crisis, ensuring that more children gain access to targeted therapies during the critical window of neurodevelopment where they have the greatest life-changing impact. |
|---|---|
| ISSN: | 2045-2322 |