616 - An Aggregate and Deidentified Analysis of Real-World Performance of an Artificial Intelligence-based Diagnosis Aid for Autism Spectrum Disorder

Autism spectrum disorder (ASD) assessments are time-intensive, and families may wait up to 18 months between initial screening and final specialist diagnosis. The American Academy of Pediatrics recognizes that primary care pediatricians are critical in identifying, diagnosing, and managing ASD in children. Existing ASD diagnostic tools are used almost exclusively in specialty care settings and innovative diagnostic aids suitable for use in the primary care setting are not yet widely available. Canvas Dx is an FDA authorized software as medical device designed to support ASD diagnosis in children ages 18 through 72 months. The accuracy of Canvas Dx was established in comparison to blinded specialist diagnosis in a multisite, prospective, active comparator cohort study. Early post-market authorization prescription and output data is needed to shed light on device performance in real-world settings.

Objective:

Evaluate the performance of an FDA authorized software as medical device designed to support ASD diagnosis in a real-world setting.

Design/Methods:

A de-identified aggregate data analysis of the initial 95 Canvas Dx prescriptions fulfilled post-market authorization was conducted to evaluate the proportion of determinate device outputs (ASD Positive or ASD Negative); key prescriber characteristics; key patient characteristics. The study (PR015) received an IRB exemption from Advarra.

Results:

50.5% of children (48/95) (CI- 40.1%, 60.9%) received a determinate output (77.1% were ASD Positive and 22.9% were ASD Negative). Fulfilled prescriptions were generated from 61 unique prescribers across 13 states. The average age of children who received an ASD positive output was 36.8 months (13.2 months earlier than the current average age of ASD diagnosis in the United States). Confirmatory analysis of larger datasets is required; however, this finding is promising and highlights the potential of the device to support early diagnosis. While follow-up analysis powered for statistical inference on covariates is needed, there was no evidence of device performance inconsistency across patients’ sex. While girls only comprised 28.4% of the sample analyzed, 31.3% of children who received a determinate output were female.

Conclusion(s):

Early Canvas Dx prescription and output data analysis suggests device use is feasible in real world settings and could expand early primary care diagnostic capacity.