AI-supported early autism identification as public health infrastructure: economic implications for the United States...

Artificial intelligence (AI)-supported tools may shorten autism spectrum disorder (ASD) identification pathways, but their public health value depends on whether earlier identification is followed by timely intervention.

This study developed a calibrated, scenario-based cost-of-illness model to examine how AI-supported early autism identification could affect the timing and magnitude of US societal ASD-related costs under explicitly stated assumptions. The model reconstructed a business-as-usual baseline from 2011 cost cells, calibrated annual per-person cost growth to reproduce a published 2025 national burden benchmark of approximately $461 billion, and projected costs from 2025 to 2050 using a cohort-based stock model. AI-supported tools were represented as accelerators within screening, triage, referral, and clinician-led diagnostic pathways.

Under the Base deployment scenario, cumulative discounted net cost remained positive through 2050 at approximately $52 billion, while annual net costs declined sharply toward zero. The High deployment scenario reached annual net savings by the mid-2040s. Extended-horizon analysis indicated that cumulative fiscal payback may emerge after the primary 25-year policy window. Prevalence, per-person cost growth, and population growth mainly scaled national dollar totals; discounting changed the present value assigned to delayed savings; and the adult trajectory-shift assumption strongly influenced payback timing.

These findings are conditional projections, not estimates of realized savings from observed AI deployment. Their interpretation depends on timely intervention access, adequate service capacity, equitable implementation, and durable reductions in downstream support needs. Calibration provides a transparent published baseline for scenario analysis; it does not validate future projections.