Remote Machine-Learning-Assisted Diagnostics, Autism, and Attention-Deficit/Hyperactivity Disorder: The Challenges of Technology and Regulation

Abstract

In the next decade, the evolution in diagnostic medicine for Autism Spectrum Disorder (ASD) and Attention-Deficit/Hyperactivity Disorder (ADHD) will prove to be as monumental as the evolution from radiography to cross-sectional imaging was to radiology in the 1970’s. Clinical assessments delivered through telemedicine, digital screeners built on mobile platforms, and behavioral analyses constructed by machine learning have been introduced into clinical practice at an accelerated pace, leaving the reimbursement and regulatory infrastructure struggling to keep up. This thesis interrogates that infrastructure. I ask whether the intersection between scientific validation, Food and Drug Administration (FDA) authorization, and Centers for Medicare and Medicaid Services (CMS) coverage policy is sufficient to ensure the safe, effective, and equitable provision of digital diagnostics for neurodevelopmental disorders to children in the U.S. This thesis is divided into eight chapters. Chapter 1 outlines the problem. Chapter 2 reviews the history of ASD and ADHD diagnostic practice from the mid-20th century to DSM-5-TR. Chapter 3 dives into the clinical evidence for current digital diagnostic tools and the importance of interpreting sensitivity, specificity, and predictive values with real-world prevalence in mind. Chapters 4 and 5 serve as case studies for recent regulation featuring the new category: Cognoa’s Canvas Dx, the first AI-assisted diagnostic tool for ASD to receive De Novo classification from the FDA, and EarliTec Diagnostics’ EarliPoint Evaluation, the first 510(k)-cleared device in the category. Chapter 7 reviews FDA regulation for software as a medical device and its history. Chapter 8 examines reimbursement considerations for digital diagnostic tools. The final chapter outlines recommendations for the FDA, CMS, and Congress. I argue that the scientific and regulatory infrastructure exists for AI-powered diagnostic tools in children’s neurodevelopment, but the reimbursement and surveillance infrastructure does not. The consequence of this results in validated diagnostic aids that clear FDA authorization can still fail to reach the children demographic for whom they were built while many unvalidated products can reach families under this guise of wellness or information tools. This thesis proposes an argument grounded in evidence on the gaps in U.S. policy and changes in regulation and reimbursement that could close those gaps.