With UpDoc clearance, FDA sets path for care augmented by AI

The U.S. Food and Drug Administration has granted clearance to an artificial intelligence agent, and reading the tea leaves offers hints for how the FDA intends to regulate the growing market for AI in clinical and wellness applications. 

UpDoc, a startup based at Stanford Medicine, last month announced FDA 510 (k) clearance for software as a medical device for patient-facing large language models (LLMs). The company aims to address a longstanding problem – coordinating care between in-person visits – with an AI agent capable of adjusting medication dosing, ordering lab tests, or otherwise checking in with patients.  

Specifically, noted the Nixon Law Group, the FDA clearance supports prescription management and medication monitoring for patients with Type 2 diabetes. A 2023 JAMA Network Open paper found the technology “significantly improved time to optimal insulin dose, insulin adherence, glycemic control, and diabetes-related emotional distress” compared to the standard of care. (UpDoc was founded three months after the clinical trial described in the paper was completed.) 

Clinical AI, not an AI doctor

In a LinkedIn post, UpDoc CEO Dr. Sharif Vakili explicitly said the FDA-cleared technology isn’t an “AI doctor,” adding that “reducing the practice of medicine to a chatbot fundamentally misunderstands and undercuts what physicians do every day.” Instead, he referred to the technology as “clinical AI” capable of augmenting a doctor’s care and ensuring care continuity.  

In a Healthcare Innovation interview, Vakili framed the technology’s role in the context of remote patient intervention. This care model takes remote monitoring a step further to “close the loop” that otherwise have required an in-person clinical encounter. Within the scope of UpDoc’s 510(k) clearance, the closed loop is an insulin dose adjusted within parameters set by human clinicians.  

As Vakili described, the physician creates a treatment plan, then directs the clinical AI to implement it, communicate with the patient, and recommend dose adjustments or follow-up tests as indicated in the treatment plan. By incrementally adjusting doses, rather than waiting months for the next available in-person appointment, patients can “continue that therapeutic momentum” and improve outcomes, he said in the interview. 

A blueprint for regulation – and development

According to the Nixon Law Group, UpDoc’s FDA clearance may provide a “regulatory blueprint for clinical AI” that need not require new regulatory processes. (This is distinctly different from software for general wellness, which the FDA said can make claims about sustaining or improving health provided there’s no reference to specific diseases or conditions.) 

For starters, UpDoc’s clearance was granted to a specific indication, which remains consistent with the FDA’s approach to software, device, and drug approval. In addition, LLM outputs are based on previously determined clinical decisions; the model itself isn’t deciding whether to change an insulin dose. 

The law firm also said UpDoc’s path to clearance provides a model for digital health developers. Separating the LLM’s interface from the back-end “clinical decision-making engine” seems to ease the path to validation. However, organizations must pay close attention to product and user interface design to ensure they align with their intended FDA regulatory classification. 

To that end, Vakili told Healthcare Innovation he sees potential for UpDoc and other clinical AI tools for many types of chronic disease management, not just Type 2 diabetes. Serial entrepreneur, investor, and asthma patient Halle Tecco agreed. Reacting to the UpDoc announcement on LinkedIn, she pointed to the potential to feed biometric and environmental data to an AI assistant empowered to dynamically adjust medication doses – and help her avoid asthma attacks and preventable emergency room visits. 

Brian Eastwood is a Boston-based writer with more than 10 years of experience covering healthcare IT and healthcare delivery. He also writes about enterprise IT, consumer technology, and corporate leadership.