AIHealthcare Analytics

Overview

Research published in npj Digital Medicine explores using large language models to scale medical device regulatory science—applying AI to automate or augment the compliance and regulatory review processes that govern AI medical device approval. This represents a meta-application of AI: using AI to regulate AI.

Research Approach

The research examines how LLMs can assist with: parsing and interpreting regulatory guidance documents, reviewing device submissions for completeness and compliance, identifying precedents in prior device authorizations, and generating regulatory documentation. These applications could significantly reduce the time and cost of regulatory review for both industry and the FDA.

Potential Impact

If LLMs can reliably assist with regulatory science tasks, the FDA could potentially process more device submissions faster, reducing the backlog that slows AI medical device market entry. Industry sponsors could use LLM tools to improve submission quality, reducing rejection rates and review cycles.

Limitations and Risks

LLM application to regulatory science introduces risks: hallucinated regulatory precedents, misinterpretation of guidance, and over-reliance on AI review that misses novel safety issues. The same validation challenges that apply to clinical AI apply to regulatory AI—and errors in regulatory AI could have downstream patient safety consequences if they contribute to approval of unsafe devices.