A Precedent for Regulating Clinical AI Models

For decades, academic hospitals developed and ran their own diagnostic tests without any regulatory review. A test built and used inside a single laboratory (laboratory developed test [LDT]) was generally regarded as a clinical service rather than a medical device and, therefore, outside the purview of the US Food and Drug Administration (FDA). Clinical artificial intelligence (AI) is now developing along the same lines. Academic medical centers are training AI models using their own electronic health records and incorporating these tools into existing workflows, in which the AI is used only inside the single academic system where it was developed. The FDA accommodation that allowed LDTs to flourish accelerated some local clinical innovation but sometimes kept these advances from reaching the broader marketplace. Clinical AI faces a similar crossroad.