An existing network of clinical laboratories could be used to regulate artificial intelligence in health care, according to a research paper in NPJ Digital Medicine and another in the Journal of the American Medicine Informatics Association. How so? The Food and Drug Administration approves tests for diagnosing disease and illness. The Centers for Medicare and Medicaid Services with input from the FDA and the Centers for Disease Control and Prevention ensures those tests are used safely under a 1988 law establishing lab oversight standards. Labs overseen by the agencies could also help regulate not only FDA-regulated AI, but tools outside the FDA's scope like the AI doctors use to take notes, the authors say. How it could work for AI: Clinical AI Ops units, as the researchers call them, could validate AI in health systems where they’re deployed, evaluating data quality, modeling performance, running risk assessments and ensuring AI is tailored to specific patient populations. This method allows flexibility to validate both discrete algorithms and whole platforms, the NPJ Digital Medicine authors write. That’s important, since some health tech developers sell whole platforms on which they can design custom tools for specific providers. One way to vet an algorithm, the Journal of the American Medicine Informatics Association paper suggests, would be to send a batch of fake patient cases to multiple sites using a specific AI model and compare the results across sites. Another bonus of using the lab model for AI: Labs and services would be available to all health systems regardless of size or resources. What’s next: Federally regulated labs would have standardized performance metrics and testing protocols. Although there isn’t agreement yet on what those standards should be, labs could potentially adapt labs’ existing testing framework. The lab model isn’t perfect — it failed to root out Theranos’ fraudulent blood tests. Most of the company’s products were lab-developed tests, which are self-validated, bypassing FDA oversight. But researchers think there’s an opportunity to learn from those lessons and create a new federally regulated AI lab model to ensure local health systems use AI safely.
|
No comments:
Post a Comment