The authors of the report, Jayson Marwaha, MD, and Joseph Kvedar, MD, mention an October 2021 study that analyzed 65 trials that evaluated the clinical benefits of AI. The study found that there was no clinical benefit to the AI tools in 40 percent of cases and that machine learning had no benefit over traditional statistical risk calculators, despite most of the tools performing well during development and validation.
Clinical effectiveness may be held back by lack of provider trust, interoperability issues and poor implementation, Drs. Marwaha and Kvedar argue.
Instead, they argue that when AI is integrated into systems, it should be paired with real-world interventions and workflows to be successful. They also suggest that all AI researchers should use real-world data to inform the efficacy of their technology, as does the FDA.
