Abstract
The FDA should create functional regulations for the growing number of machine learning medical devices. The healthcare system is increasingly using these devices for diagnosis. Machine learning devices trained on biased data sets are susceptible to furthering certain types of bias and generating flawed outcomes. The FDA should require ML medical devices to include a label that describes the demographics of the tested population. If manufacturers fail to include this information, the FDA could determine the label false or misleading under §502 of the FD&C Act and stop sales of the device. After approval, the FDA should use §814.89(2) and §519 to require manufacturers to report and evaluate the real-world performance of their approved devices. These reviews should include studies for clinical validation or model evaluation and model testing. While addressing bias in diagnostic medical machine learning devices will take more than the FDA, the agency should continue to support efforts to find an effective way to mitigate and measure bias.
Recommended Citation
Charli Beam,
Machine Learning-Based Medical Devices: the FDA’s Regulation, Requirements, and Restrictions,
35 J.L. & Health
419
(2022)
available at https://engagedscholarship.csuohio.edu/jlh/vol35/iss3/5