Current Standards of Monitoring Models in Healthcare Settings

The amount of FDA‐approved artificial intelligence and machine learning devices in clinical applications lags drastically behind the vast literature on novel methodologies being developed for healthcare use. Only 3.4% of these approved models have associated predetermined plans for necessary updates. Other regional regulatory organizations, despite most often being decentralized and not as easy to track, do not appear to cover the remaining deficit.The increasingly popular operationalization of deep learning methods in other fields does not seem to be reflected to the same proportion within the medical field. Based on the high‐risk scenarios of medicine, the limited trust and verifiability for most devices are likely to be major obstacles that help explain this gap. Monitoring tools can be implemented to analyze model outcomes by providing a means to reflect on performance and verify results. In this paper, we discuss the current frameworks surrounding device monitoring and highlight what challenges are presented when incorporating model monitoring in the healthcare setting.