Researchers have found limitations in the FDA evaluations of medical AI devices. The evaluations are often retrospective and not conducted in multiple clinical sites, leading to concerns about the effectiveness, reliability, and safety of these devices.
In a review published in Nature, researchers emphasized the need for FDA evaluations to include prospective studies and assessments in multiple clinical sites. With the increasing use of AI algorithms in healthcare, there is a growing need for industry-wide standards and safeguards to ensure patient safety.
While the FDA has taken steps to improve its oversight of AI medical software, researchers found that most approved AI devices underwent only retrospective studies. This raises questions about the quality of test data, transparency, bias, and algorithm monitoring in practice.
The review of 130 AI devices approved by the FDA showed that the majority of devices were evaluated through retrospective studies, with none of the high-risk devices undergoing prospective studies. This lack of prospective studies hinders the full characterization of the impact of AI decision tools on clinical practice.
Moreover, researchers found that a significant number of approved devices were evaluated at a small number of sites, limiting the geographic diversity of the evaluations. This lack of multi-site assessments can impact the understanding of algorithmic bias and reliability, as well as the generalizability of the devices across different populations.
To address these limitations, researchers emphasized the importance of conducting prospective studies with comparison to standard care, as well as implementing post-market surveillance of AI devices to monitor unintended outcomes and biases. By evaluating AI devices in multiple clinical sites and ensuring transparency in the evaluation process, the FDA can improve the reliability and safety of medical AI devices.
In conclusion, overcoming the limitations in FDA evaluations is crucial for the successful integration of AI into healthcare delivery and improving patient outcomes. By encouraging multi-site evaluations and post-market surveillance, the FDA can enhance the effectiveness and safety of medical AI devices for the benefit of patients and healthcare providers.