Nearly half of the AI-based medical devices approved by the US Food and Drug Administration (FDA) have not been trained on real patient data, according to a new study.
The study, published in Nature Medicine, finds that 226 of the 521 devices authorised by the FDA lack published clinical validation data.
“Although AI device manufacturers boast of the credibility of their technology with FDA authorisation, clearance does not mean that the devices have been properly evaluated for clinical effectiveness using real patient data,” says first author Sammy Chouffani El Fassi, an MD candidate at the University of North Carolina, USA.
“With these findings, we hope to encourage the FDA and industry to boost the credibility of device authorisation by conducting clinical validation studies on these technologies and making the results of such studies publicly available.”
Artificial intelligence-based tools have proliferated in medicine in recent years, particularly in diagnosis and prognosis.
Researchers have used machine learning tools to help with diagnosing and predicting a huge range of conditions, including endometriosis, lung and prostate cancer, vertigo, and rare diseases.
As a general rule, these tools are machine-learning programs that have been trained on patient datasets, so their algorithms can provide solutions to new problems when they occur.
In this study, the US team of researchers examined the FDA’s official “Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices” database.
“Using these hundreds of devices in this database, we wanted to determine what it really means for an AI medical device to be FDA-authorised,” says Professor Gail Henderson, a researcher at the University of North Carolina’s Department of Social Medicine.
Of the 521 devices in this database, just 22 were validated using the “gold standard” – randomised controlled trials, while 43% (226) didn’t have any published clinical validation.
Some of these devices used “phantom images” instead – computer-generated images that didn’t come from real patients.
The rest of the devices used retrospective or prospective validation – tests based on patient data from the past or in real-time, respectively.
“We shared our findings with directors at the FDA who oversee medical device regulation, and we expect our work will inform their regulatory decision making,” says Chouffani El Fassi.
“We also hope that our publication will inspire researchers and universities globally to conduct clinical validation studies on medical AI to improve the safety and effectiveness of these technologies. We’re looking forward to the positive impact this project will have on patient care at a large scale.”
In Australia, the Therapeutic Goods Administration (TGA) requires software that uses AI to provide information about the data it’s trained on, as well as a justification for “how this data would be appropriate for the Australian population and sub-populations for whom the AI is intended to be used”.
Medical devices with AI must also meet more general clinical evidence guidelines.
Discussion about this post