0
0
A recent study conducted by the University of North Carolina (UNC) at Chapel Hill reveals a concerning gap in the clinical validation of artificial intelligence (AI) health devices authorized by the US Food and Drug Administration (FDA). The study found that 43% of the 521 AI health tools approved by the FDA between 2016 and 2022 lack publicly available clinical validation data, raising questions about the transparency and reliability of these technologies.
Urgent Need for Public Data and Clearer Standards
The study underscores the urgent need for more public data and the establishment of clearer standards around AI medical devices. “The number of devices that we found clinical validation data on — a lot of people think, oh, that’s the big news,” said Sammy Chouffani El Fassi, the study’s lead author and an MD candidate at the UNC School of Medicine. “But [that’s] not as [big] as realizing that there’s no standard yet.”
This lack of standardization in assessing AI health devices is problematic, as it leaves clinicians and other stakeholders uncertain about which devices are truly effective and which could benefit from further validation. While the absence of publicly available data does not necessarily indicate a lack of validation, it does create uncertainty that could hinder the adoption of new AI tools in clinical settings.
Confidential Data and Regulatory Challenges
According to Troy Tazbaz, director of the FDA’s Center for Devices and Radiological Health, Digital Health Center of Excellence, regulators review “thousands, if not tens of thousands of pages” of information about each AI tool, much of which includes real patient data. However, this information is often confidential, contributing to the lack of publicly accessible clinical validation data.
The study’s findings highlight the need for greater transparency and clearer communication between regulators, manufacturers, and clinicians. “Physicians probably won’t trust a device if it wasn’t exposed to confounding variables in real-world implementation,” Chouffani El Fassi noted, emphasizing the importance of real-world testing and validation.
The Role of Clinicians in Advancing AI Validation
The study presents an opportunity for clinicians and academic institutions to play a more active role in validating AI health devices. Chouffani El Fassi suggested that clinicians could incorporate the validation of AI tools into their training or collaborate with public-private partnerships like the Coalition for Health AI to conduct prospective validation studies.
Nigam H. Shah, PhD, chief data scientist for Stanford Health Care, who was not involved in the study, highlighted the rapid increase in AI medical device authorizations, which surged from an average of two per year in 2016 to 69 per year in 2022. While this growth reflects the potential of AI in healthcare, it also poses challenges for regulators and underscores the need for robust validation processes.
Inconsistent Language and the Need for Standardization
The UNC researchers also identified inconsistencies in the language used to describe different methods of evaluating AI health tools. The FDA, academic researchers, and manufacturers often use different terms for prospective, retrospective, and clinical validation, leading to confusion and varying levels of evidence quality.
To address this issue, the researchers proposed a set of standardized definitions for each validation method. This “clinical validation standard” aims to guide stakeholders in testing AI devices and provide clear information to potential users about the level of validation a tool has undergone.
Tazbaz noted that most public-facing summaries of device authorizations are written by manufacturers rather than the FDA. Standardizing the language used in these summaries could help better categorize the clinical validation of AI tools and improve their adoption in healthcare.
Conclusion
The UNC study highlights significant gaps in the transparency and standardization of AI health devices, emphasizing the need for more public data, clearer standards, and greater involvement from clinicians in the validation process. As the use of AI in healthcare continues to grow, addressing these challenges will be crucial to ensuring that these technologies are both effective and trusted by the medical community.
