FDA Watch: FDA Moves Slowly on AI-Driven Lab Devices

Artificial intelligence (AI) is beginning to take the healthcare world by storm, but its use is still only rising slowly at best in the laboratory sector, according to data from the U.S. Food and Drug Administration (FDA).

To date, only 40 tests or laboratory-related devices with AI or machine learning components or features have been approved by the FDA.¹ By contrast, there have been about 950 devices with some form of AI or machine learning approved by the agency in total, with the vast majority relating to radiology.

However, the list is somewhat overoptimistic. About half of the approvals were from more than five years ago, and one dates to 1995, when dialup modems were still common and the appearance of the first smartphones was more than a decade away, suggesting that AI either might have been used in some form of marketing material and not in the device itself, or the AI performs on an extremely basic level. That approval was for a cervical cytology slide reader. The FDA description of the approval suggests future iterations of the device contain more advanced technology.²

Otherwise, there have been only nine laboratory-related devices in the past three years with AI or machine learning features that have been approved by the FDA, records show.¹

“The agency has treaded carefully here and probably with some merit,” said Bruce Carlson, publisher of the Eye on IVD newsletter. He explains that though claims of AI systems are generating buzz—including in IVDs—IVDs must have a high performance bar for AI to consistently hit. He added that the past approvals of many tests have been what he calls “AI lite,” or they are marketed as “smart” devices but really don’t have the components most people associate with AI.

AI use in other laboratory products

AI has been seeping into laboratory products that do not necessarily require regulatory approvals. In early September, for example, Baltimore-based Sapio Sciences launched ELaiN, an AI-powered “lab assistant” that uses natural language processing to allow technicians to verbally interface with Sapio’s laboratory information management system (LIMS) or electronic laboratory notebook (ELN). The platform was launched in a beta version last November and is now available for purchase.³

“Through a natural language ‘chat,’ ELaiN can help a medicinal chemist synthesize a new compound, or help a researcher design a complex experiment including complicated plate layouts,” Sapio said in a statement.³

AI use in laboratory tests and platforms expected to grow quickly

Meanwhile, there is significant evidence that AI-powered lab tests and platforms are going to be mushrooming in number—along with FDA reviews and approvals—soon.

According to Politico, the U.S. Department of Health and Human Services (HHS) has spent $129 million on AI and AI-related purchases over the past five years. The FDA spent $32 million, the second-highest total among agencies under the HHS umbrella.⁴

A recently published study in the Journal of the American College of Radiology concluded that, based on the venture capital funding curve for AI-related startup companies and other investments, the FDA will be approving as many as 350 AI-powered medical devices annually by 2035. There were 69 such approvals in 2022, according to the study.⁵

How the FDA will be regulating such devices remains to be seen. The agency has issued numerous publications on the topic, but they have generally been vague. In June, it issued a brief document about one of its research efforts that aims to help stakeholders “determine and use least burdensome metrics for appropriate evaluation of AI-enabled medical devices.” The document added that as part of evaluating such devices, “tasks such as classification, estimation, image segmentation, time-to-event analysis, and detections/localization of abnormalities each may need to use different performance metrics. Furthermore, within each task, there are nuances as to how the output is presented to the user, and how the data is collected and structured. These factors affect what kind of performance metric(s) are best suited to assess device performance.”⁶

Possible hurdles to AI in IVDs, and outlook

According to a 2021 issue brief by the Pew Charitable Trusts, one of the FDA’s biggest challenges is parsing whether AI-powered software in a medical device can help clinicians make informed decisions. If the device powers much of the decision-making process, it must be regulated as a medical device. Many future devices may be straddling the fence in terms of whether they are an information source or catalyst in that process.⁷

It also remains to be seen how the lawsuits recently filed by the American Clinical Laboratory Association and the Association for Molecular Pathology challenging the FDA’s plan to regulate laboratory-developed tests will impact the agency’s future decision-making process on AI-powered devices. The ACLA declined to comment for this story, while AMP did not respond to a request seeking comment.

Carlson believes that because of the agency’s history of caution in the approval process, AI products related to the pathology process are the likeliest kinds of devices to be approved in the near term.

“Pathology is an area where AI has a past performance record and can do well,” he said, adding that imaging in general is where the new technology has the broadest applications.

Over the near term, Carlson predicts most FDA approvals for AI lab devices will be focused on pathology interpretation assistance. Devices providing more general interpretive assistance on testing may also get approvals, but, “I see these as slower to be approved,” Carlson said.

References:

1. 1. https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices

1. 1. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P940029

1. 1. https://www.sapiosciences.com/news/sapio-sciences-advances-the-worlds-first-ai-powered-lab-assistant/

1. 1. https://www.politico.com/newsletters/politico-pulse/2024/09/13/inside-hhs-ai-purchases-00178902

1. 1. https://pubmed.ncbi.nlm.nih.gov/37843483/

1. 1. https://www.fda.gov/medical-devices/medical-device-regulatory-science-research-programs-conducted-osel/evaluation-methods-artificial-intelligence-ai-enabled-medical-devices-performance-assessment-and

1. https://www.pewtrusts.org/en/research-and-analysis/issue-briefs/2021/08/how-fda-regulates-artificial-intelligence-in-medical-products

********

Key laboratory test approvals by the FDA in the past month include the following decisions:

New CE Marks and Global Certifications

Significant EU and/or global approvals announced during the period: