Eye on FDA

AdvaMed, the world’s largest medical technology association representing device, diagnostics and digital technology companies, released a white paper that reviews the current landscape of artificial intelligence (AI)-based applications and products in the health care sector, and identifies steps to accelerate the use of AI in medical technologies. The trade association highlights two main types of tasks that AI is uniquely well suited to tackle (1) identifying and analyzing patterns in patient charts that practitioners might miss; and (2) automating repetitive routine tasks. AI is being incorporated into a range of technologies in the health care sector, and AdvaMed’s white paper focuses primarily on AI/machine learning (ML)-enabled medical devices, which are regulated by the FDA. AdvaMed anticipates that FDA will likely need to issue additional guidance to keep pace with development of AI models, including for adaptive models and approaches to mitigating bias. AdvaMed endorses FDA’s use of “Predetermined Change Control Plans” (PCCPs), which permit manufacturers to outline approaches to future modifications as part of an initial submission, and states that PCCPs should evolve to allow for greater post-market modifications for adaptive algorithms. The trade association also calls for domestic and international harmonization of requirements, including development of common AI standards to advance safe, secure and trustworthy use of AI.

Researchers from the National Institutes of Health (NIH) have developed an algorithm that harnesses AI to help accelerate the process of matching potential volunteers for relevant clinical research trials. The algorithm, called TrialGPT, is intended to help clinicians navigate the vast range of clinical trials available to patients by identifying potential matches and providing a summary of how that person meets the criteria for study enrollment. The team of researchers used a large language model (LLM) to develop an innovative framework for TrialGPT and compared the algorithm to the results of three human clinicians who analyzed over 1,000 patient-criterion pairs. The team also conducted a pilot user study, where two human clinicians reviewed six anonymous patient summaries and matched them to potentially suitable clinical trials. When clinicians used TrialGPT as a starting point, they spent 40% less time screening patients and maintained the same level of accuracy. The research team was selected for the Director’s Challenge Innovation Award, which will allow the team to further assess the model’s performance and fairness in real-world clinical settings. The researchers “anticipate that this work could make clinical trial recruitment more effective and help reduce barriers to participation for populations underrepresented in clinical research.”

On September 26, 2024, FDA published a compilation of commonly used terms in the digital health, artificial intelligence (AI), and machine learning space and their definitions, available here.¹   FDA intends the glossary to be used for general education purposes. The definitions are either directly from, or adapted from, various public sources, including consensus standard organizations and published literature, and FDA includes references to those sources in the glossary.