Introduction
The US Food and Drug Administration (FDA) has developed a comprehensive framework to streamline the regulatory process for artificial intelligence (AI) and machine learning (ML) enabled medical devices. This framework aims to ensure the safety and effectiveness of these devices while promoting innovation in healthcare technology.
Description
Devices that incorporate artificial intelligence (AI) and machine learning (ML) may require premarket review through various pathways [3], such as 510(k) [3], De Novo [3] [5], or Premarket Approval (PMA) [3], depending on their risk classification and intended use [3]. The FDA has established specific regulations for Software as a Medical Device (SaMD) [3], which include criteria for determining whether software falls under device regulations [3]. Recently, the FDA released final guidance aimed at streamlining the approval process for modifications to AI/ML-enabled medical devices, emphasizing a forward-thinking approach to ensure their safety and effectiveness while promoting the development of these technologies in healthcare. This guidance highlights a “least burdensome” approach for developers, allowing for iterative improvements without extensive additional submissions for each modification [4].
A key component of this framework is the Predetermined Change Control Plan (PCCP), which manufacturers are encouraged to submit alongside their marketing approval applications. The PCCP should detail planned modifications, necessary testing for safety and effectiveness [3] [4], and a risk mitigation strategy [4]. It is essential for manufacturers to state that the device has an authorized PCCP and to describe any modifications made. The FDA will review the PCCP as part of the marketing submission [4], facilitating ongoing safety and effectiveness assessments without requiring separate submissions for each change [4]. This framework supports iterative changes to devices with “locked” algorithms, underscoring the importance of version control and maintenance. Recommendations for summarizing changes to the PCCP and maintaining version control are included [2], with confirmation that automatic implementation of modifications will be considered for AI-Driven Software Functions (AI-DSFs) [2]. Modifications made in accordance with an authorized PCCP will not necessitate a new marketing submission [5], allowing the FDA to focus on reviewing significant modifications while encouraging the submission of documents with tracked changes [5].
To assess the long-term safety and effectiveness of AI/ML-enabled devices [3], the FDA may require Real-World Evidence (RWE) [3]. Manufacturers must specify whether changes will be implemented automatically or manually and whether they will apply uniformly across devices or vary by clinical site or patient characteristics [1]. Device labeling must inform users about the incorporation of machine learning and the existence of an authorized PCCP [1], indicating potential software updates and performance changes [1]. Additionally, communication plans for users regarding implemented modifications are recommended [2].
Addressing algorithmic bias and fairness is crucial to ensure equitable healthcare outcomes [3]. Manufacturers are expected to maintain transparency regarding the development [3], validation [3] [4], and performance of AI/ML algorithms [3]. Considerations regarding performance across various demographic factors [5], such as race [5], ethnicity [5], and age [1] [4] [5], are essential in the ongoing development and monitoring of these devices [5]. Data quality and patient privacy are critical considerations in the development of AI/ML-based devices [3], and rigorous validation and testing are necessary to confirm their safety and effectiveness [3]. Furthermore, ethical considerations [3], including algorithmic bias and transparency [3], must be addressed [3].
By proactively tackling these challenges and collaborating with regulatory authorities [3], manufacturers can effectively navigate the regulatory landscape and successfully bring innovative AI/ML-enabled medical devices to market [3]. For instance, a manufacturer could propose to retrain an AI model used in an intensive care unit to reduce false alarms while maintaining sensitivity [1]. Under a PCCP [1], the manufacturer could collect data [1], retrain the model [1], and update the device’s label without needing a new marketing submission [1], as long as the changes align with the PCCP [1]. Recommendations for monitoring real-world device performance and procedures for updating users on safety and effectiveness are also provided [2]. A webinar is scheduled for January 14, 2025 [5], to provide further insights into the guidance [5], supporting the FDA’s commitment to expediting the deployment of new devices while maintaining a science-based regulatory framework for AI and ML-powered medical devices [5].
Conclusion
The FDA’s framework for AI and ML-enabled medical devices represents a significant advancement in regulatory practices, balancing innovation with safety and effectiveness. By facilitating iterative improvements and addressing ethical considerations, the framework supports the development of cutting-edge healthcare technologies. This approach not only accelerates the deployment of new devices but also ensures that they meet rigorous standards, ultimately benefiting patients and healthcare providers.
References
[1] https://www.medtechdive.com/news/fda-final-guidance-predetermined-change-control-plans-ai/734608/
[2] https://www.kslaw.com/news-and-insights/fda-publishes-final-predetermined-change-control-plan-guidance-for-ai-enabled-device-software-functions
[3] https://www.jdsupra.com/legalnews/navigating-the-complex-regulatory-9503041/
[4] https://www.techtarget.com/HealthtechAnalytics/news/366616658/FDA-finalizes-AI-enabled-medical-device-guidance
[5] https://www.healthcareitnews.com/news/fda-finalizes-ai-enabled-medical-device-life-cycle-plan-guidance
