FDA’s Troy Tazbaz, Director, and John Nicol, PhD, Digital Health Specialist at the FDA’s Digital Health Center of Excellence (DHCoE), recently proposed “A Lifecycle Management Approach toward Delivering Safe, Effective AI-enabled Health Care.“ They emphasize the importance of ensuring AI-enabled medical devices are safe, effective, and trustworthy. While they acknowledge that AI’s adaptability can improve performance, they also recognize that it poses risks such as reinforcing biases and potentially harming patients, particularly underrepresented populations. This approach addresses the growing interest and challenges in implementing artificial intelligence (AI) in healthcare.
Managing the AI Lifecycle
To manage these complexities, Tazbaz and Nicole propose leveraging Lifecycle Management (LCM) principles, which have been fundamental in software development since the 1960s. They introduce the concept of an AI lifecycle (AILC), mapping traditional Software Development Lifecycles (SDLCs) to the specifics of AI software development. This mapping highlights key activities for each phase, from data collection and management to model building, tuning, and post-deployment monitoring, aiming to ensure AI systems meet real-world needs while managing risks.
The AI lifecycle management wheel shows the agile nature of the framework. While the interconnected chevrons on the right provide further details of the lifecycle phases. Please note the feedback looks across the chevrons representing the lifecycle phases:
Below each chevron, the framework includes suggested technical and procedural considerations to be addressed in each AI lifecycle phase.
Is Industry Ready?
85% of life science industry professionals rate their organizations capacity to adequately address the data management and privacy concerns associated with training large language models as moderately/not adequate according to Axendia’s research “The State of Generative AI in Life Sciences: The Good, The Bad and The Ugly.” This casts a shadow over the full realization of generative AI’s potential in the life sciences industry.
Organizations should consider a strategic approach to fortify the data privacy frameworks in the context of GEN AI utilization. This entails initiating comprehensive training programs aimed at equipping AI teams with the necessary expertise to address GEN AI’s distinctive data privacy considerations. Concurrently, it is imperative to establish data governance structures to adeptly navigate the intricacies of GEN AI integration. Cultivating a corporate ethos that prioritizes data privacy across all operational levels is essential, ensuring that each employee is cognizant of their role in data protection.
In Brief
The AILC concept provides a comprehensive framework to guide the development, evaluation, and monitoring of AI in healthcare. It incorporates technical and procedural considerations for each lifecycle phase, emphasizing the importance of standards to ensure quality, interoperability, ethical practices, and transparency.
By adopting the AI lifecycle framework, the industry can better navigate the complexities of AI software development. This structured approach not only helps manage risks but also ensures AI systems are safe, effective, and trustworthy. Collaboration and continuous refinement of these concepts are essential for fostering innovation and trust in AI technologies. The DHCoE invites all stakeholders to engage in this crucial effort, driving progress and improving healthcare outcomes through responsible AI integration
DHCoE welcomes your comments and feedback related to AI use in health care and has requested interested parties email: digitalhealth@fda.hhs.gov, noting, “Attn: AI in health care,” in the subject line.
Related Content
FDA on Artificial Intelligence Across the Product Lifecycle
Artificial Intelligence Has the Attention of Regulators
10 Regulatory Commandments of Artificial Intelligence
To discuss how to leverage the AI lifecycle management approach in your organization, click on this link to schedule an Analyst Inquiry on this topic
The opinions and analysis expressed in this post reflect the judgment of Axendia at the time of publication and are subject to change without notice. Information contained in this post is current as of publication date. Information cited is not warranted by Axendia but has been obtained through a valid research methodology. This post is not intended to endorse any company or product and should not be attributed as such.