Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products

Overview

The U.S. Food and Drug Administration (FDA) issued the draft guidance titled “Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products” to address the rapidly evolving landscape of artificial intelligence (AI) and machine learning (ML) technologies in pharmaceutical development. This document, identified by docket number FDA-2024-D-4689, provides non-binding recommendations to sponsors and other interested parties on how to effectively use AI to generate data and information that will be submitted to the FDA to support regulatory decisions. The guidance is particularly focused on ensuring the credibility, reliability, and trustworthiness of AI models when their outputs are intended to influence critical regulatory determinations regarding the safety, effectiveness, or quality of drugs and biological products throughout their lifecycle. It acknowledges the transformative potential of AI in accelerating drug discovery, optimizing clinical trials, and enhancing manufacturing processes, while simultaneously highlighting the critical need for robust validation, transparency, and bias mitigation strategies to maintain public trust and patient safety.

The scope of this draft guidance is comprehensive, encompassing AI applications across various stages of the drug product lifecycle, including nonclinical development, clinical trials, manufacturing, and postmarketing surveillance. It is specifically applicable when the AI model's output directly supports a regulatory decision. The FDA's initiative reflects a proactive approach to integrating advanced technologies responsibly into the regulatory framework, aiming to foster innovation while upholding stringent standards for medical product evaluation. By providing a structured framework for credibility assessment, the FDA seeks to offer clarity to the industry, enabling developers to navigate the complexities of AI integration with greater confidence and ensuring that AI-driven insights are robust and reliable enough for regulatory scrutiny. This guidance is a foundational step in establishing a harmonized understanding between regulators and industry on the acceptable use and validation of AI/ML in drug development.

Definitions

Within the context of this FDA draft guidance, several key terms are implicitly or explicitly defined to ensure a common understanding for stakeholders. Artificial Intelligence (AI) generally refers to machine-based systems that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. These systems typically perceive environments, abstract perceptions into models through automated analysis, and use model inference to formulate options for information or action. Machine Learning (ML) is presented as a subset of AI, encompassing techniques used to train AI algorithms to improve performance at a task based on data.

A critical concept introduced is the Context of Use (COU), which defines the specific role and scope of an AI model in addressing a particular question of interest within drug development. The COU is paramount because the credibility assessment of an AI model is inherently tied to its intended application. Credibility Evidence refers to any data or information that supports the trustworthiness and reliability of an AI model's output for its specified COU. This evidence is crucial for demonstrating to regulatory bodies that the AI model produces reliable, reproducible, and unbiased outcomes. The guidance also implicitly touches upon concepts like Bias, referring to systematic errors in an AI model's output that could lead to unfair or inaccurate results, particularly across diverse patient populations, and the need for Transparency, which involves clear documentation and explainability of how AI models function and arrive at their conclusions.

Governance and Institutional Framework

The governance of AI/ML in drug development within the United States primarily falls under the purview of the U.S. Food and Drug Administration (FDA), specifically through its centers such as the Center for Drug Evaluation and Research (CDER) and the Center for Biologics Evaluation and Research (CBER). This draft guidance is a direct output of the FDA's commitment to ensuring the safety and effectiveness of drugs while fostering innovation. The FDA acts as the central regulatory authority, responsible for developing and implementing a risk-based regulatory framework that promotes the responsible adoption of AI technologies. The agency's role extends to evaluating submissions that incorporate AI components, providing recommendations, and establishing engagement pathways to work with sponsors and interested parties to navigate the evolving technological landscape.

The FDA's approach to governance is characterized by its issuance of guidance documents, which, while non-binding, represent the agency's current thinking and provide critical recommendations to the industry. The institutional framework encourages open dialogue and public comment periods, allowing stakeholders to provide feedback that informs the finalization of such guidance documents. This collaborative model aims to ensure that regulatory policies are robust, adaptable, and reflective of both scientific advancements and practical industry needs. Furthermore, the FDA's engagement in collaborations with international bodies, such as the European Medicines Agency (EMA), on developing guiding principles for AI in drug development underscores a broader commitment to harmonizing regulatory approaches and fostering global best practices.

Key Focus Areas

This draft guidance primarily focuses on establishing a robust framework for assessing the credibility of Artificial Intelligence (AI) and Machine Learning (ML) models used in drug and biological product development. A central tenet is the proposed risk-based credibility assessment framework, which guides sponsors in evaluating and documenting the trustworthiness of AI models for specific contexts of use (COUs). This framework is designed to ensure that AI-generated data and insights are sufficiently reliable to support regulatory decision-making regarding the safety, effectiveness, and quality of medical products. The emphasis is on proactive planning, rigorous execution of validation strategies, and comprehensive documentation of results to demonstrate the AI model's fitness for its intended purpose.

Another key focus area is data governance and documentation. The guidance implicitly stresses the importance of high-quality, well-managed data for training, validating, and monitoring AI models. It highlights that the reliability of AI outputs is intrinsically linked to the integrity and appropriateness of the underlying data. Consequently, sponsors are encouraged to implement robust data governance practices, including data collection, curation, and management, to minimize bias and ensure reproducibility. Furthermore, transparent and thorough documentation of the AI model's design, development, training, validation, and performance is crucial for regulatory review. This includes clearly defining the problem the AI model addresses, its COU, the risks associated with its use, and the plan for assessing its credibility, along with detailed records of the execution and evaluation of that plan.

Implementation Framework

The implementation framework proposed by the FDA's draft guidance centers around a risk-based credibility assessment for AI models. This framework is not prescriptive regarding specific AI approaches but rather provides a structured methodology for sponsors to demonstrate the reliability of their AI models in supporting regulatory decisions. At its core, the framework requires sponsors to clearly define the problem the AI model is intended to solve and its specific Context of Use (COU). This initial step is crucial for tailoring the subsequent assessment activities to the model's intended application and potential impact on regulatory outcomes. The framework then progresses to a comprehensive risk assessment, where potential risks associated with the AI model's use are identified and evaluated.

Following the risk assessment, sponsors are expected to develop a detailed credibility plan outlining how the model's trustworthiness will be established and validated. This plan should encompass the methods, data, and metrics to be used for evaluation. The execution of this plan involves rigorous testing and analysis, with meticulous documentation of all procedures, results, and any deviations from the original plan. The final stage of the framework involves evaluating the adequacy of the AI model, ensuring it meets the required standards for its intended use and that the evidence gathered sufficiently supports its credibility for regulatory decision-making. This iterative process emphasizes transparency, reproducibility, and a clear understanding of the AI model's limitations and performance characteristics to instill confidence in its outputs for regulatory purposes.

Monitoring and Evaluation

The FDA's draft guidance implicitly underscores the importance of ongoing monitoring and evaluation throughout the lifecycle of AI models used in drug development. While the seven-step credibility framework primarily addresses pre-submission validation, the dynamic nature of AI models necessitates continuous oversight to ensure their sustained reliability and performance. This involves establishing mechanisms for post-approval updates and monitoring using real-world performance data. The guidance recognizes that AI models can be inherently adaptive and may require periodic adjustments or retraining. Therefore, sponsors are encouraged to consider processes that facilitate these updates while maintaining transparency and ensuring that any modifications do not compromise the model's established credibility.

Effective monitoring and evaluation strategies should include robust data governance practices to track the quality and relevance of input data over time, as data drift or changes in real-world conditions could impact model performance. Regular performance assessments, potentially against independent reference datasets, are crucial to detect any degradation in accuracy, identify emerging biases, or ensure the model remains fit for its evolving context of use. The goal is to establish a lifecycle management approach for AI models, where their performance is continuously assessed, documented, and, if necessary, updated to ensure they consistently produce reliable and unbiased outcomes that support regulatory decision-making. This proactive approach to monitoring helps maintain confidence in AI-driven insights from discovery through post-market surveillance.

Penalties, Liability, and Appeals

As a draft guidance document, the FDA's “Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products” primarily provides recommendations and does not, in itself, establish binding legal requirements or specify penalties for non-compliance. Guidance documents represent the agency's current thinking and do not create or confer any rights for or on any person, nor do they bind the FDA or the public. Therefore, direct penalties, liability provisions, or appeal processes specifically tied to this draft guidance are not outlined within the document. The legal implications of using AI in drug development would instead fall under existing statutes and regulations governing drug and biological product approval, manufacturing, and post-market surveillance.

However, failure to adhere to the principles and recommendations outlined in this guidance, particularly concerning the credibility and reliability of AI models used to support regulatory submissions, could indirectly lead to regulatory actions. For instance, if an AI model's output, submitted as part of a drug application, is found to be unreliable, biased, or inadequately validated, it could result in delays in approval, requests for additional data, or even rejection of the application. The FDA's existing enforcement mechanisms for ensuring drug safety and efficacy would apply if AI-driven decisions lead to adverse outcomes. Therefore, while the guidance itself does not impose penalties, compliance with its recommendations is critical for successful regulatory engagement and avoiding potential complications under broader FDA regulations.

Relationship to Other Instruments

This FDA draft guidance on AI/ML in drug development is situated within a broader ecosystem of regulatory and policy instruments, both domestic and international, that address artificial intelligence in healthcare. Domestically, it complements other FDA guidances related to AI, such as those concerning Artificial Intelligence in Software as a Medical Device (AI/ML SaMD), which focus on AI used in medical devices rather than drug development. The agency has also published "Guiding Principles of Good AI Practice in Drug Development" in collaboration with the European Medicines Agency (EMA), which outlines overarching principles like human-centric design, risk-based approaches, and data governance. While the "Guiding Principles" offer a high-level framework, this specific draft guidance provides more detailed recommendations for establishing the credibility of AI models in the context of regulatory decision-making for drugs and biological products.

Internationally, this FDA guidance aligns with and contributes to a growing global effort to regulate AI in healthcare. The European Medicines Agency (EMA) issued its own Reflection Paper on AI in drug development in September 2024, discussing principles for AI use from discovery to post-authorization and advising developers on good practices, benefits, and risks. Similarly, the UK's Medicines and Healthcare Products Regulatory Agency (MHRA) has released guidance on AI use in medical devices and health products. These international efforts, often developed through collaborative initiatives, aim to foster a harmonized understanding and approach to AI regulation, facilitating cross-border innovation while ensuring patient safety. The FDA's guidance, therefore, represents a significant step in establishing a consistent and robust regulatory stance on AI in drug development, both within the U.S. and in conjunction with global partners.

International Alignment

The FDA's draft guidance on AI/ML in drug development demonstrates a clear commitment to international alignment, recognizing that artificial intelligence innovation and regulation are global endeavors. The agency has actively engaged in collaborations with international counterparts, most notably with the European Medicines Agency (EMA), to develop shared principles and foster a harmonized approach. For instance, the FDA's Center for Drug Evaluation and Research (CDER) and Center for Biologics Evaluation and Research (CBER) collaborated with the EMA to develop "Guiding Principles of Good AI Practice in Drug Development," which outlines 10 key considerations for industry and product developers. This collaboration underscores a mutual understanding that consistent regulatory expectations across jurisdictions can facilitate innovation while ensuring patient safety on a global scale.

The FDA's guidance is also informed by and contributes to broader international discussions on AI governance. While the FDA's document specifically focuses on credibility and risk in regulatory decision-making for drugs, it shares common themes with other international regulatory frameworks, such as the EMA's Reflection Paper on AI in drug development and the UK MHRA's guidance on AI in medical products. These documents collectively emphasize the importance of transparency, validation, human oversight, and a risk-based approach in AI-based systems. By aligning its recommendations with these global principles, the FDA aims to reduce regulatory fragmentation, support the development of globally applicable AI solutions, and ensure that advancements in AI in drug development benefit patients worldwide through a consistent standard of safety and efficacy.

Implementation Timeline

Compliance Checklist

Sources and References