FDA Proposes Framework to Enhance AI Models in Drug Development

The U.S. Food and Drug Administration (FDA) has recently released a draft guidance document titled, "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." This guidance explains how the FDA plans to make sure that artificial intelligence (AI) models integrated in the drug development process are as reliable, valid, and transparent as possible. The increasing use of AI in the pharmaceutical domain can enhance regulatory decision making, but this possibility comes with the obligation to ensure that these models are capable of efficiently aiding in AI regulatory decisions. "AI" is in every conversation; how does this affect business decisions in the industry and other decision-making regarding drugs and drug development?

The goal of this article is to highlight the most important parts of the framework proposed by the FDA and analyze what it could mean for drug development and regulatory actions, including how pharmaceutical companies can comply with the AI regulations and actively support the FDA’s recommendations.

Why AI is Necessary for Submission of Drugs and Biological Products?

One of the areas AI is gradually being integrated into is the development of drugs. This includes clinical trials, drug safety evaluations, and post-market surveillance. With the pharmaceutical sector embracing AI-powered technologies, the FDA understands that it will need to have detailed and nuanced AI policies for regulatory actions. The FDA’s draft guidance attempts to meet these issues by ensuring that AI models are fit for purpose and that scientific integrity is not compromised in any manner.

The remarks propose that AI will enhance the productivity and accuracy of the drug development process but only if the models are accepted as robust, transparent, and validated. This guidance states that AI applications need to be vetted and tested with a framework that factors in the impact to public health and patient safety.

Key Considerations in the FDA’s Draft Guidance

1) Defining the Context and Purpose of AI Models: The guidance emphasizes the importance of defining the context of use for AI models. Developers should clearly state the intended use of AI in regulatory submissions, whether it's for assessing drug effectiveness, safety monitoring, or aiding with clinical trial design. Companies are encouraged to clarify how they intend to utilize AI models and ensure these applications meet regulatory requirements.

2) Trustworthiness of AI Models: The FDA expects regulatory submissions to include information that justifies the dependability of AI-enabled decisions. This include:

• Data sources: Specifying the location of the data, training AI models, and determining their quality.
• Model development and validation: Explaining the construction, methodologies, and algorithms used to develop and validate AI models.
• Performance testing: Showing model results from different data sets and for different regulatory conditions.

In each case, it is important to maintain transparency to ensure AI models are seen as reliable and effective in regulatory decision-making.

3) Risk-Based Approach to AI Model Evaluation: The FDA supports the evaluation of AI models based on the credibility model. This suggests measuring the models’ ability to impact people’s health and their engagement in the regulatory process. The depth of scrutiny and validation needed differs according to the range of the AI model’s application and the possible consequences of erroneous decision-making.

More rigorous validation would be needed for models designed to evaluate safety concerns than for those designed to carry out secondary analyses. FDA recommends that risk be assessed at any stage of development.

4) Data Governance and Quality: The guidance applies the same standards for traditional clinical data to the data used for training and retraining AI models, and therefore the data put into it has to be of a very high standard. This includes correctness, consistency, and absence of bias. Developers need to account for traceability of the data throughout the AI lifecycle to transform and analyze the data.

The guidance states that the model should handle missing or incomplete portions of the input data, while at the same time bearing in mind that procedures meant to achieve these aims should not compromise the regulatory model integrity or the model quality.

Implications for Pharmaceutical Companies and AI Developers

The framework proposed by the FDA focuses on strategic elements for pharmaceutical companies and AI developers interested in the application of AI in their regulatory submissions. Following the guidance of the FDA ensures that sponsors prepare sufficient AI models which are well documented and transparent. Here are a few notes to remember:

1) Supporting Documentation and Validation: Sponsors must be ready to justify any AI model they intend to use, explain its purpose, and detail how it was developed and validated. If the model is shown to be credible, robust, and clear, it will more likely be accepted.

2) Engagement with FDA on Regulatory Issues: The FDA is willing to actively engage sponsor representatives wishing to propose studies to discuss proposed study designs and applications of the AI model. This dialogue at the beginning of the study will make it easier to recognize regulatory issues and make the review process simpler and faster to bring drugs to market.

3) Collaborative Trust: To trust AI models, transparency is crucial. Developers should be encouraged to build collaborations with the FDA, as well as provide unrestricted documentation and data, and deal with model performance and data quality concerns.

4) Risk-Based Decision Making: The FDA’s risk-based framework is based on the sponsors’ evaluation of the assumed risks of AI models and the validation process of the model. AI models with greater potential risks will require extensive validation processes, while those with lower risks will undergo less extensive scrutiny.

How Maxis Clinical Sciences Can Help?

As AI progresses in its involvement in drug formulation, decision, and clinical trials, Maxis Clinical Sciences remains committed to help sponsors with the evolving regulatory changes.

We actively support the FDA’s initiative and provide the following assistance to pharmaceutical and biotech companies:

• Evaluation and Documentation of AI Models: We aid in the development and documentation of AI models, as per FDA requirements, which ensures that your models are well documented, validated, and transparent.
• Data Quality and Traceability: Our experts guarantee that all data for training AI models is of the highest standard, properly managed, and fully traceable throughout the data life cycle.
• Regulatory Strategy and Guidance: We provide consulting services to help? companies communicate with the FDA sooner, which ensures that their AI models are compliant and in good standing with regulatory and industry expectations.
• Assistance in Conducting Clinical Trials: To help execute clinical trials, we assist in introducing AI technology and ensure the model's output is credible, valid, and adheres to regulations.

Looking Ahead

A diverse group of stakeholders should engage with the draft guidance to enable the FDA to further refine its approach to AI in drug development. Pharmaceutical and biotech companies can use AI within the FDA’s framework to improve drug development processes without compromising public health and safety.

The time has arrived when AI can and is being used in medicine and we can only hope that under the FDA's guidance, we will establish a more effective, open, and trustworthy method of regulation.

Sponsors are encouraged to keep updated on the latest developments in AI applications for regulatory submissions and related activities.

Reference

Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products - https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-artificial-intelligence-support-regulatory-decision-making-drug-and-biological