FDA Issues Whitepaper on Using AI & ML in Drug Development

Recent, rapid technological innovations combined with increasingly robust information management systems and advanced computing abilities are transforming the way drugs are developed and used. This ever changing ecosystem presents unique opportunities and challenges, especially in the context of Artificial Intelligence (AI) and Machine Learning (ML).

In Spring 2024, the Food and Drug Administration’s (FDA’s) Centre for Drug Evaluation and Research (CDER), in collaboration with other likeminded organisations, compiled a discussion paper emphasising it’s commitment to working domestically and internationally to ensure that the full potential of these innovations is realized for the benefit of the public. The FDA also acknowledges that AI/ML is already being integrated into clinical trial design, DHTs, and RWD analytics. Submissions to the FDA referencing AI/ML have surged, with over 100 in 2021 alone, ranging from various therapeutic areas and stages of drug development. Currently, the most common stage for incorporating AI/ML in submissions is during the clinical development and research phase.

In this article, we endeavour to summarise the wide-ranging whitepaper, which, over the course of 33 pages, sets out the FDA’s aim to understand the use of AI and ML in drug development. They seek not to do this in terms of FDA guidance or policy but instead, to form the basis of interim communications and considerations with stakeholders, academia, researchers and technology developers, to promote mutual learning and initiate discussions. It is important to note that the FDA views this as a collaborative process and is seeking feedback on the opportunities and challenges of utilizing AI/ML in the development of drugs and medical devices to help inform the regulatory landscape in this area.

In the FDA discussion paper, three main topics are covered:

Landscape of current and potential uses of AI/ML
Considerations for the use of AI/ML
Next steps and stakeholder engagement

Current and Potential Uses of AI/ML in the Drug Development Process

Drug discovery

This is an area where AI/ML plays a crucial role. It assists in identifying, selecting, and prioritising suitable biological drug targets, as well as screening compounds and aiding in drug design.

Nonclinical research

Nonclinical research involves in vitro and in vivo studies that aim to advance potential therapeutics toward clinical research in humans. These studies occur at various phases of drug development, including before clinical trials, in parallel with clinical development, and even in post-marketing settings. AI/ML can be leveraged in nonclinical research using data from:

Pharmacokinetic (PK) and pharmacodynamic (PD) studies in animals.
Exploratory mechanistic studies in animal models.
Organ-on-chip and multi-organ chip systems.
Cell assay platforms.

Applications of AI/ML include evaluating toxicity, exploring mechanistic models, and developing in vivo predictive models. For instance, AI/ML algorithms like recurrent neural networks can enhance complex PK/PD data analysis, benefiting both nonclinical and clinical drug development.

Clinical research

Clinical research involves a series of phases of clinical trials to assess drug safety and effectiveness in increasing numbers of human subjects. AI/ML plays a significant role in streamlining and advancing clinical research. It:

Analyses data from interventional (clinical trials) and non-interventional (observational studies) studies.
Informs the design of non-traditional trials (e.g., decentralised trials) using RWD.
Analyses data from DHTs used in clinical studies.
Improves trial conduct and operational efficiency.

Some of the current and potential applications of AI/ML in the design and conduct of clinical research include in:

Recruitment
Selection and Stratification of Trial Participants
Dose/Dosing Regimen Optimisation
Adherence
Retention
Site Selection
Clinical Trial Data Collection, Management, and Analysis
Clinical Endpoint Assessment

Post-marketing Safety Surveillance

Pharmacovigilance (PV) refers to the science and activities related to detection, assessment, understanding, and prevention of adverse events associated with drug use, including medication errors and product quality issues. During the post-approval period, PV involves reporting adverse events associated with human drug and biological products. Individual Case Safety Reports (ICSRs) serve as crucial data sources for potential drug safety issues during post-market surveillance. These reports include clinical information such as suspect products, temporal details related to product use, and adverse events in a patient’s medical history. Ensuring complete and accurate ICSRs is essential for understanding a drug’s safety profile. Due to the growing volume of ICSR submissions, regulatory agencies are exploring AI/ML applications to process and evaluate these reports. There are potential opportunities for the use of AI/ML in:

Case processing
Case evaluation
Case submissions

Advanced Pharmaceutical Manufacturing

In drug development, ensuring safety, effectiveness, identity, strength, quality, and purity is critical. Advanced analytics using AI/ML in pharmaceutical manufacturing can enhance process control, increase equipment reliability and throughput, monitor early warnings for process control issues, detect recurring problems, and prevent batch losses. AI/ML can be integrated with other advanced technologies like analytical process technology and continuous manufacturing to realise Industry 4.0, to create a well-controlled, hyper-connected, digitised ecosystem for manufacturers. Additionally, AI/ML can improve supply chain reliability by forecasting demand, analysing production schedules, mitigating disruptions, and optimising inventory. Ultimately, the use of AI/ML in pharmaceutical manufacturing can be grouped into:

The optimisation of process design
Advanced process control
Smart monitoring and maintenance
Trend monitoring which covers the drug manufacturing life cycle from design to commercial production.

FDA Experience with AI/ML for Drug Development

The FDA currently supports innovative AI/ML development through the CDER AI Steering Committee (AISC), which coordinates AI/ML use in therapeutic development. Collaborating with CDRH and DHCoE, the FDA offers consults on AI/ML drug submissions and is developing frameworks for AI/ML-based devices. The FDA has organised workshops, held a Patient Engagement Advisory Committee meeting, and fostered research partnerships to evaluate the safety and effectiveness of AI/ML products. CDER has developed the Innovative Science and Technology Approaches for New Drugs (ISTAND) Pilot Program. The model-informed drug development (MIDD) pilot programme uses AI/ML to enhance clinical trial simulations, dose optimisation, and safety evaluations.

The FDA’s Sentinel Initiative, including CDER’s Sentinel System, CBER’s Biologics Effectiveness and Safety (BEST) system, and CDRH’s National Evaluation System for health technology (NEST), is exploring AI/ML to enhance post-market safety surveillance. The FDA’s 5-year strategic plan for the Sentinel System includes using linked claims and electronic health records (EHR) data with advanced analytics. The Sentinel System Innovation Center’s four-pronged approach focusing on.

The CBER BEST system and CDER are leveraging AI/ML to improve data analysis and enhance postmarket safety surveillance. The BEST system aims to utilise AI/ML to analyse EHRs for predicting adverse events and generating real-world evidence on product efficacy. CDER is developing AI tools like the Information Visualisation Platform (InfoViP) to detect duplicate ICSRs, classify ICSRs by level of information quality, and visualise clinical event timelines. AI/ML methods are also being used to automate adverse event identification in drug product labelling. Another AI-based tool that focuses on drug product labelling is Computerised Labelling Assessment tool (CLAT), which serves to automate the review of label and labelling (e.g. prescribing information carton and container labelling). Through various programmes, the FDA has been engaged with industry to gather feedback on AI/ML applications in pharmaceutical manufacturing.

Considerations for the Use of artificial intelligence/machine learning (AI/ML) in Drug Development

AI/ML has significantly impacted drug development and is continuously evolving. However, AI/ML algorithms can amplify errors and biases present in the underlying data. This raises concerns about the generalisability and ethical implications of extrapolating findings beyond controlled testing environments. Additionally, the complexity or proprietary nature of these algorithms often limits their explainability due to not being fully transparent.

Discussion of Considerations and Practices for artificial intelligence/machine learning AI/ML in drug development

FDA is considering approaches to provide regulatory clarity around the use of AI/ML in drug development. Whilst its seen in the overarching standards and practice for the use of AI/ML, in the context of drug development the use of AI/ML may raise specific challenges. The FDA will therefore initiate discussions with stakeholders and solicit feedback in three key areas outlined below to help inform future regulatory activities:

Human-led governance, accountability, and transparency
Quality, reliability, and representativeness of data;
Model development, performance, monitoring, and validation

Many of the overarching principles and standards related to the characteristics of trustworthy AI can help inform considerations or key practice areas for the application of AI/ML in the context of drug development. The use of AI/ML in drug development raises challenges related to human-led AI/ML governance, accountability, and transparency; data considerations; and model development, performance, monitoring, and validation. Transparency and documentation across the entire product life cycle can help build trust in the use of AI/ML. In this regard, it may be important to consider pre-specification and documentation of the purpose or question of interest, context of use, risk, and development of AI/ML.

The V&V 40 risk-informed credibility assessment framework can aid in model development, performance, monitoring and validation. Engaging with the FDA early in the process can also help address these considerations effectively around the level of evidence and record-keeping required for the verification/validation of AI/ML models.

Summary

The FDA has released an initial discussion paper to explore the use of AI/ML in human drug and biological product development. This is to engage stakeholders on the use of AI/ML in drug development and so far has focussed on workshops and other mechanisms for addressing questions. Formal meetings and programs like the ISTAND Pilot Program and Real-World Evidence Program facilitate communication on relevant AI/ML methodologies. Communication and engagement with patients and the public for AI/ML in drug development is important to ensure patient-centered approaches.

Moving forward, the FDA will continue to seek feedback and collaborate with stakeholders to discuss further considerations for the safe implementation of AI in drug development workflows. Ultimately, this feedback could provide the foundation for future framework/guidance.

To discuss this with our regulatory experts, contact us at hello@dlrcgroup.com or visit www.dlrcgroup.com for more information on our award-winning services.