Generative AI-Enabled/Augmented Clinical Trials: The Road Ahead

Commentary
Article

The current clinical development model is shifting toward a generative AI-augmented proactive approach supported by real-world data for real time evidence.

Image credit: Kaikoro | stock.adobe.com

Image credit: Kaikoro | stock.adobe.com

With the recent advances in generative AI (GenAI) with large language models (LLMs), GenAI-enabled or augmented process improvement has garnered the attention of bioengineering and biomedical scientists, as well as regulatory authorities. The FDA has recently released two discussion papers on “Artificial Intelligence in Drug Manufacturing” and “Using Artificial Intelligence and Machine Learning in the Development of Drug and Biological Products.”1

As a result, existing commercial business practices in clinical developments especially in clinical trials will evolve rapidly and significantly impact the clinical development model. In this essay, we attempt to explore the implication of AI-enabled/augmented planning, implementation of clinical trials, and data analytics from the prospective of clinical development model.

From accelerated computing platform to GenAI-enabled/augmented informatics infrastructure transitions

Drug developments have been a slow, tedious, and costly process. A new drug coming to market would take 10 years and a billion dollars with a 90% failure rate which, until recently, remains mostly a trial-and-error process.2

Enter 2023, two simultaneous developments—namely accelerated computing and GenAI—are emerging in computing and information technology.3 These likely disruptive advances are a result of the exponential increase in computing power from innovative graphics processing unit (GPU) chip designs and a deep learning approach for training with existing data for classification, categorization, regression analysis, and pattern recognition based on business and/or use data.

Predictive AI is a form of computer-assisted machine learning (ML).3,4 The GenAI is inspired by the neural networks of the human brain, which is capable of multimodal reasoning for inference augmenting human decision-making when presented in training with a variety of formats of data—including structured data i.e., patient information in standard formats—for example, numerical and ordinal entries in clinical trial registries or unstructured ones, such as clinical notes, patient communications, medical imaging, audio, and video, etc.4-6

Both technological developments might have far-reaching implication for humanity, especially in the context of biomedical sciences, drug, and clinical development. GenAI is being rapidly adapted for healthcare and medicine, initially in transforming the clinical trial landscape, through the offer of a range of new ideas and creative solutions to accelerate drug development.7-9

GenAI could enhance overall efficiency through sensible and timely decision-making resulting in more accurate predictive analytics and workflow management of the clinical trial process, significantly improving patient outcomes. The GenAI-based model has demonstrated subtleties in mathematic reasoning10 and in the United States Medical Licensure Examinations (USMLE) soft skills assessment.

The recent release of the GenAI of LLMs of GPT-4 passed the assessment achieving close to 90% accuracy, implying that AI has the remarkable potential to meet the complex interpersonal, ethical, and professional demands intrinsic to the practice of medicine.11 Other notable progressions in clinical developments areGenAI-derived investigational medicinal products (IMPs), such as halicin and adaucin.12,13

Informatics is at the forefront of a fourth wave of industrial revolution. This information revolution is being driven by several creative disruptive forces starting from the internet for connectivity, the rise of data-driven insights in predictive analytics, predictive AI of deep learning with ML, and now the GenAI-enabled/augmented inference for process improvements with automations armed with robotics facilitated further by seamless human-machine interactions.9,14,15

Thereby, the technological evolution of AI movement from pattern recognitions of predictive AI to deep ML of inference for reasoning of GenAI powerful represents a quantum jump in terms of the powerful technical capability of computing algorithms, offering a future of productivity gains with greater performance improvements.

Regulatory oversight and policy formulation in GenAI-enabled/augmented clinical trials

In recent years, AI/ML has been increasingly explored to facilitate drug development and clinical trials. The two recent discussion papers on AI in drug discovery and clinical development from the FDA focus on the future regulatory framework anchored in ML/AI/GenAI deep learning and the associated challenges for regulatory decision making.

While data processing in classical ML follows conventional statistical modeling, the newer deep learning approach in the GenAI can manage a vast amount of scientific and clinical data with conceptually different maneuvers and in multiple dimensions. It emulates neural networks of the human brain by modeling multiple successive layers of data representation using transformations with back propagation steps, which allows it to adjust the weights of each layer for a given task toward expected outputs.4 This would make it highly useful for many potential use cases in clinical trials leading to greater efficiency and outcome improvements.1,9,13

It is likely that emerging insights from the reasoning capability of GenAI will have a significant impact on current clinical development models, with far-reaching implications for all stakeholders involved in clinical developments, including regulatory agencies, clinical investigators, industrial sponsors, contract research organizations (CROs), and patients and advocates.

In the end, human biology is complex with the involvement of many systemic and environmental risk factors in the pathophysiology of disease process. Human cognitive reasoning for decision is limited to handling in any time merely a score of variables. With that, it is perceivable that GenAI-enabled/augmented modeling could facilitate the decision-making process, although it currently remains a significant challenge in practice.

Nonetheless, it is foreseeable that the intensive data-driven process in clinical development could be assisted by GenAI-enabled/augmented decision. The accelerated computer platform and GenAI-deep learning algorithms may enhance the efficiency of the clinical trial process and provide real-time insights into real-world patient data from clinical trials, empowering patients with personalized, patient-centric shared decision supports.

To be clear that the semantic nuances with respect to whether the GenAI algorithms could enable or augment for the human decision process in clinical studies is debatable. In our opinions, the difference may simply lie in enabling which tends to be more autonomous while augmentation that could imply integrated human judgments in the decision loop.

Health care data ecosystems

The data-information-evidence-knowledge (DIEK) framework defines information as the data in a clinical context which also implies it should be interpretable for acceptable evidence or, in other words, clinically actionable.16

In the United States, the National Academies of Sciences, Engineering, and Medicine, and National Institutes of Health, as well as American Medical Association, are promoting the sharing of scientific and clinical trial data. The benefits to data sharing are substantial for knowledge growth in the medical sciences and to facilitate the development of treatments and products capable of improve human health.17-19

CROs are operating in a niche environment linking sponsors, clinical investigators, and health authorities. Its clinical development model is science-based and data-driven. The clinical and patient data ecosystems are multidimensional and is designed for being protective of patient privacy and safety.

The platform infrastructure and data governorship for their use should be fit-for-purpose, meeting the requirements in GenAI, which will likely be integrated with proprietary and customized GenAI algorithms. These need to be further validated according to best practices and continuously optimized by the follow-on hybrid real-world scientific and clinical data for incremental performance improvements.

Returning to the concerns about patient data safety and privacy, legal and regulatory mandates for meeting stringent government requirements means that industrial sponsors and clinical investigators must make every effort to ensure their use of protected health information (PHI) is appropriate and fit-for-purpose according to professional judgment acting in the best interest of the patients.

Potential shift of GenAI-enabled/augmented all stages of the clinical development model

The standard fee-for-service model of clinical development is usually focused on time in hours and efforts, i.e., activities with clients and investigators in executing the study plan according to approved protocols providing a range of quality services, and less on aligning payment terms with how efficiently they manage clinical trials.11 GenAI is likely to shift the clinical development model to a more data-driven and patient-centric trajectory path scoping to fundamentally change the current approach in clinical developments due to increasingly and readily available real-world data and high demands for resources in clinical trials to keep pace with high volumes of novel drug discoveries from GenAI.12 

Under these circumstances, it is expected that customized GenAI algorithms would be trained with data in the hybrid ecosystems from public domains, such as de-identified clinical trial data and proprietary sources such as that derived from intellectual properties, in addition to other forms of PHI in academia and industries. This should be an important characteristic of the next generation data ecosystems in configuration that is invaluable constitutional assets for future clinical development model. The GenAI algorithms for decision would create higher values if substantiated in validations and operational across all functions and stages of clinical developments from drug discovery and pre-clinical study to clinical trials.

Therefore, the current clinical development model is shifting toward a GenAI-augmented proactive approach supported by real-world data for real time evidence. The decision lag time will be shortened and the clinical trials would be more insightful in clinical contexts, in a sense that the full spectrum of relevant variables would be fully embedded in the GenAI algorithm for greater efficiency and performance improvements, accelerating clinical developments from pre-clinical to clinical stages with substantial cost-saving.

The road ahead

Overall, GenAI would have significant impacts on drug discovery and downstream clinical developments, transforming future clinical development models following clinical data sharing initiatives with predictive analytics approaches augmented by multimodal reasoning anchoring on real-world data empowered by GenAI deep learning algorithms.

About the Authors

Dave Li, MD, PhD, is a principal consultant and Clinical Research Physician with the KCR Consulting. He is a medical oncologist and regulatory scientist, and an expert in molecular medicine, immuno-oncology, and clinical informatics. He was on the faculty of Johns Hopkins Medicine and served as a medical officer with the US/HHS FDA before joining KCR. He obtained his medical degree from the Sun Yat-sen University, and MSc/PhD at the University of Texas M.D. Anderson Cancer Center at Houston, Texas.

Anna Baran, MD, is the chief medical officer at KCR and oversees all stages of clinical trial operations and KCR Consulting services. She brings expert clinical and medical experience to KCR, holding past positions in endocrine and immunology specialties. Baran has a medical degree and a postgraduate degree in healthcare management and is an expert in all stages of drug development.

References

  1. US Food and Drug Administration. FDA Releases Two Discussion Papers to Spur Conversation about Artificial Intelligence and Machine Learning in Drug Development and Manufacturing. https://www.fda.gov/news-events/fda-voices/fda-releases-two-discussion-papers-spur-conversation-about-artificial-intelligence-and-machine
  2. Duxin Suna D, Wei Gaoa W, and Hu H et al. Why 90% of clinical drug development fails and how to improve it? Acta Pharmaceutica Sinica B Acta Pharm Sin B. 2022;12(7): 3049–3062. doi: 10.1016/j.apsb.2022.02.002. PMID: 35865092
  3. MIT Technology Review Insights: The great acceleration: CIO perspectives on generative AI Report. https://www.databricks.com. Databricks 2023.
  4. Chollet F (ed.). Deep Learning with Python November First edition 2017 by Manning. ISBN 9781617294433
  5. IBM, What is AI inferencing? https://research.ibm.com/blog/AI-inference-explained
  6. Copeland M. What’s the Difference Between Deep Learning Training and Inference? August 22, 2016. https://blogs.nvidia.com/blog/difference-deep-learning-training-inference-ai/
  7. Liu Q, Huang R, and Hsieh J et al. Landscape Analysis of the Application of Artificial Intelligence and Machine Learning in Regulatory Submissions for Drug Development From 2016 to 2021. Clin Pharmacol Ther. 2023 Apr;113(4):771-774. doi: 10.1002/cpt.2668. PMID: 35707940
  8. Glass L, Shorter G and Patil R.IQVIA White Paper: AI in Clinical Development. https://www.iqvia.com/-/media/iqvia/pdfs/library/white-papers/ai-in-clinical-development.pdf
  9. McKinsey & Company. How artificial intelligence can power clinical development. https://www.mckinsey.com/industries/life-sciences/our-insights/how-artificial-intelligence-can-power-clinical-development
  10. Bischoff M. AI Matches the Abilities of the Best Math Olympians. Scientific American. January 17, 2024. https://www.scientificamerican.com/article/ai-matches-the-abilities-of-the-best-math-olympians/
  11. Brin D, Sorin V, and Vaid A et al. Comparing ChatGPT and GPT-4 performance in USMLE soft skill assessments. Sci Rep. 2023 Oct 1;13(1):16492. doi: 10.1038/s41598-023-43436-9. PMID: 37779171.
  12. Marchant J. Powerful antibiotics discovered using AI. Nature. 2020 Feb 20. doi: 10.1038/d41586-020-00018-3. PMID: 33603175.
  13. Cesaro, A., de la Fuente-Nunez, C. Antibiotic identified by AI. Nat Chem Biol 19, 1296–1298 (2023). https://doi.org/10.1038/s41589-023-01448-6. PMID: 37821719.
  14. Bleys J, Fleming E, and Mirman H et al. CROs and biotech companies: Fine-tuning the partnership. June 9, 2022. https://www.mckinsey.com/industries/life-sciences/our-insights/cros-and-biotech-companies-fine-tuning-the-partnership
  15. Maharaj N. Two Wharton graduates create AI startup that predicts drug trial outcomes. The Daily Pennsylvanian 12/03/23. https://www.thedp.com/article/2023/12/wharton-artificial-intelligence-biophy-drug-trials
  16. Dammann O. Data, Information, Evidence, and Knowledge: A Proposal for Health Informatics and Data Science. Online J Public Health Inform. 2018; 10(3): e224. Published online 2019 Mar 5. doi:10.5210/ojphi.v10i3.963, PMID: 30931086.
  17. Institute of Medicine. 2015. Sharing Clinical Trial Data: Maximizing Benefits, Minimizing Risk. Washington, DC: The National Academies Press. https://doi.org/10.17226/18998.
  18. NIH. Data Management and Sharing Policy. https://sharing.nih.gov/data-management-and-sharing-policy
  19. Flanagin A and Curfman G. Data Sharing and the Growth of Medical Knowledge. JAMA 2022 Dec 27;328(24):2398-2399. doi: 10.1001/jama.2022.22837. PMID: 36469324.
Recent Videos
© 2024 MJH Life Sciences

All rights reserved.