LANDRICH GROUP
The Transformative Role of AI in Clinical Trials
Software and life sciences are converging to accelerate drug and device development.
Innovations in machine learning (ML) have transformed early-stage drug discovery during the past five years. Now, breakthroughs such as Generative Artificial Intelligence (Generative AI) offer the potential to reduce costs, delays and risks traditionally associated with clinical trials
It wasn’t always such a hopeful story. For decades the rate of progress in software and the life sciences seemed to be on divergent paths. A Nature article famously coined the term eRoom’s Law, which is Moore’s Law spelled backwards, intended to draw the contrast between the two disciplines.
The axiom Moore’s Law was named after Intel founder Gordon Moore to describe exponential improvements in computation. The complexity of integrated circuits has doubled roughly every two years since the 1960s. Drug development on the other hand, according to eRoom’s Law, was experiencing a decades-long deceleration. The number of therapies approved per inflation-adjusted R&D spending decreased by half every decade in the past 50 years prior to the 2012 Nature publication.
Breaking eRoom’s Law
Advances in AI/ML have reversed the decades-long deceleration in successful clinical trials. The number of Biologics License Application approvals in the past 5 years is more than double the rate as when the ERoom’s Law article was published in 2012. Last year, the FDA approved 55 novel therapeutics in 2023, the second highest count in the past 30 years, including 38 New Molecule Entities (NME) and 17 Biologics License Applications (BLAs).
One example of the symbiotic relationship between software and therapeutic enhancement is the shift to next-generation genomic sequencing based on Bayesian analysis and prioritization algorithms. Another is the general movement towards Structure Based Drug Design, which leverages another AI technique called Deep Learning to identify, simulate and optimize an exponentially larger pool of potential drug candidates.
Impact on Medical Devices
The role of AI/ML in clinical trials for devices is even more rapid and transformative than drug development. In 2015, the FDA only authorized 6 AI-enabled devices annually. Last year, the agency authorized 221 devices. The vast majority of devices authorized in the past ten years focus on imaging and radiology, where AI techniques can improve imaging and assist triage, analysis and diagnostics. Emerging fields include robotics and augmented reality (AR), where AI might provide analytics and data visualization for a surgeon treating a patient.
Over 95 percent of devices approved by the FDA in the past decade have gone through the 510k pathway for moderate-risk devices that can prove a product is “substantially equivalent” to existing approved devices.
The Rise of Generative AI
In the past year, we’ve seen breakthroughs in transformer models that leverage Large Language Models (LLM) of information to search, synthesize and interpret enormous sets of unstructured data. These are known as Generative AI because they have the ability to generate a completely new information based on this unstructured data, such as a written answer to a question asked in natural human language via tools like Claude, Google Gemini and Chat GPT. Generative AI paves the way for even greater innovation and productivity in clinical trials.
Here are some key trends in clinical AI usage:
Trial Design
Initial design is the one-way door that sets the foundation for everything that follows, including years of effort and millions in investment. The characteristics of the patient population. The formulations and administration of therapy. The data to be collected, measured and weighted for evaluation. What if it was possible to simulate different design scenarios up front before making these considerable real-world commitments?
Software platforms like HINT (Sequential Predictive Modeling of Trial Outcome), InClinico and Trial Pathfinder let trial sponsors input criteria like patient eligibility, electronic health records (EHR) and target molecule structures to assess pros and cons of setting different parameters in the trial design. This enables a more data-driven approach to risk/reward trade-offs prior to final decisions.
Patient Recruiting
Studies show that globally 80 percent of clinical trials fail to recruit on time. New tools provide the ability to accelerate the search and selection process by sifting through vast amounts of data. Tools like ObvioHealth use AI to scan EHR records for specific inclusion and exclusion criteria. Unlearn.ai has created a “digital twin” using machine learning and statistical models to forecast a prospective patient performance in a clinical trial. These forecast clinical outcomes can then be included as covariates in an analysis of the estimated treatment effect.
Data Management.
One of the most important and resource consuming aspects of any large-scale trial, often drawing of vast stores of public data and scientific journals. Now there are a number of ways to search and analyze information critical to optimizing all aspects of a successful trials. Firms like Intelligent Medical Objects, Clinidigest and Autotrial use AI technologies like NLP and Large Language Models (LLM) to navigate historical and real-world data sources through simple co-pilot interfaces initially popularized with Chat GPT.
Regulatory Considerations
The FDA has approved hundreds of AI/ML-related biologic, drug and device applications in recent years. The agency has involved multiple departments, including the Center for Biologics Evaluation and Research (CBER), Center for Drug Evaluation and Research (CDER), Center for Devices and Radiological Health (CDRH) and Office of Combination Products (OCP), in efforts to stimulate collaboration and innovation in this area
A few specific areas where the FDA has provided guidance:
-
CBER workshops and public discussions on use of AI/ML in biological products promoting core principles like data integrity, transparency of algorithms, cybersecurity, health equity and elimination of bias.
-
CDRH guidance on submissions for AI/ML-enabled medical devices.
-
CDER framework for advanced manufacturing standards.
-
A coordinated discussion paper from CBER, CDER, CDRH and OCP to provide a holistic overview of agency strategy for regulatory guidance in this dynamic field.
Landrich Insights into Clinical AI
During 2024, the Landrich Group has committed to understanding how AI is rapidly transforming clinical research and product development. David Petrich, VP of Quality and Regulatory Affairs attended the RAPS conference in Long Beach, CA, which focused on AI as a source of efficiency in regulatory operations.
Petrich and Landrich CEO Tina Landess also recently attended the Global Expert Alliance EVOLVE 2024 Summit in Santa Clara, CA, an event focused on the convergence of AI and Deep Tech. This summit brought together innovators, entrepreneurs, and thought leaders to explore the groundbreaking ways these technologies are transforming industries, including healthcare.
Conclusion
We are entering a period of digital transformation in the conduct of clinical trials. Advances in the AI/ML have the potential to improve the speed, efficiency and effectiveness of key activities like trial design, recruitment and data management.
The FDA is already approving devices, biologics and drugs developed with these technologies and has demonstrated a commitment to ongoing guidance. Manufacturers, investors and clinical professionals can expect the synergies between software, hardware and life sciences will continue to expand opportunities to deliver and approval of new medical tools and therapies.
Glossary:
Artificial Intelligence (AI) – The field of computer science involving intelligence exhibited by machine software and hardware.
Machine Learning (ML) – A branch of AI that enables machines to learn from experience, patterns and statistic models to make predictions. For instance, processes millions of images of cats and other animals to predict whether an image is a cat. Or ingesting images of photographs, sketches and paintings to classify an image as a painting.
Generative AI – the ability to create entirely new information based on learning and analysis, such as creating an original painting of a cat, based prior learning and analysis of various forms of images including cats and paintings.
Deep Learning – A more advanced form of machine learning that incorporates more forms of data for more complex analysis such as synthesis vast quantities of genomic information to find a pattern relationship between genetic variants and disease occurrence.
LLM – A powerful Generative AI system based on Deep Learning that is capable of recognizing and responding to human language like Chat GPT or Google Gemini.
Moore’s Law – The observation by Intel co-founder Gordon Moore that the number of the transistors on a microchip doubles every two years, effectively doubling the computing capability available at the same cost.
ERoom’s Law – An observation in a 2012 Nature Magazine article that while the productivity and cost/benefit of computing has increased approximately every 2 years since the 1960s, the biopharma industry experienced the opposite trend, with the rate of drugs approved per $1 billion in inflation-adjusted dollars decreased by half every 9 years during the same time period.
Next Generation Sequencing – A technology that leverages AI/ML to simultaneously sequence many fragments of genetic material, reducing the time and cost required to effectively sequence DNA and RNA.
Structure Based Drug Design - A drug discovery and design approach that used AI/ML systems to computationally model the 3D structure of target proteins to model new therapies.
Sources:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7342339/
https://unlearnai.substack.com/p/digital-twins-in-clinical-trials
https://arxiv.org/abs/2307.14522
https://arxiv.org/pdf/2304.05352.pdf
https://aclanthology.org/2023.emnlp-main.766.pdf
https://www.fda.gov/media/167973/download?attachment
https://www.medtechdive.com/news/fda-ai-medical-devices-growth/728975/