Before AI Can Transform Clinical Trials, It Needs More Patient Data
By Professor Maddy Parsons, King’s College London
AI is becoming impossible to ignore in discussions about the future of drug development. New tools promise to improve everything from patient recruitment and trial design to biomarker discovery and treatment selection. The expectation is that AI will help us run faster, efficient, and more successful clinical trials.
But there is a danger that we will focus too heavily on the technology itself and not enough on the information that underpins it.
In a recent episode of Hypothesize That!, UCB Head of Global Clinical Development Graham Price and I debated a core hypothesis: Precision medicine will fail to deliver on its future promise because outdated trial frameworks keep us stuck in the past.
The Hidden Limitations Of Traditional Trial Models
For decades, clinical trials have been designed primarily to answer a specific question: Is a treatment safe and effective? That objective remains essential, but advances in technology now give us an opportunity to learn far more from every study than we have in the past. This is the key piece that needs addressing, before we can start to maximize the opportunities of AI in clinical trial practice.
Trials typically focus on collecting the information needed to meet regulatory endpoints. While entirely appropriate for demonstrating efficacy and safety, this approach can miss opportunities to generate a deeper understanding of disease biology. Valuable insights into underpinning mechanisms and how these relate to patient response to therapy are often left unexplored.
Understanding Why Patients Respond Differently
We know that patients respond differently to treatment for a wide range of reasons. Genetics play a role, but so do factors such as age, ethnicity, environment, disease stage, and broader biological variation. Yet we still do not fully understand why some patients benefit from a therapy while others experience limited or no response.
While we’re learning whether a treatment works against predefined endpoints, we’re not capturing information about why some patients respond, why others do not, and what that can teach us about disease and future trial design. To reach this level of understanding, we need a combination of more granular biological information from each patient and smarter ways to utilize and interpret the data we have. Applying new AI models to this data will accelerate our ability to decode and treat every patient.
Why AI Is Only As Strong As The Data Behind It
AI is exceptionally good at identifying patterns within large and complex datasets. However, it cannot uncover information that has never been collected. If the underlying data fails to capture the biological diversity of patients or the mechanisms driving treatment response, even the most sophisticated AI algorithms will have limited value.
The good news is that our ability to generate this critical, underpinning data has advanced dramatically in recent years. New technologies now allow us to examine tissues, cells, and disease processes at an unprecedented level of detail. We can combine information about gene expression, protein activity, molecular pathways, cellular interactions, and tissue architecture to build a much richer, unbiased picture of the biological basis of disease. Patient-derived models and organoids can also help us to investigate how biological features influence treatment response.
As part of our chat, Price also discussed how AI can help bring efficiency into our early interpretations of clinical data.
“I think AI probably pulls the efficiency and speed lever when it comes to data interpretation. Collecting data from the internal and external world, medical writing, predicting patient response and modelling trial designs are all areas of work that take a long time. But once the data is there, AI can really accelerate things in the future,” he said. “For example, there’s potential for AI models to take a look at molecular structure and mechanisms of action, and with this information, start selecting a spectrum of populations where this might work, which is already increasing the number of indications in patient types you can treat with one new medical entity.”
A New Vision For Precision Medicine And Trials
This thinking is reflected in initiatives such as the Smart Trials Hub at King's College London, which was established to address two of these persistent challenges in clinical research: our limited understanding of why patients respond differently to therapies, and the underrepresentation of diverse populations in clinical trials.
The Hub brings together tissues, clinical information, advanced molecular profiling from diverse patient populations, and patient-derived models to create a deeper understanding of disease biology — uncovering why some individuals respond to therapies and others don’t. When we generate high-dimensional datasets across diverse patient populations, we aim to identify the molecular features associated with treatment responses and understand why those responses vary between individuals. We can also harness this data to uncover new biological targets for intervention when current therapies fail.
Importantly, AI is part of this approach, but the starting point is the biological evidence that feeds the AI engine. AI and machine learning then help us integrate and interpret these complex datasets, revealing patterns that would otherwise remain hidden.
These approaches create an opportunity to rethink the role of clinical trials themselves. Rather than viewing trials solely as a mechanism for testing drugs, we should increasingly see them as opportunities to learn about human biology. Every patient enrolled in a study represents a chance to better understand disease mechanisms, identify new therapeutic targets, and determine why treatments work differently across populations.
Advances in molecular profiling, data science, and translational research mean we can now ask more ambitious questions.
That shift in thinking is particularly important in the field of precision medicine. Precision medicine is often presented as creating a unique treatment for every patient; a more practical goal is understanding broader biological characteristics that define different patient groups and using that knowledge to match therapies more effectively.
Redefining Success In Clinical Trials
The next generation of trials must maintain a focus on outcomes, while also gathering and generating as much knowledge as possible. That means collecting richer biological data, embracing diversity, and treating every clinical trial as an opportunity to learn. Success should not be measured solely by whether a study reaches its endpoint, but also by how much we learn about disease biology and patient response along the way. Increasing the amount of data we retrieve from every patient will massively accelerate and enable machine learning tools to spot patterns and derive new hypotheses for validation. Importantly, achieving this goal will require close partnership between industry and academia to effectively harness emerging innovations and secure data sharing to maximize accuracy of AI models.
AI will undoubtedly become an increasingly important part of clinical development. But better algorithms alone will not solve the challenges facing clinical research. We need to understand where AI can provide true value in accelerating and improving clinical trial outcomes through biology-informed decisions.
About The Author:
Maddy Parsons is professor of cell biology and dean of research excellence frameworks at King’s College London. She has over 20 years’ experience leading interdisciplinary discovery research teams in the development of spatial biology approaches to understand human disease. Maddy also works closely with many leading biopharma industry partners to harness her discoveries for patient benefit. She is director of three imaging core facilities — including the recently established Smart Trials Hub — and leads several national and international organizations and initiatives to enable open access to advanced imaging technology. Maddy has received several awards in recognition of her work including election to the Academy of Medical Sciences.