Artificial intelligence (AI) in oncology is evolving quickly - from headline-grabbing breakthroughs to the steady, rigorous work that turns potential into real-world impact. While Flatiron is an industry leader in AI-powered technology across EHR and oncology real-world data (RWD), we are also pushing the boundary of cutting-edge scientific methods . Our research is anchored in Digital Twin Predictive Modeling, focusing on high-impact areas like patient stratification, control arm benchmarking, covariate adjustment, and more. By bringing together our world-class RWD and AI expertise, we are working to bring transformative impact to oncology clinical research and, ultimately, patient care.
Unmet needs
Despite tremendous medical advances in the past couple of decades, there are still huge unmet needs for cancer patients. Developing innovative medicines are facing real pressure - trials are expensive, time-consuming, and probability of success continues to be low. Innovations for clinical trial design and execution have moved from nice-to-have to urgent. Many innovative ideas, like external control arms, once considered exploratory — has become standard practice. But there are more that we can do, in particular with predictive modeling.
What we mean when we say "predictive modeling"
The core idea of predictive modeling is to transform historical data into foresight. We use advanced methods to learn the complex patterns within vast datasets, and use these patterns to make predictions. Predictive models also allow us to simulate outcomes under different conditions, the foundational concept for digital twins. In oncology research, prediction might mean estimating a patient's risk of disease progression, forecasting the next lab measurement, or assessing the probability of responding to an alternative therapy. These aren't abstract outputs. They're inputs to decisions—about how to design a trial, how to select a patient population, or how to interpret signals from the clinic.
Predictive modeling draws on a variety of methods, ranging from traditional statistical methods, machine learning, deep learning, and more recently generative AI and large language models. What makes predictive modeling really exciting right now isn't just the methods themselves. It's what those methods can do when applied to RWD of the quality, scale, and depth that we have today.
The timing is right
RWD has come a long way. The richness of longitudinal, structured and unstructured clinical data from electronic health records, the kind that captures what actually happens to patients across diverse care settings, has grown substantially. So has our ability to work with the data—curate them and extract clinically meaningful variables with the level of accuracy and precision that research requires.
That matters because the value of predictive modeling is fundamentally dependent on the quality of the data they are trained on. While models trained on low-quality data will have limited utility, those that are built on high-quality, rigorous data can generate reliable insights essential for research and patient care.
What Flatiron is doing in this space
Our team has been doing meaningful work at the frontier of predictive modeling. At ESMO AI and the upcoming ISPOR Annual Meeting and ASCO Annual Meeting, we present research on predictive models powered by Flatiron data—including work on digital twins.
We're still early in uncovering the full potential of predictive modeling in oncology, and that's part of what makes this work exciting. The questions are hard, the stakes are high, and the opportunity to accelerate drug development and improve patient care is real.
If you're thinking about how predictive modeling could apply to your research priorities, we'd welcome the conversation. Visit our ISPOR landing page to explore our latest research, or come find us in person at the conference to talk through what these methods could mean for your work.
***
Jacqueline Law is VP of Scientific Engagement and Applied Research at Flatiron Health, where she leads a team focused on advancing AI and real-world data methods for oncology research.
