The technology around clinical trials is advancing, but teams are facing the same fundamental questions: Who should we recruit? What outcomes can we expect, and how might they vary across patient subgroups? How many patients do we need to recruit? And how can we anticipate and account for inherent differences between trial participants and real-world patients?
For Clinical Development teams, answering these questions has traditionally required substantial time, data, and analytical work before the first patient is even enrolled. In recent years, increasingly robust real-world data (RWD) has helped streamline parts of that process, providing insight into patient populations, treatment patterns, and outcomes outside the trial setting. But that evidence was largely retrospective: a window into what happened in the past. AI is now changing that equation, turning RWD into a forward-looking evidence layer that can help teams explore different scenarios before enrollment begins and minimize many of the risks traditionally associated with trial design.
Let’s take a closer look at this important evolution and what it means across the clinical development lifecycle, focusing on one particularly powerful emerging approach: digital twins.
Retrospective evidence: strengths and limitations
RWD first established its value in clinical development as a source of retrospective evidence. Researchers have used — and continue to use — RWD very successfully to characterize disease natural history, treatment patterns and standards of care, describe patient populations, assess protocol feasibility, understand biomarker prevalence and testing patterns, and answer other foundational questions about the real-world treatment landscape.
Those questions remain essential. Understanding who real-world patients are, how they are diagnosed and treated, and how care has evolved in the past provides a critical foundation for developing the target product profile (TPP) for a new therapy, defining inclusion and exclusion criteria for a new trial, and assessing enrollment feasibility. Increasingly robust RWD and self-serve intelligence platforms like Flatiron Telescope have made this work faster and more iterative. Rather than commissioning a new study every time a question changes, researchers can query large RWD datasets themselves, build and refine cohorts, test assumptions, assess feasibility, and benchmark standards of care in a matter of minutes or hours.
These capabilities have fundamentally changed how researchers explore evidence. But they remain anchored in the past. When teams need to anticipate how the treatment landscape might evolve over the course of a trial, which is increasingly likely in today’s fast-changing oncology environment, they often end up extrapolating from historical data and making assumptions about what the future will look like.
AI is quickly changing that equation.
What AI unlocks: from hindsight to foresight
AI and predictive modeling add a new dimension to RWD: the ability to anticipate what is likely to happen, not just describe what has happened. By learning from historical patterns across large, longitudinal datasets, validated AI-based models can generate risk and prognostic scores, predict outcomes under current standards of care, and estimate cohort-level trajectories such as progression risk, event rates, and expected control-arm outcomes. These AI-based models are intended to complement randomized clinical evidence by providing additional context for trial design and interpretation, making them a valuable decision-support tool.
Just as importantly, these models can help explain why outcomes vary. By identifying the baseline characteristics that most strongly predict different outcomes, including ECOG status, biomarkers, sites of metastases, lab values, and patient history, AI can help teams understand which factors matter most and use those insights to inform population stratification and trial design tradeoffs.
One AI modeling technique making a big impact already on clinical development pipelines — and translating predictions into highly practical insights — is digital twins.
Digital twins: from prediction to action
What are digital twins, exactly? The term refers both to outcome prediction models built from large-scale RWD and to the virtual representation of a patient cohort generated by those models. Using baseline characteristics from patients enrolled in a trial, digital twins can be leveraged to estimate likely outcomes under different treatment scenarios, helping Clinical Development teams inform trial design and interpret results.
Digital twins applications are relevant across the drug lifecycle where development costs and timelines are taxing. By improving patient selection, strengthening trial design, and providing better context for outcomes, digital twins can help de-risk development programs and improve confidence in key clinical development decisions.
Here are four applications where digital twins are already adding substantial value:
- Patient stratification: Who should we recruit?
Not every patient responds to treatment in the same way. Baseline characteristics can drive meaningful variation in outcomes across real-world oncology populations. Digital twins can generate prognostic scores that help teams identify relevant patient subgroups and refine trial populations and design accordingly. - Control-arm benchmarking: What should we expect from a control arm?
Before enrollment, teams need realistic expectations for how comparable patients are likely to perform on standard of care. Digital twins trained on RWD can model expected outcomes for a comparable cohort, helping teams pressure-test inclusion and exclusion criteria and establish more informed control-arm assumptions during trial design. - Counterfactual control arms: How do we interpret single-arm data?
In early phase trials or when a randomized control arm is impractical, such as in rare cancers or settings with ethical constraints, teams still need context for interpreting single-arm results. Digital twins can generate counterfactual outcomes: estimates of what the trial cohort might have experienced under standard of care, providing a baseline comparison for observed outcomes and enabling interpretation of early signals to support go / no-go decisions. This is especially important in phase 1 / 2 settings when single arm studies are more common and identifying early efficacy signals are important. - Covariate adjustment: How can we mitigate prognostic imbalance?
Patient heterogeneity can introduce statistical noise and reduce trial power. Prognostic scores generated by validated digital twins AI models can be used to adjust for differences in baseline risk, reducing unexplained variation and improving statistical power in alignment with regulatory guidance. This helps teams detect treatment effects in smaller or heterogeneous populations without sacrificing rigor.
Prediction requires the right data foundation
Digital twins and other modern AI techniques are helping Clinical Development teams make more informed decisions across the development lifecycle, from protocol design and feasibility to trial execution and evidence generation — and in the process build a connected, dynamic evidence strategy capable of evolving with their assets — but these breakthroughs are only possible with the right RWD foundation. The data needs to be deep enough to account for all important variables, scalable to support robust, representative models, and of the highest quality to build confidence in the results.
Flatiron's Panoramic datasets, for instance, now cover longitudinal data for 6M+ de-identified global cancer patients, with unmatched clinical details and integrated molecular insights for 1M+ of them. Not only is the data extracted under expert oversight and thoroughly checked for accuracy, it’s also validated every step of the way against established clinical findings thanks to rigorous replication and benchmarking analyses.
With that robust foundation in place, we’re hard at work developing agentic tools and advanced analytical and modeling capabilities to expand what RWD can do for Clinical Development teams like yours around the world. Whether you’re looking to explore our retrospective data on your own, partner with our experts to design your next trial, or even co-develop new predictive capabilities alongside our research sciences team, we’re here to help you build the evidence strategy you need to support your development pipeline.
Please reach out to our Clinical Development experts below and let’s talk about next steps.
.png)
