I've spent my career as an epidemiologist working at the intersection of real-world data and oncology—first studying cancer etiology and outcomes, and now at Flatiron Health, where I collaborate with biopharma clinical development teams to build the evidence strategies that underpin some of the most consequential decisions in drug development.
Working closely with these teams, it is inspiring to see the current pace of innovation in oncology research. Every June, there are trial readouts met with applause on oncology’s biggest stage: ASCO, serving as a preview of the shift in the treatment paradigm that will follow shortly thereafter. Just last week thousands stood to enthusiastically cheer the trial results for a new potential therapy that showed a doubling of survival among pancreatic patients during this year’s ASCO plenary session. The promise of AI to accelerate early-stage discovery suggests this trend in innovation will only intensify as more novel therapies progress through the development cycle toward human trials. In order to meet, or better yet, further catalyze the innovation we are seeing in clinical development, real-world evidence (RWE) generation is due for a revision.
Each stage of clinical development has distinct needs, and a modern evidence strategy requires access to accurate and clinically current insights at each specific inflection point across the development timeline. No single study can be commissioned early and repurposed indefinitely, but rather, we need a connected approach that keeps pace with the research questions and clinical practice as they both evolve.
There are clear opportunities for the right RWE architecture to enable clinical development teams across their lifecycle to pivot with more confidence, capitalize on possibilities to generate robust insights, and increase the likelihood of success throughout the lifecycle. Below I’ll walk through some examples showing how a connected strategy can be the ultimate catalyst for translating clinical potential into meaningful gains in patient outcomes.
Getting Early Decisions Right, Quickly
For programs with multiple plausible development pathways, early decisions around target product profile refinement and trial design carry long-term consequences. When cross-functional teams meet to weigh in on important decisions, answering one question often surfaces new ones, requiring rapid evidence-based answers in order to keep the development timeline on track.
Consider what this may look like for a prostate cancer program: a clinical development team runs monthly governance meetings focused on quantifying the unmet need that their novel therapy can address, despite recent market approvals. The questions shift weekly as new signals emerge and the target population is refined. The team needs to know whether the eligible population’s unmet need is in the post-ARPI treatment landscape, if patient outcomes differ based on HRR-mutation status, if taxanes or ARPIs or RLTs should be considered as a combination partner. Each week the focus narrows.
To truly understand if their investigational therapy has the potential to improve upon the existing standard of care, teams typically answer these questions using two approaches:
- The first relies on existing literature, which often means extracting insights from a population that doesn’t directly correlate the target population, or worse, is not recent enough to reflect the current treatment landscape.
- The second approach requires queuing a dataset request, designing a large study in order to address every potential question, and finally running analyses over many months.
What teams need at this stage isn't a commissioned study, but rather the ability to iterate on population assumptions in near-real-time. Analytics platforms that enable natural language querying of high-quality RWD, like Flatiron Telescope, are emerging to fill exactly this gap. Using Flatiron Telescope, teams can evaluate population feasibility, examine treatment distributions across subgroups, and iterate on study design assumptions in days rather than weeks or months. Additionally, monthly data refreshes with a 60-day recency (or more frequent) mean you're working against a contemporaneous picture of clinical practice, not a static snapshot from a prior year.
Establishing the Baseline
Once a development path is taking shape, it’s often the right time to commission natural history studies to establish an evidence backbone for phase III trial design through post-launch activities. This is the right instinct, but the timing and anchoring of these studies matters more than is commonly acknowledged.
Using bladder cancer as an example, the accelerated approval of enfortumab vedotin + pembrolizumab (EV+P) by the U.S. Food and Drug Administration in 2023 significantly reshaped the first-line treatment landscape. Where treatment decisions were once largely split based on cisplatin eligibility, the pool of patients who can receive a highly active regimen upfront has broadened and reduced reliance on traditional platinum-based chemotherapy as the default starting point. Now EV+P is showing promise even earlier in the treatment journey as a perioperative therapeutic option, and as a result, clinicians are rethinking how and when to use regimens after progression on EV+P, and treatment sequencing strategies are beginning to shift.
The consequence is that a natural history study drawing on data from two or three years ago—sometimes even less—may not accurately characterize the patient population the trial will actually enroll. The standard-of-care benchmarks that comparator arms were built against no longer exist in the same form. That's not a data quality problem. It's a recency problem. And in fast-moving indications, recency is a scientific and clinical requirement.
Flatiron's Panoramic data, with longitudinal, EHR-derived clinical data updated with two-month recency across more than five million de-identified patient records, is what makes a current, credible natural history analysis possible. Our bladder cancer Panoramic dataset spans 50,000 patients and captures clinical depth with recency to enable timely understanding of patients in the real-world, allowing teams to optimally position both emerging assets in an increasingly complex treatment paradigm.
Surfacing the Emerging Signals
The deep patient characterization gathered during natural history studies often yield critical evidence on the frequency and prognostic or predictive impact of emerging biomarkers on clinical outcomes. This information is essential as programs move from concept to design, raising a related challenge: how do you evaluate whether an emerging biomarker-defined population is worth building into a trial design before committing to the full study infrastructure?
In breast cancer, this is increasingly relevant as researchers explore alterations that aren't yet part of routine clinical testing, such as RB1, FGFR1/2, CCND1 mutations, and others, but may carry real prognostic and predictive significance. A development team may want to understand whether patients positive for a particular alteration show meaningfully different outcomes in real-world practice before committing to a biomarker-stratified development program.
Earlier on in development, if a team has identified a candidate biomarker signal, understanding what it means clinically, from co-occurring alterations to how it maps treatment response in a real patient population, requires data that goes well beyond biomarker positivity status. Flatiron's multimodal datasets, the Flatiron Health–Foundation Medicine Clinico-Genomic Database (CGDB) and the Flatiron Health–Caris Life Sciences Clinical-Molecular Database (CMDB), link deep molecular information with our industry-leading clinical data to provide answers that surface important signals for clinical development teams.
For later stage research questions, especially one interested in scale or identifying a larger cohort of positive patients, AI-enabled extraction can surface a more expansive and robust population across a variety of testing platforms. Structured fields often don't capture non-standard of care biomarkers. AI applied to unstructured EHR data can surface these signals at scale, extracting test results and associated clinical context that wouldn't be reachable through any structured query. For exploratory outcome stratification in an emerging biomarker-defined population, it's a capability that didn't exist a few years ago, and one that's becoming increasingly relevant as the number of viable targets continues to grow.
Increasing Trial Success
The importance of a dynamic RWE strategy continues well beyond the initial trial design, offering opportunities to further increase the likelihood of trial success and accelerate timelines. Advancements in analyzing large, quality datasets with clinical depth on the patient journey like Flatiron Panoramic data using novel machine learning-based predictive models have unlocked new ways for RWE to support trials.
Digital twins, one example application of building patient-level prediction models from large cohorts, are transitioning from methodological concept to powerful analytic tool. Statistical approaches spanning penalized regression to deep learning can be developed using Flatiron's real-world data with clinically validated endpoints, and our teams have demonstrated their ability to reproduce trial-observed outcomes with remarkable accuracy. These powerful predictive models can be used to develop prognostic scores for risk stratification as a part of trial design or even as covariate adjustment to increase statistical power of the trial. This type of application can help clinical development teams mitigate against potential underenrollment issues, better detect treatment effect in pre-defined subgroups, and more generally increase likelihood of trial success given the improved ability to detect a treatment benefit by reducing the surrounding noise. In findings presented at ASCO this year, these models have supported sample size reductions in the range of 9–21% while maintaining statistical power—a gain that can be the difference between a study that leads to approval for patients and one that stalls.
This predictive modeling application extends beyond sample size efficiency and ability to detect treatment effects during the trial analysis itself. Earlier on in the development cycle, for a program sitting on a promising phase I or II single-arm signal, simulating what a phase III would look like under different design assumptions based on digital twins creates an evidence-informed basis for go/no-go decisions that compliments phase I or II data. In the right situations, these types of insights can improve upon external control arms to ensure the conclusions drawn and decisions made are grounded in robust evidence.
The new RWE playbook for clinical development
Each of these capabilities can address distinct evidence needs at various stages of the clinical development timeline, and suggests that the RWE playbook for clinical development isn't just about access to datasets—it’s about timely and high-quality data, and adopting a dynamic mindset to derive the insights that are needed most at a given timepoint in the development cycle. The innovative oncology clinical development teams navigating today’s fast-moving treatment landscapes aren't treating evidence generation as a series of isolated, ad-hoc studies. Instead, they are building a connected, dynamic strategy that evolves alongside their clinical questions from day one. By treating RWE as a continuous catalyst rather than a static evidence source, teams can pivot with confidence, maximize their probability of success, and ultimately bring life-saving therapies to patients sooner.
At Flatiron, we are making sure that the RWE derived from global Panoramic data to our linked multimodal datasets, and using platforms like Telescope, is designed to answer the most important research questions at each stage by learning from the experience of people with cancer.
If you're thinking through how to structure an evidence approach for a current or upcoming oncology program, we'd welcome the conversation. Reach out to Flatiron's Evidence Solutions team to learn more.
