On April 7, Friends of Cancer Research (Friends) hosted a public meeting, “Application of External Control Arms in Oncology Drug Development” to explore opportunities to advance the use of external control arms (ECAs) in oncology drug development. ECAs are constructed from external data sources, such as prior clinical trials or real-world data (RWD), and can provide valuable comparative evidence for evaluating new therapies, particularly when traditional randomized controlled trials (RCTs) are impractical, including settings with small or hard-to-enroll patient populations and where the standard of care (SOC) remains limited or associated with poor outcomes. The event featured preliminary findings and insights from the Friends ECA Pilot Project, a multistakeholder research partnership aimed at advancing best practices for generating rigorous, reliable evidence through ECAs. Throughout the meeting, discussion focused on how ECAs can be implemented to support regulatory decision-making and expand timely patient access to promising therapies.
One-Pager: “ECA Pilot Project Overview and Preliminary Findings”
Presentation Slides: “Evaluating External Control Arms: Results from a Collaborative Pilot Study“ by Bernat Navarro, Friends
Key Takeaways
- Data Feasibility: The ability to generate a credible ECA depends fundamentally on the quality, completeness, and relevance of the underlying data. Complete or nearly complete baseline information is essential to operationalize eligibility criteria and achieve balance with the target trial control arm. Data missingness remains a significant barrier to ECA implementation and distinguishing whether data are structurally missing or simply not collected is critical for assessing fitness for use.
- Methodological Rigor: Approaches such as propensity score matching and weighting, alongside multiple imputation by chained equations (MICE), are important for addressing confounding and missing data. However, these techniques cannot compensate for the absence of key prognostic variables, underscoring the importance of robust data collection at the outset.
- Evolution Toward Modern Evidence Generation: Evidence generation is shifting toward more integrated use of clinical trials and external data. ECAs are already providing value in settings such as rare cancers and diseases with limited or poor-prognosis SOC. Further progress will be supported by more prospective use of ECAs, including hybrid trial designs and improved integration of longitudinal data, digital health tools, and biomarker-driven approaches.
- Regulatory Alignment and Prospective Planning: Effective use of ECAs requires early and iterative engagement with the U.S. Food and Drug Administration (FDA). ECAs should be incorporated into trial design from the outset rather than applied retrospectively. While regulatory flexibility exists, evidentiary standards for demonstrating effectiveness remain unchanged. Future trial designs should better reflect real-world clinical practice to support more generalizable populations and enable the use of robust external comparators.
Keynote Fireside Chat with Michael McCaughan, Prevision Policy and Amy Abernethy, Highlander Health
Lessons from Early Pilots and the Importance of Collaboration
Moderator Michael McCaughan and Dr. Amy Abernethy kicked off the fireside chat by reflecting on the evolution of RWD, including prior efforts such as the Friends RWD Pilot 1.0 Project and the COVID-19 Evidence Accelerator led by Friends and the Reagan-Udall Foundation. Dr. Abernethy emphasized how these early pilots demonstrated that diverse stakeholders, including biotechnology companies, regulators, and academic experts, could align under a shared protocol to define a use case, a methodological approach, and discuss the future direction of RWD applications. She noted that a key driver of progress has been incorporating real-time feedback, such as input from the Oncology Center of Excellence, which established early parameters around trustworthy RWD methodologies.
Advancing Data Quality and Analytical Guardrails
Dr. Abernethy highlighted the maturation of RWD inputs, noting the transition from rudimentary claims and electronic health record (EHR) data to longitudinal, multimodal, high-quality datasets including patient reported outcomes (PROs) and genomics, which together to capture the patient’s full clinical trajectory over time instead of isolated snapshots. She noted that as data have evolved, so too have the required epidemiological and statistical methodologies. Dr. Abernethy stressed the need for clear analytical guardrails. Rather than waiting for the FDA to publish exhaustive guidance, industry can proactively showcase what good looks like. This includes clearly documenting data curation (e.g., data linkage, deduplication, and the translation of unstructured notes into high-quality variables) and how analytical methods are defined and reported. This transparency can show support how science and the methods evolve and help reviewers interpret the data in a consistent way. This transparency can support consistent interpretation and build confidence in RWD-based evidence
Optimal Use Cases for ECAs
Addressing the optimal deployment of ECAs, Dr. Abernethy noted that the clear frontrunners are rare diseases and rare cancers, where conducting traditional RCTs is statistically and practically infeasible. Beyond rare disease settings, ECAs can support decision-making across the product lifecycle, including early development (e.g., internal go/no-go decisions), post-approval confirmatory studies, and prospective registries. In rare disease settings, historical data curated by patient advocacy groups or distributed global investigators can be aggregated to establish common controls, provided the endpoints align with variables that hold direct clinical meaning for patients.
Session 1 – Methodological and Data Insights from the ECA Pilot Project
Key Takeaway
The pilot findings underscore that credible ECAs require not only robust data and appropriate analytical methods, but also closer alignment between trial design and real-world clinical practice.
Session 1 provided an overview of the ECA Pilot Project (hereafter referred to as “the Pilot”), including its design, data sources, analytic approach, and key findings. It highlighted methodological and operational lessons from the project, and established a shared foundation for subsequent discussions on drug development applications and policy considerations. The session was moderated by Elizabeth Garret-Mayer (American Society of Clinical Oncology), with panelists Ruthie Davi (Medidata, a Dassault Systèmes company), Janet Espirito (Ontada), Brad Karalius (AstraZeneca), Cassadie Moravek (Pancreatic Cancer Action Network), Jennifer R. Rider, (ConcertAI), and C.K. Wang, Verana Health (COTA).
Presentation by Bernat Navarro: Friends ECA Pilot Project Overview and Preliminary Findings
Bernat Navarro, Friends’ Senior Science Policy Analyst and ECA Pilot Project Lead, presented preliminary results and outlined the project’s key objectives: assessing reproducibility, characterizing data and methodological variability, and defining criteria for fit-for-purpose ECAs supporting regulatory decisions. He highlighted the use of the RESOLVE trial control arm as the target alignment population. This arm comprised adult patients with metastatic pancreatic ductal adenocarcinoma (mPDAC) receiving first-line treatment of gemcitabine and nab-paclitaxel. In total, 9 data partners contributed data to the Pilot Project. The presentation included initial findings from six partners that applied a shared statistical analysis plan and independently constructed an ECA.
Addressing Data Attrition and Missingness
The Pilot highlighted the challenges of data attrition when applying RCT inclusion/exclusion (I/E) criteria to external data sources. Jennifer R. Rider and Janet Espirito explained that attrition is driven by two distinct sources: true exclusion based on eligibility criteria (e.g., patients presenting with abnormal hematologic or hepatic laboratory values) and exclusion driven by data unavailability (e.g., the required data was not captured in RWD or fell outside the ascertainment window.)
Rider warned against automatically excluding patients with missing data, as this can introduce selection bias and drastically reduce the cohort size. C.K. Wang proposed tiering I/E criteria, where core clinical disease-related characteristics are prioritized and ancillary data (e.g., other medications the patient takes) are considered lower-tier elements for which data missingness may be acceptable. The working group acknowledged that understanding real-world clinical practice is necessary to determine whether unrecorded laboratory values represent a deviation from SOC or simply operational variability in data capture.
Achieving Baseline Balance and Methodological Flexibility
To emulate randomization and manage data limitations, the working group implemented multiple imputation by chained equations (MICE) for key prognostic variables. Ruthie Davi explained that while MICE correctly estimated variability, the computational burden of propensity score modeling limited the number of imputation iterations compared to standard approaches for handling missing outcome data.
The pilot permitted data partners flexibility in their balancing methodologies, utilizing both propensity score matching and weighting. Davi noted that propensity score matching was favored by some project partners due to its conceptual similarity to traditional RCTs. Karalius observed that variations in matching techniques or the inclusion of more complex doubly robust machine learning models offer only marginal benefits compared to the fundamental need for sufficiently rich baseline risk factor information. The pilot demonstrated that when comprehensive baseline data is available, external controls closely mimic the randomized control, effectively mitigating unmeasured confounding.
Designing Clinical Trials to Emulate the Real-World
A recurring finding was the inherent divergence between clinical trial populations and routine oncology care. For instance, C.K. Wang noted that patients enrolled in clinical trials tend to have more favorable prognoses due to trial parameters such as the clinical stability required for lengthy screening periods, which can span up to 6 weeks. In contrast, patients with metastatic PDAC who are treated in community settings face a median survival of just 2 months.
To build credible ECAs moving forward, Karalius and Moravek advocated for “pragmatic trials” with I/E criteria that emulate real-world populations and clinical care. This would minimize the need to force RWD to match stringent and exclusionary RCT parameters. Designing pragmatic trials could minimize control arm dropout rates and directly facilitate the construction of more generalizable ECAs.
Session 2 – Translating Evidence to Practice: Integrating ECAs into Oncology Drug Development
Key Takeaway
Fit-for-purpose use of ECAs is highly context-dependent, with increasing evidentiary expectations as ECAs move from descriptive use toward supporting regulatory decisions—making data quality, transparency, and early FDA engagement essential.
Session 2 explored the integration of ECAs into oncology drug development, including the need for early evidence planning and proactive engagement with regulators. The panel discussed strategies to overcome core operational, data, and methodological barriers, highlighting the potential of hybrid trial designs and robust data curation to enable more efficient, patient-centered clinical trials. The session was moderated by Ashita Batavia (J&J Innovative Medicine), with panelists Pete Ansell (AbbVie), Jaclyn Bosco (IQVIA), Marie Bradley (FDA), Neal J. Meropol (Flatiron Health), and Jane Perlmutter (Patient Advocate).
Defining Fitness-for-Purpose and Regulatory Integration
Ashita Batavia guided the panel through the practical applications of ECAs across the oncology drug lifecycle. Jaclyn Bosco delineated the hierarchy of ECA applications, emphasizing that different levels of analytical rigor are required depending on the intended application. ECAs used for descriptive purposes (e.g., characterizing treatment patterns) require less stringent methods, whereas ECAs used for contextualization (e.g., interpreting single-arm findings without formal comparison) require a higher degree of rigor. The most stringent requirements apply when ECAs are intended to support regulatory decision-making, where formal comparative analyses and high-quality data are essential. Marie Bradley provided insight into the FDA’s multidimensional fitness-for-use evaluations. The FDA assesses relevance—whether the data source captures the target disease, treatment, and clinical context—and longitudinality (the ability to follow patients continuously without large gaps in care due to fragmented healthcare systems). Sponsors must provide quantitative assessments of missingness for key prognostic factors, ensure that clinical outcomes are measured objectively and consistently, and submit patient-level data to the FDA for independent reanalysis—a process often complicated by data use agreement and ownership constraints.
Categorizing the Core Barriers: Operational, Data, and Methodological
Bosco synthesized the primary barriers to ECA integration into three core domains:
- Operational Barriers: Securing third-party data access and navigating prolonged contracting phases are routinely underestimated logistical hurdles that delay evidence generation.
- Data Barriers: Evaluating completeness and structural missingness remains the paramount challenge. Pete Ansell noted that modern therapies, such as antibody-drug conjugates, rely heavily on molecular biomarker expression (e.g., protein overexpression), information that is frequently absent in RWD.
- Methodological Barriers: Calculating effective sample sizes and accurately determining lines of therapy—particularly when treatment sequences are incompletely captured in EHR data—require advanced epidemiological approaches.
Neal Meropol highlighted that not all datasets are curated equally. Sponsors must rigorously understand how unstructured data is extracted—evaluating human abstractors against natural language processing and large language models on a parameter-by-parameter basis. Meropol also pointed to the potential for ambient scribes and clinical transcripts to extract nuanced side effects and decision-making logic, which are not captured in standard EHR data collection.
Hybrid Trial Designs and Patient Reported Outcomes (PROs)
To bridge the gap between real-world external controls and internal randomization, the panel strongly endorsed hybrid externally controlled trials. Bradley confirmed the FDA is actively funding demonstration research on adaptive borrowing approaches, which allow statisticians to adjust the influence of RWD based on its comparability to the internal control. This hybrid structure increases statistical efficiency, mitigates Type 1 errors, and helps isolate sources of bias by comparing the small internal control against an external cohort.
From a patient advocacy perspective, Jane Perlmutter championed hybrid and external designs as a mechanism to accelerate trial accrual. Because oncology patients are highly motivated to receive investigational therapies, hybrid designs that guarantee a higher probability of placement on the treatment arm by reducing the number of patients on the internal control arm would improve trial enrollment. Perlmutter also emphasized the critical gap in capturing PROs within RWD. As cancer therapies expand, RWD must systematically capture tolerability and patient experience, not just overall survival (OS) and severe adverse events.
Scaling ECAs Across the Development Lifecycle
Ansell outlined the utility of ECAs beyond late-stage regulatory filings. In early Phase 1 and Phase 2 development, ECAs allow sponsors to benchmark novel molecule efficacy against expectations, identifying specific clinical or molecular subpopulations that demonstrate heightened sensitivity. This data can be used to optimize subsequent Phase 3 study designs by narrowing the target population, shrinking the required sample size, and improving drug development efficiency. Furthermore, in settings where the SOC is rapidly evolving, such as multiple myeloma, contemporaneous RWD can provide vital context for trials using an SOC that became outdated during the enrollment period.
Session 3 – Policy Perspectives and Regulatory Priorities for ECAs
Key Takeaway
Advancing broader use of ECAs will depend on shifting from retrospective, opportunistic data use toward more prospective and shared data strategies, supported by iterative and transparent engagement with FDA.
Session 3 examined the policy landscape and regulatory priorities for ECAs, emphasizing the need for early and iterative communication with the FDA. The panel advocated for a shift toward prospective data collection and the use of pooled or shared control arms to address limitations of retrospective data, while still maintaining the rigorous statutory requirements for demonstrating substantial evidence of effectiveness. The session was moderated by Clay Alspach (Leavitt Partners and ACRO), Amy Comstock Rick (FDA), Joe Franklin (Biotechnology Innovation Organization), Mark Lee (N-Power Medicine), and Donna Rivera (Canal Row Advisors).
Aligning Stakeholder Expectations and Regulatory Toolkits
The final panel focused on the macro-policy environment and alignment between sponsors, RWD organizations, and the FDA. Joe Franklin highlighted that while all stakeholders agree on the utility of ECAs, achieving alignment on the operational, data, and design complexities remains difficult. Sponsors face significant investment decisions when pursuing novel evidence generation, weighing the cost of data procurement against the scarcity of formal meeting times with the FDA. Franklin argued that the current regulatory toolkit—comprised of formal guidance documents, structured meetings, and case studies—is insufficient to provide the regulatory certainty sponsors seek. The key enabler for more routine ECA deployment will be establishing an interactive, iterative feedback loop between sponsors and FDA reviewers, allowing for real-time methodological refinements and adjustments.
The “6 Cs” of Externally Controlled Trial (ECT) Implementation
Donna Rivera distilled six regulatory and methodological challenges of implementing ECAs into clinical trials:
- Complex Design: The multidisciplinary nature of ECAs requires integrated expertise across epidemiology, biostatistics, and clinical medicine to navigate unprecedented trial architectures.
- Clinical Context: ECAs must be appropriate for the specific disease state.
- Completeness of Capture: Data must be fundamentally fit-for-use, providing measurable endpoints and baseline covariates without compromising structural integrity.
- Comparability of Populations: Cohorts must be comparable, taking into account temporal alignment (e.g., index dates) and real-world practice patterns (e.g., geographic impact on SOC, changing SOC).
- Confounding and Bias: Sponsors must pre-specify protocols, statistical analysis plans, and sensitivity analyses to demonstrate robustness.
- Clinical Benefit: The data must be interpretable and able to demonstrate a sufficient magnitude of benefit to meet the high statutory bar for substantial evidence of effectiveness.
The Shift Toward Prospective ECAs and Shared Controls
Mark Lee challenged the industry’s reliance on opportunistic, retrospective data approaches. Instead, sponsors should pivot toward prospective use of ECAs. By anticipating clinical settings where patients may refuse randomization—such as minimal residual disease (MRD) positive cohorts where recurrence is almost guaranteed—sponsors can collect RWD and design ECAs that mirror trial components, including I/E criteria, blood draws for circulating tumor DNA, and imaging modalities and schedules aligned with RECIST evaluations. Lee also identified crossover trials in accelerated approval pathways as strong candidates for prospective ECAs. When patients randomized to the control arm cross over to the investigational arm upon progression, the OS endpoint is contaminated. A contemporaneously gathered, uncontaminated ECA could help preserve the OS analysis, negating the need for a redundant confirmatory RCT following approval.
Amy Comstock Rick spoke about the Center for Drug Evaluation and Research (CDER) Rare Disease Innovation Hub’s work to make these approaches more feasible and introduced the concept of pooled and shared controls. To maximize the utility and longevity of data, companies operating within specific disease areas should pool control arm or natural history data to serve as a communal ECA resource for future trials.
Navigating Regulatory Flexibility and Certainty
In rare disease populations, the infeasibility of executing randomized, placebo-controlled trials requires a pragmatic approach; failure to generate evidence is not an acceptable endpoint. However, Amy clarified a critical regulatory dynamic: sponsors frequently conflate requests for regulatory flexibility with demands for regulatory certainty. While the FDA may exercise flexibility based on clinical context and unmet need, the statutory requirement for demonstrating substantial evidence of effectiveness remains unchanged. Furthermore, sponsors must not operate under legacy assumptions; FDA guidelines continually evolve alongside scientific advancements, and historical acceptability does not guarantee contemporary approval. To successfully navigate this evolving landscape, sponsors, patient advocates, and regulators must abandon siloed development and fully commit to prospective, collaborative evidence planning.