Studies Using External Control Arms Gaining Ground
By Deborah Borfitz
July 16, 2021 | The willingness of healthcare authorities to accept external control arms (ECAs) in regulatory submissions depends largely on how and when they are used, according to real-world data (RWD) experts from Merck, Bristol-Myers Squibb, and Ikaika Health who presented at the recent DIA 2021 Global Annual Meeting. Aaron Kamauu, M.D., vice president of real-world data services at Ikaika Health and an industry-leading advisor on leveraging RWD in clinical research, says ECAs are “not quite ready for primetime” only in terms of their supporting role in new drug approvals.
For label extensions and, especially, post-marketing studies, ECAs have been extensively and successfully used over the past few years, reports Ravinder Dhawan, vice president of outcomes research at Merck who leads the company’s real-world evidence global oncology group. Several proposed frameworks—including one produced by the Duke Margolis Center for Health Policy in 2017 and another by the U.S. Food and Drug Administration (FDA) in 2018—provide useful context for potential uses and limitations of ECAs.
External control arm is an umbrella term for any control that is not randomized and is used as a reference for interpreting experimental data, explains session moderator Zoe Li, director of life sciences at COTA Healthcare. It is alternatively called an historical control, a concurrent control, or a synthetic control, she says, adding that the speakers are all referencing data coming from real patients.
ECAs have a “long history” in regulatory decision-making and are used when randomization is unethical or unfeasible, says Leah Burns, global real-world evidence strategic advisor at Bristol-Myers Squibb (BMS). They can also serve as a benchmark in health technology assessments.
Advances in data access, quality, methodologies, and societal influence may broaden the use cases for ECAs, continues Burns. If data confidence is relatively high, an ECA could become an option when a randomized controlled trial (RCT) is impractical to do—for example, if not enough patients are willing to enroll or the development cost of a drug is prohibitive but the medicine would address an unmet medical need.
In the right scenario, a single-arm clinical trial with an ECA can replace a traditional, two-arm RCT on a New Drug Application with the FDA, says Kamauu, but the use case is “fairly limited to rare indications with unmet medical need and high morbidity and mortality.” While he is optimistic about the use of ECAs later in the drug lifecycle, he adds, there are “not [yet] a ton of examples.”
Potential And Pitfalls
One central difficulty is how to match RWD with clinical trial definitions, says Kamauu. Clinical outcomes often change over time, and definitions can also differ across studies. Stakeholders therefore tend to be more comfortable and confident in using ECAs when they can “make the ethical argument” for them.
Readiness of ECAs for regulatory use is influenced by the “totality of evidence” that amasses as a drug is being developed, says Burns. Data sources are also less familiar early in the drug lifecycle, requiring more documentation (e.g., about its quality).
“RCTs are still the gold standard and that will continue to be the case … [because of the] profound public health implications,” says Dhawan. But electronic medical records (EMRs) and registries are allowing the parallel collection of RWD to help address channeling and selection biases in RCTs.
In the future, ECAs may well address the need for biomarker-defined populations in single-arm studies, he says. But first, researchers need to demonstrate RWD sources can reproduce results produced by clinical trials.
Varying degrees of rigor are required to remove as much bias as possible from RWD sources, says Kamauu, quickly adding that he is “more afraid of unknown, unreported bias.” Secondary data sources are not captured in a de novo way and the context for its initial collection, which impacts data relevance and reliability, may not be fully known.
Just because data has a field, its meaning cannot be assumed, says Kamauu. “Different health systems provide different care in different ways.”
Another context to keep in mind is the degree to which surrogate clinical definitions, versus unknowns of the data, are acceptable, Kamauu says, because there are going to be tradeoffs.
A lot of work has been done in recent years to improve “variable missingness,” says Burns. Flatiron Health has a dataset BMS uses in oncology when looking at overall or progression-free survival, for example, and it pulls from different data sources to increase the sensitivity and specificity of those variables.
More hybrid approaches to variable definitions will be taken in the future, Burns says. Decentralized collection of data, for example, could involve nurses interviewing patients at home to improve data reliability.
“Data infrastructure is still a struggle,” says Dhawan, referencing unstructured data in EMRs and less-than-ideal data curation technology. EMRs are also not tied together with claims, genomics, and imaging data. Better data connectivity would help raise the confidence level of regulators in the use of ECAs, including for new drug approval purposes, he adds.
While many sophisticated registries exist in the U.S., and some in Europe, “the vast majority of countries are still struggling,” says Dhawan. “Regulatory agencies around the globe need to be jumping up and down.”
One exciting development for addressing the connectivity issue, Dhawan says, is a “tokenization” process introduced by some data companies to follow patients on their journey across a single integrated dataset. If participants in clinical trials get tokenized, that data could theoretically be linked with real-world evidence later.
Kamauu is likewise enthusiastic about the prospect of longitudinal data access and “more comprehensive use of patient populations.” Innovative methods of bringing disparate data sources together in high-confidence ways (i.e., from the same patients) are under development and several companies are not limiting themselves to data that is easy to access, he says.
In the interim, the Duke-Margolis Center is working to harmonize terminology and definitions, Kamauu says. The meaning of “high-quality data” currently lacks consensus. Over the longer term, ECA trials could broaden to incorporate endpoints to fill evidentiary gaps of interest to healthcare payers, he notes.
The value of ECAs include being able to answer questions quickly when they come up as well as to generate data that may be needed internally about a disease area, says Burns.
ECAs can also help increase the number of trial participants who get in the treatment arm and reduce development timelines and overall costs, Dhawan adds.