Data from past clinical trials has provided researchers with a good starting point to create more ECAs.
Although randomized control trials (RCTs) have long been considered the gold standard of clinical trial design, many life sciences companies are beginning to utilize external control arms (ECAs) when designing studies to support regulatory submissions.1 This follows a trend wherein more and more companies are incorporating real-world data (RWD) and real-world evidence (RWE) into clinical trial design. This trend gained traction during the COVID-19 pandemic, when physical distancing requirements meant that trials needed to be run partially or fully remotely.2 Because life sciences companies were unable to recruit patients and conduct studies in-person, many turned to existing RWD data sources—such as claims and electronic health record (EHR) data—to bolster their research and offer complementary data.
As ECA studies become more common, companies are beginning to recognize the advantages that they offer. With a variety of RWD sources such as prior clinical trials, medical claims, and electronic health records, researchers have a wealth of data that they can leverage to create ECAs.
Although RCTs with placebo or standard of care control arms are the traditional approach—and placed at the top of the evidence hierarchy—they are not always ethical or feasible. If researchers are investigating a potentially life-saving treatment they believe will be particularly effective, it can be unethical to randomize the trial participants to a placebo arm. With ECAs, this is not an issue. Because the comparison group is outside of the study, there are no patients in the study who miss out on active treatment.
ECAs can be particularly useful in the rare disease and rare oncology spaces. Because these disease types have very small patient populations, it can be impractical or prohibitively difficult to find enough patients to enroll in an RCT, which can lead to sample sizes that are too small to obtain meaningful results. When faced with a small patient population, enrolling study participants can be challenging and can result in delays to the study timeline. In these circumstances, incorporating an ECA using RWD can allow researchers to expand the sample size, ensure all study participants have access to the new medication, and accelerate the study timeline.
Researchers can also create a synthetic control arm as an alternative to an ECA constructed using RWD. A synthetic control arm is a type of ECA that enables researchers to generate a large comparator arm for a study. Using machine learning, researchers can generate synthetic patient data and create ‘digital twins’ or hypothetical patients whose characteristics are identical to the actual patients in the study. By creating digital twins, (triplets, or quadruplets,) researchers can investigate the impact of the treatment on the actual study participants while also predicting outcomes for the synthetic patients who did not receive the treatment. This technique is especially useful for rare diseases, where patient populations are so small that even obtaining enough RWD to generate RWE is challenging. Creating a synthetic control arm essentially gives researchers a larger quantity of data to study that increases statistical power even when the treated patient population is small.
Both external and synthetic control arms provide additional data for analysis, helping to expand the interpretability of the results and increase the overall impact of the study. With FDA increasingly open to the use of RWE as a factor in decision-making, this ability to provide meaningful comparative analysis can strengthen regulatory submissions for life sciences companies and support market access with payers.3
Currently, ECAs are most often used in oncology. For example, Roche utilized an ECA when investigating the drug alectinib as a treatment for non-small cell lung cancer (NSCLC). Alectinib first received accelerated FDA approval to treat this form of lung cancer, and it was later conditionally approved in the EU. However, the EU required additional evidence of alectinib’s effectiveness relative to the drug that was the standard of care at the time for metastatic NSCLC. Instead of waiting for Phase III clinical trial results (which would have slowed down the approval timeline), Roche utilized an external control arm to provide the necessary evidence.4 The use of an ECA advanced coverage of Roche’s drug by 18 months in 20 EU countries.5
Another study showcasing the value of ECAs can be seen in Pfizer and Merck’s investigation of the drug avelumab for metastatic Merkel cell carcinoma (mMCC), a rare type of skin cancer.6 In this case, an RCT was not practical for several reasons. The rare nature of the disease meant that researchers had limited patient data, and the short patient survival times typical of the cancer precluded recruiting enough patients, not to mention the ethical concerns of denying treatment for those in the control group. When seeking approval of avelumab, Pfizer and Merk incorporated an ECA constructed with EMR data into their submission. This approach was successful, and avelumab was approved for treating mMCC in 2018.
These examples illustrate how ECA studies can accelerate the time to approval and avoid potential challenges inherent in RCT study design. ECAs are an important tool to speed access to critical treatments in diseases with high unmet need. Although ECAs are currently primarily used in oncology, their potential impacts stretch into many other disease spaces, and we will undoubtedly see their use increase in the coming years.
To create an ECA, researchers can employ a technique called propensity score matching (PSM). This process involves taking existing RWD, such as EHR or claims data, and applying the inclusion and exclusion criteria used for the trial. With this technique, researchers can identify individuals outside of the study who match study participants. They can then compare the study participants who receive the treatment to these individuals outside of the study who do not, balanced according to characteristics that might influence either treatment or outcome.
The purpose of PSM is to try to reduce confounding variables that could be attributed to basic differences between the control group and the study group, as opposed to the impacts of the treatment. To ensure that individuals are closely matched, the key is to apply select inclusion and exclusion criteria. The goal is to line up the control arm as closely as possible with the treatment arm in demographics, medical history, co-morbidities, and other characteristics.
PSM can be done using data analytics software that is capable of mapping RWD sources, applying inclusion and exclusion criteria, and identifying external patients whose characteristics match the treatment group. The ideal analytics platform contains RWD from a variety of sources including claims, EHR, registry, and more, and is purpose-built for healthcare data. Such software allows for the closest matches between the treatment and control group and thus offers the most meaningful results.
Like PSM, generating synthetic data and digital twins (triplets, or quadruplets) can be accomplished using data analytics software with machine learning capabilities. Before running any predictive analyses, the model must be trained with significant amounts of data. Once adequately trained, the machine learning model can be used to generate synthetic data and predict outcomes for the digital patient copies. Effective analytics software will simplify this complicated process and enable researchers to conduct meaningful studies using both real external and synthetic control arms.
When assembling a regulatory submission in industry, the data must be relevant and reliable, and the analytics must be transparent and reproducible. Fortunately for life sciences companies, the top analytics platforms enable regulatory-grade submissions using RWD and RWE. Because ECAs are still a relatively new phenomenon, regulatory bodies such as the FDA can be skeptical of the data and the platforms used to conduct the analysis. Using an analytics platform that is well-known in the industry and that automatically captures all the methods used can mitigate issues around quality and ensure that all aspects of the submission are fully transparent and thoroughly documented.7
To set themselves up for success with regulatory submissions involving ECAs, researchers must have a deep understanding of the data and be able to clearly articulate their research questions. The data used to construct the ECA must be accurate, complete, and relevant to the research question. Employing effective data analytics software as described earlier will also equip researchers with regulatory-grade data and insights that lead to strong and acceptable submissions.
FDA has published four draft guidances on how it will consider real-world data (RWD) in regulatory decision-making, including the use of electronic health records and registry data. More guidance documents are planned to detail study designs that incorporate external control arms.8
What does the future hold for ECAs? To date, ECAs have been used in studies of rare disease and in oncology, where single-arm trials are the most practical or ethical option. No marketing applications in oncology have used an external control arm as part of the primary efficacy analysis. Instead, ECA data have been used to establish the natural history of the disease or in a comparative efficacy and safety analysis.9 The movement in 2022 is towards a true contemporaneous comparator arm, wherein RWD sourced from EHRs, administrative claims and registries are expected to be as broad and as deep as clinical trial data.
RWD data—because they were created for a different purpose than biopharmaceutical research—are by nature messier than clinical trial data. Some outcomes are captured less frequently or not at all; concepts of interest may be recorded in a non-standardized way or exist only in physician notes rather than in a coded field. One way to address data ‘missingness’ is to link various data sources to capture more fully the longitudinal patient journey through the healthcare system and to check the accuracy of self-reported information. There are several companies that assist with this data linkage by assigning unique identifiers called tokens to each patient in each data source. The tokens assigned to one source can be matched to a token in another source so as to be able to identify a single patient who exists in, for example, both a claims data source and a registry.10
This ‘messiness’ of an RWD-informed ECA can introduce some error in the estimation of treatment effect, but if the effect size is large enough, the difference (favorable or unfavorable) will manifest. This makes an ECA a useful option for new and promising treatments but not for those likely to demonstrate only a modest benefit. Moving forward, ECAs are likely to play a larger role in assessing benefits and risks for treatments that receive accelerated approval or in diverse patients who were not included in pivotal clinical trials.11
As more life sciences companies begin to utilize comparator arms, both researchers and patients will see numerous benefits. First, ECAs are more ethical and acceptable to patients because there is no risk of being assigned to placebo or standard of care. This is most important when the treatment cannot be blinded, and patients know that they are not receiving a promising new therapy. Second, ECAs can help to shorten study timelines and reduce costs. A shorter study timeline means increased speed to actionable insight for life sciences companies. When investigating life-saving treatments, this is especially critical. When researchers can complete studies, uncover insights, and make regulatory submissions more quickly, these treatments get into the hands of the patients who need them sooner. Ultimately, ECA-supported studies can enable faster approval of therapies and lead to more lives save or improved.
Meg Richards, PhD, Executive Director of Solutions, Panalgo