NEW YORK – Drugmakers, academic oncologists, and even regulators are becoming more accepting of clinical trials using synthetic, or external, control arms when evaluating the benefit of precision oncology drugs, recognizing that gold-standard randomized-controlled trials aren’t often feasible.
In interventional medicine, it is widely accepted that the most unbiased way of proving a treatment’s benefit is to compare its efficacy to a placebo or control arm in a randomized-controlled clinical trial. For precision oncology treatments meant for increasingly rare, biomarker-defined patient populations, however, such trials aren’t often possible.
External control arms are catching on as an alternative, where outcomes among on-trial patients treated with an investigational drug are compared against the outcomes of patients treated with different interventions outside of that clinical trial, be it past studies or in the real world.
“The use of external control arms isn’t new, but what’s changing is how they’re being used in the context of regulatory decision-making,” said Mark Stewart, VP of science policy at Friends of Cancer Research. According to Stewart, the approach has historically been more of a supplementary tool to help provide additional context on a treatment’s activity.
“Most of the instances where [synthetic control arms] have been used are for informal analyses and benchmarking,” he said. The approach hasn’t really been used to produce “statistical comparisons where you would make a judgment on the superiority of your investigational agent.”
But as more drugmakers are developing precision oncology products and taking them through regulatory review, “there are just a lot of instances where it makes sense to justify” the use of a synthetic control arm, in Stewart’s view.
Submissions to regulators
There’s a growing list of targeted oncology drugs on the market, for which sponsors submitted data from synthetic control arms to the US Food and Drug Administration as part of the regulatory application package, said Vivek Subbiah, an associate professor of investigational cancer therapeutics at the University of Texas MD Anderson Cancer Center.
For example, AstraZeneca submitted synthetic control arm data with its application for Koselugo (selumetinib) for an extremely rare pediatric tumor called neurofibromatosis type 1 (NF1).
Janssen also submitted data to the FDA from a synthetic control arm — drawing on real-world electronic health record data from the Flatiron Health database — in its application for Balversa (erdafitinib) as a treatment for bladder cancer patients with FGFR alterations, Subbiah said. Janssen had conducted a single-arm trial to demonstrate how patients responded to the targeted agent, but without a comparator arm, it’s not possible to demonstrate an overall survival advantage with the drug. To provide the FDA an idea of overall survival benefit, Janssen submitted data from a synthetic control arm.
The overall survival comparison against a synthetic control arm didn’t end up allowing Janssen to claim that its drug enables bladder cancer patients to live longer compared to another treatment. The drug’s label doesn’t mention an overall survival benefit against a synthetic control arm and cites some concerns about the methodology for the external control arm used. Still, the totality of the data submitted to the agency helped Janssen net an accelerated approval for Balversa.
Other new drug applications in oncology, including for Amgen’s Blincyto (blinatumomab) for CD19-positive B-cell precursor acute lymphoblastic leukemia, have included synthetic control arm data, too. And in the EU, the European Commission decided to expand the label for Roche’s Alecensa (alectinib) for ALK-positive non-small cell lung cancer, factoring in external control arm data from the Flatiron Health database.
Justifying the approach
Perhaps the timeliest example of a precision oncology trial where a synthetic control arm could be justified is that of Roche and Blueprint Medicines’ RET inhibitor Gavreto (pralsetinib) for RET fusion-positive advanced NSCLC.
As is typical with accelerated approvals, the FDA told Gavreto’s sponsors that it would need to conduct post-market studies to confirm the drug’s benefit, particularly in the first-line treatment setting where it’s unclear how Gavreto compares to treatment with chemotherapy with or without Merck’s Keytruda (pembrolizumab).
Accordingly, in early 2020, Roche launched the global randomized, controlled Phase III study, AcceleRET Lung, which was requested by the FDA.
In the two years since, the study has struggled to recruit its targeted 226 patients, which may seem like a small number of patients to enroll in a Phase III trial, but is a significant lift given just 2 percent of NSCLC patients harbor RET fusions. Finding these patients requires broad access to next-generation sequencing panels, access to which is not equitable and can be limited based on reimbursement challenges and patient’s place of care.
What’s more, the COVID-19 pandemic upended clinical research almost immediately after the trial launched. Plus, the fact that the drug is already approved for this patient population in US and Europe — albeit on a conditional basis — made it tough to find patients willing to be randomized to the chemotherapy or chemo-immunotherapy control arm.
“With the amazing [Phase I/II data], why would they want to go on a clinical trial randomizing them to chemotherapy?” Subbiah asked hypothetically.
Also, patients in many countries outside the US and Europe wouldn’t have access to Gavreto outside of the clinical trial, but many of the study’s 108 locations are in regions where the drug is already commercially available.
That trial is still ongoing, but given its recruitment challenges, Subbiah, Roche, and a team of researchers across the US and UK, separately performed a synthetic control arm analysis, in which they compared Gavreto’s overall survival, progression-free survival, and time to treatment discontinuation (TTD) against real-world data from patients in the Flatiron Health database.
With the scarcity of RET fusion-positive NSCLC patients, even the synthetic control arm analysis proved difficult. But Subbiah and colleagues took an approach they hope other drugmakers and researchers will pursue for precision oncology drugs they’re developing in extremely rare patient populations or when randomized-controlled clinical trials are impractical or infeasible.
The approach involved using two different cohorts of real-world data to build the synthetic control arms. The first included just 10 patients in the Flatiron Health-Foundation Medicine Clinico-Genomic database with documented RET fusion-positive advanced NSCLC. These patients matched the ARROW clinical trial population, which allowed researchers to perform a direct comparison and show Gavreto numerically improved on the endpoints of interest compared to real-world data, even though the synthetic control arm contain too few patients to achieve statistical significance.
The second external control arm, which was much larger, included patients treated in the real world whose RET fusion status was unknown. These patient datasets — which were taken from electronic health records in Flatiron Health’s database — included 686 patients treated with first-line Keytruda and 1,270 patients treated with Keytruda plus chemo.
Of course, the fact that it wasn’t known whether these patients had RET mutations made the comparison imperfect. But combining the two sources — the small, but well-matched data, and the large, but less well-matched data — painted a fuller picture than using one or the other.
“This allowed us to maximize the sample sizes of the real-world cohorts, translating into a much higher statistical power,” Subbiah said.
In the larger dataset, too, patients treated with Gavreto appeared to benefit most.
To address the possibility of bias in using synthetic controls — one of the main concerns with using real-world data — Subbiah and colleagues used various statistical methods, including quantitative bias analyses and tipping point-based biased analyses.
“Real-world data [bias] has always been cause for concern for decision markers,” Subbiah said. In randomized-controlled clinical trials, stringent criteria ensure that, at baseline, patients in different treatment arms reflect one another.
Due to the lack of such controls in real-world data, both regulators and health technology assessment agencies have consistently said that any conclusion made using such information “should be supported by analyses that quantify the impact of potential sources of bias,” Subbiah said. Trials that use synthetic controls or real-world data, in other words, need to do more than show the ways they adjusted for bias; they need to show just how much of an effect the bias in real-world data could have on the outcomes.
In a study published last month in Nature, Subbiah and colleagues explained the statistical methods they used to quantitatively assess the impact of bias on their findings. The system used, Subbiah said, could function as something of a guide for future trials taking this approach.
Friends of Cancer Research has also put out a number of guidance documents, outlines, and case studies in an effort to help researchers address statistical biases with real-world data and improve the way the field uses synthetic control arms. Last year, for example, Stewart and colleagues put out a paper in the Journal of Biopharmaceutical Statistics that walked through the models and equations researchers used to address biases in a NSCLC synthetic control arm case study.
As for confirming Gavreto’s benefit to the FDA, Sreeram Ramagopalan, the global head of real-world evidence at Roche, said that the question as to whether the firm will present the synthetic control arm data depends on how things go with the lagging recruitment in the Phase III randomized-controlled trial.
“We may submit this study as part of the [post-marketing] commitments to the FDA, but they specifically wanted randomized-controlled trial data,” Ramagopalan said. “We will have to review how the randomized-controlled trial is going with recruitment challenges with COVID and make a call as to whether we submit this [synthetic control arm data] in addition to any randomized-controlled trial data.”
Seeing as the synthetic control arm analysis has further validated Gavreto’s benefit in this patient population, Subbiah predicted that patients may be even less keen on getting randomized to the control arm in that trial now.
Considering data quality
As drugmakers and researchers get better at adjusting for bias using statistical methods, both Subbiah and FOCR’s Stewart think that the field should also improve the breadth, diversity, and quality of real-world data sources. This could go a long way to making regulators and health technology assessment agencies even more accepting of synthetic control arms.
“The use [of synthetic control arms] as well as how they’re being used in the context of a regulatory decision, is likely going to grow, particularly as, over time, we have higher quality data available,” said Stewart.
Right now, there are variations in quality across available sources of real-world data. According to Subbiah, investigators can pull these data from health records, medical claims, public surveys, technology companies, wearable devices, and all sorts of sources.
For the Gavreto study, he and his colleagues used the data in health records and databases collected by Flatiron Health and Foundation Medicine — two companies that Roche, Gavreto’s sponsor, owns. Roche’s acquisition of these data-centric companies, and its subsequent focus on improving real-world data collection, suggests it sees the use of such information playing a greater role in not just its own drug development efforts but across the industry at large.
Other life sciences companies are also making big bets on the increasing importance of real-world data and growing acceptance of synthetic control arms. Caris Life Sciences Chief Medical Officer Michael Korn explained that the molecular profiling company has been expanding its partnerships with the biopharma industry in recent years expecting they’ll use its cache of real-world data to design clinical trials, “build cohorts to understand standard-of-care benchmarks or areas of unmet need, and identify ideal patient populations.” These real-world datasets, Korn believes, will only become more critical to “increasing the probability of technical and regulatory success of therapeutics.”
In Korn’s view, however, even though synthetic control arms are more common than they once were in drug development, they are still rare and tend to be used “on niche opportunities for label expansion.” But this could change as the data quality improves, he said, and noted that Caris has also placed a big emphasis on improving the quality of its real-world data.
Cancer centers and researchers can access Caris’ real-world data if they join the company’s Precision Oncology Alliance. As of earlier this month, the network boasts nearly 70 cancer centers and academic institutions.
The company’s data have not yet been used in a synthetic control arm and submitted as part of regulatory package for a drug, but Korn thinks it’s just a matter of time. “The desire and demand is apparent in our discussions with investigators and biopharma,” he said.