Consensus Builds for Real-World Evidence to Support Multi-Cancer Screening Tests

NEW YORK – Advocacy groups are making progress in fostering collaboration among stakeholders around genomic multi-cancer early detection and screening tests, highlighting growing consensus around the promise and pitfalls of such tests and the potential for real-world evidence to help speed their path to regulatory approval.

The nonprofit organization Friends of Cancer Research recently released a whitepaper developed via discussions among industry, regulatory bodies, and clinicians, and held a virtual meeting to discuss this work at the end of last month.

Multi-cancer early detection (MCED) tests have emerged from advances in blood-based multiomic and epigenomic analyses, and have been lauded as holding the potential to dramatically change the landscape of cancer treatment and mortality.

Speaking on a virtual meeting panel, Seema Singh Bhan, senior VP of public policy at Thrive Earlier Detection (part of Exact Sciences), said that while cancer treatment advancements such as immunotherapy and targeted therapies have impacted patient outcomes, the reality remains that "over 70 percent of cancers are detected at late stage," where curative treatment is impossible.

"To paraphrase Dr. Bert Vogelstein, no therapy is as effective as earlier detection," she added.

"Single-organ cancer screening is credited with asymptomatic detection of approximately 17 percent of cancers annually [but] despite the progress made by these exceptional single-organ screening innovations, there's still a great unmet need," she added.

Bhan highlighted that there is currently "no clear pathway" for MCED tests. "The science and the technology were not in existence when policies around prevention and early detection were made decades ago for cancer screening, and as a result, this does not simply fit into the [existing] single organ screening framework," she said.

Challenges include the lack of a reimbursement pathway without direct congressional action, and the daunting prospect of demonstrating clinical utility in a pan-cancer setting.

Wendy Rubinstein, director of personalized medicine at the US Food and Drug Administration, praised the FOCR group's "careful thinking" about real world data in this regard, adding that the National Cancer Institute has also convened an MCED trial team focused on defining the key assessment questions needed to evaluate performance characteristics, safety, and clinical outcomes for this new generation of screening tests. For example, NCI plans to launch its own pivotal multi-arm, multistage randomized trial to evaluate multiple MCED screening tests.

"We can anticipate that in the short term that test developers will bring their technology to [the FDA] for premarket evaluation, and that prospective clinical studies, possibly supplemented by real-world evidence, will support initial approvals," most likely for the top five or 10 highest incidence tumor types, Rubinstein said.

That said, patients are susceptible to a long list of cancer not covered by those initial intended use cases, where physicians won't have a solid grasp of the costs and benefits of diagnostic workups, she added.

"Back of the napkin calculations" make it clear that traditional clinical trials cannot address all these questions in the right time frame for an affordable cost, Rubinstein said. But "if we organize our healthcare data to function as a learning system, we can aggregate the experiences of patients, and that could mean real-world data serving as valid scientific evidence," Rubinstein said.

"I think it might be OK not to know all of these answers at initial approval if the benefit risk assessment is favorable," she said. "But unless we plan, we won't even know the answers five or 10 years from now."

According to Girish Putcha, senior VP of reimbursement at Freenome, many of the limitations of traditional clinical trials are "fairly obvious."

"First, the duration of follow-up for interventional studies can be long and variable, given different natural histories for different cancers," he said. "Second, especially for the less prevalent cancers … performing such studies in the intended use population can require tens, if not hundreds, of thousands of patients."

"And then finally, as we know, registrational clinical trials are carefully controlled and so generally do not reflect real-world medical practice. And one example that applies here is that adherence to screening can be quite different in the real world. And as many have shown, this can have a very, very significant impact on the clinical utility of such tests," he added.

That doesn't mean that real-world data doesn't have pitfalls.

"The mantra goes like this: randomized clinical trials with disease-specific mortality as the endpoint," said Ruth Etzioni, a biostatistician at the Fred Hutchinson Cancer Research Center.

"The reason for this is that the randomized trial is going to avoid selection issues that you inevitably have in the observational setting. And … it's not only a matter of those patients who elect to receive screening [being] different from those who don't, but also what happens afterwards? Because remember, that screening can only be beneficial as part of a continuum of care. You have to have proper follow-up and you have to have appropriate treatment."

Randomization also harmonizes the analysis of outcomes such that it can correct for lead-time bias, she said. Measuring mortality from a single time point — the start of the trial — allows researchers to be confident that any improvement is real, rather than an artifact.

Sam Roosz, CEO of data firm Crescendo Health said that the FOCR group had some "lively discussion and debate" about how to even define the term real-world data in this specific context.

"What I would put forward is that when we talk about a real world for MCED studies, really what we're talking about is the clinical data about patients participating in a study that rests outside of the trial site," he said. "Patients might receive the screening at one institution, but the possible future institutions where the patient might be subsequently diagnosed and managed [are myriad]."

Investigators face the challenge of trying to track whether patients have been diagnosed with cancer — whether diagnoses represent true positives, and whether the lack thereof represent true negatives — amidst a potentially huge follow-up landscape.

"That's exactly the promise that real-world data offers. We gain the opportunity with a lot of the recent advancements of interoperability and data liquidity to be able to blend together the best of both worlds," Roosz said.

Looking forward, "we have to continue the efforts to expand on tools for interoperability and increase liquidity of data in the ecosystem to make these data accessible for researchers. And we also need to do the work to define and validate the measures that we're going to use from the real-world data and confirm that we're going to be able to get strong signal and handle the messiness that's inherent in these datasets as we bring it into regulatory context," he added.

Etzioni said that the statistical community has stepped up, developing methods that can help correct for selection factors. But lead time remains "very difficult to account for in observational studies."

Another concept discussed in the FOCR white paper is the use of late-stage events, essentially the rate of cancers diagnosed at late stages, as a surrogate endpoint for disease-specific survival.

"We have to recognize that this is definitely going to be an important surrogate," Etzioni said, "but if we look back at the screening trials that we have conducted, we find that the reduction in late-stage incidence does not predict the observed reduction in mortality across trials, so we have some work to do here as well."

According to Rubinstein, the position of regulators is that real-world data could potentially be considered as the primary evidence in the post-market setting for additional indications, after tests gain initial approvals for a subset of prevalent cancers using more traditional studies.

Evidence from clinical practice could be used to understand the sensitivity of these tests in rare cancers, or it could be used as supplemental evidence in younger populations where the relative paucity of cancer diagnoses might lead to otherwise wide confidence intervals in traditional randomized controlled trials. "We could even gain a better understanding of safety on a cancer specific basis for diagnostic workups, where most of the risk probably lies," Rubinstein said.

Rubinstein argued that it isn't a moment too soon for these discussions and harmonization efforts as MCED tests are already making pre-regulatory waves, available to clinicians and patients right now in the case of Grail's Galleri, and likely in the near future from other companies including Exact Sciences, Guardant Health, and Delfi Diagnostics.