NEW YORK – With the success of drugs like Merck's pembrolizumab (Keytruda) and Loxo Oncology and Bayer's larotrectinib (Vitrakvi), many companies that make oncology therapies are now trying their hand at developing histology-agnostic cancer drugs. But getting these treatments into patients' hands isn't simple — they must be tested in specifically designed clinical trials, approved by regulators, and paid for by insurance companies.
Further, the successful development of an increasing number of these drugs is likely going to require a retooling in how clinical trials are done, how regulators and payors use real-world evidence after a drug is approved to add or remove indications, and how academic centers, researchers, and clinicians collect such evidence.
According to Keith Flaherty, co-leader of the Dana-Farber Harvard Cancer Center developmental therapeutics program and professor of medicine at Harvard Medical School, the work to develop a new clinical trial paradigm must begin with academia. Cancer drugs being trialed in multiple indications has been the standard going back at least a decade, he said.
However, companies have generally started with one indication in Phase I, and then moved to other tumor types for which there is preclinical evidence to support that a drug might be useful in Phase I/II. It's only been in the last three to four years, especially with the advent of immunotherapies, that multiple indications or tumor types have been studied from a trial's start.
But there's still a very poor understanding of how to use preclinical models such as cell lines or animal models to predict which tumor types would be more responsive or less responsive to a given compound in humans, Flaherty said. He believes it's the responsibility of academia to develop better preclinical models with these kinds of tumor-agnostic clinical trials in mind.
"In addition to conventional cell lines and conventional xenograft models, there's been a lot of work in the field to try to develop more patient-derived xenografts, so we have better representation of the whole spectrum of human patient tumors in the mouse model space," he added. "[There's also] organoids, which are a complex mixture of cell types, not just the tumor cells, and less complex systems where it's just a patient tumor sample directly taken to the laboratory and exposed to therapies or screened for sensitivity. And all of those types of efforts really have only been pursued in earnest in the past five years."
He also believes that academia should take the opportunity to work on arms of tumor-agnostic cancer trials that have been "left behind along the drug development path" by pharma companies. For example, he said, there are a number of cases where genetically defined subpopulations of certain cancers are ignored in the clinical development of a promising drug.
"Oftentimes, that decision is made strategically, from an efficiency [viewpoint] of clinical development, as well as a potential return on the cost of development in terms of the market," Flaherty said. "Those rare or sparsely distributed populations are left behind, never to be revisited. That's where with next-generation platform trials that the NCI is currently scheming up, I'd say there's a role for publicly funded clinical research to potentially pitch in. If we think there's a therapy and it's clearly valuable in one tumor type that's biomarker defined, and more rare subtypes have been left behind, then why not pick that low-hanging fruit? It may only be a couple percent of all cancer patients, but that's not so bad."
On the regulatory side, Flaherty said the US Food and Drug Administration has been "pioneering" in the area of histology-agnostic cancer therapies and had prepared for the possibility of such drugs long before larotrectinib was approved.
"It's the fundamental premise [of the trials] that in some cases there may be homogeneity of effect across tumor types, and in other cases there may be heterogeneity of effect," he said. "The FDA really wants to know that, in terms of the outcome of these studies. If there's a tumor type that shouldn't be included in a broad approval, this is a key point that they want evidence to be able to help address."
He cited BRAF-mutated tumors and the development of Genentech's vemurafenib (Zelboraf). Though vemurafenib was found to work well on melanoma and other BRAF-mutant cancers, it failed in clinical trials to treat BRAF-mutated colon cancer.
Flaherty also noted that some clinical trials may only have one or two patients representative of a given tumor type. If one patient with one cancer responds and another patient with a different cancer doesn't respond to a drug being tested, that's not enough evidence to say the drug should definitely be approved or rejected for those indications, he said.
"This is where the FDA has been in open dialogue with the field about how to think about accumulating that evidence, at the time of initial approval and then for [continuing] approval," he added. "With larotrectinib, they laid out the expectation in terms of how many patients they wanted to see additional data from, with a handful of different tumor types to confirm this concept that there was a broadly active drug regardless of a tumor type. And so, you're going to have to look at the data as it rolls in to try to judge: Do we have enough evidence that this is having its activity broadly and in a way that doesn't depend on the cell of origin or the lineage of the tumor?"
Indeed, at a panel on the challenges and opportunities of histology-agnostic trials at the American Society of Clinical Oncology's annual meeting in Chicago earlier this month, the FDA's Gideon Blumenthal said the main prerequisites for approvals of these drugs were "a detailed understanding of the scientific biologic underpinnings of the disease under study, clinical data showing large magnitude and consistency of effect in patients with rare and refractory cancers, limited therapeutic options, and unmet need."
He noted that the FDA was able to make the shift in its thinking when it came to approvals of larotrectinib and pembrolizumab because of these prerequisites, but also because of the drugs' large magnitude of effect, including complete responses in patients with no other options.
Flaherty also noted larotrectinib and pembrolizumab's high response rates, adding, "Imagine if you have a drug with a 20 percent response rate. You've got 10 different tumor types enrolled in a study. Where do you draw the line in terms of which tumor types you include [on a label] with a response rate like that, bouncing around above and below that number per tumor type?" he added.
Though such an example has yet to hit the desks of regulators at the FDA, it may one day happen — and this is where the need for a new kind of clinical trial paradigm and an increased use of real-world data will come into play.
Randomized trials, which are currently the gold standard for drug development, aren't really possible for tumor-agnostic therapies. "A randomized trial that's tumor-agnostic, it's going to be tricky to run, because if you have 20 different tumor types being included, what do you randomize patients to?" Flaherty said. "What's the standard of care?"
Blumenthal, who serves as deputy director for the agency's Oncology Center of Excellence, acknowledged at ASCO that randomized controlled clinical trials are unlikely for these kinds of drugs.
Instead, Memorial Sloan Kettering Cancer Center medical oncologist David Hyman said during the panel discussion that one tool that could be very useful testing tumor-agnostic hypotheses is the basket study, where patients are enrolled on the basis of common genomic alterations, regardless of cancer type.
"We refer to these often as tumor-agnostic studies, but I think that really misses a nuance in these clinical trials," he said. "We go into these studies with a very real recognition that we are often going to find tumor lineage-dependent differences in efficacy. So, these studies may hope that a therapy turns out to be tumor-agnostic. But they're not blind to the possibility of identifying tumor lineage-specific differences in efficacy, and that really is part of the biostatistical design of these studies from the outset."
According to Flaherty, real-world evidence of utility would also come into play in this context. Researchers have already demonstrated the potential of real-world data to explore the efficacy of precision oncology drugs in rare populations. In this case, such evidence could be used to either include more indications on a drug's label, or to remove indications that had been previously approved.
Before that can happen, though, a lot of work must be done to really capture real-world evidence in a robust way. "There are precious few health systems that have engineered themselves to be able to capture that data holistically. It's really early days in terms of agreeing on what that data should look like, how you would extract it from an electronic health record, how you include or exclude certain patients from analysis," Flaherty said.
"Is it simply sufficient to say if they got the drug, we count them? What if they weren't biomarker-selected? What if they had certain comorbidities that would have precluded them from entering a prospective clinical trial? There are all sorts of devilish details that really haven't yet been unpacked in oncology for use of real-world evidence for post-marketing approval," he said.
With respect to post-marketing data and real-world evidence, Blumenthal also questioned what types of trials would generate the most useful kinds of information in the post-marketing setting. Traditional clinical trials or registry observational studies? And what types of controls should be incorporated — randomized controls, historical controls, contemporary external controls?
"What level of evidence is necessary to modify or omit specific histologies if accumulating data suggests no responses?" he further asked. "Obviously, if a given tumor type [shows] zero responses out of three, you wouldn't necessarily want to restrict the indication at that point. But if there's a whole body of evidence suggesting that there's resistance with a given tumor type, maybe you want to carve those indications out."
Health systems must also see the value in it for themselves and their patients in order to participate in such a system. "There has to be some way of thinking about compensating centers for that evidence," Flaherty said. "The payors could get involved in that, certainly, but it could also be the pharma companies. It's obviously in their vested interest for that data to be generated as well."
At the ASCO discussion, Lee Newcomer — Principal of Lee N. Newcomer Consulting and a former senior vice president of oncology and genetics at health insurance giant UnitedHealth — concurred, noting that payors should start covering large-panel genetic testing for cancer patients in order to harvest and gather data.
He called on pharma companies to build the necessary infrastructure, adding, "Pharma is going to need to get together and develop a clinical trial information service, or database, or website that is fast, updated, and efficient. Because you need to be able to plug in the genomic report, the 10 clinical variables that are relevant here, and find out quickly, right then and there, what tests are available, geographically sorted next to you."
Blumenthal also acknowledged during his talk that there are still many uncertainties around histology-agnostic drug trials. "Absolute certainty is not possible for every biomarker, histology, and drug combination," he said. "So, the questions with early clinical development are what are some of the key early indicators to potentially signal a successful tissue and age-agnostic approach? What statistical methods could be used to analyze activity and dose, including borrowing information across tissue types?"
Having conducted many such trials in the past five years, Hyman said he's learned several lessons that could shed some light on what kinds of data will be useful in the long run. For example, he said, cancer dependency maps that show the relationships between genomic alterations are a valuable resource. He also noted that negative data has a lot of utility because it can help to stop ineffective prescribing of cancer drugs.
Hyman also noted the importance of using real-world evidence to improve the development of new cancer drugs and to improve the way currently approved drugs are used to treat patients. "Existing traditional clinical paradigms are unsuited to answer all the questions that our patients need [answered]," he concluded.