This article has been updated with additional information about a collaboration between Flatiron Health and the FDA.
CHICAGO (GenomeWeb) – A database of real-time, continuously updated genomic and clinical information on cancer patients can be a useful tool for precision medicine research, Foundation Medicine and Flatiron Health showed in data presented at a major medical meeting.
The two firms announced last year that they were launching the so-called Clinico-Genomic Database, integrating deidentified clinical data from cancer patients, such as diagnosis, treatment, and outcomes information from Flatiron's EHR data platform, with detailed genomic data from Foundation Medicine's registry comprising information for more than 100,000 cancer patients' samples.
At the American Society of Clinical Oncology's annual meeting here Monday, Gaurav Singal, VP of data strategy and product development at Foundation Medicine, presented analysis of anonymized information in this Clinico-Genomic Database from more than 1,600 non-small cell lung cancer patients. These are NSCLC patients in Flatiron's electronic medical records database that were tested for alterations in more than 300 genes and tumor mutational burden status using the FoundationOne next-generation sequencing test.
Out of 1,600 NSCLC patients largely seen at community practices, 576 had alterations that are targetable by a National Comprehensive Cancer Network recommended drug. Roughly 50 percent actually went on a targeted therapy, according to Flatiron, but the firm didn't provide a more definitive figure stating that the data are still evolving. Without further details about patients' stage of disease and prior therapy lines, "the exact number [or percentage that got a targeted drug] is not that informative,” a spokesperson said.
Median overall survival for patients with a targetable alteration in the database was 35 months compared to 19 months for those without such an alteration. This finding confirms what was already known, Christophe Le Tourneau from Institute Curie said during the meeting, "that survival of patients with driver mutations in lung cancer, is longer than patients who don't have any driver mutations."
Studies have shown that cancer patients treated with drugs targeting tumor markers fare better than those who aren't. A meta-analysis of 340 Phase I studies showed that in treatment arms employing precision medicine approaches, median progression-free survival was 5.7 months versus just under 3 months for those not getting such care.
But a randomized-controlled trial, called SHIVA, led by Tourneau and colleagues and published in The Lancet two years ago, has shown the opposite. In that study involving heavily pretreated cancer patients, molecularly targeted drugs prescribed off label didn't improve progression-free survival compared to when doctors chose which treatment to give.
As a result, some experts have cautioned that better prognosis in patients with targetable mutations could be driving the improved responses some studies have found with genomically guided treatments when comparing them to standard approaches. "One question people have been asking is: Does precision medicine matter?" said Singal. "It's a hard question to answer because it's so multifactorial and part of what makes it tricky is that people who have driver mutations or mutations identified by sequencing, they might be fundamentally different from those without those mutations."
The study by Singal’s team shows that having a driver mutation does benefit patients in terms of better prognosis. But he said it also shows, based on real-world information from the Clinico-Genomic Database, that receiving a treatment targeting a driver mutation prolonged survival and on average these patients lived more than year longer than those who didn't get a genomically guided drug.
Patients who got on a targeted treatment based on a tumor marker had a median survival of 42 months compared to 28 months for those who didn't receive NCCN-recommended therapy.
Vineeta Agarwala, director of product management at Flatiron Health, noted that studies reporting limited utility of genomic profiling may be because they're looking only at patients who get on targeted therapy for indications not supported by guidelines. For example, almost 20 percent of all NSCLC patients have EGFR mutations, and those patients are largely being identified by tests that interrogate just that gene. So, excluding cases like this, studies would estimate a lower benefit from NGS-guided precision approaches because they are evaluating the ability of NGS testing to get patients to the right drug who wouldn’t get there via other testing.
Ultimately, Singal and colleagues presented this data at ASCO in order to demonstrate that the Clinico-Genomic Database can be broadly used to do outcomes research, improve variant interpretations, inform drug development efforts at pharmaceutical companies, and even gather real world evidence that can be submitted to the US Food and Drug Administration. Life sciences translational and outcomes research groups, and academic partners will have access to the database, a spokesperson said.
However, Le Tourneau cautioned that when using a real-world databases like this, one should be cognizant of the fact that EHR data isn't always robust, and that certain conclusions may be extrapolated from proxy measurements. For example, in this analysis, Flatiron looked at patients' date of death with the EHR to derive overall survival estimates, and the amount of time a patient remains on treatment as a proxy for progression-free survival.
Still, there is growing interest in using real-world data in a variety of research settings. At this same meeting, the FDA said it would draw on real-world evidence within ASCO's CancerLinQ big data initiative to track how genomically targeted drugs and immunotherapies are being used in clinical practice, particularly off-label, and compare outcomes to data from clinical trials.
In a collaboration with the FDA, Flatiron retrospectively analyzed deidentified EHR data from community based oncology practices to track PD-L1 testing patterns in the context of prescribing immunotherapies. Data presented at the ASCO meeting showed that PD-L1 expression testing was low among patients receiving Keytruda or Opdivo; most patients weren't tested using FDA approved companion tests; and even when companion testing is required, as with Keytruda, most patients who weren't tested or negative received the drug anyway.
Collecting data on testing patterns of this type, as well as on outcomes from patients who have had treatments guided by genomic profiling, are of interest to payors. Last year Palmetto had considered requiring NGS test providers to submit data to registries as a condition for coverage in NSCLC, and had discussed the strategy with a number of health tech firms, including Flatiron.
"But payors are only one part of the ecosystem starting to think this way," said Singal. Drug developers, researchers, and providers are all "starting to appreciate that real-world data collected thoughtfully and integrated meaningfully could prove to be a critical source of evidence generation in the not-too-distant future."
For example, the database can provide insights on the pathogenicity of genomic variants and help clinicians get a better understanding of what patients with specific genomic alterations look like in the real world. "Are they smoking? What's their gender distribution? Where are they? What treatments are they receiving and how are they faring on them? It's really not known and there really isn't good literature supporting those questions," Agarwala said.
Additionally, the Clinico-Genomic Database sheds light on one potential benefit of comprehensive genomic profiling using next-generation sequencing, a technology that analyzes many genes at once and that payors consider investigational. Analysis of the database suggested that up to 30 percent of single-marker testing on EGFR, ALK, and ROS1 yielded a false-negative result.
The majority of biomarkers that the NCCN recommends testing for personalizing treatment for NSCLC aren't done via NGS-based comprehensive genomic profiling, but through single-marker tests. "This is analysis that is otherwise hard to put together," Agarwala said. "But for these patients who underwent Foundation's test, the EHR also contained evidence of testing done by another lab for those biomarkers."
Researchers identified five patients who were deemed negative for an ALK rearrangement by another lab, but who, after testing through Foundation, received ALK-targeted treatment and had at least a partial response.
The database also showed that 21 percent, or 340 NSCLC patients in the database received immunotherapy. Agarwala noted that using the database, researchers can see whether these therapies are extending patients' lives and how long it takes for their disease to progress. "We're starting to look for predictors of immunotherapy response," she said.
PD-L1 expression is a biomarker for identifying which patients are likely to fare best on immunotherapy, though it doesn't clearly delineate responders from nonresponders. At a medical meeting earlier this year, Foundation discussed using the database to look for additional evidence around biomarkers hypothesized to be associated with immunotherapy response, including tumor mutational burden, which many experts believe may be a more precise predictive marker than PD-L1 expression.
Foundation has been reporting tumor mutational burden as part of its NGS test. And last year at this same meeting the company presented data from more than 60,000 patients' samples and showed that tumor mutational burden was particularly high in samples with mismatch repair deficiency.
To date, there hasn't been a large-scale evaluation of the predictive value of TMB in the real-world setting, Agarwala said. However, the information in the database does appear to show that TMB predicts response to immunotherapy. Foundation has said it plans to seek premarket approval for FoundationOne, which includes analysis of microsatellite instability, tumor mutational burden, and major mismatch repair genes as a pan-cancer, universal companion diagnostic.
Experts who aren't yet convinced that the current evidence backs genomic profiling of all cancer patients would likely point out that the analysis of the Clinico-Genomic Database was restricted to NSCLC patients, a population that has a number of precision oncology options, and therefore, the findings from this study can't be broadly applicable in cancer. Furthermore, despite the availability of genomic profiling and precision medicine approaches in cancer, median five-year survival in advanced NSCLC is still low ― 5 percent for Stage IIIb and 14 percent for Stage IIIa disease.
Genomic profiling doesn't always benefit cancer patients, Agarwala acknowledged. Nearly a third of NSCLC patients are too sick at the time of their diagnosis to receive systemic therapy, and others may have alterations that aren't well understood or have targeted treatments associated with them. But even if they might not get precision medicine based on their tumor genomic profile, Agarwala said that this information can improve understanding of the mutational landscape of cancer and advance research on new drugs.