Skip to main content
Premium Trial:

Request an Annual Quote

At ASCO, Real-World Data Shows Potential to Improve Precision Oncology


This article has been updated to note that Foundation Medicine and Flatiron Health's clinico-genomics database has around 50,000 profiles.

CHICAGO (GenomeWeb) – At the American Society of Clinical Oncology's annual meeting, researchers demonstrated the potential of real-world data to explore the efficacy of precision oncology drugs in rare populations, predict how long lung cancer patients will live, and identify gaps in genetic testing.

For example, researchers from Roche, drawing on real-world data from Flatiron Health and Foundation Medicine’s clinico-genomics database, compared the efficacy of its investigational non-small cell lung cancer drug entrectinib against Pfizer’s crizotinib (Xalkori) in 1 percent of NSCLC patients with ROS1 fusions and showed that patients receiving the newer drug may fare better in terms of progression-free survival. 

Researchers at the US Department of Veterans Affairs have combined clinical and genomic data collected from veterans who were treated within the healthcare system for lung cancer and built the initial version of an algorithm that could potentially be used to predict patients' survival at diagnosis. Integra Connect, a provider of cloud-based technologies that help physician practices meet cost and outcomes metrics of value-based reimbursement models, demonstrated how real-world evidence can identify lung cancer patients who weren't receiving guidelines-based genomic profiling to direct targeted treatment decisions.

In the US each year, 1.7 million people are diagnosed with cancer, but only 3 percent enroll in clinical trials. Real-world data is information collected outside of a clinical trial setting, often with the help of electronic health systems and other data repositories. Although data collected within EHRs and similar databases can be a valuable resource for informing patient care, institutions use a variety of EHR platforms and store information using different terms, and as a result the data cannot be easily integrated and queried for research purposes.

Recent efforts to integrate and clean up EHR and genomics data sets across institutions by companies like Flatiron and others have now given researchers access to more real-world data than they previously had access to. As the numerous abstracts presented this week at the ASCO annual meeting show, researchers are using real-world data to glean insights that they couldn't garner in traditional clinical trials.

"One of the roles of real-world data is to answer questions that might not be answerable" via a traditional study, said Foundation Chief Data Officer Gaurav Singal, who played an integral role with Flatiron experts to develop the clinico-genomics database, which combines clinical data from about 50,000 cancer patients treated at 200 cancer centers and their genomic test results.

The study presented at the ASCO annual meeting by researchers from Roche, Flatiron, and other institutions demonstrated the ability of real-world data to compare treatments in a rare, molecularly defined patient subpopulation. Roche has evaluated the efficacy of entrectinib in ROS1-positive patients in three, single-arm Phase I/II studies, and submitted the data to the US Food and Drug Administration as part of its registrational evidence for the drug.

Meanwhile, back in 2016, in order to garner FDA approval for crizotinib as a treatment for ROS1-positive metastatic NSCLC, Pfizer submitted data from a single arm trial involving 50 patients with this tumor marker. "It was hard enough to get the initial studies [of these drugs] done, and because of the rarity of ROS1 fusions, a comparator study may never get done," said Singal, who wasn't an author on this particular real-world study.

Moreover, the differences in enrollment criteria between the original single-arm studies involving entrectinib and crizotinib don't enable a side-by-side efficacy comparison. "You may be comparing apples to oranges, and that really becomes tough as a provider," he said.

Instead, researchers described in an ASCO abstract how they retrospectively identified a real-world cohort of 69 patients who received crizotinib within the clinico-genomics database between January 2011 to June 2018 and who had the same inclusion and exclusion criteria used to enroll 53 patients in the entrectinib studies. The study then compared these two cohorts and showed that entrectinib reduced the risk of disease worsening or death by 56 percent and patients, on average, stayed on entrectinib longer than on crizotinib, a median of 14.6 months versus 8.8 months. Median overall survival for patients receiving entrectinib was not reached at nearly 16 months of follow up, while median overall survival for patients receiving crizotinib was 18.5 months.

The FDA is using real-world data to improve understanding of the efficacy and safety of medical products, and last year at ASCO presented data it had gathered with the help of Flatiron's clinico-genomics database. The agency also recently granted a priority review designation to entrectinib as a treatment for metastatic ROS1-positive NSCLC, and adult and pediatric patients with NTRK fusion–positive locally advanced or metastatic solid tumors. The agency is slated to make a decision by Aug. 18, and the real-world data comparing entrectinib and crizotinib "are under discussion with the FDA," a spokesperson from Roche subsidiary Genentech said.

At the same meeting, Nathanael Fillmore from the VA Boston Healthcare System discussed efforts to develop a prognostic model for determining one-year mortality using real-world data from around 350 NSCLC patients who had enrolled within the VA’s precision oncology program between 2015 and 2017 and passed away by the end of September 2018. These patients were molecularly profiled for alterations in around 100 cancer-linked genes, were treated at the VA, and had information stored in its electronic health records system.

These real-world data are stored in the VA's precision oncology data repository, which the Massachusetts Veterans Epidemiology Research and Information Center (MAVERIC) uses to build predictive models that may be used to advance decision support tools and inform patient care. The NSCLC survival algorithm Fillmore and colleagues are developing with this data is an early attempt at providing more personalized outcomes predictions for veterans.

NSCLC is one of the leading causes of cancer deaths worldwide, and as a result there is "wide interest in developing prognostic models to predict mortality following diagnosis using real-world data," said Fillmore, director for machine learning and predictive analytics within MAVERIC. The initial data presented at the ASCO annual meeting showed this algorithm’s promise in being able to predict one-year lung cancer survival, including in patients with early and late-stage disease. Statistical analysis identified that the total number of mutated genes patients had (a proxy for tumor mutational burden), and whether they harbored mutations in TP53, were among the most important features in the algorithm. Additionally, mutations in KRAS, STK11, TSC2, ATM, PIK3CA, and CDKN2A, were also important features.

However, this algorithm does not yet incorporate important sources of information, such as the treatments patients received, which obviously impacts survival. Researchers are still refining the model.

Robert Doebele, an associate professor in the division of medical oncology at the University of Colorado School of Medicine who reviewed the VA study, pointed out a few other weaknesses of the algorithm. For example, he noted that the follow-up time frame of the study to develop the algorithm might be too short to accurately predict survival. Moreover, merely factoring in whether patients had mutations in the tested genes doesn't accurately account for how these markers are impacting survival by suppressing or driving tumors, he said.

Though the VA model doesn't yet factor in the effect of targeted treatments, Doebele noted that survival estimates would be impacted by whether there are or aren't approved therapies targeting specific genetic mutations. "Going with this idea that all genes and mutations are not equal, you can have variations in the same oncogene," he added. "Subclasses of mutations exist within each oncogene type" and impact survival differently. For example, ALK mutations aren't significant in lung cancer, but patients with ALK fusions will respond to ALK inhibitors.

One of the criticisms of real-world data, particularly in studying precision oncology approaches, is that it can be biased by who receives genetic testing and whether the results were ultimately used to guide treatment. Doebele cited a study that found that only 60 percent of lung cancer patients in community practices received EGFR and ALK testing from 2013 to 2015.

"One limitation of this study is that only those who were genetically tested were included in the VA cohort," he said, noting that the model should account for what percentage of VA NSCLC patients received testing and therapy directed by results. "This data may be the most impactful in terms of telling us where we're failing to implement guideline-recommended therapies."

Integra Connect is using real-world data to do just that. The company recently launched a solution called PrecisionQ to help doctors deliver precision care by providing them with real-world data insights on practice patterns, treatment outcomes and adverse events, and how interventions are impacting total cost of care.

At the meeting, researchers from Integra and Guardant Health presented a study in which they used the Integra Connect database, containing electronic medical records and claims data from around 600,000 cancer patients, to evaluate whether advanced lung cancer patients were being assessed for mutations in four genes — EGFR, ALK, ROS1, and BRAF — and if they were receiving targeted treatments and immunotherapies as recommended by guidelines.

Researchers manually reviewed the charts of 1,200 advanced NSCLC patients treated across five community oncology practices since January 2017 and found that only 22 percent were tested for all four genes — 54 percent were tested for mutations in EGFR; 51 percent for ALK rearrangements; 43 percent for ROS1 fusions; and 29 percent for BRAF mutations.

"Despite the onset of next-generation sequencing, close to 50 percent of the patient population is not getting tested,” said Ash Malik, president of the life sciences division at Integra and an author on the ASCO abstract. "When we did qualitative surveys with physicians, turnaround time and difficulty collecting tumor samples were at the top of the list for why this [undertesting] happened." Reimbursement pressures on testing may also be an exacerbating factor, he added.

Of the 163 patients who were found to have a targetable alteration in one of these genes, 55 percent did not end up on targeted treatment. Specifically, of the 84 patients who had targetable alterations in EGFR or ALK and had no evidence in the database of receiving targeted treatments, 37 percent were prescribed immune checkpoint inhibitors. Twenty four percent of the patients who got immunotherapy had results on targetable alterations before they received it, while 13 percent received test results after they got on immunotherapy.

The data demonstrate underutilization of genomic testing and targeted treatments, and use of off-label immunotherapy. "Further research is needed to identify strategies to increase testing in advanced NSCLC patients to provide physicians with the information needed to make optimal treatment decisions," they wrote in the abstract.

An interesting finding in the real-world data was that the median time to result for blood-based somatic tests was four days versus 14 days for tissue-based tests. Other studies have shown that the turnaround time for results from tissue-based next-generation sequencing panels can be as long as a month. Patients may not want to wait that long to get on targeted treatments based on test results and instead, ask for immunotherapy. Blood-based testing is a newer technology, Malik acknowledged, but it has promise particularly in advanced NSCLC since these tumors may shed more material in the blood.

As the use of liquid biopsy testing grows, real-world data can highlight to physicians its advantages, he said. While Integra's goal isn't to encourage tissue testing or blood testing, the data gives doctors visibility on who should be tested and flag the opportunities for guidelines-based treatment based on the results. This data, when considered in the context of the high cost of immunotherapy and potential adverse events that require active management, may help physicians make better value-based care decisions, Malik said.

Although the use of real-world data is growing, Singal noted that it is also important to recognize the drawbacks, primarily that it isn't collected in a randomized, controlled trial setting. But there are biases in clinical trials too, for example, patients tend to be carefully selected to meet the study criteria and not reflect the characteristics of those in the community.

"It's important to understand and characterize the biases and be deliberate in the steps you take toward analyses, and don't jump to conclusions before you're ready," he said, adding that the community will come up with methods to adjust for the biases in real-world data.

Researchers from Flatiron and Foundation validated their clinico-genomics database recently by confirming that data drawn from the repository could confirm previously known clinical and genomic features of NSCLC patients, including specific genetic mutations, PD-L1 expression, and tumor mutational burden, and how these features associated with treatment response. In a publication on the study, researchers outlined the limitations of this real-world data, but also envisioned it could be broadly useful in drug development, clinical trial design, and eventually, clinical decision making. 

Just a few years ago, doing this kind of large-scale research was unthinkable, Singal recalled. It was a rite of passage for medical students to painstakingly extract data from patients' charts at a single institution, often to answer just one specific research question. The variety of abstracts presented at the ASCO meeting this year signals to him that groups are leveraging these big datasets amassed from hundreds of centers to explore how best to take care of patients.

"It's less about the results [of real-world data studies] at this point than about the fact that this conversation is even happening," he said.