Real-world evidence (RWE) has emerged as a useful tool for industry and regulators in oncology—to date, the FDA has accepted submissions using real-world data (RWD) and evidence to support drug efficacy and effectiveness claims, especially in rare tumor types. It’s also been used to create comparator arms in single arm oncology trials.
But how can clinicians leverage RWE to inform their decision-making on how to best treat patients with cancer? We discussed that with Deb Schrag, M.D., M.P.H., Chief of the Division of Population Sciences and attending physician in the Gastrointestinal Oncology Center at Dana-Farber Cancer Institute, and professor of medicine at Harvard Medical School.
As both an observational researcher, whose research focuses on evaluating and improving the quality and effectiveness of cancer care delivery, and an oncologist, Dr. Schrag is uniquely positioned to address the impact of RWE in oncology clinical care. Among her top recommendations for advancing RWE is the development of standards: According to Dr. Schrag, RWE will remain a “companion contributor” to evidence packages until all RWE stakeholders embrace a common standards language.
Responses have been edited for clarity and length.
Q: Where do you see the greatest potential for RWE in cancer care?
A: I think of RWE’s potential areas of impact in three buckets: patterns of care, toxicity, and comparative effectiveness of treatment interventions.
- Patterns of care examine which patients receive treatment, which do not, and what factors prevented them from receiving treatment. RWE can give additional context to help us learn more about these care gaps, which is critical to developing quality care metrics and measurements.
Real-world evidence is also inclusive of oncology “edge cases,” including elderly patients and others underrepresented in clinical trials. Cancer is a disease of the elderly, and to better understand these populations, real-world data is an essential tool to inform how to approach care.
To do this, researchers can leverage data from administrative, Medicare, and commercial claims. These provide large RWD sample sizes inexpensively and efficiently, and, when analyzed, identify treatment patterns and durations that can work well in broader populations. Granular data sets like electronic health records (EHRs) provide richer data and greater detail on individual patients, but the tradeoff is a smaller sample size.
- Secondly, RWD highlights toxicities within large populations. For example, what percentage of 75-year-olds receiving a dose of immunotherapy end up in the hospital 30 days later? If you don't see shifts in metrics like hospitalization rates and emergency department visits in the data, then you can be cautiously reassured that exposure to these new agents isn't dramatically changing complication rates. If you see major differences, you can drill down and ask why.
- Finally, RWD helps to evaluate comparative effectiveness. EHR data provide access to biomarkers and other factors that might influence a patient’s response to treatment. For example, the drug trastuzumab targets patients with human epidermal growth factor receptor 2 (HER2)-positive breast cancer. Without access in the RWD to HER2 status, it's hard to evaluate the drug’s effectiveness.
The important thing is choosing the correct data set to answer your question: Sometimes claims data alone lack critical variables—in cancer, it's often the stage of the disease that is left out. Sometimes administrative data are crude and lack detail. However, with EHR data, you can develop preliminary estimates of effectiveness, particularly in niche populations. It's absolutely critical to realize that observational data—the preferred and more accurate term for RWD—can never be used to measure causality. Observational data are far superior to no data, but they are not a substitute or replacement for clinical trials. They can, however, help us design more strategic and efficient randomized controlled trials.
Q: Where does RWD affect clinician decision-making today?
A: We use RWD to identify gaps in care delivery, and will continue to do so. The faster we can obtain RWD, the more readily we can create high quality data visualizations and place that data into decision makers’ hands. The better we communicate this data, the easier it will be for stakeholders to take action.
RWD also impacts how we evaluate toxicity. Immunotherapy, for example, was approved for lung cancer patients, but there was concern about using it to treat patients with autoimmune disorders. After FDA approval, RWD from patients treated with immunotherapy was cautiously reassuring, as it suggested that off-label use and scope creep did not result in any evidence of danger. However, “cautiously reassuring” is not the same as “positive compelling evidence” and “definitive benefit.” It's evidence of a lack of irreparable acute harm.
Real-world evidence has been better at demonstrating no irreparable acute harm in off-label contexts than in showing that drugs work as well for 80-year-olds as they do for 60-year-olds. For that, you need more detailed data.
I think we'll get there, both as statistical methodologies improve and as the ability to manipulate clinical data for these purposes expands. I'm optimistic, but also sensitive to the need for rigorous randomized controlled trials.
Q: You’ve mentioned that the field of oncology would greatly benefit from standards around data collection and quality. Which would you consider top priority?
A: We need more tools and guidelines to enable structured reporting for all oncology scans—not just for the presence of cancer, but also for the cancer’s progression. We also lack data standards to understand a drug’s clinical benefit and the patient’s progress. It's amazing how much ambiguity exists around those fundamental questions.
The field has made some progress. Oncology researchers from Kaiser Permanente have developed a model for data standards called the Prognostic Information System, or PRISM. PRISM retrospectively evaluates endpoints from EHRs and allows us to calculate priority metrics like real-world progression-free survival, real-world response, real-world recurrence, disease-free survival, and progression-free survival.
When a radiologist interprets a mammogram, they give it a Breast Imaging Reporting and Data System (BI-RADS®) score from zero to six to indicate the likelihood of breast cancer detected. Though it’s not a perfect system, it’s a helpful standard: a mammogram with a BI-RADS of 1 indicates no evidence of cancer (a zero often means there was an error with the test), and a BI-RADS of six means the patient has a confirmed breast cancer diagnosis. We now have Prostate Imaging Reporting and Data System (PI-RADS®) for prostate cancer scans, as well, which operates on a similar one-to-five evaluation system.
Q: How do you envision the future of RWD use in clinical care?
A: We’re in early days here, but we work with the data we have. We may tell an 85-year-old patient that a clinical trial has shown a drug works well in 60-year-old patients. Or, we explain we have a lot of data for patients treated with a drug, but it’s unclear how this treatment will work when combined with their bad rheumatoid arthritis.
Patient care remains a combination of common sense and good principles of clinical medicine.
The real-world data sources available to physicians have certainly evolved over the years. Now, we can look at the data to identify patients with a specific set of conditions and see whether particular drugs work for them or not. However, we cannot yet go to a database and quickly identify a treatment’s benefit for a large group of people with a particular problem, and the complementary data sources clinicians use in decision-making—data in case series, anecdotal reports, and RWE—are imperfect and fraught with selection issues.
As the power of computing improves, so will the ability to quickly aggregate data sources and perform specific queries, like the effect of chemotherapy drugs in niche populations. We’ve seen just how powerful data can be in the Covid-19 pandemic—the faster we get good data, the faster we can solve problems. Having standards facilitates data sharing, which is why we are so focused on developing community consensus around standards.
We don’t yet have RWD at our fingertips to answer some of those questions, but we’re getting closer.