How can AI and RWD work together to improve understanding of chronic diseases? Q&A with Kathryn Starzyk, Head of RWE at OM1
OM1, a new Aetion data partner, curates deep clinical real-world data (RWD) from electronic medical records (EMRs) and other sources, then builds artificial intelligence (AI) models to go even deeper, fostering a more complete understanding of treatment patterns and outcomes in a range of therapeutic areas.
In a recent interview, Kathryn Starzyk, epidemiologist and Head of Real-World Evidence (RWE) at OM1, shared what makes OM1’s data unique, and how she and her team partner with biopharma and health systems to generate RWE that helps researchers better understand patient journeys in disease areas including rheumatology and immunology.
Kathryn’s career started at the National Institutes of Health and Sanofi Genzyme, then, during her tenure at Outcome Sciences (now IQVIA), her interest in observational data grew as she led the development and execution of global patient registries. Now, Kathryn works cross-functionally across epidemiology, data science, informatics, and product teams to lead RWE initiatives at OM1.
Read on to hear more from Kathryn, including how she supports biopharma clients as they adopt RWE, and how RWD can inform the understanding of COVID-19’s impact on chronic disease management.
Responses have been edited for clarity and length.
Q: How does OM1 curate data, and what differentiates these data from other RWD sets?
A: At OM1, our goal is to understand the patient journey as completely as possible. Rather than having siloed data sets that you can take off the shelf and combine—or not combine—we feel strongly that you should start with longitudinal, deep clinical data, and look at outcomes in a standardized way across the representative population.
We love to go deep in specific therapeutic areas, which is a must if you’re going to spend time and resources to get meaningful data out of things like the semi-structured and unstructured parts of EMRs. Anyone who has worked with EMRs knows that the data aren’t always straightforward, and it takes quite a bit of effort to make them useful from an analytic point of view.
Once we have that clinical depth backbone, then we figure out the additional data we need to complete the puzzle. Is it linked to administrative claims? Is it socioeconomic data? Is it better sources of ascertainment of death? Are there AI models that we can build on the observational data to improve our understanding of these patients? In the end, the answers depend on the questions you’re trying to answer with the data.
Q: Which of OM1’s data sets is your favorite to work with, and why?
A: Right now, I’d have to say rheumatology and immunology, because of our ability to go deep. They are the furthest along in maturity, but the way we look at it, they’re growing and evolving all the time. For me, it’s exciting from a scientific point of view to be able to ask how we can extract more from our data sets. How can we learn more? How can we source additional data when we see unmet needs in a therapeutic area?
Autoimmune diseases are particularly complicated from an analytic point of view. In one of our lupus registries, for example, we see so much heterogeneity in the population, and the more we look at it, the more diversity we see in the way rheumatologists and other specialists manage patients’ disease. There’s so much still to be learned from the real-world data that we’re never going to get from clinical trials with highly selective populations; the patients in the real world are so different.
Q: How do your clients typically work with RWD and RWE?
A: It’s evolving. We work with both biopharma and health systems, but for biopharma, interest in RWE is growing at different rates in different segments of organizations. Sometimes we work with experienced scientists who are comfortable with RWD and RWE, and know exactly how RWE can support their goals. Then there are other cases where someone has heard of RWE, but they’re not quite sure how or where to apply it.
RWE isn’t new for HEOR or epidemiologists, regardless of where they sit within a company, but recently we’ve seen a lot more interest from R&D—in early stage development and in late stage. For early stage, they’re thinking about protocol optimization, for example, or testing out inclusion or exclusion criteria. They’re thinking about the heterogeneity of a patient population and the relevant outcomes, and how they can meet criteria for regulatory bodies.
We’ve learned that it’s important for biopharma to think proactively about evidence generation, and start building an evidence base early. They need to consider what additional data they can collect along with the required endpoints, and how these data can support conversations about reimbursement or comparative effectiveness post-launch—or how to integrate these data into synthetic or hybrid control arms in phase II or III.
Biopharma’s receptivity to using AI modeling on top of the RWD varies depending on the stakeholder and the use case. On the R&D side, they tend to see AI for exploratory applications as less risky; we sometimes use AI modeling to help cast a broader net as we identify a patient population, for example. I don’t blame clients for being hesitant to use AI or thinking of it as a black box, but the key is making people comfortable with the metrics around the models used. When it starts to make sense clinically, then it’s easier for them to accept.
Q: Are there any challenges you’ve observed biopharma clients face as they adopt RWE? How have you helped them troubleshoot?
A: The main challenge is often their comfort level with RWD and RWE. Some people may have had a bad experience with RWE, so solving their problem means educating them about what good real-world evidence looks like. It helps to have tangible, established standards for quality and validity when showing people where RWE fits into their strategy. It also helps that regulatory bodies, including the FDA, as well as payers or pharmacy benefit managers, are starting to articulate their expectations for RWE more clearly.
The best way we’ve found to troubleshoot with clients is to share concrete examples. We walk through examples of why data was fit for purpose, we explain the critical points of understanding data quality and how it met business needs. We share why a study met the requirements of a regulatory body or payer, or why it was accepted by a publication. It’s helpful when clients can get their arms around a real use case.
Q: How can RWD and RWE support research in disease areas like rheumatology and immunology?
A: Rheumatology and immunology are chronic disease areas, and part of the reason we chose to concentrate on them is because there is a lot of ambiguity around the comparative effectiveness of treatments, and also durability of effectiveness, where real-world data and real-world evidence are critical.
As an example, we have a rheumatoid arthritis (RA) registry. There are a ton of treatment options for RA—plenty of both conventional and biologic disease-modifying antirheumatic drug (DMARD) options. But when deciding on a treatment, it always comes down to the needs of the individual patient, and knowing when it’s time to switch to a different therapy. We do a lot of work analyzing switching behavior, and reasons for a switch in therapy, in the real-world setting.
Another good example is in multiple sclerosis (MS). Again, there are a lot of treatment options and many disease modifying therapies, but there’s a need for comparative effectiveness analyses to understand the real-world outcomes that matter to patients—subtle, day-to-day factors that impact quality of life, and which aren’t captured in most data sources. We can see these with real-world data.
Q: How can RWD support research and development in light of the challenges brought on by COVID-19?
A: All drug development has been impacted, to some degree, by COVID-19. It’s going to affect how long it takes to complete a trial, whether a trial can even be completed, and, ultimately, how trials are analyzed during this period of alternative delivery, both of interventions and assessments.
For us, there’s been a flurry of interest around whether some of the immunotherapies in our data sets are potential therapies for COVID-19, and whether we’re seeing higher rates of hospitalization or positive COVID-19 cases among high-risk populations, like the autoimmune or respiratory disease populations.
Eventually, we’ll be able to use our RWD to understand how COVID-19 has impacted the management of some of the chronic diseases we target, such as RA, lupus, and MS, and how these patients interact with the health system. These patients might be affected by drug shortages that are exacerbated by COVID-19, or patients that were accustomed to getting infusions in-clinic may be transitioning to self-injectable or oral drugs. We’ll be able to see in the data whether there are implications for compliance and longer-term outcomes.
Q: In your opinion, what are the most exciting opportunities for RWE today?
A: I’m most excited about the AI applications of real-world data. If we can get away from the black box perception of AI, apply the methods in a thoughtful way, communicate the relevant metrics, and help people understand how the models work, it can be a very powerful tool.
For example, we model things like disease activity in patients with autoimmune diseases, and it’s exciting to think that we’ll be able to access those measures at more frequent time points and in more patients—things that are really difficult to get from structured data, but help you get a larger view of how a patient population is doing. And for us, because we work on the health care system side as well, it’s exciting to see that this knowledge can have an effect on patient management and improving outcomes.