In this episode of Eureka!, McKinsey’s podcast on innovation in life sciences R&D, hosts Navraj Nagra, Alex Devereson, and Meredith Langstaff speak to Ashita Batavia, head of hematology and oncology data sciences, R&D, at Johnson & Johnson Innovative Medicine (J&J). They discuss how real-world data can better inform clinical trials, the shifts in mindsets and capabilities needed to support AI in biopharma, and how to navigate the complexity of clinical trials with more-pragmatic trial designs. An edited version of their conversation follows.
The need for more, higher-quality data
Alex Devereson: There’s been a huge increase in the volume of patient and disease data, and AI methods available to life sciences practitioners have become more sophisticated. What do you see as the biggest opportunities and challenges because of that?
Ashita Batavia: Having more data is exciting, but it’s important to maintain a high quality of data as the volume grows. When I started working in the real-world data space some years ago, we always focused on the top-of-the-funnel number and noticed significant attrition due to the data quality. As we start to stitch together larger data sets to create fit-for-purpose data sets, good data quality is really going to matter.
The breadth and depth of data are important, as is aligning the data we obtain for our use cases so it’s fit for purpose. There are several ways merge and harmonize different data sets—including different systems and different electronic health records (EHRs)—so you’re not double-counting patients or multimodal data. The conversation may start to shift even more, but the quality of the data should be emphasized.
Alex Devereson: What can help the industry increase the use of some of that data?
Ashita Batavia: Data has become more accessible, and it’s easier to generate. We know how to use data and analytics; now it’s more about putting those data sets together to approach a use case that’s top of mind. A little bit more magic has to happen to do that. The biggest enabling factor is talent and having access to it. It’s almost like having trilingual talent because they need to understand the data and be able to work with the data scientists while understanding the science, industry, application, and medicine.
On top of that, the business has an overarching strategy with all these externalities at play. So how can they achieve their strategy as efficiently as possible while managing costs? Real-world data isn’t necessarily always an inexpensive answer. Navigating all these factors and getting the right talent to realize the value are the ways to make the magic happen.
The talent and capabilities that can push biopharma forward
Navraj Nagra: What kind of talent are you looking for as you build out your team at J&J? What qualities do you think are critical to help move the organization’s goals forward?
Ashita Batavia: That profile shifts. There are basic necessary attributes, such as an understanding of data science and advanced-analytics principles. Even if you’re not programming, you can still interpret the outputs and help the teams sift through the information to glean insights. You need to understand the context in industry and, regardless of the industry, be able to communicate insights to diverse profiles, such as translational scientists, trialists, physicians, medical-affairs professionals, and commercial colleagues.
It’s rare to find a person who has, say, practiced medicine, has been a trialist who understands data sciences, and has a business strategy lens. But if a candidate has a couple of those skills, they can upskill and learn the rest. When it comes to talent, there are some intangible qualities that are valuable. For me, that’s an unstoppable need to learn, grow, and decipher, which is so important in a space like this, where the ground’s not yet steady and we’re still figuring out where AI can be applied.
The best data set today and the best data set in six months may not belong to the same company. So the ability to continuously scan and seek out opportunity where there is nothing but white space also belongs to a special type of individual. I’ve seen people on my team who have no formal data sciences background or science background, yet they thrive because they have an intangible desire to learn. The best team contains diversity of experience.
Meredith Langstaff: What are some capabilities that need to be built across biopharma to adapt to AI and data while tackling the challenges that come with those new capabilities?
Ashita Batavia: The one capability that doesn’t get as much airtime as it should is explainability: knowing what the data is, what the AI application is doing, why it’s doing it, and how we’re able to use it. Explainability should not be reserved for a variety of individuals who are involved in the work. The entire continuum needs to know that explainability is critical to impact.
Beyond that, it’s important to fit new insights, knowledge, or processes into existing workflows. If you try to lend insights to different parts of trials without paying attention to the current ways of working, it can create tension and friction that is counterproductive.
Privacy is another one. Given how fast-paced the space is, there has to be a balance between doing impactful work within your company based on business priorities and being a participant in the external environment to shape how data is used because, as the pathways for certain data or analysis become clear, that knowledge needs to be disseminated.
How a pragmatic trial design helps navigate complexity
Alex Devereson: Your white paper with Friends of Cancer Research discusses the pragmatic elements of oncology clinical trials. In your current role at J&J, what are some of challenges you see in increased complexity across oncology trials, and how do you foresee this being mitigated in the future?
Ashita Batavia: Oncology trials are becoming more complex in terms of the rigor of the inclusion/exclusion criteria, the quantity of tests, the frequency at which they’re collected, and the reporting requirements. Trial activities have shifted to larger academic centers and places where the demographics of the patients may not mirror the demographics of the disease. A lot of more diverse patients from different socioeconomic backgrounds tend to get care at community centers that aren’t always included in the conversation, especially when trials become more complex. Part of the spirit behind pragmatic design and moving toward simplification is to help us democratize access to trials and bring more communities and different providers into trials.
When we talk about pragmatic design, we mean being able to modularize design elements. One of the best examples of a pragmatic trial was sponsored by the Oncology Center for Excellence. During the project, we talked about expanding the inclusion/exclusion criteria, which allowed us to take patients with a lower performance status than we might otherwise take into a trial.
We also have to think about efficiencies and the resources we need to have versus those that are nice to have, which has implications at a business level in terms of how much we can spend and how we can deploy resources in a complicated environment. We scrutinized the schedule of assessments for a trial. So how many labs are you collecting, and how often are you collecting them? It’s best to have the collection cadence mirror what would happen in routine care so the patient isn’t obligated to come in for additional visits or blood draws. We also have a lot of primary and secondary endpoints, so there may be opportunities to streamline those. For trials in which we use drugs that are known and have side-effect profiles that we’re familiar with, maybe we can be less stringent with adverse effects (AE) reporting and focus more on serious AE, as opposed to more mild effects if there’s already ample literature on them.
It’s phenomenal that we can go back to regulators with these design proposals early and often to get feedback and find answers that we can bring to patients in their communities.
Navraj Nagra: Pragmatic trials have existed for several decades, yet there aren’t many examples of them being used in regulatory decision-making, especially in oncology. How can more trials incorporate these pragmatic elements?
Ashita Batavia: The biggest change has been the encouragement and interest of regulators. When they push in that direction and ask smart questions with us and respond to our efforts in the space, that’s what’s going to move the needle. And I think we’re starting to see it. As more of these trials go in front of regulators and gather feedback, we can better understand what is and isn’t working and adapt accordingly.
For another project I’m leading, we’re developing novel statistical methodologies that line up endpoints in the real-world data and the clinical-trial data so we can enhance the acceptability of external control arms and real-world evidence. Regulators were willing to provide support for this kind of research, and we were able to get academics involved in the broader team, which creates opportunities to advance the field.
The power of using real-world data to inform clinical trials
Meredith Langstaff: What is the role of real-world data across clinical trials, specifically trials in the oncology space? What challenges can real-world data address?
Ashita Batavia: Early on, we used real-world data to help us understand the populations we used for a trial, so they can help us understand the current standard of care and unmet needs. That supplementary research can help us down the road when we’re looking to submit applications for regulatory approval. In larger trials, the real-world data can help you understand what centers to prioritize and where you can have the most efficiency for enrollment.
Also, real-world data that is collected retrospectively can help us find patients with a subset of the criteria, so we can flag them for closer evaluation before they are offered a trial or screening. This step can help accelerate enrollment and can streamline collection of care through technology interfaces.
We’ve also been able to use diagnostic-testing-based imaging algorithms to find patients. With the retrospective data, this effort can provide a lot of value for drug development. For example, if there is a promising new therapy in oncology or a rare disease, you don’t want the standard of care to be random. An external control arm can populate the trial with prospective, real-world data and provide important context to improve the efficacy of a trial.
Real-world data can also be used to shape post-marketing commitments, safety signals, and clinical-trial tokenization. There’s a wealth of applications of real-world data across that drug development continuum.
Using early endpoints to accelerate drug development
Meredith Langstaff: We have seen some good research on the use of novel endpoints to accelerate approvals for treatments, including the recent FDA decision approving the use of minimal residual disease as an early endpoint. What impact does this have across oncology?
Ashita Batavia: If you think back, for most clinical trials in oncology, the main metric we cared about was the survival of the patient. Thankfully, for a lot of diseases, the time to that endpoint is getting longer because our drugs are getting better. So in the spirit of that, we’ve moved toward these surrogate endpoints. The big one in myeloma, for example, is progression-free survival (PFS). As treatments become more effective, the time to PFS is stretching years, which creates different challenges. If you have a new therapy that looks promising, you don’t necessarily want to wait years to get that therapy approved and for one of these endpoints to mature.
If we can find earlier signals, earlier endpoints, then we have a better opportunity to understand whether this drug is delivering a benefit faster. And if you’re the patient waiting for a new treatment, that’s fantastic. Having these endpoints doesn’t mean the trial stops—it continues until those endpoints are validated and fully accepted. We want to help patients. We can start to do that faster. And that’s the spirit behind all of these early endpoints.
The role of algorithms and AI in prescreening
Navraj Nagra: We recently read about using histopathology deep-learning algorithms for prescreening, which is a different angle on the clinical-trial applications of AI methods. What was your approach to this method, and how did you mitigate some challenges you faced?
Ashita Batavia: That algorithm was homegrown within J&J for patients with bladder cancer. When you have cancer and the tumor is found, a doctor will cut off a piece of the tissue and look at it under a microscope as part of confirmation. The staining that happens is this hematoxylin and eosin staining. When you look at that through a microscope, you can essentially take a picture of what you see. To this digital pathology image we could apply an algorithm that we built and validated to predict the presence or absence of a qualifying mutation for a trial, which pathologists are unable to do.
Also, as you can imagine, when you take out a piece of tumor, that tissue is extremely precious—there’s not very much of it, and you can’t go back in easily and get more. And molecular tests are expensive and take several days to a week to turn around. We tuned the algorithm to have an appropriate sensitivity and specificity to avoid the risk of excluding someone who might qualify from the trial. With the algorithm, we were able to get a good directional answer in minutes for pennies on the dollar, and we were able to prescreen and accelerate the time to an answer or a screening decision for some patients.
Patient finding could be another application for the algorithm. Providers can potentially run the algorithm in the background on the appropriate subset of patients to identify patients that could be a good fit for a specific trial.
However, the movement of these algorithms from research use to clinical practices is a whole other pathway that needs to be disambiguated, and this is where explainability comes back into play to make sure the test is reimbursable, for instance. For example, when we were figuring out how to send that insight back to the provider, it ended up being a bigger challenge than expected because pathologists aren’t used to receiving the molecular insight, so they didn’t understand how to report back this research finding. So we had to work through the mechanics to make efficient clinical-care pathways.
Alex Devereson: Are there any other examples of impacts you’ve seen using these AI-based biomarker detection algorithms on patients? And are there any other challenges you see with encouraging their uptake?
Ashita Batavia: The breadth of what’s possible in this space is exciting and a bit intimidating. There are so many different other opportunities to take big steps forward in medicine that are ripe for some sort of AI to be involved. For example, in lymphoma, one way for a radiologist to track response is to pick a lymph node on a scan and follow it over time. There’s more than just the one lymph node, but doing any kind of analysis of all the lymph nodes or the tumor metabolic volume is extremely time-consuming. If you could start deploying some of the AI sciences for volumetric analysis, it could be impactful.
Or another application we’ve looked at is for echocardiograms for AL amyloidosis, which is a difficult medical diagnosis that isn’t typically the first one that jumps to mind when the patient is in front of you. These algorithms can help identify that patient population easier and faster.
I don’t necessarily believe there is an algorithm that can solve every one of these obscure conditions, but we also aren’t sophisticated enough yet to know which diagnoses are able to have algorithms that work.
Fostering diversity in clinical trials
Meredith Langstaff: Tell us about a passion project of yours at J&J.
Ashita Batavia: I care a lot about diversity and making sure that access to trials is democratized. That means different things depending on the community and people. We spend a lot of time thinking about how trials can fit in the scope of racial and ethnic diversity, but there’s also sexual orientation and gender identity data to consider. That’s an area I feel passionately about and where we can be more inclusive. J&J also ran a degendered, transgender-inclusive prostate cancer study, which I was proud of.
I appreciate being able to do these sorts of collaborative projects, even when they fall outside of advanced analytics and the data sciences, and being around people in pharma who care about doing the right thing and finding the right solutions based on real-world data.
I think a lot about fill rates for certain variables, such as racial and ethnic and gender variables. Are gender variables being collected? Is there a field for it? If there is, is it being populated? More important, is it accurate? Who populated it: the patient, or somebody else? Having that conversation shift over the next few years is important to me. And that’s one area I’m excited about moving forward.
