Real-world data (RWD) and Artificial Intelligence (AI) are evolving fast, and with the integration of unstructured data, large language models (LLMs), and multimodal data sources, the next generation of RWD promises deeper, more holistic patient insights than ever before. By combining clinical notes, patient-reported outcomes, and social determinants of health, next-gen RWD can deliver a complete view of the patient journey.
To explore how AI and machine learning (ML) are accelerating this transformation, I spoke with Elnaz Alipour, Patient Care Gaps and Customer Segmentation Lead at Pfizer, Jake Cohen, AI Tech Fund Lead at Memorial Sloan Kettering, and Rahul Das, VP of AI Solutions at Norstella. Here’s how they’re applying next-gen RWD in their organizations today, the challenges of deploying AI responsibly, and their predictions for what’s next.
Q: How do you define next-gen RWD, and how does it manifest in your work today?
Elnaz Alipour: Next-gen RWD helps us understand barriers to treatments in multiple phases of the patient journey, examining disease signs, prescriptions, and reported outcomes.
Jake Cohen: The two main areas where we incorporate next-gen RWD are in backend hospital operations, having to do with billing and scheduling, and in AI drug discovery, including clinical trials and translational research.
Rahul Das: At Norstella, we’re using next-gen RWD across the life cycle of drug development. Empowering our biopharma partners to identify patients is the main goal, from identifying patients for HEOR epidemiology type of research, to clinical trial patient recruitment, to market access efforts.
Q: What are some barriers or challenges that you’ve faced internally in trying to apply some of the newer AI and ML approaches?
EA: There are always concerns about data privacy, as well as regulatory concerns, in terms of what’s allowed and not allowed to be done with the data. We need to go through an approval process, so the time from developing a solution to being able to deploy it can be quite long.
JC: MSK thinks a lot about how to deploy AI in a way that’s responsible and ethical. Just a few months ago, our team released a major paper on AI governance. Across the organization, we’re ensuring that AI is of the highest quality and prioritizing value for patients. We’ve been partnering with life sciences companies for over a century, running clinical trials with industry partners to ensure that we can continue to do that and not slow down the process at all, but on the contrary, actually demonstrate that we’re able to operationalize those novel AI models.
Q: Building on that, how do you make sure that what you generate from AI modeling is something that you can operationalize and translate into something impactful?
JC: We’re doubling down on the use of AI and we have applications from across the institution: clinical trials, pathology, radiology, and surgery. In each of those, we’ve developed a framework that evaluates how AI is operationalized and deployed. There are gates that the models have to go through and risk levels that we assess to make sure that the AI is ready for the use case and fit for the purpose that it’s being used for.
EA: Everything we develop has a clear purpose and process. We ask our stakeholders what they need, what questions they’re trying to answer from the evidence generation side or the market access team on the medical side. It’s about inherent and close collaboration with the stakeholders, both on the HCP sides outside of our organization and with our field team.
Q: In the next five years, what are your predictions for real-world data as it pertains to AI?
RD: The main challenge right now in our industry is the fragmentation of the data coming from different types of modalities. In the next five years, we need to improve how we represent a patient journey, combining multimodal data and then layering AI and machine learning on top of that. That would automatically lead to faster insights.
EA: I expect there will be higher barriers to data integration due to data value and privacy concerns, but on a positive note, I also think there will be progress in patient phenotyping and tailored treatments if we can overcome data aggregation challenges.
JC: By 2030, I’d love to see a complex biomarker identifying patients for drugs and bringing together all sorts of multimodal data; not just one gene, for example, but multiple different sources of data labs. Having patient-level data and notes in the hands of teams that are actually taking these drugs to the patient is going to be very powerful.
