Leveraging Real-World Evidence to Optimize Trial Design

ACT: Are there any opportunities for these AI models to leverage data such as real-world data (RWD)? If so, how?

Kondru: As we all know, there is really a vast amount of unstructured, real-world evidence data out there including billions of patient notes. This data has been untapped for clinical trials for many years, and in the past few years, there is really a focused effort by the life sciences community to use AI (artificial intelligence) to be able to understand the patient journey in real-world evidence and be able to map that into clinical trials. This has tremendous consequences in really creating the protocols and optimizing the protocols for the patient population. This is one area. Second is reaching the diverse patient populations. Diversity and inclusion has been one of the critical issues in clinical trials. Having the accurate representation of participants in a clinical trial is necessary for us to deploy the drugs that come out onto our wider population, and so identifying optimal endpoints, patient eligibility criteria, and treatment regimens are important within real-world evidence, and they're also important in clinical trials. By having this connection, I feel like we can actually deliver best value for our participants. I talked about patient identification and recruitment a few minutes back—real-world evidence gives you the best pool of information for patients across our ecosystem, so connecting those sorts of information is really critical, in my opinion. I also want to highlight a topic which is kind of important is, how do you analyze these large sets of data using large language models? There is federated learning, anomaly detection, causal interference—there's a lot of thinking that goes into looking at this unstructured text that is prevalent in real-world evidence. Creating a structured outcome for it and being able to impact clinical trials while following the real-world evidence data sets.