Transforming Real-World Data (RWD) Strategies with AI: Key Insights
This panel discussion brought together experts from across the industry to explore the intersection of AI and RWD and highlighted its tremendous potential and possible challenges.
By on Nov 27, 2024Transforming Real-World Data (RWD) Strategies with AI: Key Insights
As artificial intelligence (AI) continues to revolutionize the healthcare and pharmaceutical industries, it's becoming increasingly important for pharma executives to understand how it can help their RWD strategies.
This panel discussion brought together experts from across the industry to explore the intersection of AI and RWD and highlighted its tremendous potential and possible challenges.
The Role of Unstructured Electronic Health Record (EHR) Data in Real-World Research
When choosing data sources for real-world research, it's essential to consider factors like diversity, source fidelity, and the extent to which the data has a longitudinal dimension.
Jordan Overcash of Veradigm emphasized the importance of having data from multiple specialties and practices to ensure health equity representation, and she stressed the value of getting data directly from the original sources to maintain continuity and the accuracy of interpretation.
“So, where is the data actually coming from? Are you getting it from the originators of the data, or is it passing through multiple different structures, layouts where that continuity of the data, the actual interpretation of that data could be shifting many times before it actually gets in your hands?” she says.
Lisa Prettypaul-Lodhia from Pfizer raised the need to know how data is actually used in real-world settings; for her, it is important to understand how healthcare providers (HCPs) are trained to use data, especially as they are frequently short on time.
“…what are the roles in the multidisciplinary team, especially when you’re talking about the handoffs between a medical assistant to a nurse, to a doctor? I think it’s worthwhile to think about that dynamic… because it’s going to look very different from one system to another,” Lisa added.
She also pointed out the value of unstructured data in understanding the whole patient, including information about how they live day-to-day—their family setup, and socioeconomic factors that might impact care decisions.
Kelly Zhou emphasized this interplay, noting, “There’s a balancing act between the rich information we can gain from doctor notes and the risk of provider burnout. Systematic approaches to harness this data are essential to improve outcomes while respecting provider time.”
Balancing Rich Information and Provider Burden
A key challenge in leveraging unstructured EHR data is balancing the wealth of information available with the potential for increased provider burden. Jordan noted that adding more structured data fields can lead to longer charting times and less patient engagement.
“The longer it takes them [HCPs] to chart, the more click boxes we give somebody, the more they’re looking at their screen instead of looking at the patient… the experience starts walking away… then maybe the patient doesn’t want to come back because they didn’t feel like they were actually engaging with the provider,” Lisa said.
She suggested using AI to mine unstructured data on the back end rather than burdening providers with additional data entry tasks.
John Diaz-Decaro of Moderna emphasized the importance of considering continuity of care and the reality that patients change jobs and insurance, leading to enrollment/disenrollment from datasets.
“People obviously lead real lives, and so there might be some missing data in some of these data sets as well,” he said.
This underscores the need to choose the right dataset for each specific research question.
Risks and Challenges of AI in Healthcare
While AI offers tremendous potential, it's crucial to be aware of the risks and challenges. John highlighted the issue of "hallucination" in large language models, where they may provide inaccurate information. This is especially concerning for those trying to gain clinical insights from the analysis. He also pointed out the inherent bias in many datasets, which can affect AI outputs.
Jordan stressed the importance of having clinical staff work closely with data science teams during the modeling process to ensure accuracy and proper interpretation of results. She noted that Veradigm aims for at least 93% accuracy in capturing enhanced data elements from unstructured sources.
Kelly Zhou pointed out the evolving regulatory landscape for AI in healthcare, stating, “Regulatory-grade data—or fit-for-purpose data—is key to ensuring AI is used responsibly. Ongoing validation processes are essential, particularly as new data challenges arise.”
Regulatory Considerations for AI in Healthcare
The regulatory landscape for AI in healthcare is rapidly evolving. John referenced recent FDA publications, including a JAMA article outlining the FDA's perspective on regulating AI in healthcare. Key points include:
- The need for ongoing validation and evaluation before and after model deployment
- Alignment with the FDA's lifecycle approach
- Forthcoming guidance on using AI to support regulatory decisions for drugs and biologics
Jordan emphasized the importance of using AI responsibly as a tool or "copilot," rather than removing humans from the process entirely, especially given the current regulatory uncertainty.
Lisa added, "As we become more patient-centric, especially in healthcare, the sources of data become even more important, especially when we know that depending on who you are and where you live, access to healthcare varies so much. The messiness of the data and how you code it really speaks to the need here for us to be very intentional about how we collect the data, how we read the data, and how we analyze it.”
Ensuring Accuracy and Reliability of AI Models
Validation processes for AI models vary depending on the specific use case and model type. Jordan stressed the need to establish a framework for each model that includes checkpoints, iterations, and ongoing accuracy assessments as new data is introduced.
“You need to lay out a plan… who is going to start the process? Where does the check come in? How do we do the iterations? How do we make sure along the way that we have what we need? And we’re holding that accuracy? And the biggest part after we have great accuracy… is how do you make sure that it continues to hold that accuracy as more data comes at it,” she said.
Reliability can be challenging for more ambiguous questions. Jordan noted that models may need to be fine-tuned for specific use cases and may not generalize well to other scenarios without additional work.
The Future of AI and RWD in Healthcare
Looking ahead, the panelists envisioned an exciting future for AI and RWD in healthcare. Some potential developments include:
- Increased integration of data from wearables and other smart devices
- Greater patient empowerment and ownership of health data
- More holistic views of patient health, incorporating lifestyle and environmental factors
- Improved personalization of care based on comprehensive data insights
Kelly Zhou reflected on the broader implications of these advancements, stating, “AI has the potential to integrate multimodal data sources, offering unprecedented insights. However, as we look to the future, we must ensure that technological advancements respect the humanity of patient care.”
Key Takeaway: Use AI as a Tool, Not a Replacement
The overarching message from the panel was that AI should be viewed as a powerful tool to enhance RWD strategies, but not as a replacement for human expertise and oversight. As Jordan Overcash stated:
"AI is a tool. It is something that we can use, we can leverage, but we need to use it responsibly and make sure that we're sitting within the regulations, that we're respecting the patient and the conversations they are having with the provider to ensure our use of it helps sponsors get the answers they need but is respectful to our patient population."
By approaching AI with this mindset, pharma executives can harness its potential to transform RWD strategies while maintaining the critical human elements of healthcare and drug development.