From silos to insights: How generative AI is ‘democratizing’ biopharma data insights
The demand for real-world evidence is growing
Over the last decade, real-world evidence (RWE) has emerged as a critical component for every stage of drug development and commercialization. Today, RWE plays a pivotal role in regulatory decision-making, market access and reimbursement, clinical development, safety monitoring, and competitive differentiation. As its scope and importance continue to expand, so do the demands on the scientists and organizations who work with real-world data (RWD).
Recently, vice president of Real-World Data Activation at Pfizer, Jennifer Webster, noted:
A key challenge comes from the breadth of work that can be done using real-world data. A RWE scientist might be asked to complete a standard treatment patterns study for publication in the morning, a quick market sizing exercise in support of business development at noon, and an analysis using real-world genomics data in the afternoon. Last time we counted, there were more than 80 real-world evidence use cases. No one has all of those skills, and no one can keep up with the new opportunities.
Many organizations have made significant investments in building internal capabilities to meet the growing demand for RWE. These efforts include cloud-based advanced analytics platforms, self-service cohort builders, and dedicated internal teams of RWE scientists, statisticians, and programmers who design and execute analyses and studies.
Despite these investments, the demand for RWE continues to outpace the internal capacity of biopharma companies. To keep up with the rapidly evolving requirements of RWE, existing tools and processes must be modernized.
The limitations of existing technologies
In recent years, a variety of self-service tool suites have emerged across the industry, each offering two main value propositions. First, these tools aim to accelerate cohort definition and feasibility analyses, as well as automate routine tasks (e.g., descriptive statistics for cohorts of interest). Second, they empower a broader range of stakeholders to analyze real-world data through intuitive point-and-click interfaces, eliminating the need for coding in languages like SQL or R.
While these tools have succeeded in accelerating cohort definition and producing commonly requested analyses, they have not fully realized the promise of democratizing RWE. Their point-and-click interfaces often demand extensive training and present a steep learning curve. Moreover, many vendors offer these solutions as hosted services outside the biopharma environment, introducing challenges with integration, data control, and scalability, particularly due to per-user licensing models. As a result, the vision of truly “democratized” RWE generation remains elusive.
Enter Generative AI: changing the way we generate RWE
The technology landscape for generating RWE is undergoing a dramatic transformation with the arrival of generative artificial intelligence (GenAI), which is reshaping business practices across industries. And biopharma is no exception.
GenAI is revolutionizing how companies interact with RWD by allowing users to “talk to their data” in a natural, conversational manner. However, it’s not as simple as asking a foundational large language model (LLM) a question and instantly receiving the answer. While foundational LLMs excel at natural language reasoning, they can be prone to hallucinations and often lack a native understanding of data schema, clinical code systems, and temporal logic. Additionally, they are not designed to meet the audit, traceability, and nuanced demands of RWE. Consequently, RWE requires a purpose-built GenAI solution.
Introducing RWE Agent
Deloitte has worked with Amazon Web Services (AWS) to develop RWE Agent, a sophisticated conversational assistant designed to empower a broad range of stakeholders to analyze RWD, generate insights, and ultimately realize the vision of democratizing RWE.
Given the complexity and nuances of RWD and RWE, we implemented a multi-agent architecture featuring specialized agents tailored to specific tasks such as rules, reasoning, and analytics. When a user submits a question in natural language, a supervisor agent breaks it down into smaller components and assigns each part to the appropriate specialized agent. These agents then collaborate to understand the prompt, complete their respective tasks, and then pass the work to the next agent, ensuring a seamless workflow and a high level of accuracy.
Multi-agent architecture
How RWE Agent utilizes a multi-agent architecture to enable democratization of RWE.
The RWE Agent architecture consists of three main components:
- Multi-agent orchestration: Deployed within clients’ AWS environments, RWE Agent leverages foundational models available through Amazon Bedrock to enable advanced reasoning and orchestrates agent-driven workflows using LangChain.
- Data democratization: Amazon Bedrock Knowledge Bases provide robust support for multiple datasets in their native formats. The onboarding service interprets dataset semantics, building the knowledge base by integrating Amazon Bedrock foundation models with AWS Lambda.
- Agent user interface: RWE Agent includes a conversational application tailored for RWE scientists. This interface is built on Amazon Elastic Kubernetes Service (Amazon EKS) and utilizes foundational AWS services such as Amazon Relational Database Service (Amazon RDS), Amazon Simple Queue Service (Amazon SQS), Amazon Simple Notification Service (Amazon SNS), and Amazon CloudWatch.
By deploying RWE Agent within a biopharma company’s AWS environment, clients retain complete control over their data while seamlessly integrating with existing systems and workflows, thanks to its fully serverless architecture enabling easy deployment within client environment.
Key differentiators
RWE Agent is engineered for accuracy, transparency, and scalability, offering several key capabilities that set it apart:
- Zero-ETL (extract, transform, and load): RWE Agent works directly with data in its native format, eliminating the need to convert to a common data model and thereby saving time and resources.
- Handles complexity: It can process highly complex inclusion and exclusion criteria across various clinical domains—including temporality—making it ideal for advanced use cases.
- Continuous learning: The solution offers intelligent suggestions based on prior RWE analyses conducted within the organization, maximizing the value of historical studies.
- Human in the loop: Critically, RWE Agent incorporates a “human in the loop” at key stages, allowing users fine-grained control to review and refine important details such as agent-selected medical codes and generated SQL queries, all without the need for repeated prompting. This approach accelerates output, enhances transparency, and builds confidence in the solution’s accuracy.
Unlocking next-gen RWE with Deloitte and AWS
Real-world evidence generation is rapidly evolving, with GenAI innovations like RWE Agent opening new possibilities for biopharma. These technologies make it possible for more stakeholders to work directly with real-world data, ensuring greater transparency, scalability, and speed. As a result, organizations can create significant efficiencies in insight and evidence generation, and produce high-quality, more actionable insights that drive better patient outcomes and informed strategic decisions.
If you are interested in moving beyond traditional approaches and want to see firsthand how generative AI can transform your use of real-world evidence, now is the perfect time to take the next step. Connect with our team to learn how RWE Agent and other generative AI solutions can modernize your RWE capability.
