Huron Accelerates the Administration of Clinical Trials Using Generative AI on AWS

Huron specializes in strategic consulting, technology, and analytics solutions for clients primarily in healthcare and higher education. It helps large research organizations reduce
risk, manage costs, and address compliance for clinical trials. 

In particular, Huron provides expertise in coverage analysis, which determines who will pay for each billable item or service that a study requires. “What’s really important to our clients
is the ability to start clinical trials as quickly as possible,” says Cody. “It’s a challenge that organizations all over the country face. And the obstacles include the components that surround coverage analysis, such as the clinical trial budget and the contracting piece.” 

For coverage analysis, Huron consultants require specialized knowledge to support informed searches of multiple medical documents. Typically, they must manually search complex documents one by one to identify clinical guidelines for billing determinations. Those documents often contain important unstructured data such as patient records. “Our consultants wanted to shift the focus from having to comb through all these documents and extract information manually,” says Cody. “We want users to home in on critical thinking with heavy thought-provoking tasks that drive the coverage analysis.”

At first, Huron’s AI team experimented with a proof of concept that was based on predictive modeling, which uses algorithms to analyze historical data and predict outcomes. But Huron realized that the complexity of unstructured data and the need for more nuanced answers warranted another approach. So it shifted to generative AI, first experimenting with a large language model (LLM) from a leading AI provider and building the support infrastructure from scratch. The team had to write code for chunking—the breaking down of complex, lengthy documents into smaller segments for the model to process—and manage the vector store database that housed metadata. 

In March 2024, Huron’s AI team opted to try a proof of concept using Amazon Bedrock on its own, as opposed to the homegrown LLM-based solution. To develop the prototype, Huron met weekly with the AWS solutions team and collaborated with the AWS Generative AI Innovation Center and strategy experts with deep experience in generative AI techniques. Huron discovered that it can deliver more relevant and accurate responses using the model family of Anthropic’s Claude in Amazon Bedrock. Moreover, the company maintains full control over the data of its initial generative AI solution through built-in Amazon Bedrock security and privacy protocols, such as the encryption of data in transit and at rest. 

Additionally, Huron’s AI team improves efficiency and reduces complexity by using Amazon Bedrock native capabilities for handling chunking, vector storage, and information retrieval. By replacing four custom scripts with a single API call, the team has streamlined workflows and significantly minimized the risk of errors. Through the built-in functionality of Amazon Bedrock Knowledge Bases, which lets organizations deliver more relevant, accurate, and customized responses, Huron’s AI team has cut in half its development and debugging efforts for the content vectorization process. Huron further improves search functionality using Amazon OpenSearch Serverless, which delivers search and log analytics without provisioning and adjusting resources. “The benefits of Amazon Bedrock go beyond the model itself,” says Clara Liu, director of Huron’s AI Capability Center. “It’s the ease of use and incorporation into the whole solution.”

In August 2024, Huron began to deploy its solution that is powered by Amazon Bedrock, significantly shortening turnaround times for launching clinical trials. “We can type a question, type a follow-up question, and query all those documents at once,” says Cody. “And that’s the perfect picture of creating efficiency. We no longer have to search documents manually.”

Now that it has standardized a foundation for the use of generative AI, Huron plans to continue improving its internal processes and offer its generative AI solution to additional public projects. “Organizations want to get patients on clinical trials quickly, especially for novel treatments that are making a difference in patients’ lives,” says Cody. “Anything that facilitates the administration of a clinical trial is a step in the right direction. We can support that in a very innovative way that can make a difference not only for the organization but ultimately for the patients.”