Exclusives

Amazon Web Services (AWS) used its AWS Summit New York keynote on July 26 to unveil new generative artificial intelligence (AI) and machine learning (ML) offerings, including AWS HealthScribe, a new generative AI-powered service that it said automatically creates clinical documentation.

HealthScribe leverages speech recognition and generative AI to automatically create preliminary clinical documentation from patient-clinician conversations, according to AWS. 3M Health Information Systems, Babylon Health and ScribeEMR are among the customers and partners that AWS said are “looking forward to using” the new service.

AWS also announced the expansion of its fully managed foundation model (FM) service Amazon Bedrock to include the addition of Cohere as an FM provider and the latest FMs from Anthropic and Stability AI, as well as a new capability for creating fully managed agents in only a few clicks—”a game-changing feature for builders, with no expertise required,” according to AWS.

“What do customers need to do to unlock the value of generative AI for their use cases? The first thing is you need access to best in class foundational models  … because there is going to be no one model to rule them all,” Swami Sivasubramanian, VP of databases, analytics and ML at AWS, told attendees and those viewing online. “It’s about choosing the right model for the right job,” he said.

“Then the customers need the ability to securely customize these models with that data and then they need easy to use tools to democratize generative AI within their organizations and improve employee productivity,” he explained. “And underpinning all of this is you need to keep your cost and latency low with purpose built ML infrastructure.”

AWS is “delivering all of this to our customers through Amazon Bedrock,” he said. “With Bedrock, customers can easily build and scale gen AI applications. It’s a selection of industry leading FMs, all with a simple” application programming interface (API) without managing any infrastructure. Amazon Bedrock makes it easy to customize these foundational models with your data.”

HealthScribe is powered by Amazon Bedrock and “makes it faster and easier for healthcare software providers to integrate generative AI capabilities into their application,” AWS said in a news release.

“HealthScribe enables responsible deployment of AI systems by citing the source of every line of generated text from within the original conversation transcript, making it easier for physicians to review clinical notes,” AWS said.

“Generative AI is quickly transforming many industries, including healthcare and life sciences,” according to AWS. “As interest in generative AI continues to grow, healthcare software vendors are looking to leverage this technology in their clinical applications to solve common pain points for clinicians in the healthcare industry,” the company said, adding: “One of the most common issues is compiling clinical documentation after every patient-clinician discussion. This is important for compliance, quality measures, and reimbursement, but it is also a complex, multi-step process that takes time away from seeing patients. While many of these healthcare software providers use text to speech and natural language processing (NLP) to streamline this process today, generative AI has been the missing piece to help these applications go from recorded discussions to concise clinical documentation.”

But AWS conceded that “working with generative AI is complex, and integrating multiple AI systems into a cohesive solution requires significant engineering resources.”

In order to “build these generative AI capabilities, a provider must train or fine-tune their own LLM to generate accurate clinical documentation, which requires access to in-demand AI experts, massive amounts of carefully annotated healthcare data, and significant compute capacity. Even then, an LLM for healthcare needs to be specially trained to understand complex medical terminology across different specialties (e.g., general medicine, pediatrics, or orthopedics), to be capable of understanding, analyzing, and summarizing free-flowing discussions, as well as recognizing prescription names and dosages. To ensure these solutions are working properly, software providers must also build with responsible AI in mind, including designing the solution so that clinicians can trace the origin of any generated text to mitigate the risk of errors or hallucinations. Healthcare software providers must also dedicate engineering time and resources to ensuring these systems meet the stringent security and privacy requirements of the healthcare industry. Because of these barriers, it is challenging for healthcare software providers to bring AI-powered solutions to market quickly, despite the potential benefits to both clinicians and patients.”

By integrating AWS HealthScribe into a clinical application, healthcare providers are able to “leverage built-in text-to-speech capabilities to create robust conversation transcripts that identify speaker roles and segment transcripts into categories (e.g., small talk, subjective comments, or objective comments) based on clinical relevance,” according to AWS. “The application can then use AWS HealthScribe’s NLP and generative AI capabilities to extract structured medical terms, such as medical conditions and medications, and generate discussion-based notes that include relevant details.”

AWS also announced the general availability of AWS Entity Resolution, an analytics service powered by ML that the company said “helps organizations easily analyze, match, and link related records stored across applications, channels, and data stores.”

AWS Entity Resolution uses customizable workflows that AWS said “leverage rule-based and ML techniques to join related consumer, business, and product information.”