Transforming Healthcare Document Processing with Generative AI: A Collaboration Between Myriad Genetics and AWS

Addressing Challenges in Medical Documentation Management

Unpacking the Bottlenecks in Healthcare Operations

The Role of Amazon Bedrock and Generative AI in Streamlining Processes

Technical Implementation and Optimizations

Enhancing Document Classification with Generative AI

Automating Key Information Extraction for Efficiency

Results and Business Impact

A Future-Ready Approach to Document Processing

Hands-On Experience: Explore the Solution Yourself

About the Authors

Acknowledgements

Transforming Healthcare Document Processing with Generative AI: A Case Study of Myriad Genetics

This post was co-authored by Martyna Shallenberg and Brode McCrady from Myriad Genetics.

In a rapidly evolving healthcare landscape, organizations face tremendous pressure to efficiently manage complex medical documentation while providing quality patient care. For companies like Myriad Genetics, a leader in genetic testing and precision medicine, the challenge of processing thousands of healthcare documents daily is substantial.

Myriad Genetics’ Revenue Engineering Department handles vast amounts of data across various specialties, including Women’s Health, Oncology, and Mental Health. Each day, the team processes crucial documents such as Test Request Forms, Lab Results, Clinical Notes, and Insurance Certifications. Traditionally, these documents underwent manual classification, requiring significant time and resources to ensure accuracy and compliance, often resulting in slow workflows and increased operational costs.

To combat these challenges, Myriad sought a more effective solution. This blog post will delve into how Myriad Genetics partnered with the AWS Generative AI Innovation Center (GenAIIC) to enhance their document processing operations using advanced AI technologies, particularly Amazon Bedrock and Amazon Nova foundation models.

Identifying the Bottlenecks in Healthcare Operations

Efficient document processing is vital for regulatory compliance and the delivery of timely patient care in the healthcare sector. However, Myriad’s previous setup, which combined Amazon Textract for Optical Character Recognition (OCR) with Amazon Comprehend for document classification, encountered several limitations:

• Operational Costs: Each page processed averaged around 3 cents, culminating in monthly expenses of approximately $15,000 per business unit.
• Classification Latency: The average processing time stretched to 8.5 minutes per document, causing delays in prior authorization workflows.
• Manual Information Extraction: Critical data extraction relied heavily on skilled human analysts, consuming significant time and resources.

Myriad’s objectives became clear: reduce costs, streamline processing times, automate information extraction, and scale operations across various business units and document types.

Harnessing Amazon Bedrock and Generative AI

The solution lay in adopting advanced large language models (LLMs) capable of understanding the intricacies of healthcare documents. Amazon Bedrock, a fully managed service offering access to various high-performing LLMs, provided Myriad with the tools necessary to revamp their document processing workflow.

Utilizing Amazon’s innovative models:

• Amazon Nova Pro: Designed for efficient document classification.
• Amazon Nova Premier: Equipped with advanced reasoning capabilities for precise information extraction.

A Comprehensive Solution Overview

We implemented a structured solution using AWS’s open-source GenAI Intelligent Document Processing (GenAI IDP) Accelerator. This scalable, serverless architecture converts unstructured documents into structured data, allowing simultaneous processing of multiple documents without overloading downstream services.

Key Features of the Accelerator

1. Concurrent Processing: Managed by utilizing DynamoDB to track concurrent processing jobs.
2. Efficient Text Extraction: Amazon Textract analyzes text and layout components of documents.
3. Advanced Classification: Leveraging LLM prompts for identifying and categorizing documents.
4. Key Information Extraction: Extracting relevant medical data using specialized prompts.

Through this process, Myriad’s operations team successfully integrated the GenAI IDP Accelerator into their existing workflow, significantly enhancing processing capabilities.

Overcoming Document Classification Challenges

Myriad’s existing classification system achieved a commendable 94% accuracy rate, yet issues persisted due to the structural and formatting similarities among various documents. The solution involved refining prompt engineering techniques for enhanced accuracy using semantic analysis. Key strategies included:

• Format-Based Classification: Training LLMs to recognize specific formatting cues that differentiate similar document types.
• Negative Prompting: Instructing the model on characteristics to avoid misclassifying documents, such as specifying clear distinctions between different forms.

This renewed focus on prompt optimization resulted in improved classification performance, achieving a remarkable 98% accuracy while reducing processing costs by 77%.

Automating Key Information Extraction

With mounds of information needing extraction, Myriad faced another challenge in automating KIE. Employing the GenAI IDP Accelerator, the team tackled various complexities associated with healthcare forms, particularly:

• Checkbox Recognition: Enhancing Amazon Textract’s capabilities to distinguish marking styles.
• Visual Context Learning: Utilizing multimodal analysis to guide the model through concurrent examination of textual and visual form layouts.
• Chain of Thought Reasoning: Implementing sophisticated reasoning processes to assess ambiguous or complex extraction scenarios with higher accuracy.

Overall, this automation initiative not only improved extraction accuracy—maintaining a 90% success rate—but also significantly reduced labor-intensive requirements.

Tangible Results and Business Impact

The new generative AI-driven solution yielded significant improvements for Myriad Genetics:

Document Classification Performance

• Increased accuracy from 94% to 98%.
• Reduced costs from 3.1 to 0.7 cents per page.
• Decreased processing time from 8.5 to 1.5 minutes per document.

Automated Key Information Extraction Performance

• Maintained 90% extraction accuracy.
• Achieved processing costs of 9 cents per page with a turnaround time of 1.3 minutes.

Myriad expects to save up to $132,000 annually in document classification costs and enhance overall workflow efficiency across various business units.

Looking Forward

Myriad Genetics is excited to roll out this innovative document processing solution, starting with the Women’s Health division and subsequently expanding into Oncology and Mental Health. As noted by Martyna Shallenberg, Senior Director of Software Engineering at Myriad, "Partnering with the GenAIIC has been a transformative step forward… This has been an excellent example of how innovation and partnership can drive measurable business impact."

Conclusion

The AWS GenAI IDP Accelerator has empowered Myriad Genetics to tackle complex document processing challenges with unprecedented efficiency. By focusing on strategic prompt engineering, proper model choice, and comprehensive automation strategies, Myriad has set a new benchmark in the healthcare sector.

For organizations looking to refine their document processing capabilities, consider exploring the GenAI IDP Accelerator for a hands-on experience in transforming your workflows through generative AI technology.

About the Authors

Martyna Shallenberg and Brode McCrady from Myriad Genetics are at the forefront of implementing AI-driven solutions in healthcare, collaborating with AWS to optimize processes and enhance patient care through innovative technology.

Acknowledgements

We wish to extend our gratitude to the team members and collaborators who contributed significantly to this project, ensuring a successful partnership between Myriad Genetics and the AWS GenAI Innovation Center.

Ready to explore? Reach out to the AWS GenAIIC team today to see how you can personalize and optimize the GenAI IDP Accelerator for your unique business needs!