Transform healthcare revenue cycle management with Amazon Bedrock AgentCore

Healthcare organizations face significant challenges in optimizing their Revenue Cycle Management (RCM) processes while maintaining compliance and accuracy. The financial impact is substantial, with healthcare organizations losing approximately 5% of their net patient revenue due to claim denials.

The challenge: Manual RCM creates revenue bottlenecks

From the moment a patient schedules a procedure to the final payment reconciliation, traditional RCM workflows are manual, error-prone, and create bottlenecks that delay payments and impact cash flow. The error rate is particularly concerning in diagnostic procedures, where a study shows 35% contain errors, resulting in more than $20 billion in delayed or lost reimbursements.

The RCM process involves document processing, medical coding, claims generation, and claims submission. Medical records, biopsy reports, and procedure notes must be reviewed, then clinical staff can assign ICD-10 and CPT codes. Generating claims involves creating claim packages, requiring extensive manual assembly that is then sent to payors for validation and payment.

This approach, with its manual touchpoints and process gaps, can lead to patient frustration, delayed payments, increased administrative costs, and potential revenue leakage. Healthcare organizations need a solution that can orchestrate these complex workflows, while maintaining the accuracy and compliance required in healthcare.

Solution: Multi-agent architecture using Amazon Bedrock AgentCore

Our solution uses Amazon Bedrock AgentCore to create a sophisticated multi-agent solution that automates the RCM process. Strands Agents work collaboratively to process healthcare data, make intelligent decisions, and automate RCM workflows. From intelligent document extraction using Amazon Bedrock Data Automation to automated Electronic Interchange Data (EDI) claims submission using AWS B2B Data Interchange. Figure 1 illustrates this architecture and shows our implementation of the automated RCM process.

RCM workflow

Let’s walk through the seven steps as outlined in the architecture diagram:

1. Data ingestion
2. The RCM Orchestrator Agent
3. The Documentation Agent
4. The Medical Coding Agent
5. The Claims Assembly Agent
6. The Claims Submission Agent
7. External systems integration

1. Data ingestion begins when patient data enters the system through:

• Amazon EventBridge: Orchestrates real-time data flow between services, triggering agent workflows based on data events
  • AWS HealthLake: Stores FHIR-formatted clinical data including patient records, diagnoses, and procedures
  • Amazon Simple Storage Service (Amazon S3): Manages document intake for unstructured medical documentation

Amazon EventBridge orchestrates real-time data flow between AWS HealthLake and the RCM Agent Orchestrator. When an encounter is completed in the clinical system, AWS HealthLake updates the encounter record and emits events that Amazon EventBridge captures. It also triggers the AWS Lambda function to call the RCM Orchestrator Agent. This event-driven architecture verifies the RCM process continues when an encounter is ready for billing, enabling near real-time processing without manual intervention.

2. The RCM Orchestrator Agent serves as the central coordinator, managing the workflow between specialized agents and handling exceptions.

Key capabilities:

• Natural language orchestration: Interprets complex multi-step RCM requests in plain English
• Dynamic workflow reasoning: Adapts workflows based on context, exceptions, and real-time conditions
• Intelligent task routing: Routes tasks to appropriate specialized agents based on content analysis
• Response synthesis: Combines outputs from multiple agents into coherent, actionable responses
• Exception handling: Uses reasoning to handle edge cases and unexpected scenarios

The Orchestrator Agent acts as the conductor overseeing all the complex multi-agent workflows. It understands when to involve multiple agents, how to resolve conflicts, and when to escalate to humans.

3. The Documentation Agent serves as the entry point for the individual RCM agent workflows, processing various medical documents and extracting structured data.

Key capabilities:

• Processes biopsy reports, radiology reports, procedure notes, physician summaries, and anesthesia reports
• Uses Amazon Bedrock Data Automation for intelligent document processing
• Achieves extraction accuracy across document types
• Validates extracted data against clinical guidelines

4. The Medical Coding Agent applies clinical expertise to assign accurate medical codes, facilitating compliance with coding standards and payor requirements.

Key capabilities:

• Analyzes clinical documentation for accurate ICD-10 and CPT code assignment
• Cross-validates codes against payor-specific requirements
• Implements AI-powered quality assurance checks
• Provides coding recommendations with confidence scores

5. The Claims Assembly Agent assembles comprehensive claims packages, verifying all required documentation and data elements are included for successful submission.

Key capabilities:

• Validates patient demographics, provider credentials, and facility information
• Confirms insurance verification and procedure code accuracy
• Performs format and compliance checks before submission
• Generates EDI 837P claim structures automatically

6. The Claims Submission Agent handles the complex process of submitting claims to various payors, while managing different submission requirements and formats.

Key capabilities:

• Performs pre-submission readiness checks
• Manages clearinghouse routing and transmission
• Handles duplicate and resubmission checks
• Provides near real-time submission tracking
• Uses AWS B2B Data Interchange to handle the complex EDI transactions required for healthcare claims submission, supporting multiple payer formats and requirements

7. External systems integration provides providers with a way to use any EDI connectivity solution to transfer EDI documents and acknowledgements to and from Amazon S3.

This architecture uses domain-specific knowledge bases that provide critical context for agent decision-making:

• Compliance rules repository: Contains regulatory requirements and compliance guidelines
• Medical coding rules: Stores current ICD-10, CPT, and HCPCS coding standards

The multi-agent RCM solution also incorporates Amazon Bedrock AgentCore capabilities, such as AgentCore Identity, AgentCore Memory, and AgentCore Observability.

Security and compliance

The solution implements a robust security and compliance framework that safeguards sensitive healthcare data through multiple layers of protection. This includes end-to-end PHI encryption, role-based access controls, and automated de-identification processes for analytics.

Implementation recommendations

To implement this multi-agent RCM solution, we recommend:

• Assessment phase: Evaluate current RCM processes and identify automation opportunities
• Pilot implementation: Start with document extraction and medical coding agents
• Gradual expansion: Add claims generation and submission capabilities
• Full deployment: Implement complete multi-agent workflow with orchestration
• Optimization: Continuously improve agent performance based on real-world feedback

Conclusion

The multi-agent RCM solution using Amazon Bedrock AgentCore represents a significant change in how healthcare organizations can approach revenue cycle management. By using specialized AI agents that work collaboratively, healthcare organizations can transform manual, error-prone processes into intelligent, automated workflows. They can improve accuracy, reduce costs, and accelerate revenue realization.

As healthcare organizations continue to face challenges in optimizing operations, while maintaining quality care, AI-powered RCM solutions will become essential tools for better financial outcomes and efficiency.

For more information about implementing multi-agent systems with Amazon Bedrock AgentCore, visit the Amazon Bedrock documentation or contact an AWS Representative to find out how we can help accelerate your business.

Additional resources

• Automating paper-to-electronic healthcare claims processing with AWS
• Guidance for Identifying Diagnosis Codes from Clinical Notes on AWS
• AWS HealthLake documentation: for FHIR data management
• AWS B2B Data Interchange documentation: for EDI transaction handling
• AWS for Healthcare & Life Sciences