Redesigning HEDIS reporting using AI Agents (in AWS)

Reference AWS architecture using AI agentic architecture

HEDIS (Healthcare Effectiveness Data and Information Set) is a standardized set of performance measures used by health plans (Payers) and providers to evaluate the quality of care and services delivered to patients. Maintained by the National Committee for Quality Assurance (NCQA), HEDIS includes over 90 measures across domains like preventive care, chronic condition management, behavioral health, and patient experience. It plays a critical role in value-based care, payer contracts, accreditation, and public reporting.

However, HEDIS reporting today is often inefficient and labor-intensive. Data is typically siloed across claims, EHRs, and patient surveys, requiring manual chart reviews, complex mapping to standard codes, and extensive data validation. The hybrid method — combining administrative and clinical data — adds to the complexity. Delays, inaccuracies, and audit risks are common, making real-time quality improvement difficult. This inefficiency hinders proactive care gap closure and limits providers’ ability to act swiftly on performance metrics that impact reimbursement and patient outcomes.

To build HEDIS (Healthcare Effectiveness Data and Information Set) measures using LLMs (Large Language Models) and Agentic workflows in EMR (Electronic Medical Records) and RCM (Revenue Cycle Management) systems, we can design a modular architecture in AWS that integrates multi-source data (claims, EHR, surveys, and hybrid methods), applies AI agents for measurement computation, and ensures explainability and compliance.

🧠 Conceptual Overview

1. LLMs Role

Extract: Unstructured data (clinical notes, survey text)

Normalize: Coding terms to HEDIS-aligned formats (e.g., ICD-10, CPT)

Infer: Missing clinical events (e.g., identifying missed screenings)

Explain: Generate justifications for care gaps or measure inclusion/exclusion

2. Agentic Workflow Role

Agentic workflows orchestrate multiple specialized agents:

Data Ingestion Agent: Ingests from claims, EHR, surveys

Normalization Agent: Maps to clinical ontologies (SNOMED, LOINC)

Measure Calculation Agent: Applies HEDIS logic (e.g., denominator, numerator)

Gap Closure Agent: Recommends actions or flags alerts to providers

Audit Agent: Explains rationale, ensures traceability for compliance

🚀 Data Ingestion & Normalization

• Amazon HealthLake / Athena / Glue: Ingests and normalizes EHR + claims data
• Amazon S3: Stores raw survey responses and documents
• Amazon Comprehend Medical: NLP on unstructured text

🔄 Processing Layer (Agents & LLMs)

• Amazon Bedrock (for LLMs like Claude, Titan, or fine-tuned models)
• AWS Step Functions: Agentic workflow orchestration
• Lambda / SageMaker: Hosts lightweight or custom agents
• Amazon Neptune / OpenSearch: For relationships (e.g., social determinants, provider mapping)

📊 HEDIS Engine

• Custom HEDIS Engine in SageMaker or Lambda:
• Apply logic for ~90+ HEDIS measures
• Calculates denominator, numerator, exclusions
• Flags care gaps and quality improvement opportunities

🧾 Reporting & Monitoring

• QuickSight / Looker Studio: HEDIS dashboards
• CloudWatch + EventBridge: Workflow observability and triggers
• Audit logs in S3 / DynamoDB: For compliance and CMS audits

🧬 Data Flow

Workflow for the proposed re-architecture

[Claims + EHR + Surveys + Hybrid Sample]

↓

[HealthLake / Glue / Comprehend Medical] → structured FHIR resources

↓

[Agentic Workflow (Step Functions)] → orchestrates data agents

↓

[LLM-powered Measure Engine (Bedrock)] → computes HEDIS measures

↓

[Care Gaps Identified] → RCM, EMR, Quality Teams

↓

[Dashboards / Notifications / Audits]

✅ Benefits

An LLM and agentic workflow-powered HEDIS architecture delivers significant benefits. It is automated, scalable, and auditable, reducing manual chart reviews and enabling consistent, standards-based reporting. Real-time quality monitoring allows providers to identify and close care gaps proactively, improving outcomes and performance scores. LLMs add value by extracting insights from unstructured clinical data, inferring missing information, and generating explainable patient engagement summaries. The use of modular AI agents ensures flexibility — easily adapting to annual updates in HEDIS specifications without disrupting the system. This modern approach transforms quality reporting from a retrospective compliance task into a dynamic, intelligence-driven quality improvement engine.