Skip to content

unsigned_notes_plan.md

Latest commit

 

History

History
177 lines (149 loc) · 5.63 KB

unsigned_notes_plan.md

File metadata and controls

177 lines (149 loc) · 5.63 KB

Azalea Health API - Unsigned Notes Processing Plan

Overview

Create an application to fetch patient charts for encounters with >1 unsigned notes from Comprehensive Medical Clinic's Azalea Health EHR, process them with AWS Bedrock AI, and update the EHR with recommendations for provider review.

Current Setup

Implementation Plan

Phase 1: Authentication ✅

  • ✅ OAuth 2.0 client credentials implementation
  • ✅ Environment switching (sandbox/production)
  • ✅ Token management with system/. scope
  • ✅ Comprehensive error handling

Phase 2: Find Encounters with Unsigned Notes ✅

  1. ✅ Fetch recent encounters (status=in-progress or finished)
  2. ✅ Check for Encounter Signature Extension
  3. ✅ Get DocumentReferences for each encounter
  4. ✅ Filter encounters with >1 unsigned/preliminary documents
  5. ✅ Pagination support for large datasets
  6. ✅ Detailed logging of discovery process

Phase 3: Collect Patient Chart Data ✅

For each qualifying encounter, collect:

  • ✅ Patient demographics
  • ✅ Encounter details
  • ✅ Clinical notes (all types)
  • ✅ Diagnoses (Conditions)
  • ✅ Allergies
  • ✅ Medications
  • ✅ Vital signs
  • ✅ Problem list

Phase 4: AI Processing & Storage ✅

  1. ✅ AWS Bedrock integration for AI processing
  2. ✅ Clinical context building for accurate AI recommendations
  3. ✅ SQL Server database for tracking processed notes
  4. ✅ Structured prompt engineering for medical documentation
  5. ✅ Error handling and fallback mechanisms

Phase 5: EHR Updates ✅

  1. ✅ Update DocumentReference with AI recommendations
  2. ✅ Comparison format for provider review
  3. ✅ Preserve original content with AI suggestions
  4. ✅ Clear labeling of AI-generated content

Technical Architecture

Components Implemented:

  1. DatabaseManager Class

    • SQL Server connection management
    • Table initialization
    • CRUD operations for processed notes
    • Error handling and logging

  2. BedrockAIProcessor Class

    • AWS Bedrock client initialization
    • Clinical context extraction
    • Prompt engineering for medical notes
    • Response parsing and formatting

  3. Document Update System

    • FHIR DocumentReference updates
    • Base64 encoding for content
    • Comparison format generation
    • PUT operation implementation

  4. Comprehensive Logging

    • File and console output
    • Multiple log levels (INFO, ERROR, DEBUG)
    • Detailed operation tracking
    • Audit trail for compliance

API Endpoints Used

Core Resources

  • /Encounter - Find encounters
  • /DocumentReference - Get clinical notes and update them
  • /Patient - Demographics
  • /Condition - Diagnoses & problems
  • /AllergyIntolerance - Allergies
  • /MedicationRequest - Medications
  • /Observation - Vital signs

Key Filters

  • Encounters: ?status=in-progress,finished&date=ge[date]&_sort=-date
  • Documents: ?encounter=[id]&category=clinical-note
  • Conditions: ?patient=[id]&category=encounter-diagnosis,problem-list-item
  • Observations: ?patient=[id]&category=vital-signs

Environment Variables Required

# SQL Server (optional)
SQL_CONNECTION_STRING=<your-connection-string>

# AWS Bedrock
AWS_ACCESS_KEY_ID=<your-key>
AWS_SECRET_ACCESS_KEY=<your-secret>
AWS_DEFAULT_REGION=us-east-1

# Optional
BEDROCK_MODEL_ID=anthropic.claude-v2

Database Schema

CREATE TABLE processed_notes (
    id INT IDENTITY(1,1) PRIMARY KEY,
    encounter_id VARCHAR(255) NOT NULL,
    patient_id VARCHAR(255) NOT NULL,
    document_id VARCHAR(255),
    original_text NVARCHAR(MAX),
    ai_enhanced_text NVARCHAR(MAX),
    recommendations NVARCHAR(MAX),
    processed_at DATETIME NOT NULL,
    status VARCHAR(50) DEFAULT 'pending',
    created_at DATETIME DEFAULT GETDATE(),
    updated_at DATETIME DEFAULT GETDATE()
)

Output Format in EHR

Documents are updated with:

[Original Note Content]

========== AI ENHANCED VERSION (For Review) ==========
[AI Enhanced Version]

========== AI RECOMMENDATIONS ==========
[Specific Recommendations]

========== END AI SUGGESTIONS ==========
Note: Please review the AI suggestions above and edit as needed before signing.

Security Considerations Implemented

  1. ✅ Environment-based configuration
  2. ✅ Secure credential storage via environment variables
  3. ✅ HTTPS-only communication
  4. ✅ Comprehensive error handling without exposing sensitive data
  5. ✅ Audit logging for all operations
  6. ✅ Sandbox testing before production

Current Status

  • ✅ Core functionality complete
  • ✅ AWS Bedrock integration implemented
  • ✅ SQL Server tracking implemented
  • ✅ Document update capability added
  • ✅ Comprehensive logging throughout
  • ✅ Sandbox testing environment configured
  • ✅ Production-ready architecture

Usage Summary

  1. Set environment to sandbox in config.json
  2. Configure environment variables (.env file)
  3. Run python main.py
  4. Monitor logs for progress
  5. Review updated documents in EHR
  6. Check SQL Server for tracking data
  7. Switch to production when ready

Future Enhancements

  1. ⏳ Automated scheduling (cron/Windows Task Scheduler)
  2. ⏳ Web dashboard for monitoring progress
  3. ⏳ Bulk approval interface for providers
  4. ⏳ Analytics on AI recommendation acceptance rates
  5. ⏳ Support for different note types (SOAP, H&P, etc.)
  6. ⏳ Multi-organization support
  7. ⏳ Real-time processing via webhooks