AWS Comprehend Medical PHI engine (Tier 3)

[sarvanithin]

Overview

Add AWS Comprehend Medical as a Tier 3 PHI engine. It handles complex clinical narrative text (discharge summaries, clinical notes) where regex and Presidio miss entities embedded in natural language.

API

AWS Comprehend Medical DetectPHI:

import boto3
client = boto3.client("comprehendmedical")
result = client.detect_phi(Text="Patient John Smith...")
# Returns entities with Type, BeginOffset, EndOffset, Score

What to build

• medguard/guardrails/phi.py — AWSComprehendPHIEngine class
  • Implements PHIEngineProtocol
  • Maps AWS entity types to medguard PHIMatch format
  • Requires pip install medguard[aws] (boto3 already in optional deps)
• Add engine: "aws" option to PHIConfig

Entity type mapping

AWS type	medguard entity_type
NAME	NAME
SSN	SSN
DATE	DOB
PHONE	PHONE
EMAIL	EMAIL
ID	MRN
ADDRESS	ADDRESS

Files to modify

• medguard/guardrails/phi.py — add AWSComprehendPHIEngine
• medguard/config.py — add aws to PHIConfig.engine Literal
• tests/test_phi.py — add tests with mocked boto3

Acceptance criteria

• AWSComprehendPHIEngine.analyze(text) returns correct PHIMatch list
• Graceful ImportError with helpful message when boto3 not installed
• boto3 mocked in tests (no real AWS calls)
• All existing PHI tests still pass