Amazon Comprehend Medical

Pass-through endpoints for Amazon Comprehend Medical - detect entities, PHI, and medical ontology links in clinical text, in native AWS format (no translation).

Feature	Supported	Notes
Cost Tracking	✅	For the sync text operations listed below
Logging	✅	works across all integrations
End-user Tracking	❌	Tell us if you need this
Streaming	❌	Not offered by the Comprehend Medical API

Just replace https://comprehendmedical.{aws_region_name}.amazonaws.com with LITELLM_PROXY_BASE_URL/comprehendmedical 🚀

LiteLLM signs the forwarded request with SigV4 using the proxy's AWS credentials, so clients only need a LiteLLM virtual key.

Quick Start

1. Set AWS credentials and region in the proxy environment

export AWS_ACCESS_KEY_ID=""
export AWS_SECRET_ACCESS_KEY=""
export AWS_REGION_NAME="us-east-1"

2. Start the proxy

litellm

# RUNNING on http://0.0.0.0:4000

3. Call a Comprehend Medical operation through the proxy

curl -X POST 'http://0.0.0.0:4000/comprehendmedical/DetectEntitiesV2' \
-H 'Authorization: Bearer sk-1234' \
-H 'Content-Type: application/json' \
-d '{"Text": "Patient is taking 40mg of atorvastatin daily for hyperlipidemia."}'

The operation name in the URL is one of the supported sync text operations: DetectEntitiesV2, DetectPHI, InferICD10CM, InferRxNorm, or InferSNOMEDCT. Other operations (e.g. the async batch job APIs) return a 400 listing the supported set. See all Comprehend Medical operations

Usage with the AWS SDK (boto3)

Point the SDK's endpoint_url at LITELLM_PROXY_BASE_URL/comprehendmedical. The proxy reads the operation from the SDK's X-Amz-Target header, per the AWS JSON 1.1 protocol.

import boto3

client = boto3.client(
    "comprehendmedical",
    region_name="us-east-1",
    endpoint_url="http://0.0.0.0:4000/comprehendmedical",
    aws_access_key_id="placeholder",
    aws_secret_access_key="placeholder",
)
client.meta.events.register(
    "before-send.comprehendmedical.*",
    lambda request, **kwargs: request.headers.__setitem__("x-litellm-api-key", "sk-1234"),
)

response = client.detect_phi(Text="John Smith was admitted on 2026-08-01.")
print(response["Entities"])

The SDK still signs the request locally with the placeholder credentials, but LiteLLM discards that signature, authenticates the call with the LiteLLM virtual key from the x-litellm-api-key header, and re-signs the request with the proxy's AWS credentials.

Cost Tracking

Spend is computed from the request's Text length: Comprehend Medical bills per started 100-character unit with a 1-unit minimum. LiteLLM applies the first-tier on-demand price per unit for each operation:

Operation	Price per unit
DetectEntitiesV2	$0.01
DetectPHI	$0.0014
InferICD10CM	$0.0005
InferRxNorm	$0.00025
InferSNOMEDCT	$0.0075

Requests are logged with model comprehendmedical/{Operation} and provider comprehendmedical, and spend shows up in the usual places (SpendLogs, key/team budgets, logging integrations). Operations outside this table are rejected with a 400, so no Comprehend Medical call can slip past spend tracking.