Comprehend Medical

🧠 What is Amazon Comprehend Medical?

Amazon Comprehend Medical is a HIPAA-eligible NLP service designed to extract clinical information from unstructured medical documents such as doctor's notes, prescriptions, and EHRs (Electronic Health Records).

✅ It identifies PHI (Protected Health Information), medical conditions, anatomy, medications, dosage, and treatment plans from free-form text.

---

🩺 Key Use Cases

Use Case	Description
🧾 EHR processing	Extract structured info from free-text medical records
🧑‍⚕️ Clinical data mining	Identify patterns in patient histories
💊 Medication reconciliation	Detect drug names, dosages, frequency, and route
🔍 PHI de-identification	Automatically detect PII/PHI from clinical text
📑 Clinical trial eligibility	Match patients to trials based on diagnoses and history
📚 Medical document search	Index patient records for search based on conditions or medications

---

🔍 Extracted Entities by Comprehend Medical

Category	Description
MEDICATION	Drug names, dosages, frequency, route, etc.
CONDITION	Diagnoses, symptoms, and diseases
ANATOMY	Body parts and anatomical terms
TEST_TREATMENT_PROCEDURE	Medical procedures or treatments
PROTECTED_HEALTH_INFORMATION (PHI)	Names, IDs, dates, addresses
TIME_EXPRESSION	Durations, frequencies, timeframes
ICD-10-CM, RxNorm	Coding for diseases and drugs (optional SNOMED)

---

🧪 Example: Input and Output

Input:

“John Smith was prescribed 5 mg of Lipitor once daily for high cholesterol.”

Output:

[
  {
    "Type": "MEDICATION",
    "Text": "Lipitor",
    "Attributes": [
      { "Type": "DOSAGE", "Text": "5 mg" },
      { "Type": "FREQUENCY", "Text": "once daily" }
    ]
  },
  {
    "Type": "CONDITION",
    "Text": "high cholesterol"
  },
  {
    "Type": "PHI",
    "Text": "John Smith"
  }
]

---

📊 API Operations

Operation	Description
DetectEntitiesV2	Extract medical concepts from text
InferICD10CM	Map text to ICD-10-CM medical codes
InferRxNorm	Map medications to RxNorm identifiers
DetectPHI	Detect and classify PHI (Protected Health Information)

📎 Supports plain text input of 10–20 KB per call.

---

🌐 Supported Language

• English only (as of 2024).

---

💰 Pricing (2024)

API Operation	Price per unit (1 unit = 100 characters)
DetectEntitiesV2	$0.0015 per unit
DetectPHI	$0.0015 per unit
InferRxNorm	$0.0015 per unit
InferICD10CM	$0.0015 per unit

🧠 Free Tier: Not included in AWS Free Tier (unlike Amazon Comprehend).

---

🔒 HIPAA Eligibility & Security

Compliance Feature	Status
HIPAA Eligible	✅ Yes (BAA required)
Encryption	TLS in transit, KMS at rest
IAM & CloudTrail	✅ Yes
PrivateLink Support	✅ Yes (via VPC endpoint)

---

🧑‍💻 Sample Python Code (Boto3)

Detect Entities in Clinical Text

import boto3

client = boto3.client('comprehendmedical')

response = client.detect_entities_v2(
    Text="Patient was prescribed 10 mg of Lisinopril for hypertension."
)

for entity in response['Entities']:
    print(f"{entity['Type']} → {entity['Text']}")

---

Detect PHI

response = client.detect_phi(
    Text="John Smith was seen at St. Joseph Hospital on 2023-06-12."
)

for phi in response['Entities']:
    print(f"PHI: {phi['Text']} - Type: {phi['Type']}")

---

🧱 Terraform Support

Amazon Comprehend Medical has no direct Terraform resources, but you can:

• Set up IAM roles and policies

• Automate via Lambda, Step Functions, or S3 triggers

Example IAM Policy

resource "aws_iam_policy" "comprehend_medical_access" {
  name = "ComprehendMedicalPolicy"
  policy = jsonencode({
    Version = "2012-10-17",
    Statement = [{
      Effect   = "Allow",
      Action   = [
        "comprehendmedical:DetectEntitiesV2",
        "comprehendmedical:DetectPHI",
        "comprehendmedical:InferRxNorm",
        "comprehendmedical:InferICD10CM"
      ],
      Resource = "*"
    }]
  })
}

---

🔁 Integrations

AWS Service	Usage
S3	Store patient records or extracted entities
Lambda	Run entity extraction on document upload
Step Functions	Build ETL pipelines for medical text analysis
Athena	Query extracted structured data
OpenSearch	Search by medication, conditions, etc.
Amazon Translate	Translate non-English text first (limited use)

---

✅ TL;DR Summary

Feature	Amazon Comprehend Medical
HIPAA eligible	✅ Yes
Entity types	MEDICATION, CONDITION, PHI, ANATOMY, etc.
Language support	English only
Code standard support	✅ RxNorm, ICD-10-CM
Real-time API	✅ Yes
Pricing	~$0.0015 per 100 characters
Free Tier	❌ Not available
Terraform support	⚠️ Only IAM-based setup

---