Quick Start

Python
import fitz  # PyMuPDF
import re
from datetime import datetime, timedelta
import random

def anonymize_pdf(input_path, output_path):
    doc = fitz.open(input_path)
    
    # Common PHI patterns
    patterns = {
        'names': r'\b[A-Z][a-z]+ [A-Z][a-z]+\b',
        'mrn': r'MRN:?\s*\d{6,10}',
        'ssn': r'\b\d{3}-\d{2}-\d{4}\b',
        'phone': r'\b\d{3}[-.]?\d{3}[-.]?\d{4}\b',
        'dob': r'\b\d{1,2}/\d{1,2}/\d{4}\b'
    }
    
    for page in doc:
        for pattern_name, pattern in patterns.items():
            matches = page.search_for(pattern)
            for match in matches:
                page.add_redact_annot(match, fill=(0, 0, 0))
        page.apply_redactions()
    
    doc.save(output_path)
    doc.close()

Workflow

Progress:

• Inventory Documents - Catalog all files by type (PDF, DOCX, images)
• Backup Originals - Create secure copies before processing
• Identify PHI Categories - Patient names, MRNs, DOBs, addresses, phone numbers
• Set Anonymization Rules - Replace vs redact, fake data generation
• Process by File Type - Use appropriate tools for each format
• Quality Check - Manual review of sample files
• Validate Medical Content - Ensure clinical information remains intact

1. Create folder structure:

   portfolio/
   ├── originals/     (secure backup)
   ├── anonymized/    (portfolio ready)
   └── scripts/       (automation tools)
   

2. Install required tools:

   Bash
   pip install PyMuPDF python-docx Pillow pytesseract

3. Run anonymization pipeline:

   • PDFs: Text replacement + OCR for scanned docs
   • DOCX: Find/replace with python-docx
   • Images: OCR + overlay/blur sensitive areas

4. Generate consistent fake data:

   Python
   fake_names = ["Patient A", "Patient B", "John D.", "Jane S."]
   fake_dates = generate_shifted_dates(original_date, -30, +30)  # ±30 days
   fake_mrns = ["MRN: 12345", "MRN: 67890"]

Examples

Example 1: PDF Report Input: "Patient: Sarah Johnson, DOB: 03/15/1985, MRN: 987654321" Output: "Patient: Patient A, DOB: 04/12/1985, MRN: 12345"

Example 2: DOCX Letter Input: "Dear Dr. Smith, Regarding your patient Robert Wilson..." Output: "Dear Dr. [Physician], Regarding your patient Patient B..."

Example 3: Lab Image Input: Lab result with patient name in header Output: Same lab values, header shows "Patient C" with blurred original text

Best Practices

• Consistent replacement - Same patient should get same fake name across all documents
• Preserve medical timeline - Shift dates by same offset to maintain relative timing
• Keep clinical relevance - Age ranges, approximate dates matter for medical context
• Use blackout for handwriting - OCR often misses handwritten PHI
• Maintain document formatting - Headers, layouts should look professional
• Test with colleagues - Have others review for missed identifiers

Common Pitfalls

• Inconsistent anonymization - Same patient having different fake names across documents
• Missing image-embedded text - Scanned documents need OCR processing
• Forgetting metadata - Document properties often contain original author/patient info
• Over-anonymization - Removing medically relevant information (age, general location)
• Signature blocks - Your own signature/credentials should remain intact
• Date logic errors - Creating impossible timelines (discharge before admission)
• Manual review skip - Automated tools miss context-dependent identifiers