Client Background
We aim at building a tool for risk adjustment to improve the efficiency of the medical coding process using NLP, ML, and DL capabilities. It extracts clinical insights from unstructured text, improves risk score accuracy and operational efficiencies and thereby provides the best care to patients.
Client Need
- Extract data from unstructured and nonstandard document formats. This includes chart pages which are scanned charts, lab reports, patient chart history, etc.
- Build a medical record review service to assist coders with PHI identification and patient and provider information extraction using OCR and NLP techniques
Our Solution
- Developed an OCR module for training a custom tesseract model to extract text from patient’s medical charts
- Built NLP Engine constituting Entity extraction module and ICD extraction modules
- Entity extraction module – A pipeline to extract patient information like patient name, age, gender etc. and provider information provider name, title, facility, etc. Leveraging regular expressions and advanced deep learning approaches
- ICD extraction module – A pipeline to extract ICD codes
- Trained using a state-of-the-art multi-label classification model to extract ICD details and corresponding annotations
- Implemented regex logic to extract AS-IS ICD codes apart from annotation based ICDs
Tools & Technologies
Key Benefits
- 90% increase in accuracy rate with 5000+ documents
- 60% decrease in time required for synthesis of clinical guidelines
- Increased output rate using NLP (15-16 charts per day) compared to manual coding (10-11 charts per day)
- Reduced overall administration costs