Related Experiment Videos
Automating the assignment of diagnosis codes to patient encounters using example-based and machine learning
Serguei V S Pakhomov1, James D Buntrock, Christopher G Chute
1Division of Biomedical Informatics, Department of Health Sciences Research, Mayo Clinic, Rochester, MN, USA. pakhomov.serguei@mayo.edu
Summary
An automated coding system accurately classifies over two-thirds of clinical diagnoses, significantly reducing manual coding efforts. This system enhances efficiency in medical practices by automating diagnosis categorization for billing and research.
Area of Science:
- Medical Informatics
- Health Information Management
Background:
- Manual diagnosis classification is resource-intensive and requires specialized personnel.
- Accurate coding is crucial for medical billing, research, and data analysis.
Purpose of the Study:
- To develop and evaluate an automated system for classifying clinical diagnoses.
- To improve the efficiency and accuracy of the medical coding process.
Main Methods:
- Developed an automated coding system utilizing a certainty-based approach.
- Employed a database of 22 million manually coded entries for high-certainty classifications.
- Used a Naïve Bayes classifier for lower-certainty classifications requiring manual review.
- Applied standard information retrieval metrics (precision, recall, f-measure) for evaluation.
Main Results:
- Over 48% of EMR problem list entries were automatically classified with 98.2% accuracy (f-score).
- An additional 34% were classified with 93.1% accuracy.
- The system achieved high precision and recall across different certainty levels.
Conclusions:
- The automated system successfully codes over two-thirds of diagnoses with high accuracy.
- Implementation at Mayo Clinic reduced manual coding staff from 34 to 7 verifiers.
- The system offers a significant improvement in efficiency and resource allocation for medical coding.