De facto diagnosis specialties: Recognition and discovery

Area of Science:

• Health Informatics
• Medical Data Analysis
• Machine Learning in Healthcare

Background:

• Accurate documentation of medical specialties is crucial but often lacking or inaccurate in healthcare institutions.
• Diagnosis histories contain valuable information about the skills and judgments of healthcare providers.

Purpose of the Study:

• To leverage diagnosis histories for recognizing existing medical specialties and discovering unlisted ones.
• To identify de facto diagnosis specialties within the Health Care Provider Taxonomy Code Set (HPTCS) and uncover new specialties.

Main Methods:

• Utilized similarity and supervised learning models to recognize listed de facto diagnosis specialties.
• Introduced a discovery-evaluation framework employing semi-supervised and unsupervised learning models to identify unlisted specialties.
• Collected two datasets of diagnosis histories from a large academic medical center for analysis.

Main Results:

• Identified 12 core de facto diagnosis specialties within the HPTCS that are highly recognizable from diagnosis histories.
• A semi-supervised model discovered a breast cancer specialty, and an unsupervised model identified an obesity specialty.
• Evaluation confirmed high recognizability for the discovered specialties, comparable to listed ones.

Conclusions:

• Diagnosis histories are a viable source for recognizing and discovering medical specialties.
• The proposed framework effectively identifies both listed and potentially unlisted de facto diagnosis specialties.
• This work has implications for improving healthcare provider data accuracy and resource allocation.