Classification of user queries according to a hierarchical medical procedure encoding system using an ensemble...

Area of Science:

• Medical Informatics
• Computer Science

Background:

• The Swiss Classification of Surgical Interventions (CHOP) is essential for daily clinical practice, quality assurance, and billing.
• Manual CHOP code selection involves expert knowledge and catalog searches, which can be time-consuming and prone to errors.

Purpose of the Study:

• To investigate the feasibility of automatic CHOP code generation from text queries.
• To develop an accurate and efficient automated system to support manual CHOP classification.

Main Methods:

• Evaluation of different classification strategies, including multi-class non-terminal and per-node classifications.
• Utilizing hierarchical features, stacked autoencoders, and Bayesian aggregation for improved classification performance.
• Implementing threshold adaptation to enhance the reachability of per-node classifiers.

Main Results:

• Per-node binary classification demonstrated superior performance, achieving an F1-micro score between 92.6% and 94%.
• Hierarchical prediction using per-node classifiers resulted in a high exact match rate for single code assignment.
• Threshold adaptation significantly increased F1 measures at code levels 3-6 from 6% to 89%.

Conclusions:

• The hierarchical structure of CHOP encoding is valuable for both classifier training and representation learning.
• Per-node binary classification with hierarchical feature utilization offers a flexible, accurate, and efficient solution for automated CHOP code generation.
• Automated CHOP code generation can streamline healthcare administrative processes and improve data accuracy.