Using data mining techniques to explore physicians' therapeutic decisions when clinical guidelines do not provide...

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Data Mining in Healthcare

Background:

• Clinical guidelines are essential for evidence-based practice but often have knowledge gaps.
• Unaddressed clinical situations limit comprehensive patient care.
• A novel approach is proposed to identify and fill these gaps using data mining.

Purpose of the Study:

• To develop and demonstrate a data mining method for identifying knowledge gaps in clinical guidelines.
• To explore physicians' therapeutic decisions to generate evidence-based recommendations.
• To fill identified knowledge gaps in type 2 diabetes management guidelines.

Main Methods:

• Analysis of French national type 2 diabetes guidelines to identify gaps.
• Extraction of patient records for uncovered clinical conditions from a hospital database.
• Application of C5.0 decision-tree learning to physician prescriptions.
• Development of decision-trees for treatment type, class, INN, and dose.
• Comparison of learned rules with updated guideline recommendations.

Main Results:

• 27 rules were extracted from 463 patient records.
• Rules were identified for treatment type (11) and pharmaco-therapeutic class (13).
• Limited rules for drug name and dose due to insufficient patient data.
• Extracted rules demonstrated similarity to newer guideline versions.

Conclusions:

• The method effectively complements guideline recommendations with practice-based knowledge.
• It serves as a valuable tool for guideline development and evaluation.
• Physician practice, as reflected in prescriptions, can sometimes precede guideline recommendations.