Discovering interpretable medical process models: A case study in trauma resuscitation.

Area of Science:

• Health Informatics
• Process Mining
• Medical Process Analysis

Background:

• Understanding actual medical processes ("work-as-done") is crucial for improving patient outcomes.
• Existing process mining methods often produce complex or incomplete models of medical activities.
• Interpretable models are needed to accurately represent critical steps and variations in care.

Purpose of the Study:

• Introduce TAD Miner, a novel Trace Alignment-based Process Discovery method.
• Develop interpretable process models for complex medical procedures, focusing on pediatric trauma resuscitation.
• Compare TAD Miner's performance against state-of-the-art algorithms.

Main Methods:

• Applied TAD Miner to activity logs from 308 pediatric trauma resuscitations.
• Discovered process models for key resuscitation goals (e.g., IV access, intubation).
• Evaluated models quantitatively (complexity, accuracy) and qualitatively (expert review), comparing with Inductive Miner and Split Miner.

Main Results:

• TAD Miner generated process models with lower complexity and higher interpretability.
• Model fitness and precision were comparable to existing methods.
• Identified errors and optimal locations for tentative steps in expert-driven models, leading to revised, improved models.

Conclusions:

• TAD Miner effectively creates accurate and interpretable models of complex medical processes.
• The method enhances the understanding of "work-as-done" in clinical settings.
• Improved process modeling can lead to better patient care and optimized medical interventions.