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

Keyi Li, Ivan Marsic, Aleksandra Sarcevic, Sen Yang, Travis M. Sullivan, Peyton E. Tempel, Zachary P. Milestone, Karen J. O'Connell, Randall S. Burd

Abstract: Highlights•Introduced TAD Miner, a data-driven process model discovery approach for complex medical processes based on trace alignment.•TAD models feature a backbone process, concurrent activities, and uncommon-but-critical activities. Compared to the models discovered by the state-of-the-art methods, TAD models are more interpretable and achieve comparable accuracy.•Discovered process models using 308 real trauma resuscitation cases for five resuscitation goals. The models enhanced the understanding of complex medical processes.•Identified the errors and the best positions for the tentative steps in knowledge-driven process models.