Medical Workflow Modeling Using Alignment-Guided State-Splitting HMM

Abstract

Process mining techniques have been used to discover and analyze workflows in various fields, ranging from business management to healthcare. Much of this research, however, has overlooked the potential of hidden Markov models (HMMs) for workflow discovery. We present a novel alignment-guided state-splitting HMM inference algorithm (AGSS) for discovering workflow models based on observed traces of process executions. We compared the AGSS to existing methods using four real-world medical workflow datasets and a more detailed case study on one of them. Our numerical results show that AGSS not only generates more accurate workflow models, but also better represents the underlying process. In addition, with trace alignment to guide state splitting, AGSS is significantly more efficient (by a factor of O(n)) than previous HMM inference algorithms. Our case study results show that our approach produces a more readable and accurate workflow model that existing algorithms. Comparing the discovered model to the hand-made expert model of the same process, we found three discrepancies. These three discrepancies were reconsidered by medical experts and used for enhancing the expert model.

Learning alternative ways of performing a task

“…This type of representation is potentially limited to represent concurrency, duplicate actions, hierarchical activities, or model OR-splits/joins in a mined model. For instance, Yang et al (2017) proposed a method to split duplicate activities in the final model in order to gain expressiveness. This approximation could be introduced in future versions of our algorithm.…”

“…The generated model represents the main flow of activities in the process at hand. Process mining has been mainly applied to business management (van der Aalst et al, 2007) but also to other fields such as healthcare (Mans et al, 2008;Yang et al, 2017). Although the previous techniques are able to explain the processes in an organisation, they cannot be directly applied for the purpose of supervision: the models may contain non-essential activities for carrying out the process and they are not able to capture the different ways (if any) of performing a process since only one model is built.…”

“…Dependency graphs are hence a more concise non-univocal simplification of an activity sequence. Similar formalisms based on graphs can be found in the workflow learning literature (Yang et al, 2017;Günther & van der Aalst, 2007;Agrawal et al, 1998) (van der Aalst & Van Hee, 2004) with applications in bioinformatics (Yan et al, 2005), social-network analysis (Carrington et al, 2005) or web navigation (Papadimitriou et al, 2010). However, we take advantage of the properties derived from the graphs to propose an inductive method that is able to generate more than one model, each one identifying a different form of performing the task showed by the examples.…”

Process mining applications in the healthcare domain: A comprehensive review

“…S. Yang, Li, et al (2017) proposed an HMM-based algorithm to investigate the trauma resuscitation process mined from video recordings. In the same year, M. Zhou et al (2017) and S. Yang, Zhou, et al (2017) tested their trace alignment-based algorithm on a trauma resuscitation dataset for the extraction of the process model. Mertens et al (2018) proposed an algorithm called DeciClareMiner that combines process and decision mining to extract a process model from past executions.…”

Process-Oriented Iterative Multiple Alignment for Medical Process Mining

“…Another study derived insights from a consensus sequence alignment of the laparoscopic cholecystectomy workflow [5]. Alignment information has also proven useful in guiding workflow discovery [15].…”