Detecting Deviations In Clinical Pathways Using Process Mining

The clinical pathway is a fine-grained and standardized clinical process defined for a group of patients with similar diagnosis and prognosis.In daily practice,it is not realistic for all clinical processes to follow the predefined clinical pathway.The mismatch between the actual clinical process and the ideal clinical pathway is defined as clinical practice deviation.By studying clinical practice deviation,researchers can identify the execution problem in clinical pathways and then refine pathways or improve practice.Care quality is improved in this procedure.Currently,there are two approaches to study clinical practice deviations.One approach is prospective study of clinical practice deviations.This approach requires care-givers manually record deviations.By using this approach,detailed reason of deviation can be recorded for further analysis.However,this manual work increases the workload of care-givers.Therefore,the completion rate is not satisfying and the data recorded are not always reliable.Another approach is retrospective study.Computer programs will automatically analyze deviations from history data recorded in information systems.This approach would save great amount of human effort and it is more objective.Therefore,it is considered as a promising direction of clinical practice deviation research.However,this approach relies on a reliable deviation detect results.Deviation detecting is critical to retrospective clinical practice deviation studies.However,it is a challenge to detect the deviations,because clinical pathways are normally complex and highly flexible.Rule-based approaches represent critical care activities as rules and then test clinical data against these rules.This approach works best for expected deviations.But it is not possible to detect unexpected deviations by using rule-based approaches.Process model based approaches,such as naive flow charts,Markov Chain models,directed diagram,and Petri Net,are all based on the optimal alignments between event logs and directed diagrams.However,there is a major challenge of applying this approach to complex and flexible clinical pathway deviation detecting,since current optimal alignments theories are either unreliable or not feasible.This dissertation dedicated to the studied of the optimal alignments algorithm.Based on the proposed algorithm,this dissertation studied two types of typical clinical pathways,which are complex care processes and flexible care processes.Correlations between clinical pratice deviations and their possible reasons are also studied in this dissertation.The research work is detailed as follows.1.Systematically reviewed the develop and application of clinical pathway,specified the background,purpose and scope fo deviation studies,analyzed the importance of deviation detecting to continuous quality improvement of healthcare.Reviewed different types of deviation detecting approaches,analyzed the advantage and necessity of studying process mining based deviation detecting approach and their limitations.2.The computerized representation model of clinical pathway is the basic requirement of deviation detecting studies.This dissertation analyzed the representation requirements of typical clinical pathways.The advantage and limitation of several modeling languages were compared.The Business Process Modeling and Notation language is selected as the representation language for clinical pathway after the comparison.Clinical pathway concepts were then mapped to the Business Process Modeling and Notation elements.3.The optimal alignments between an event trace and directed graph is the key problem of deviation detecting.There is yet no mature study on this topic.This dissertation transformed the optimal alignments problem into the shortest path searching problem.The search space and cost function of the optimization had been studied.This dissertation contributed a robust optimal alignments algorithm with good performance and accuracy.It contributes to the process mining domain and serves as the theoretical foundation of this study.4.This dissertation studied a deviation detecting algorithm based on typical control flow for complex clinical pathways.We firstly studied the state space generation of BPMN model,the transformation from state space to directed diagrams,and the translation of clinical pathway models to directed diagrams.Then with the directed diagrams,deviation detecting can be applied in typical control flow clical pathway models.To evaluate the algorithm,experimental event traces with different deviation ratios were tested.The result showed the proposed algorithm could sovle deviations detecting problem effectively.5.This dissertation studied a deviation detecting algorithm based on task-time matrix model for flexible clinical pathways.To apply optimal alignments algorithm,the BPMN task-time matrix model was firstly transformed into a special directed graph with Gateway Pairs.In this way,multiple tasks in one time unit are kepted in the Gateway Pair constructor,which avoided low performance in statespace generating approach.With the Gateway Pair directed graph,an online optimal alignments searching algorithm was developed.In practical case,the method developed could solve the deviations detecting problem effectively.6.This dissertation developed a clinical pathway deviations detecting and analysis system.Two case studies were carried out:weaning pathway in Maarstricht University Medical Center in the Netherlands and unstable angina treatment pathway in Chinese PLA General Hospital.In the weaning protocol case,the system discovers an added deviation,which is the raising PEEP value.This finding helped care-givers to redesign their clinical pathway.In the unstable angina case,the system discovered the correlation between prescribing nitrate drug and angina event,adding Metformin and diabetes.These findings served as strong evidence for specifying pathways for patients with co-morbidities.The results indicate that the approach proposed in this study can effectively detect comprehensive deviations and find possible reasons for the deviations,serving for continuous care process quality improvement.