The analysis of clinical pathways from event logs provides new insights about care processes. In this paper, we propose a new methodology to automatically perform simulation analysis of patients’ clinical pathways based on a national hospital database. Process mining is used to build highly representative causal nets, which are then converted to state charts in order to be executed. A joint multi-agent discrete-event simulation approach is used to implement models. A practical case study on patients having cardiovascular diseases and eligible to receive an implantable defibrillator is provided. A design of experiments has been proposed to study the impact of medical decisions, such as implanting or not a defibrillator, on the relapse rate, the death rate and the cost. This approach has proven to be an innovative way to extract knowledge from an existing hospital database through simulation, allowing the design and test of new scenarios.
In health-care, activities such as consultation, imaging examination or surgery are parts of a clinical pathway (CP). The design of a CP is a major challenge to better understand the impact of treatments on the whole journey of the patient. Health authorities intend to propose standardization of care processes for various operational purposes: organization of care activities, assignment of human resources, reducing practice variability, minimizing delays in treatments or decreasing costs while maintaining quality. The large amount of data collected from a CP by an Hospital Information Systems is valuable because it may reveal important patterns of the CP, allowing the creation of formal models that can be simulated. This paper provides a methodology in order to analyze and simulate such CPs using existing databases on the national level.
In a previous study (Prodel et al. 2015), we proposed a new approach to discover CPs from a national hospital database using Process Mining (PM) (Van der Aalst 2011) and Integer Linear Programming. The objective was to create the most representative process model of the event log under a constraint on the model’s size. The metric for assessing representativeness was based on the frequency of events in the log and of direct transitions between events. Decision variables were the choices of events to keep in the nodes of the model and the arcs to keep between these nodes. In the literature, CP analysis of recorded data was mainly done using either Data Mining or Process Mining techniques. Such approaches receive an increasing attention in the field of Medical Informatics. The next step of this research consists in proposing a model that can be executed using simulation and in testing what-if scenarios. Scenarios can be related to various decisions, such as a change in the medical treatment of certain patients, the launch of new medical devices supposed to be more effective to cure certain diseases, or a change in hospital activities’ financing.
In this paper, we propose a new methodology to automatically build a simulation model of patients’ CP from a national hospital database using Process Mining techniques. Such methodology may be applied using any database as data input, and may be applied for any cohort of patients, which constitutes the main scientific contribution of this paper. Simulation of Clinical Pathways brings new knowledge and allows the evaluation of scenarios through design of experiments. Target users of our approach are numerous:
- hospital managers: predict the results of investments in new care services or management strategies;
- health-care practitioners: test the relevancy of new treatments at certain steps of the care pathway of the patients under study;
- pharmaceutical firms: extrapolate the impact of a new drug or a new medical device on the patient care pathway by taking into account the cost of hospital stays.
Global methodology of the study