Health systems are dynamic and complex, continually evolving through interactions among stakeholders with competing interests. This complexity challenges the accurate evaluation of interventions within healthcare systems, especially in relation to health economic modeling. The principles of health systems engineering, such as process modeling and simulation, are increasingly being applied to overcome these challenges by providing structured approaches to analyze and optimize health systems and processes. Individual-level simulation models may, in theory, fully capture this complexity. However, their development may be challenging and time consuming, in particular when many different patient trajectories need to be defined. With the increasing availability of large and detailed real-world clinical datasets reflecting actual patient trajectories, process mining (PM) techniques can provide a data-driven approach to discover and quantify existing care processes, providing a robust means to study process behavior. The discovered processes reflect current care delivery and support the implementation of realistic care pathways in the structure of health economic simulation models. This is the starting point for evaluating the impact of new health technologies based on their position within this care pathway. In this paper, we investigate how PM can be used to analyze clinical data/event logs to support (semi-) automated simulation model development. We model a Generalized Stochastic Petri Net (GSPN) containing only immediate transitions informed by process discovery applied to event logs. This paper presents a proof-of-concept, and the event logs are generated by a wellknown state transition micro-simulation of a hypothetical disease in health economics - known as the Sick-Sicker model. This application is essentially a reconstruction of the mentioned microsimulation, and the results show that the introduced method accurately simulates the observed behavior in the original event logs. For example, the fitness scores of the simulated logs using the GSPN model are measured to be 1. 0, indicating a perfect fit. Our findings highlight the potential of PM to enhance the development of precise and efficient health economic models, ultimately contributing to improved healthcare decision-making and system optimization.
Faeghi et al. (Tue,) studied this question.