Process mining in the analysis of medical consultations

Abstract

Process mining is an automated technique which can be used to gain insights into processes (Van der Aalst, 2016). Object-centric process mining is a new form of process mining, with more room for the inclusion of different object types, allowing for more perspective on a process (Li et al., 2018). This is done through the use of object-centric event logs. An object-centric event log is a particular view of the available event data. The event data is gathered using the digital footprint these processes leave behind. Traditionally this is done by extracting the event data from information systems. The information systems used in hospitals only record interactions between the doctor and the system, not the conversation between the doctor and the patient. This thesis aimed to investigate the use of object-centric process mining by extracting the necessary data from medical conversations using large language models. For this purpose, the TRACED-OCEL (‘TRAnsforming ConvErsation Data into OCEL’) method was created and evaluated in a case study using data from geriatric consultations. The results show the TRACED-OCEL method was able to successfully facilitate the extraction of an object-centric event log but quality and accuracy issues do exist in the event log, and evaluating the event log is time and labor intensive. Although the potential of the TRACED-OCEL method was shown, the reasoning capacity and efficiency of current large language models are not high enough to state it is a viable alternative for process analysis in the context of the case study. Several challenges and pointers are provided for future research.