| Abstract [eng] |
Nowadays, more and more activities are being explored in the context of big data mining to gain insights and draw data-driven conclusions. Over the last 5 years, process mining has increasingly focused on operations improvement and the search of insights for process optimization [1]. Action description and modeling is helping to describe the steps required to achieve a specific result, replicate it, analyze it, and improve it. Conformance checking allows finding instances of a process, where the actions are performed more efficiently than usual or deviate from the procedure turning it into not as fast or of good quality. In the process improvement phase, key performance factors and insights are derived to help answer the questions that describe the actions. These provide insights that allow you to improve the process, understand how it works, and make data-driven decisions. The aim of this work is to describe the processes taking place in the emergency department and compare them in the context of two hospitals. Also, in a complex and unstructured process, different scenarios are clustered and key performance indicators are described. Two hospital emergency departments data from February 2017 to August 2017 were used for the study. Study identified the main features of the process activity event log. Using a heuristic and fuzzy miner algorithms, the order of events was described using a process model. With the help of conformance checking methods, the most and least common business scenarios were examined. Also, attention is paid to the most effective and longest cases. It is observed that many of laboratory or radiological tests may be redundant to establish a diagnosis of a patient. The longest visits are characterized by an excessive amount of research and uncertainty of the sequence of events. The processes taking place in the emergency departments in each hospital were clustered into 8 clusters. In the first hospital, the cluster with the highest number of cases shows the activities of patients with uncomplicated conditions. Nearly 90% of these patients were recommended for treatment at home. In the second hospital, the cluster with the highest incidence describes the activities of patients with uncomplicated conditions also. These cases were of shorter duration, with an average process time of 2.3 hours. Based on the comparison of the models, the key performance indicators were identified. They described: the percentage of patients returning to the emergency department, average number and duration of laboratory or radiological examinations performed, the number of deaths, the percentage of cases with a duration of more than a day. The average number of patients who left the hospital before the end of treatment and the average time between triage and discharge recommendations were also examined. |
