Patient Careflow Discovery

Abstract

In this study, we explored how different mining techniques can be used to gain insight into the healthcare domain. More specifically, we developed a methodology that takes a set of activity sequences from a Hospital Information System to analyze patient careflow.

We developed a data-based methodology able that provides insight into patient careflow, based on a standardized data structure from the Dutch DBC Information System and other external resources. This approach provides a set of techniques able to analyze any type of care profile, for any specialism, within any hospital and combinations of either. After an initial data collection, an event log is prepared containing high-level activities describing the logistic carepath of patients. Secondly, different types of care profiles are identified by clustering using a Partitioning Around Medoids algorithm based on the Tanimoto distance between paths. The third step is to apply classification in order to identify the main characteristics of each type of profile. As a fourth and final step, each cluster is analyzed using the Trace Alignment plugin in ProM, which allows the identification of both a cluster’s main process pattern and specific deviations from this process for individual carepaths. A variety of insightful visualization techniques allows medical specialist to interpret the results of this methodology without specific knowledge on the Data and Process Mining techniques. The insights gained from this methodology support the improvement of patient careflow in three different ways: by treating patients according to the cheapest path with the highest quality of care, by improving standardization of carepaths and by developing a robust, optimal operating schedule using predictive modeling based on the patient types defined by this analysis.