Two similar patients undergoing the same procedure might follow different pathways inside a hospital. Some of this variation is expected, but too much variation is associated with increased adverse events. Currently, there are no standard methods to establish when variability for any given clinical process becomes excessive. In this study we use process mining techniques to describe clinical processes and calculate and visualise clinical variability. We selected a sample of patients undergoing elective coronary bypass surgery from the MIMIC dataset, represented their clinical processes in the form of traces, and calculated variability metrics for each process execution and for the complete set of processes. We then analysed the subset of processes with the highest and lowest relative variability and compared their clinical outcomes. We established that processes with the greatest variability were associated with longer length of stay (LOS) with a dose-response relationship: the higher the variability, the longer the LOS. This study provides an effective way to estimate and visualise clinical variability and to understand its impact on patient relevant outcomes.
Keywords: Clinical data analytics; process mining; variability.