About. The impact of an aging population and the surge in lifestyle diseases on health service demand globally is well documented. While there is a plethora of literature that provides best practices guidelines on the provision of cost-effective care; however, such recommendations are implemented in different hospitals, each with its own unique contextual factors, resulting in continued variances in costs and quality. Instead of re-inventing the wheels by implementing, from scratch, new best practices, we can learn from others by comparing the way in which different hospitals treat their patients and by learning from those hospitals that tend to achieve better results consistently.
This case study measures the variations in the patient flows related to the treatment of patients who presented themselves with chest-pain symptoms at four different hospitals in South Australia.
Objective.This case study attempts to measure and explain the variations in clinical practices across different hospitals.
Key Question. The key questions we would like to address in this case study are:
- To what extent variations exist across hospitals, and
- why do they exist?
Data. The South Australian Health Department maintains a Clinical Reporting Repository (CRR) which stores and collates electronic medical record for each patient from a number of publicly-funded health providers. Each hospital also maintains its own information systems for managing operating theatres and tracking patient transfers between physical wards.
Our study excludes cases whereby (1) the patients involved were transferred-in from another hospital (where they had another admission immediately prior to the ACS related ED presentation), (ii) the patients had a prior ED presentation within the preceding 12 months, or (iii) the cases were not sufficiently documented (e.g., missing pathology tests). The above exclusions left us with an event log of 9997 cases, 263712 events, and 64 different event classes.
Following a number of data pre-processing steps to remove unreliable data, we are left with 3040 cases, 69872 events, and 33 event classes.
Approach. In this case study, we applied cross-organisational process mining approach to identify the similarities and differences in the patient flows of four South Australian (SA) hospitals. In particular, we replayed the replaying of logs from each hospital on the discovered patient flow models of each hospital. The resulting fitness values give us an estimate of patient flows similarity between hospitals.
Given the repetitive nature of such an analysis (i.e. doing the same type of analysis over and over again but with different combination of log and model), we also used scientific workflow technique to streamline the analysis, thus reducing potential errors. We did so by using the Rapid Miner tool that has been enriched with process mining analysis plug-ins.
In addition, we also used fuzzy animation in conjunction with manual inspections, in order to identify key differences in the patient flows across the hospitals.
Results and Impact
We have managed to identify key differences in the patient flows across the four South Australian hospitals as shown in the figure below. Key differences observed include the range of blood tests performed which vary from hospital to hospital (Observation 1 – O1), the point at which decision about admitting patients to hospital was made (Observation 2 – O2), and the various ways in which patients were moved between wards (Observation 3 – O3).
From the performance perspective, we also noticed differences in the time taken for the patient to progress from one milestone to the next. These milestones include presentation of patients at an Emergency Department (EDPres), examination of patients by doctors (DrSeen), request for admission of patients to hospital (AdmReq), the discharge of patients to a hospital ward (ED_Disch_Ward), the discharge of patients to a ward within Emergency Department (ED_Disch_AdmW/ED), and the discharge of patients from a hospital ward to home (iDisch_Home).
As shown in the table below, we notice remarkable differences in the time taken between various milestones across hospitals, e.g. the time taken from AdmReq to ED_Disch_Ward differs substantially between Hospital 3 (H3) and Hospital 4 (H4). Such insights guide us in the refinement of our comparative analysis exercise.
Impact. Through the application of cross-organisational process mining, we have managed to gain an understanding of the performance differences between hospitals and we have managed to identify differences in the patient flows across hospitals. This knowledge points us in the right direction for the next step of our analysis, that is, to understand if different patient flows indeed have an impact on performance metrics.