Process Mining at Isala Clinics

Introduction

The Isala Clinics hospital has 994 beds and is the largest non-academic hospital in the Netherlands. In total, the hospital has 7 different locations of which the biggest are Sophia and Weezenlanden which both can be found in Zwolle.

For Isala we analyzed the 5 most performed care processes of the urology department. These processes involved medically complex patients, which require an individualized treatment (bladder cancer, kidney stones), as well medically non-complex patients for which a more standardized treatment is sufficient (phimosis, hydrocele, undescended testis).

Goal of the analysis

Within the hospital substantial efforts are taken to improve the care processes. To this end, detailed descriptions of these processes are required. Here, they felt that much time was lost on interviewing people and studying patient files in order to get insights on the care processes that were running. Furthermore, large amounts of data are available within Isala IT-systems concerning these processes. As a result, they where highly interested in applying process mining and seeing the benefits of it.

In order to start applying process mining, we teamed up with several specialists of Isala and talked about the availability of data and the approach to be followed. Here it was decided to first apply process mining for the urology department. Afterwards, the results were used for attracting the interest of other medical disciplines and to further enhance the uptake of process mining within the hospital.

Event log

For the case study, data was extracted from the hospital system. Data was taken from a financial module (ChipSoft) in order to obtain for each patient the services that had been delivered by the hospital. Furthermore, additional data were taken from the bed information system in order to obtain information about the start and the end of the days of hospitalization of a patient.

The obtained dataset contained all activities performed for the 5 selected patient groups from January 2009 till december 2011. In total there were:

Process mining results

Although the application of process mining was merely explorative, the following questions were posed:

For the five patient groups we obtained the following results:

Regular behavior and process performance

As one of the first analyses for each patient group we visualized the traces using the dotted chart. Here we immediately saw that for the medically non-complex patients (phimosis, hydrocele, undescended testis) the average treatment time is quite high (around 4 months). Furthermore, there was a huge variation for these treatment times, e.g. there were patients for which treatment was completed within 1 month whereas there were also patients for which treatment was not completed within 1 year. For the medically complex-patients (bladder cancer, kidney stones) the average treatment time was far less (around 2 months) and also with less variation.

For the phimosis patients the dotted chart gives a visual overview of the actions that have taken place for each patient. On the vertical axis the different cases (i.e. patients) are shown and events are colored according to their action names. As can be seen, the process is shown using relative time, i.e. all cases start at time zero. The chart shows that there is a large variation in the total throughput time of cases.

As a next step we wanted to see where in the process much time was lost for each patient group. Moreover, as a possible reason for the waiting times, we wanted to investigate for each patient group whether many actions were required for diagnosis and treatment and the variance in this. Therefore, for each patient group, a process model was discovered showing the actions that had taken place in the diagnosis phase and the treatment phase. Moreover, for each discovered model, performed information was projected in order to identify bottlenecks. This gave the following important insights:

For the phimosis patients the discovered process is shown in case the surgical intervention is done at the operating theatre. In general, the process is as follows. First, a first visit to the outpatient clinic of urology takes place. During the visit, the medical specialists can decide that a series of diagnostic tests (e.g. urodynamic examination, echography) are required. Afterwards, the patient visits again the outpatient clinic or a preassesment takes place before finally the surgery takes place.

In total it became clear that for the medically non-complex patients there were high waiting times for surgery and the pre-operative assessment although the number of actions done before surgery is low. For the medically complex patients this is exactly the opposite. Finally, for the medically non-complex the biggest improvement can be imagined if the average waiting time for the surgery is significantly reduced. That is, if for the 3 medically non-complex patient groups the waiting time would be halved, this would mean that on yearly basis 73 patients have to wait 3.3 years (1224 days) less.

Not performed obligatory medical steps

For medically non-urgent patients, a well known rule is that before a surgery a pre-operative assessment needs to be done. In this pre-operative assessment, an anesthesiologist determines whether a patient is fit and healthy enough for anaesthetic. During our process mining analysis it became clear that for each of the five patient groups there were multiple patients for which no pre-operative assessment was registered before surgery. In particular, we found that for the phimosis patients the surgical intervention either takes place in an operating theater or at the outpatient clinic itself. In case the surgical intervention was done at the outpatient clinic itself no pre-operative assessment was done at all. This was due to the fact that only local anesthesia was given.

For the phimosis patients the discovered process is shown in case the surgical intervention is done at the operating theatre. Moreover, in total 76 patients for which there was no medical urgency have gone through this process. For each action, in the rectangle it is shown how often the action is performed. For the actions belonging to the pre-operative assessment it can be seen that the pre-operative assessment is done for 72 out of the 76 patients.

However, for the patients that had their surgical intervention at the operating theater, a pre-operative assessment is necessary. For the five patient groups it appeared that for 24 of the 446 patients (6%), no pre-operative assessment was registered.

For the missing pre-operative assessments it is important to note that they have not been registered within the care trajectory. This does not automatically mean that they have not taken place in reality (but if so, no payment is received for this work). As a next step, for the patients involved, the identifiers were given to the medical specialists in order that further investigation is possible.

Avoidance of steps

In the discovered process models for each of the patient groups it was observed that before and after surgery there were some emergency consultations either at the emergency department or at urology itself. On average, after surgery, 9% of the patients needed an emergency visit to the outpatient clinic of urology. For one patient group this appeared to be 13%. The urologists indicated that for the latter group indeed some more complications can be expected. The process mining results were concurrent with the expectations. In case the emergency treatments could have been avoided this would have saved at least 31 medical steps.

Next steps

This explorative process mining was done on systems ment for invoicing. Timestamps are of less value in such systems and may therefore be erroneous. We are convinced that when applying process mining (in healthcare) the registration quality will rise to the occasion. Nonetheless the results proved to be valuable for both the administrative and medical professional.

That is, the results presented above (and many other insights) were presented to several urologists and the manager of urology. They all were impressed by the obtained results. Furthermore, the manager was wondering why some kinds of information could not be made available by the hospital information system (e.g. the average time between the surgery and the previous examination).

As a next step, within the Isala a business case will be drawn up in order to allow for hiring staff that uses process mining for performing analysis for other medical disciplines. This illustrates the practical value of process mining in care organizations.