Process Mining at Academisch Medisch Centrum


The Academisch Medisch Centrum (AMC) hospital has 1002 beds and is one of the largest academic hospitals in the Netherlands. The hospital is located in Amsterdam. For AMC we analyzed several care processes of the surgery department for which on yearly basis many patients are received. The selected processes involved medically non-complex patients. For each of the selected care processes, the diagnosis of the patients is the following:

Goal of the analysis

Within the AMC already a pilot-study was performed in which process mining was applied for a group of gynaecological oncology patients (for more details see this publication). This pilot study made clear that process mining could provide usefull insights on care processes that are executed for medically complex patients. Moreover, the technique allowed for gettings results within a short period of time whereas with the previous method for describing processes, in which people are interviewed, more time was needed. Moreover, the previous method was also more costly.

Given the positive results obtained in the pilot project, it was decided to apply process mining in an extensive fashion during a process improvement project within the hospital. Based on the outcomes of the project it could be explored in which future projects and in which settings, process mining could be used.

Consequently, as a result of discussions with logistics professionals of the AMC, it was decided to involve the surgery department within the project. The surgery department often collaborates with multiple medical departments in order to diagnose and treat patients. Furthermore, many patients are received which only require low complex care. For these patients it is essential that the provided care is organized efficiently and performed within a short period of time. Currently, the surgery department is lacking insights into the processes that are performed for patients requiring low complex care.

Event log

For the analysis, data was extracted from the financial systems of the AMC. In these systems, it is saved for each patient which services were delivered by the hospital. More specifically, for a group of patients with a specific diagnose it is possible to identify all services that were delivered during both diagnosis and treatment.

The obtained dataset contained all activities performed for the selected patient groups from January 2008 till december 2011. In total there were:

Process mining results

The aim of the analysis was to provide a thorough process mining analysis for a medical department thereby giving concrete indications for process improvements. Furthermore, for the surgery department it was clear that there was a lack of insights into the care processes for medically non-complex patients. To this end, the following questions were posed to us by the medical professionals:

For these questions the following results were obtained:

Process performance

For the patient groups, we first visualized the process using the dotted chart. From this chart it was obvious that for all patient groups the total time for treatment is quite high (on average more than 4 months). Furthermore, there are huge variations for these treatment times.

For the hernia inguinalis patients, for which the surgical intervention took place on the daycare center, 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.

The surgery department had indicated to us that they had the feeling that for some examinations quite huge waiting times existed. For example, this has as effect that it often takes a long time before a patient is seen again on the outpatient clinic. Taking this into account, as a next step, we wanted to see for each patient group where are the bottlenecks in the process. In order to obtain reliable performance results for a process model, it is important that the obtained model is a good reflection of the behavior captured in the event log.

For the hernia inguinalis patients the discovered process is shown in case the surgical intervention is done at the daycare center. In general, the process is as follows. First, a first visit to the outpatient clinic takes place. During the visit, the medical specialists can decide that a series of diagnostic tests at different medical departments are required (e.g. radiology, general clinical laboratory) are required. Afterwards, the patient visits again the outpatient clinic or the surgical intervention takes place.

Therefore, for each patient group, we used a semi-automatic approach for discovering a process model showing the actions that have taken place in both the diagnosis and the treatment phase. The approach consisted of the following steps. First, a process mining algorithm was applied in order to discover a process model which shows the medical departments that were visited and their order. Second, the process model was adapted by hand and it was checked in the process mining tool ProM how well it reflected the behavior in the log. This second step was repeated till the model was a good reflection of the behavior captured in the log. Furthermore, performance information was projected on the model by coloring the places. This gave the following important insights:

Furthermore, for the 'lipoma and sebaceous cyst' patients it also became clear that for them the treatment trajectory is more freqently finished (44%) with a consultation by telephone compared to the other patients.


For the patient groups the medical specialists were also interested in whether the process could be performed with lower costs. One of the results was that via a dotted chart we saw for the group of patients which were suffering from general stomach complaints that there were 5 patients for which in particular many actions were performed.

Dotted chart showing that for 5 patients many actions have been registered during the entire treatment process.

More specifically, for each of these 5 patients there were more than 25 clinical days registered and more than 230 actions performed. When looking to the associated costs, it became clear that these patients consumed around 13% of the total costs that are made for the entire group of 2258 patients.

Next steps

The results presented above (and many other insights) were presented to several professionals from the surgery department. They were impressed by the results and indicated that they would like to have similar analyses for other patient groups. Furthermore, the results were also presented to other professionals within the hospital. They also indicated that the results were interesting and that they would like to use process mining in their own analyses. As a next step, it is investigated whether process mining techniqes can be used in some standard management reports that are offered to medical specialists. This illustrates the rising uptake of process mining techniques within the AMC.