Use case: ThiaperProcess in healthcare 

In a nutshell: Optimization of the management of medical care by highlighting vulnerabilities and cost generators in hospital management systems.  Allows patients and hospital staff to focus on what is truly important: treatment and recovery.

  • evaluation of clinical pathways / patient treatment journeys (transfers, referrals) through the hospital 
  • evaluation of the allocation flow of medical personnel  to various wards (to support resource planning and uncover staff issue
  • evaluation of the compliance of healthcare process flows to normative guidelines.

Process Mining for healthcare: why?

Performance optimization of healthcare systems has been shown to achieve  e.g. 6%–10% increase in speed of throughput, 20% increase in bed utilization, 5% increase in operating room output, 5%–8% annual savings in operating budget. 

Short term impacts include minimizing areas of duplication and waste (waiting times, unnecessary long chains of referrals). We expect that the small optimization steps taken towards removing redundancies and unnecessary processes  in the system would lead, in the long term, to positive changes from the patient’s viewpoint. For instance, provision of a holistic rather than myopic view of a patient’s condition would allow early detection of cases when multiple referrals are issued from differently specialized departments – the case of complicated diseases  that are rare and generally hard to spot because they affect multiple systems in the body and generally have confusing symptom picture.

The possibility to  maximize consultation time would be a way to move from I-It to an I-You doctor-patient relationship. Impaired communication largely predicts malpraxis accusations (Journal of the American Medical Association, 1997). Improvements in the quality of medical teams / doctor-patient relationships would  lead to an increase of the overall quality of life for the medical personnel (less stress & frustration) while reducing legal costs.


  • to discover and visualize clinical pathways (patient journeys) for patients in specific pathology clusters.
  • to quantify most/least frequent pathways and associated time delays 
  • to quantify the difference between the journeys of patients undergoing only lab tests, versus patients undergoing lab tests and the recommended surgical procedures.


ThiaperProcess was developed and tested for the healthcare domain with data collected from 18 private healthcare providers over 3 years.

  • obtained process maps (patient journeys) for the data filtered for clusters of pathologies of interest
  • quantified  performance indicators  (time delays, frequencies) associated to the identified patient journeys. The identified pathways and quantifiers were shown to in-house domain experts from the healthcare provider, to check if and what can be done for optimization purposes.
  • isolated the clinical pathways typically followed by patients undergoing only lab tests with the healthcare provider as opposed to patients undergoing lab tests followed by surgery (for the same pathology cluster) and were able to indicate the pathways followed by one group but not the other 


For confidentiality reasons we are not showing ThiaperProcess results obtained  for the data from the client, but for an open medical dataset. We use the real life event log of the conformance checking challenge 2019 (CCC19) to showcase how ThiaperProcess can discover medical procedure pathways and quantify performance metrics on top.

The log focuses on a medical training process: medical students learn how to install Central Venous Catheter (CVC) with ultrasound. The CVC procedure refers to installing a catheter (tube) in a central vein, aiding on delivering fluid or medications to the patient, among other uses.

The process map discovered is a birdview of all sequences of steps typically followed by the medical students while learning.

A close- up of the pathways found shows the average length of the pathway (in terms of average number of activities performed), end-to-end time delay and the frequency. The values of these indicators can further be used for clustering of the pathways found.