Finding the bridge between process mining
and outcomes in value-based health care:
An exploratory case study
Ursula AblingerStudent number: 11804718
University of Amsterdam Faculty of Science
Thesis Master Information Studies: Business Information Systems Final version: June 27th, 2018
Supervisors: Dr. Inge van de Weerd & Prof. dr. ir. H. A. Reijers Examiner: Dhr. ir. A. M. Loek Stolwijk
Abstract.
With a growing demand for high-quality care in our aging society, health care costs have been continuously rising in the last years (OECD, 2018). However, instead of focusing on the outcomes for the patients reached, reimbursement is made for single treatments performed (Kaplan & Porter, 2011). In value-based health care, value of health care is considered as the health outcomes achieved per dollar spent (Porter, 2008). In this research, the use of process mining to discover these outcomes is examined. A case study of two patient groups, head and neck cancer and kidney failure, was conducted at a Dutch academic hospital. Patient data was analyzed using process mining technology and, supported through interviews with process participants, relevant measures defined. The results showed the applicability of process mining to measure indicators in health care regarding time, number, recurrence and sequence of activities. Further, the insights of this exploratory research resulted in the development of several hypotheses. These present the relations between the indicators measured through data from health information systems, and patient outcomes that are usually measured through surveys. Further research is necessary to test these hypotheses and discover potential new outcome indicators.
Keywords. Process mining, value-based health care, outcomes in health care, health care processes.
Table of contents
1. Introduction ... 3
2. Theoretical Background ... 4
2.1. Process mining in health care ... 4
2.2. Value-based health care ... 4
3. Research methodology ... 5
3.1. Case Study context... 5
3.2. Case study procedures ... 5
4. Data analysis ... 8
4.1. Head and neck cancer ... 8
4.2. Kidney failure ... 10
5. Results ... 11
5.1. Use of process mining in value-based health care ... 11
5.2. Hypotheses proposed ... 14
6. Discussion ... 15
6.1. Applicability of the method ... 15
6.2. Practical and theoretical implications ... 16
6.3. Limitations and future research ... 17
7. Conclusion ... 17
Acknowledgements... 18
1. Introduction
Health care institutions today are facing a growing demand for care due to a higher life expectancy and at the same time increasing cost pressure and demand for productivity (Mans, van der Aalst, & Vanwersch, 2015). According to the Centraal Bureau voor de Statistiek (CBS, 2018), the national health care expenditure in the Netherlands is rising. It has increased from 60.125 billion Euros in 2006 to 83.786 billion Euros in 2016, which is the case in numerous other countries too (OECD, 2018). Besides, third parties (e.g. insurance companies) reimburse for treatments performed instead of outcomes for the patients achieved as well as patients have little responsibility for costs of health care services (Kaplan & Porter, 2011).
The fundamental issue is the non-consideration of the value created by the care process (Porter & Teisberg, 2006). Rather than analyzing the full patient care cycle and comparing it with the outcomes accomplished, accumulated costs at the specialty or department level are analyzed (Porter & Guth, 2012). Hence, the full cycle of care is poorly coordinated and analyzed (Porter & Teisberg, 2006). According to Porter’s (2008) developed framework of value-based health care (VBHC), value of health care should be considered as the health outcomes achieved per dollar spent. Despite the current shift from analyzing the volume of services to the value created, progress is still moderate (Porter, Larsson, & Lee, 2016). As crucial issue here the challenge of measuring outcomes, aside from clear indicators like survival, needs to be taken into account (Porter et al., 2016). The Harvard institute for strategy & competitiveness calls for IT solutions that display the full care cycle and allow extraction of outcome measures, process measures, and activity-based cost measures (Harvard Business School, 2018a). This is where process mining comes into play.
Process mining is a technique to identify a realistic picture of a process on the basis of event data in information systems (van der Aalst, 2011). By making use of the huge amount of data that is available in Hospital Information Systems (HIS) (Mans, van der Aalst, Vanwersch, & Moleman, 2013), the whole care pathway for a large number of patients can be discovered and visualized through process mining techniques (Mans et al., 2015). However, characteristic performance indicators in process mining are measurements like cycle time or throughput time and are not clearly value-oriented. In order to exploit the full potential of process mining in VBHC, the usage of process performance indicators for the evaluation of outcomes in VBHC is necessary. Thus, the goal of this paper is to determine in an exploratory research how process mining can support the determination of outcomes in VBHC. Following hereon, the research question is defined as:
• Which process mining measures can be used to predict outcomes in VBHC? Many (case) studies of applying process mining in a health care environment have been conducted in the last few years (Rojas, Munoz-Gama, Sepúlveda, & Capurro, 2016; Erdoğan & Tarhan, 2016). However, linking process mining to the framework of value-based health care has not gained noticeable attention in research, particularly not with an emphasis on outcomes. My paper contributes to research focusing on the measurement of outcomes in healthcare by exploring the use of process mining techniques to show the full care cycle and extract health care outcomes. The gained knowledge then can also be utilized by health care organizations and other stakeholders in care processes.
Furthermore, this study supports the so far researched application of process mining techniques in a health care setting as being a further case study in this field.
This paper is structured as follows: First, the theoretical background is provided after which the methodology and analysis of the case study are presented. This leads to the discussion and the final conclusions of this study.
2. Theoretical Background
In this research within the field of business process management, the focus is on process mining, linked to the framework of value-based health care according to Porter (Porter & Teisberg, 2006; Porter, 2008; Porter, 2009; Porter, 2010; Kaplan & Porter, 2011). 2.1. Process mining in health care
Process mining can be regarded as a discipline between data mining on the one side and process modeling and analysis on the other (van der Aalst et al., 2012). It offers a process management technique that describes the analysis of processes based on ‘event logs’. In such a log, each event refers to an activity and is related to a specific case, whereby the order of such events is crucial (Mans et al., 2015; Stahl & van der Aalst, 2011). Since in process mining factual execution data is used, an objective view on how processes are really executed is feasible, which differentiates it from traditional process discovery methods like conducting interviews (Mans et al., 2015; van der Aalst, 2011).
Considering the complex and dynamic nature of health care processes, the application of process mining techniques in this sector has been growing in recent years (Mans et al., 2015; Rebuge & Ferreira, 2011). However, the great variety of cases and large discrepancies across individual patients and institutions present a major challenge during an analysis, as standardization is nearly not feasible in this sector (Yoo et al., 2016).
2.2. Value-based health care
There are many different stakeholders (from patient to government) in the health care sector, that frequently have conflicting goals regarding e.g. access to care services, quality, costs, safety and patient satisfaction. Porter developed the framework of VBHC based on the desired, overarching goal of achieving high value for patients (Porter, 2010). The core of this framework is the definition of value as health outcomes for a patient compared to costs per patient (Porter & Teisberg, 2006; Porter, 2008; Porter, 2009).
The International Consortium for Health Outcomes Measurement (ICHOM), consisting of medical experts as well as patient representatives, has started to define minimum outcome sets for several diseases (Kelley, 2015; Porter et al. 2016). These outcomes are grouped into categories like patient-reported health & wellbeing or treatment-specific outcomes (ICHOM, 2018). However, the standardization of outcomes is still in the early stages and measuring outcomes constitutes a major challenge for health care organizations that implement the VBHC framework (Porter et al., 2016).
3. Research methodology 3.1. Case Study context
The research was executed in the form of a case study, conducted at the former VU University Medical Center (VUmc), a university hospital affiliated with the VU University Amsterdam (VUmc, 2018). Since June 7th, 2018 it is known as the Amsterdam UMC due to its administrative merger with the AMC hospital (AMC, 2018).
In the timeframe from March to June, I conducted an exploratory research that included the performance of process mining activities at the VUmc. The VUmc was currently implementing methods of the value-based health care framework according to Porter in selected care pathways. Two processes within this project, the care pathways for the diseases kidney failure and head & neck cancer, were targeted to be discovered with process mining techniques. Both units are resource-intense medical conditions and were in the phase of implementing VBHC. Therefore, these two units were considered to be appropriate for this research and constituted a representative case according to Bryman (2012).
3.2. Case study procedures
The starting point of this study comprised a literature research on the topics of process mining, specifically process mining in the health care sector and the framework of value-based health care. Due to the exploratory nature of this research, the data collection was conducted in a case study according to Yin (2014) at the hospital VUmc. Within this case, detailed examination of the kidney diseases and head & neck cancer care pathways, regarded as two units of analysis within this embedded case study (Yin, 2014), was performed. Both care pathways share the characteristic of taking a prolonged period with several treatments and consultation. For this reason, patient data over a timeframe from 2015 to 2017 was analyzed. Referring to the case study data collection techniques by Lethbridge (2005), mainly an independent analysis of data from information systems within the organization was conducted, which represents a third degree analysis. Additionally, first degree methods, for example interviews, where I was in direct contact with people, were performed to support and also check the third degree analysis.
This research project was carried out in alignment with the Business Process Management (BPM) lifecycle (see figure 1, Dumas, La Rosa, Mendling, & Reijers, 2013), with the exception of the process redesign and process implementation phases, since redesign was not being aimed at in this project (see Figure 1&2). In every stage, a different (sub)goal was pursued as presented in figure 2. Besides the process mining activities, I had 20 appointments, including unstructured and semi-structured interviews according to Yin (2014), that are summarized in table 1 per lifecycle stage. The full list of interviews conducted can be found in appendix 1.
During the first phase – Process identification – I had an unstructured interview with the domain expert for VBHC at the VUmc in order to identify and select processes fitting the purpose of this research. This led to the choice of applying process mining techniques on the care pathways for kidney disease patients in the pre-dialysis phase as well as patients with certain types of head & neck cancer (please refer to sections 4.1 and 4.2 for the detailed selection of patients).
Figure 1. Research focus within the BPM lifecycle (Dumas, La Rosa, Mendling, & Reijers, 2013) In the process discovery stage, I analyzed with means of the tool Disco the available electronic health records in the information systems, which comprise the consultation appointments, treatments, operations and other activities the patient goes through after being diagnosed with one of the case-relevant diseases. Fluxicon’s Disco is based on the fuzzy miner algorithm, which can deal with unstructured processes (Günther & van der Aalst, 2007) and is one of the mainly used algorithms in studies of health care process mining (Rojas et al., 2016). While this algorithm is also implemented in other process mining tools like for example ProM, I chose Disco for this case study due to its powerful and easy accessible filtering capabilities for explorative drill-down (Günther & Rozinat, 2012).
Figure 2. Research goals in every BPM lifecycle stage
The crucial step in the discovery stage was to identify and gather the relevant data. As suggested by Jans (2016), first, the cornerstones of the process were determined with the domain expert represented by the lead of the VBHC project. After that, together with technical experts the underlying key tables of the cornerstones were identified (Jans, 2016). In this case study, the anonymized patient number was used as unique process instance since it establishes the relations between the tables to be analyzed. The key tables used in this case study are: ‘operatie’ (surgeries), ‘opname’ (admissions),
Identification
• Goal: Identify relevant processes (care pathways)
Discovery
• Goal: Discover relevant processes with process mining
Analysis
• Goal: Analyze discovered processes
Monitoring & Controlling
• Goal: Determine utilization of process mining to measure outcomes in VBHC Scope of research
‘afspraak’ (consultations) and ‘verrichting’ (services). The patient selection was made based on DBC codes (Diagnose Behandeling Combinatie) that are used in the Dutch health care system to determine a diagnosis and treatment combination for settlement (Zorgwijzer, 2018). Patients with the relevant DBC codes in 2016 and 2017 were selected, their patient IDs pseudonymized, and the related data from the key tables retrieved.
Before I could cleanse and prepare the gathered data for further analysis in the process mining tool, it was essential to get familiar with the data provided. This took a considerable amount of time in the process discovery period. Further, I prepared an event log, where I excluded irrelevant events (e.g. administrative codes) and clustered certain activities. To validate this manual interference, I checked the steps executed with the participants of the process discovery period (see table 1). Clustering is needed in order to provide a comprehensible and readable process model (Rebuge & Ferreira, 2012). Other cleansing steps during the event log preparation were joining the data from the tables into one file and aligning timestamp formats. All the steps mentioned were executed in Microsoft Access and Excel.
After obtaining the mined process, the as-is process was analyzed (Process analysis). The as-is process resulted in the typical shape of a ‘spaghetti-like’ process model as often encountered in literature (Ferreira, 2012; Rojas et al., 2016), illustrated in figure 3. The solution was to decrease the accuracy of the paths with the fuzzy miner, so that only the main paths were still shown.
Figure 3. Spaghetti-model of the kidney failure care pathway
As next step, I used the process measures gathered through interviews conducted in the discovery period and applied them in the tool Disco. After a first analysis, further semi-structured interviews were conducted and knowledge from process responsibles and -participants (e.g. physicians, nurses) was acquired. Based on these insights, the utilization of process mining to measure outcomes in VBHC (Process monitoring and controlling) was analyzed.
After gathering the results of the process mining activities, conformance interviews were carried out with the responsible physicians in order to check the analysis.
Summed up, the steps of this research approach are as follows: 1. Determining patient groups
2. Retrieving event log
3. Cleansing and preparing event log 4. Creating process model of care pathways 5. Applying process measures
6. Conducting conformance check
Table 1. Interviews per BPM lifecycle stage.
4. Data analysis
For this study I mined and analyzed the data of patients for the two care pathways, head and neck cancer and kidney failure, for the timeframe from 2015 - 2017. The procedures followed for preparation of the data are described in the methodology section and, as announced, the final step of the approach is to assess the relevant process measures. A complete list of process measures used in this study, can be found in appendix 2. Further, the underlying process models are provided in appendix 3.
4.1. Head and neck cancer
During this research I analyzed the data of patients with a diagnosis of certain malignant tumors in the oral cavity, outlined in table 2. As a first step, the data of patients that were in the first year of their treatment was separated from the data that was generated during the follow-up period (more than one year after the diagnosis – table 3). Patients are supposed to have no more treatments in this follow-up period. In this dataset, 905 patients are in the first year of treatment, 377 are also in the follow-period. The smaller number of follow-up patients results from the limitation, that only patients who had at least one
Timeframe
BPM Lifecycle
stage
Goal Participant(s) Appointments Total
duration
07/03/2018
Process identification
Determining care pathways relevant for the analysis
Projectlead VBHC, Data analyst 1, Data
analyst 2 1 meeting ~ 60 min
13/03/2018 – 18/04/2018
Process discovery
Determining data needed, extraction of data, explanation of data, data selection & clustering of activities, subgrouping of patients, determining first process measures Projectlead VBHC, Data analyst 1, Data analyst 2 8 meetings ~ 470 min 19/04/2018 – 17/05/2018 Process analysis Validation of data selection, clustering & subgrouping, analysis of results, determining further process measures Projectlead VBHC, Data analyst 1, Data analyst 2 6 meetings ~ 300 min 17/05/2018 – 06/06/2018 Process monitoring & controlling Validation of results, determining further process measures Nurse 1, Nurse 2, Social worker 1, Patient 1, Physician 1, Physician 2 5 meetings ~ 197 min
year of treatment are in this group. There are also patients without any treatment, that were excluded from the analysis.
Head and neck cancer – 1st year of treatment
Tumor Cavum Oris Hypofarnyx Orofarnyx Speekselklieren Nasofarnyx Larynx
T re a tm en t
OP, Chemo & Radio
OP, Chemo & Radio
OP, Chemo & Radio
OP, Chemo & Radio
OP, Chemo & Radio
OP, Chemo & Radio
Only OP Only OP Only OP Only OP Only OP Only OP
OP & Radio OP & Radio OP & Radio OP & Radio OP & Radio OP & Radio OP & Chemo OP & Chemo OP & Chemo OP & Chemo OP & Chemo OP & Chemo
Only Radio Only Radio Only Radio Only Radio Only Radio Only Radio Only Chemo Only Chemo Only Chemo Only Chemo Only Chemo Only Chemo
Radio & Chemo
Radio & Chemo Radio & Chemo Radio & Chemo Radio & Chemo Radio & Chemo Table 2. Grouping of patients per tumor localization and treatment in the first year of treatment
Head and neck cancer – Follow-up (>1year)
Cavum Oris Hypofarnyx Orofarnyx Speekselklieren Nasofarnyx Larynx
Table 3. Grouping of patients per tumor localization in the follow-up period
In order to reflect the different treatments, the process models for every treatment method and tumor were generated as shown in table 2. I used the filtering function in Disco, illustrated in figure 4, that enables easy filtering of cases that include or exclude the treatments: head and neck surgery (OP), chemotherapy (Chemo) and radiotherapy (Radio).
Figure 4. Filtering in Disco
The next step was to discover relevant process measures during interviews and meetings. These process measures where then applied in the tool Disco.
The first segment of measurements in the head and neck cancer group are clinical admissions (‘klinische opname’). One admission might be usual, at least it is often unavoidable if there is an operation, but there can also be more than one. Furthermore, the number of admissions per patient might be an indicator for the treatment impact and health status of the patient. Many patients with a Nasofarnyx or Orofarnyx tumor had more than one clinical admission. They also frequently had chemotherapy, which leads to a higher amount of admissions according to interviewees.
For activities such as the length of a clinical admission, referred to as length of stay in literature (Brenner et al, 2016; Khalifa & Khalid, 2015; Kirkpatrick et al., 2015), duration is not only relevant from a cost perspective. Also for patients the necessary number of days to stay at a hospital ward is important according to ICHOM (2018). Therefore, measuring the length of stay might be a viable measure for evaluating patient’s burden of care.
Another crucial event for every patient are operations. In the head and neck cancer group, an operation can be both a diagnostic procedure as well as a treatment if the cancerous growth is removed. However, an operation can also be necessary due to complications or other (unplanned) circumstances. A distinction between a diagnostic and non-diagnostic operation could not be made with the data available. Hence I analyzed not only the number of operations. But also the amount of patients with two operations and more than two operations was retrieve since two can be still usual, including one diagnostic and one non-diagnostic intervention.
The number of visits to the emergency department is critical in health care and is considered as potentially avoidable (Wu & Hall, 2018). The amount of those visits has also been determined during this analysis. However, separating the group of patients that arrive at the emergency department and are admitted to a clinical stay at the hospital from the group that needs no further care after the emergency department visit is an even more accurate division. This was argued by interviewees since the severity of a visit to an emergency department might be higher if the patient is admitted to the inpatient clinic than when sent home on the same day.
The Dutch work group for head and neck cancer (Nederlandse Werkgroep Hoofd-Halstumoren, 2010) has agreed that for 80 % of patients with head and neck cancer the time between the first consultation in the head and neck oncology and start of the treatment should be less than 30 days in order to provide a curative treatment. This can be checked with process mining by filtering to the diagnosis activity and the treatments (radiotherapy, chemotherapy and operations as shown in table 2) and inspecting the time in between.
In the follow-up group, no treatment-specific grouping is conducted since at a certain time, patients should not need more cancer-specific treatments. Yet they still need to have checks and appointments. By investigating the events in this follow-up period for treatment and interventions, patients that need longer treatments can be identified. 4.2. Kidney failure
The second dataset consisted of patients who visit the ‘nierfalenpoli’. These patients suffer chronic renal insufficiency and visit the nephrology outpatient clinic regularly. They are in the pre-dialysis phase, which ends when the dialysis starts or a kidney transplantation is done. I prepared the dataset for these 450 patients by selecting patients with the DBC code for renal insufficiency and included the patients’ data until the first dialysis or transplantation occurs, which constitute the boundaries of this case (Yin, 2014). In contrast to head and neck cancer patients, this group does not receive a disease-specific treatment, therefore no further grouping is necessary. Due to the different structure of the care pathways, the clustered activities differ in some cases from the activities for head and neck cancer. Likewise, the measures used need to be individually determined and evaluated for this group.
I applied the same measures regarding admissions, emergency department visits and acute admissions to the data of pre-dialysis patients. Also the occurrence of operations was measured, however from another viewpoint than in the head and neck cancer group. Generally, patients in this stage of kidney failure are not planned to have admissions or operations, since they do not undergo a specific treatment. However, they might need medical help due to complications or symptoms of another medical condition. As example, I observed the most admissions outside nephrology in the cardiology
department. This was confirmed by the responsible physician since many patients with kidney insufficiency also have cardiological problems.
All in all, around half of the pre-dialysis patients had events like admissions and emergency department visits that might indicate a higher burden of care, which was regarded as plausible by the physician.
The general activities of the chronic renal insufficiency care pathway are executed in the ‘nierfalenpoli’ outpatient clinic. Patients usually visit the outpatient clinic periodically since their medical condition needs to be monitored even if there is no treatment executed. They are also informed about the possibilities of future treatments (dialysis, transplantation) in this stage. Therefore, the time between follow-up consultations or other checks can be reviewed.
Interview partners mentioned also that it is highly important for patients, to minimize their days at the outpatient clinic. This means that their different appointments with various actors (physician, dialysis nurse, nutritionist, social worker) should be scheduled for the same day if possible. However, two interviewees also argued that it depends on the specific situation and state of the patient. Three interviewees also mentioned, that patients prefer to always see the same physician or nurse at their outpatient clinic visit. This could not be measured due to missing data in the event log.
5. Results
In this chapter, first the insights gained during this research, for the use of process mining in value-based health care, are presented. Further on, the hypotheses developed based on this exploratory study are outlined.
5.1. Use of process mining in value-based health care
In current literature, there is an emphasis on measuring outcomes for patients (Porter et al., 2016). This is generally achieved through patient surveys that question patient-reported outcomes and experience (von Koch et al., 2017). According to literature, also process measures provide essential information for concrete quality improvement (Amor & Ghannouchi, 2017; El Hadj Amor & Ghannouchi, 2017; Lighter 2015; McClimans & Browne, 2012; Rubin, Pronovost, & Diette, 2001). Supporting this development, process mining enables easy assessment of these measures within the data of healthcare information systems.
Based on analysis of literature (ICHOM 2018; Porter, 2017) and the research activities executed as described in the research methodology and data analysis sections, I created a framework for the use of process mining in value-based healthcare. Porter (2017) used a model that shows the flow from ‘structure’ to ‘processes’ to ‘indicators’ and finally to ‘outcomes’ in order to illustrate a quality measurement landscape for VBHC. My framework shown in figure 5, is aligned with Porters model and explained in the following paragraph.
The first component represents the initial status of the patient (for example age, general health condition) as starting point. Also Porter described this status as a given variable that enters the process. The processes the patients go through, considered as the care pathways, are captured in the health information systems and are referred to as ‘data’ in this framework. As shown in several examples in the data analysis section, process
of the measures applied, it can be distinguished between four categories of measurement types enabled through process mining. These are time, number, recurrence as well as sequence of events such as (re)admissions and consultations. These examples of measures can be regarded as outcome indicators since certain outcomes result from the care pathway followed. These outcomes are, for example, survival rate and patient-reported wellbeing (ICHOM, 2018). The burden of care for a patient is considered as an outcome too, according to several ICHOM standard outcome sets (for example for kidney failure). Furthermore, treatment-specific outcomes need to be taken into account for the disease in focus.
Figure 5. Framework for process mining in VBHC.
The four different kinds of measurement that process mining enables, are summed up in the following sections.
Time
Various steps in a patient care path are time-sensitive. Measuring the average length of stay for a patient group or a single patient is straightforward in the process model created by means of process mining. The information is already integrated in the model and just needs to be shown with a click on the activity. As illustrated in figure 7, the total, shortest and longest duration, but also its mean and median are provided automatically. But not only the duration of an activity is captured, also the duration of the path from one activity to another is available and can provide essential information.
Figure 6. Exemplary snapshot of a clinical admission activity with time measures. Count
Counting events and activities has already been frequently used as a process measure in health care (Brenner et al., 2016; 2015 Hessels, Flynn, Cimiotti, Bakken, & Gershon, 2015; Hung & Jerng, 2014; Khalifa & Khalid, 2015; Kirkpatrick et al., 2015). Obtaining figures for certain groups and patients with distinct attributes can be complex and time-consuming. However, by capturing the process models for the relevant patient groups, the numbers of numerous activities are retrieved simultaneously. This means that, for example, the number of clinical admissions in total as well as the number of patients that had clinical admissions is presented (please refer to figure 6 ‘absolute frequency’ and ‘case frequency’).
Recurrence
Whereas many tools can be used to count events, filtering on recurrence is particularly simple with process mining. In some cases, one admission might be usual, but more than one indicates complications. Especially (unplanned) readmissions after certain events like operations are notable (Wu & Hall, 2018). The same applies to operations, where only one or two interventions should be carried out since more than that do not comply with the usual treatment. Filtering on patients with this attribute is possible through a follower determination. So if activity ‘Operatie_Keel-, neus- end oorheelkunde’ (head and neck operation) is followed by the same activity again, there is a recurrence of it. Sequence
In some cases, the direct sequence of activities is of high relevance. An illustrative example for this is the sequence of visit in the emergency department and clinical admission (‘Spoedeisende hulp’ – ‘Klinische Opname’). In the snapshot in figure 7 below, the number of kidney failure patients that arrive at the emergency department, which is 150, is shown. 38 of those patients are then admitted at a nephrology ward and 68 at another ward. The rest goes to other miscellaneous events like the next consultation
Figure 7. Exemplary snapshot of the sequence from an emergency department visit (‘Spoedeisende hulp’).
5.2. Hypotheses proposed
To my best knowledge, this study is one of the first to attempt the usage of process mining in value-based health care or for process measures in health care. Thus, based on analysis of the data and interviews presented in section 4, assumptions were made that are believed to be true. However, these assumptions need to be evaluated by further research.
The hypotheses were developed over the course of this research project. The first relation considered is between the burden of care and (re)admissions, emergency visits as well as operations. Head and neck cancer patients with more than one clinical admission, more than one operation (with a certain interval in between) or acute admissions were identified within the process mining tool. According to this analysis, most patients with a nasopharynx or hypopharynx tumor encountered these critical events. This was also confirmed by the physician, who stated that these patients undergo a more intense treatment. In the kidney failure group, according to the analysis around half of the patients in this phase visited the emergency department or had clinical admissions in the nephrology, which was also confirmed to be a plausible ratio.
Admissions are cost-intensive and interfere in patient’s daily life. But not only the number of admissions is relevant for patients, also the number of days they have to spend at the hospital is important (ICHOM, 2018). Therefore, the hypothesis is:
• A higher number of (re)admissions and extended duration of stay at the hospital increase the patients’ burden of care.
The same applies to operations, where an expected number is considered as usual and a higher number than expected indicates complications or other (unplanned) circumstances. This might increase the patient’s burden of care:
• A higher number of operations increases the patient’s burden of care.
The number of visits to the emergency department is critical in health care (Wu & Hall, 2018). Acute admissions, which means the patient visits the emergency department and a clinical admission follows, are considered as particularly critical in this research. The resulting hypothesis is:
• A higher number of emergency department visits and acute admissions increase the patients’ burden of care.
The next two hypotheses concern the malignancy of a disease. The Dutch work group for head and neck cancer (Nederlandse Werkgroep Hoofd-Halstumoren, 2010) argues that the time between the first consultation in the head and neck oncology and start of the treatment should be less than 30 days in order to provide curative treatment. The opposite would be, that the malignancy (tendency of a medical condition to become worse) increases and survival rate decreases, as formulated below:
• More time between diagnosis and first treatment decreases overall survival rate and increases malignancy.
Also treatments in the phase usually defined as ‘follow-up’ period, where no treatments are expected anymore, might indicate the treatment-specific outcome malignancy or complications with the executed treatment. Therefore, the hypothesis is made that: • Treatments in the follow-up period of a cancer therapy indicate higher malignancy
and/or complications of the treatment.
The hypotheses regarding patients in the outpatient clinic and their well-being were established on basis of the interviews with process participants. Having more appointments on the same day in the outpatient clinic instead of needing to attend these appointments on several different days is an important agenda for patients. The impact of the care process on their everyday life should be lower with less days in the outpatient clinic, and therefore patient quality of life and well-being enhanced. The resulting hypothesis is:
• Minimizing days for patients in the outpatient clinic increases patient-reported well-being.
Also the number of involved contact persons (physicians, nurses etc.) should be minimized, in order to establish trust and avoid confusion for patients which might increase their satisfaction and well-being. Based on this assumption the hypothesis is made:
• Minimizing the number of different contact persons (physicians, nurses etc.) for patients in the outpatient clinic increases patient-reported well-being.
In the future, testing of these hypotheses can be performed with availability of outcome measures and related data, for example gathered through patient questionnaires. By using this data, correlations between the outcomes measured through patient surveys and the indicators gathered through process mining can be detected. For this purpose, a clear relation between both data sources (e.g. Patient IDs) is necessary.
6. Discussion
6.1. Applicability of the method
A considerable body of research regarding process mining in health care has already been conducted where various challenges where highlighted (Mans et al., 2013; Mans, Schonenberg, Song, van der Aalst, & Bakker, 2009; Rozinat, 2011). During this research project similar challenges were encountered.
Data quality issues are a well-known problem in process mining and can be for example missing, incorrect, imprecise or irrelevant data (Mans et al., 2015). Like in other studies (Mans et al., 2009; Mans et al. 2013), some events could only be provided with a date and not a full timestamp including hour and minute. This complicates the ordering of events that are on the same day. Imprecise data was encountered in form of regular diagnostics procedures that are registered as operations and cannot be distinguished from curative or unplanned surgical interventions. On the other hand, there might also be too much information available as in the case of lab tests where the record of each laboratory
a necessity for clustering activities to be able to receive a process model that is still comprehensible. Therefore, it was aimed to create less than 30 activities as suggested to be reasonable for a process model by Dumas et al. (2013).
Besides these data issues, also the complex and dynamic nature of health care processes (Rebuge & Ferreira, 2011) needs to be taken into account. Non-routine care processes discovery is expected to result in spaghetti-like process models. Therefore, the fuzzy miner algorithm was used and seldomly occurring paths are neglected in the analysis. With an increase of high repetition activities (e.g. numerous recurring diagnostics and consultation activities for cancer therapy), the explanatory power of the model decreases.
This complex nature of health care processes is reflected in its diversity too. Both units of analysis required individual preparation of the event logs. Activities needed to be selected and clustered for each care pathway separately to obtain valuable results. A standardization of the data preparation steps is rarely achievable. But also measures need to be defined individually for every medical condition. This requirement is also reflected in the outcome standard set definition by ICHOM, which is made for every medical condition separately and currently offers 23 sets (ICHOM, 2018).
However, an overview of the whole cycle of care is regarded as valuable which leads back to the Harvard institute for strategy & competitiveness’ call for IT solutions that display the full cycle of care (Harvard Business School, 2018a), as mentioned in the introduction section. Showing the complete patient care path also reveals where else in the hospital patients are treated which is considered as particularly relevant in the kidney failure case, which is a medical condition that occurs regularly in conjunction with other diseases.
The information gained with process mining could also support the development of an IPU (Integrated Practice Unit), as suggested within the framework of VBHC that are constructed around the patient medical condition or related conditions and handle the full cycle of care as well as supporting services as nutrition, social work and behavioral health (Gaul, Brömstrup, Fritsche, Diener, & Katsarava, 2011; Harvard Business School, 2018b).
The patient care pathway can be presented easily for the whole patient group, sub-groups of it and also on a single patient level. Process mining enables simple filtering and therefore grouping of patients with specific characteristics, such as treatment attributes, within a few seconds. Further, patients that do not undergo a treatment at the institution but are diagnosed with the relevant code of a patient group, are quickly identified and can be filtered out. Also the timeframe of analysis is selectable. Then, as already outlined in section 5.1, numerous process measures regarding time, amount, recurrence and sequence of activities can be provided rapidly, since they are already integrated into the process model. Whereas many of these process measures can also be retrieved by means of other software applications, the time-saving potential while still being highly flexible in the analysis, constitutes a benefit of the use of process mining for this purpose.
6.2. Practical and theoretical implications
The findings in this research provide several implications for practice and theory. First, the use of process mining can enable health care organizations to identify and determine a variety of measures and at the same time present the complete care pathway visually.
Though, stronger visualization (‘dashboard functionality’) of performance measures used in process mining as well as usability for ongoing monitoring might be an agenda for further development of process mining technology.
Further, it was discovered that the relation between processes and outcomes in value-based health care has been neglected in literature (ICHOM, 2018; Porter et al, 2016) so far. Knowledge regarding outcomes (ICHOM, 2018) and process measures in health care (Brenner et al., 2016; Hessels et al, 2015; Hung & Jerng, 2014; Khalifa & Khalid, 2015; Kirkpatrick et al., 2015) has been extended through this research with newly introduced measures (e.g. how often the outpatient clinic has to be visited for various appointments). Finally, this study presents a linkage between processes and outcomes, by suggesting indicators for outcomes, and offers hypotheses considering the relations between these indicators and outcomes in value-based health care.
6.3. Limitations and future research
Being conducted in a considerable short timeframe of three months, this research is facing several limitations. Within this case, there were only two units of analysis within the same hospital. Future studies regarding the use of process mining in VBHC could be carried out in different wards and hospitals. The determination of measures reached no saturation and was only done by one single researcher. Through more interviews with participants in the process, more measures could be defined in further research. The two processes were mined and analyzed on a high level. A zoom into the different subprocesses of the care pathways could lead to more insights into the process.
Regarding the use of process mining for measure indicators instead of other software solutions, no comparison was included in this project.
Since this constitutes an exploratory study, suggestions for future research are a major output of this work. As explained in section 5.2, this study emphasizes several hypotheses that were not tested during this study and require further research.
7. Conclusion
In this study, I explored the usage of process mining in value-based health care with the objective to determine process-data-related measures that indicate patient-related outcomes. For this purpose, I mined and analyzed two patient care paths with the process mining tool Disco and conducted several interviews with process responsibles and participants.
The output of this case study can be categorized into two types. The first are the different kinds of measures available through process mining, which can be applied in value-based healthcare - namely time, count, recurrence and sequence. These measures are generally usable and consider the high diversity of health care processes. Secondly, this exploratory paper proposes hypotheses regarding the relations between the outcome indicators developed herein and outcomes in value-based health care. These hypotheses require further work and testing.
Process mining can be used to display the whole cycle of care that patients go through while filtering it fast according to specific attributes. However, there is still further research necessary to accelerate its use in health care. I hope that this research represents a first step towards a common, valuable application of process mining in
Acknowledgements
This research would have been impossible without the support provided by several people. First, I would like to thank Inge van de Weerd and Hajo Reijers from the Vrije Universiteit Amsterdam for the valuable feedback and guidance - their office doors were always open for discussing ideas.
I really appreciated the possibility to conduct this research at the VUmc. Therefore, I would like to express my gratitude to all participants, who took the time to help me with this project, especially Ilse Matthijssen, Gabriella Balke-Budai and Marije van der Steen.
Finally, special thanks go to my family, friends and my partner for their continuous encouragement.
References
AMC. (2018). AMC and VUmc join forces as Amsterdam UMC. Retrieved June 10, 2018, from https://www.amc.nl/web/nieuws-en-verhalen/actueel/actueel/amc-and-vumc-join-forces-as-amsterdam-umc.htm
Brenner, S. K., Kaushal, R., Grinspan, Z., Joyce, C., Kim, I., Allard, R. J., … Abramson, E. L. (2016). Effects of health information technology on patient outcomes: a systematic review. Journal of the American
Medical Informatics Association, 23(5), 1016–1036.
Bryman, A. (2012). Social research methods. Oxford University Press.
CBS. (2017, December 19). Zorguitgaven; kerncijfers. Retrieved March 19, 2018, from
http://statline.cbs.nl/Statweb/publication/?DM=SLNL&PA=83037NED&D1=1&D2=8-18&HDR=G1&STB=T&VW=T
Dumas, M., La Rosa, M., Mendling, J., & Reijers, H. A. (2013). Fundamentals of business process
management. Springer.
El Hadj Amor, E. A., & Ghannouchi, S. A. (2017). Towards KPI-Based Health Care Process Improvement.
Procedia Computer Science, 121, 767–774. https://doi.org/10.1016/j.procs.2017.11.099
Erdoğan, T. & Tarhan, A.. (2016). Process Mining for Healthcare Process Analytics. In 2016 Joint Conference
of the International Workshop on Software Measurement and the International Conference on
Software Process and Product Measurement (IWSM-MENSURA) (pp. 125–130).
https://doi.org/10.1109/IWSM-Mensura.2016.027
Ferreira, D. R. (2012). Performance analysis of healthcare processes through process mining. ERCIM News. 89, 18–19.
Gaul, C., Brömstrup, J., Fritsche, G., Diener, H. C., & Katsarava, Z. (2011). Evaluating integrated headache care: a one-year follow-up observational study in patients treated at the Essen headache centre. BMC
Neurology, 11, 124–124. https://doi.org/10.1186/1471-2377-11-124
Günther, C. W., & Rozinat, A. (2012). Disco: Discover Your Processes. BPM (Demos), 940, 40–44. Günther, C. W., & van der Aalst, W. M. P. (2007). Fuzzy Mining – Adaptive Process Simplification Based
on Multi-perspective Metrics. In G. Alonso, P. Dadam, & M. Rosemann (Eds.), Business Process
Management (pp. 328–343). Springer Berlin Heidelberg.
Hall, S., & Wu, V. (2018). Rates and causes of 30-day readmission and emergency room utilization following head and neck surgery. Journal of Otolaryngology - Head & Neck Surgery, 47(1), 1–5.
Harvard Business School. (2018a). Integrated Practice Units. Retrieved May 31, 2018, from https://www.isc.hbs.edu/health-care/vbhcd/Pages/integrated-practice-units.aspx
Harvard Business School. (2018b). Information technology platform. Retrieved March 18, 2018, from https://www.isc.hbs.edu/health-care/vbhcd/Pages/information-technology-platform.aspx Hessels, A., Flynn, L., Cimiotti, J. P., Bakken, S., & Gershon, R. (2015). Impact of Heath Information
Technology on the Quality of Patient Care. On-Line Journal of Nursing Informatics, 19, http://www.himss.org/impact-heath-information-technology-quality-patient-care.
Hung, K.-Y., & Jerng, J.-S. (2014). Time to have a paradigm shift in health care quality measurement. Journal
of the Formosan Medical Association, 113(10), 673–679.
https://doi.org/10.1016/j.jfma.2014.06.003
Jans, M. (2016). From relational database to valuable event logs for process mining purposes: a procedure. Hasselt University, Belgium. Retrieved from https://businessinformatics.be/2016/12/20/from-relational-database-to-valuable-event-logs-for-process-mining-a-procedure/
Kaplan, R. S., & Porter, M. E. (2011). How to solve the cost crisis in health care: the biggest problem with health care isn’t with insurance or politics. It’s that we’re measuring the wrong things the wrong way.(The Big Idea). Harvard Business Review, 89(9).
Kelley, T. A. (2015). International Consortium for Health Outcomes Measurement (ICHOM). Trials, 16(S3). Khalifa, M., & Khalid, P. (2015). Developing Strategic Health Care Key Performance Indicators: A Case
Study on a Tertiary Care Hospital. The 6th International Conference on Emerging Ubiquitous
Systems and Pervasive Networks (EUSPN 2015)/ The 5th International Conference on Current and Future Trends of Information and Communication Technologies in Healthcare (ICTH-2015)/ Affiliated Workshops, 63, 459–466. https://doi.org/10.1016/j.procs.2015.08.368
Kirkpatrick, J. R., Marks, S., Slane, M., Kim, D., Cohen, L., Cortelli, M., … Zapas, J. (2015). Using Value-Based Analysis to Influence Outcomes in Complex Surgical Systems. Journal of the American
College of Surgeons, 220(4), 461–468. https://doi.org/10.1016/j.jamcollsurg.2014.12.034
Lethbridge, T. C., Sim, S. E., & Singer, J. (2005). Studying Software Engineers: Data Collection Techniques for Software Field Studies. Empirical Software Engineering: An International Journal, 10(3), 311– 341.
Lighter, D. E. (2015). How (and why) do quality improvement professionals measure performance?
International Journal of Pediatrics and Adolescent Medicine, 2(1), 7–11.
https://doi.org/10.1016/j.ijpam.2015.03.003
Mans, R. S., Reijers, H. A., van Genuchten, M., & Wismeijer, D. (2012). Mining Processes in Dentistry. In
Proceedings of the 2Nd ACM SIGHIT International Health Informatics Symposium (pp. 379–388).
New York, NY, USA: ACM. https://doi.org/10.1145/2110363.2110407
Mans, R. S., Schonenberg, M. H., Song, M., van der Aalst, W. M. P., & Bakker, P. J. M. (2009). Application of Process Mining in Healthcare – A Case Study in a Dutch Hospital. In A. Fred, J. Filipe, & H. Gamboa (Eds.), Biomedical Engineering Systems and Technologies (pp. 425–438). Springer Berlin Heidelberg.
Mans, R. S., van der Aalst, W. M. P., Vanwersch, R. J. B., & Moleman, A. J. (2013). Process Mining in Healthcare: Data Challenges When Answering Frequently Posed Questions. In R. Lenz, S. Miksch, M. Peleg, M. Reichert, D. Riaño, & A. ten Teije (Eds.), Process Support and Knowledge
Representation in Health Care (pp. 140–153). Springer Berlin Heidelberg.
Mans, R. S., van der Aalst, Wil M. P., & Vanwersch, Rob J. B. (2015). Process Mining in Healthcare
Evaluating and Exploiting Operational Healthcare Processes. Springer.
McClimans, L., & Browne, J. P. (2012). Quality of life is a process not an outcome. Theoretical Medicine and
Bioethics, 33(4), 279–292. https://doi.org/10.1007/s11017-012-9227-z
Nederlandse Werkgroep Hoofd-Halstumoren. (2010). Toelichting centralisatie hoofd-hals oncologische zorg.
Hoofd-Hals-Journaal, 22(43).
OECD. (2018). Helath expenditure and financing. Retrieved March 23, 2018, from http://stats.oecd.org/Index.aspx?DataSetCode=SHA
Porter, M. E. (2008). Value-based health care delivery. Annals of Surgery, 248(4), 503–509.
Porter, M. E. (2009). A strategy for health care reform--toward a value-based system. The New England
Journal of Medicine, 361(2), 109-12. doi:10.1056/NEJMp0904131
Porter, M. E. (2010). What Is Value in Health Care? The New England Journal of Medicine, 363(26), 2477– 2481.
Porter, M. E. (2017). The strategy to transform health care and the role of outcomes. Retrieved April 5, 2018, from https://www.oecd.org/health/ministerial/policy-forum/Michael-Porter-Presentation-OECD-Health-Forum-2017.pdf
Porter, M. E., & Guth, C. (2012). Redefining German health care moving to a value-based system (Vols. 1–1 online resource (xiv, 303 pages): illustrations, portraits). Heidelberg: Springer, Retrieved from SpringerLink http://dx.doi.org/10.1007/978-3-642-10826-6
Porter, M. E., Larsson, S., & Lee, T. H. (2016). Standardizing Patient Outcomes Measurement. The New
England Journal of Medicine, 374(6), 504–506.
Porter M. E., Lee T.H., & Pabo E.A. (2013). Analysis & commentary redesigning primary care: A strategic vision to improve value by organizing around patients’ needs. Health Affairs, 32(3), 516–525. Porter, M. E., & Teisberg, E. O. (2006). Redefining health care: creating value-based competition on results.
Boston, Mass.: Harvard Business School Press. Retrieved from Table of contents http://catdir.loc.gov/catdir/toc/ecip0515/2005018631.html
Rebuge, Á., & Ferreira, D. R. (2011). Business process analysis in healthcare environments: A methodology based on process mining. Management and Engineering of Process-Aware Information Systems,
Rojas, E., Munoz-Gama, J., Sepúlveda, M., & Capurro, D. (2016). Process mining in healthcare: A literature review. Journal of Biomedical Informatics, 61, 224–236. https://doi.org/10.1016/j.jbi.2016.04.007 Rozinat, A. (2011). Process-mining-in-healthcare-case-study-no-1. Retrieved April 30, 2018, from
http://fluxicon.com/blog/2011/05/process-mining-in-healthcare-case-study-no-1/
Rubin, H. R., Pronovost, P., & Diette, G. B. (2001). The advantages and disadvantages of process‐based measures of health care quality. International Journal for Quality in Health Care, 13(6), 469–474. https://doi.org/10.1093/intqhc/13.6.469
Stahl, C., & van Der Aalst, W. M. P. (2011). Modeling business processes: a petri net-oriented approach. Cambridge, Mass.: MIT Press.
van der Aalst, W. M. P., Adriansyah, A., De Medeiros, A., Arcieri, F., Baier, T., Blickle, T., … Wynn, M. (2012). Process mining manifesto (Vol. 99). Springer. Retrieved from http://hdl.handle.net/10044/1/33041$$EView_full_text_in_Spiral_(Access_may_be_restricted) van der Aalst, W. M. P. (2011). Process Mining Discovery, Conformance and Enhancement of Business
Processes / / by Wil M. P. van der Aalst. Berlin, Heidelberg: Springer-Verlag Berlin Heidelberg.
van der Aalst, W. M. P. (2015). Extracting Event Data from Databases to Unleash Process Mining. In J. vom Brocke & T. Schmiedel (Eds.), BPM - Driving Innovation in a Digital World (pp. 105–128). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-319-14430-6_8
van der Aalst, W. M. P., Reijers, H. A., Weijters, A. J. M. M., van Dongen, B. F., Alves de Medeiros, A. K., Song, M., & Verbeek, H. M. W. (2007). Business process mining: An industrial application.
Information Systems, 32(5), 713–732. https://doi.org/10.1016/j.is.2006.05.003
von Koch, L., Flink, M., Nilsson, M., Ytterberg, C., Tistad, M., & Elf, M. (2017). The case of value-based healthcare for people living with complex long-term conditions. BMC Health Services Research,
17(1), 1–6.
Vumc. (2018). About. Retrieved March 20, 2018, from https://www.vumc.com/patientcare/about/ Yin, R. K. (2014). Case study research: design and methods (Fifth edition.). Thousand Oaks, Calif.; SAGE, Yoo, S., Cho, M., Kim, E., Kim, S., Sim, Y., Yoo, D., … Song, M. (2016). Assessment of hospital processes
using a process mining technique: Outpatient process analysis at a tertiary hospital. International
Journal of Medical Informatics, 88, 34–43. https://doi.org/10.1016/j.ijmedinf.2015.12.018
Zorgwijzer. (2018). DBC-systematiek. Retrieved April 20, 2018, from https://www.zorgwijzer.nl/faq/db
No appendix included in this version – please contact ursula.ablinger@gmail.com for further information.
