BUSINESS PROCESS MANAGEMENT CASES Digital Innovation and Business
BUSINESS PROCESS MANAGEMENT CASES Digital Innovation and Business Transformation in Practice EDITORS: Jan vom Brocke Jan Mendling
IMPROVING PATIENT FLOWS AT ST. ANDREW’S WAR MEMORIAL HOSPITAL’S EMERGENCY DEPARTMENT THROUGH PROCESS MINING Robert Andrews Suriadi Moe Wynn Arthur H. M. ter Hofstede Sean Rothwell
INTRODUCTION
Background and Motivation • Recent years have seen an increasing demand for Emergency Department (ED) services in Australia’s public hospitals without a corresponding rise in inpatient beds – EDs frequently experience overcrowding with patient’s experiencing • Patient flows have been adopted as a management strategy to systematize the processing of patients • Improving ED patient flows in terms of processing time, resource use, costs, and patient outcomes is a priority for health service professionals and is vital to the delivery of safe, timely, and effective patient care. – If patients are not moving through the system efficiently, other patients may experience delays
SITUATION FACED
Challenges • Processes in healthcare/ED are complex and semi-structured – no formal process model, many points at which different continuation paths are possible, largely driven by ad-hoc, human decision making • Patient-centric treatment plans are devised after Triage and in accordance with accepted clinical pathways • However disruptions are frequent – manifest as long-wait times in ED, delays in administering/reporting on tests, overcrowding and “boarding” of patients in ED • Prolonged Length of Stay (Lo. S) in ED is associated with poorer patient outcomes
Background and Motivation • This project involved a detailed analysis of patient flows in St. Andrew’s War Memorial Hospital’s (SAWMH) ED using a processmining methodology with the aim of providing insights into the ‘as is’ processes in the ED, particularly as these processes apply to patients who present with chest pain. • The analysis aimed to reveal process factors (such as bottlenecks, protracted activity durations, and rework loops) and context factors that affect patient flows. • The analysis outcomes then underpin evidence-
ACTION TAKEN
The Case Study Process identification • Focus was on process identification, discovery and analysis phases of BPM Lifecycle Identify research questions relevant to SAWMH Extract process-related data Data quality assessment & log cleaning Process discovery Discover ‘as is’ process model Process Monitoring & Controlling • Applied process mining methodology and techniques Process implementation Process analysis Process redesign Performance (by patient cohort) Comparative analysis
Research Questions for SAWMH • What are the key differences in the patient flows between patients who stayed in the ED for less than four hours and those who stayed for more than four hours? • How much delay was introduced to the patient flows as a result of conducting routine clinical activities, including blood tests and X-ray imaging? • What are the factors that influenced the patients’ Length of Stay?
Key Challenges • Data – Identify relevant data from Hospital IS • Encounters (patient arrival/departure), Emergency (ED patient flow milestones), Clinical (clinical observations), Orders (pathology and imaging tests) – Correlation • Matching related records to relevant Encounter – Patient may have more than one Encounter on same day – Data Quality • • • Duplicate references to same event in different logs Irrelevant events Missing events Inconsistent granularity of timestamps Multiple activities with the same timestamp Events that really represent case attributes
Key Challenges • Discovery – Generalise highly-variable, patient-centric flow data to discover readable models that capture dominant (most frequent) care pathways • Analysis – Extracting differences between flows for different cohorts was highly manual (and therefore not an efficient way to discover the process variations) – Best way to highlight differences is visually but visualisation tools for comparative analysis are lacking • Needed to develop novel visualisation techniques
RESULTS ACHIEVED
Process Discovery • Dominant (most frequent) care paths – major medical milestones • Arrival at ED • Triage • Seen by medical staff Process model describing the main patient flow • Leave ED (major milestone events). In this model, each rectangle – Admit to represents an activity, and the color density of the rectangle hospital represents the frequency of the activity. Arrows represent – Discharge hometransitions between activities, and the width of the arrow – for chest pain patients represents the frequency of the transition. The numbers on the arrows and in the boxes indicate the case frequency.
Process Discovery • Discovered patient flow models for different patient cohorts – Based on length of stay in ED Discovered process model for ED Lo. S up to 4 hours Discovered process model for ED Lo. S more than 4 hours
Process Analysis • General patient flow – There is, at a high level, a logical flow of activities (derived from the patient milestones) • Patient-centric variations occur – e. g. Medical_Assign (seen by doctor) can occur before Triage or even before Arrive • Clinical – Nursing activities form the backbone of clinical events (majority of events relate to this) – Complex model (indicating highly varied, patient-centric treatment processes) • Simple process visualisations cannot provide meaningful insights
Process Analysis • Short and Long Lo. S – Event timing for the two cohorts is similar up to ‘Blood test (Ordered)’ – Differences are discernible from ‘RN Assign Start’ • We were not able to determine the causes for this observation • Analysis would have benefited from logs recording both start and end times of events Times of milestone events (minutes after Arrive_Start event)
LESSONS LEARNED
Take Home Lessons • This project demonstrated that process mining is applicable to complex, semistructured processes like those found in the healthcare domain. • Comparative process performance analysis yielded some insights into ED patient flows, including recognition of recurring data-quality issues in datasets extracted from hospital information systems. – The templated recognition and resolution of such issues offers a research opportunity to develop a (semi-)automated data-cleaning approach that would alleviate the tedious manual effort required to produce highquality logs.
Take Home Lessons • The project highlighted the importance of hospital information systems collecting both start and end times of activities for proper performance analysis (duration, wait time, bottlenecks). • Additions to our process-mining toolset include novel comparative processperformance visualization techniques that highlight the similarities and differences among process cohorts. ‘Superimposed model’ (Pini et al. 2015) for visualizing comparative process performance. (Two events are exploded out of the model to highlight their relative temporal ordering. ).
- Slides: 20