Process Mining Thodoros Topaloglou Daniele Barone Faculty/Presenter Disclosure • Faculty: Thodoros Topaloglou • Relationships with commercial interests: – Grants/Research Support: • NSERC Discovery Grant (2006-12), PI • NSERC Strategic Network Grant: Business Intelligence Network (2008-2014), Co-PI – Speakers Bureau/Honoraria: None – Consulting Fees: None – Other: Employee of Rouge Valley Health System Disclosure of Commercial Support • This program has NOT received financial support from any Commercial Organization • This program has NOT received in-kind support from any Commercial Organization • Potential for conflict(s) of interest: None Mitigating Potential Bias • [Explain how potential sources of bias identified in slides 1 and 2 have been mitigated]. • Refer to “Quick Tips” document Understanding and Improving Hospital Processes Business Process Management • Document and catalog hospital processes using formal, visual notation like BPMN • Actively manage processes by measuring their performance • Continuously improve processes T. Topaloglou Business Intelligence • Understand operational performance by monitoring process execution • Provide process and data visibility to business users • Monitor key performance metrics RVHS Information Management Process Mining • A deeper dive into process execution to learn the structure of processes. • Find the processes or sub-processes that really get executed vs. what thought to be executed. 5 Talk Objective • The objective of this presentation is to discuss how to “understand” processes by pairing process models and data • I will also share an experience-report from the Rouge Valley Health System’s (RVHS) journey to support process based performance management through two transformative initiatives – Business process management – Enterprise business intelligence and review some of our early efforts on process mining T. Topaloglou RVHS Information Management 6 Rouge Valley Health System • RVHS is a two site hospital with 479 beds serving the East GTA community • Key facts – – – – – – – 2700 employees Over 500 physicians and 1000 nurses 122,000 ED visits in 2012-13 26,000 admissions 25,000 surgeries 3,700 births over 189,000 clinic visits • Has a corporate performance mgmt framework and corporate scorecard • Has adopted Lean as a management and quality improvement philosophy • In 2010-11, RVHS launched two transformative IT initiatives to – create a competency center in business process management, and – develop an enterprise Business Intelligence system T. Topaloglou RVHS Information Management 7 Business Process Management • If you cannot measure a process you cannot improve it • But… if you cannot “see” it you cannot measure it! • A visual notation that business and clinical users can understand T. Topaloglou RVHS Information Management lean Visual modeling BPMN 8 From Processes to Measuring Outcomes Lean meets BPM meets BI Define Reference Date Previous 7 Days 1 Total ED visits (#) 130 129.7 129.2 N/A N/A 2 ED visits CTAS I (%) 1.5% 0.6% 0.5% N/A N/A 3 ED visits CTAS II (%) 11.5% 9.7% 10.7% N/A N/A 4 ED visits CTAS III (%) 63.1% 55.5% 55.0% N/A N/A 5 ED visits CTAS IV (%) 23.1% 31.6% 31.3% N/A N/A 6 ED visits CTAS V (%) 0.8% 2.5% 2.4% N/A N/A 7 Left Without Being Seen (%) 13.1% 8.8% 7.0% 7.0% 4.0% 8 ED visits admitted (%) # Measure Emergency Department Control 10.0% 10.5% 10.5% 11.0% 11.0% 4.0 5.0 4.8 5.0 4.5 10 3.4 3.9 3.7 4.0 4.8 ED ALOS for non-admitted patients (hrs) 11 CTAS I-II non-admitted patients with LOS <= 7 hrs (%) 90.0% 88.1% 86.0% 80.0% 75.0% 12 CTAS III non-admitted patients with LOS <= 7 hrs (%) 83.1% 76.8% 79.3% 80.0% 85.0% 13 CTAS IV-V non-admitted patients with LOS <= 4 hrs (%) 90.0% 74.5% 79.6% 80.0% 85.0% 6.6 6.1 6.2 6.0 12.0 ED ALOS for admitted patients (hrs) 15 CTAS I-II admitted patients with LOS <= 8 hrs (%) 42.9% 42.9% 42.3% 45.0% 40.0% 16 CTAS III admitted patients with LOS <= 8 hrs (%) 40.0% 26.2% 22.3% 25.0% 30.0% 17 CTAS IV-V admitted patients with LOS <= 8 hrs (%) 0.0% 0.0% 11.1% 15.0% 20.0% 18 Admitted patients in ED - no IP bed at 06:00 (#) 5.0 3.3 2.7 3.0 3.0 19 IP ALOS (excl. ALC) (days) 4.5 5.0 4.4 4.5 4.0 20 IP Discharges by 11:00 (%) 10.0% 21.2% 19.8% 20.0% 30.0% 21 IP Discharges by 14:00 (%) 10.0% 25.0% 25.0% 20.0% 30.0% 22 IP Discharges (#) 40.0 29.7 31.1 27.0 35.0 23 ALC patients (#) 18.0 21.7 27.1 35.0 30.0 24 IP ALOS (excl. ALC) of patients in Unit 1 (days) 3.8 4.4 5.2 5.0 6.0 25 IP Discharges by 11:00 Unit 1 (%) 0.0% 4.2% 9.5% 20.0% 30.0% 26 IP Discharges by 14:00 Unit 1 (%) 35.7% 35.4% 35.9% 25.0% 27 ALC patients Unit 1 (#) 6.0 6.9 7.5 10 99 28 IP ALOS (excl. ALC) of patients in Unit 2 (days) 4.0 3.9 4.2 4.6 4.0 RVHS Information Management Admit & Dischg Analyse Unit 1 Improve Previous 30 Days Baseline Target 9 ED ALOS – all dispositions (hrs) 14 T. Topaloglou Metric (units) (definitions) 30.0% Rationale for BI at RVHS Evidence • Process owners need evidence to manage their business • Evidence hides in the data T. Topaloglou Intergration • Create an integrated repository of operational and clinical sources Access • Enable process owners (mgrs) to access process data and gain insights RVHS Information Management Action • Empower business users to take actions by monitoring process based performan ce metrics 10 Relevant, Real-time, Process-driven Metrics ID: 16 ER Department Manager <responsible for> Reduce the length of stay of Emergency Department CTAS III admitted patients in the Emergency Department Dimensions Time Location Provider CTAS Percentage of CTAS III admitted patients with Length of Stay (LOS) equals or less than 8 hours User Driven Business Intelligence Triage Assessment ALC-MED ALC-CCC Depart E D Leave the Emergency Department to home MED CCC L T C Home Clinical activity Patient care T. Topaloglou Infectionc Financial ontrol activity Clinical activity Patient care RVHS Information Management Infectionc Financial ontrol activity 11 From Business Objectives to Processes Improve access to care Strategic Plan QIP Corporate Scorecard HSAA CEO PBCs ED LOS < 4hrs Corp. Services Corporate Scorecard Acute Care Post-Acute ED LOS < 4hrs Admit • • • • • BI supports business goals Series of linked & cascading scorecards Scorecards as collections of metrics Metrics depend on other metrics or process KPIs Linking processes performance to metrics T. Topaloglou / December 2011 RVHS Business Intelligence Program ED ERNI process PIA Beds Medicine Discharge process 12 Actor-Goal-Indicator-Object Diagram T. Topaloglou / December 2011 RVHS Business Intelligence Program 13 Connect Strategies to Processes with AGIO T. Topaloglou / December 2011 RVHS Business Intelligence Program 14 Patient Flow Process Map T. Topaloglou RVHS Information Management 15 ED Now Dashboard T. Topaloglou RVHS Information Management 16 Process Mining • Process mining aims to discover, monitor, and improve real processes by extracting knowledge from event logs (Van Der Aalst, www.processmining.org) T. Topaloglou RVHS Information Management 17 Process Mining Tasks Wil Van Der Aalst. 2012. Process mining. Commun. ACM 55, 8 (August 2012), 76-83. DOI=10.1145/2240236.2240257 http://doi.acm.org/10.1145/2240236.2240257 T. Topaloglou RVHS Information Management 18 Process Mining in Healthcare • Event logs – ADT and Order Entry applications are rich sources of events • Process complexity – Many sources of variations • by performer, by case/patient, or practice variation. • BI applications intend to monitor variation – Process hierarchies • Multiple levels of process-subprocess relationships • BI applications typically focus on higher level processes – Process pools • There are multiple processes or initiatives active at any time • Many process metrics measure aggregate effects T. Topaloglou RVHS Information Management 19 Practical Process Mining • Process signatures are distinct data markers that correspond to execution (or not) of specific processes – e.g, CTAS 4-5 patients in the range 8-24 indicate non-departed charts! • Queries for presence of specific sequence of events in transaction (event) logs or data warehouses – if we know what we are looking for we can find it! • Abnormal results – We found that ALC designation is performed differently between sites (practice variation) because the calculated metrics didn’t match • By visualizing data and searching for patterns that can be process signatures and then find matches for these signatures – Through process mining we were able to reverse engineer actual processes and found activities in the logs were redundant e.g, not all clinic visits have to be scheduled before registered. T. Topaloglou RVHS Information Management 20 Visualization of Event Logs Action INSERTED UPDATED UPDATED UPDATED UPDATED UPDATED UPDATED UPDATED Seq_Num 1 2 3 4 5 6 7 8 T. Topaloglou Status SCH SDC PRE SDC REG SDC ADM IN ADM IN ADM IN ADM IN DIS IN Type O O O I I I I I LocationID YCCL YCCL YCCL Y9WC Y9W Y9W Y9WC Y9WC RoomID NULL NULL NULL Y910 Y910M Y928 Y910 Y910 RVHS Information Management BedID NULL NULL NULL 1 1 3 2 2 ReasonForVisit Modified_Date +/- HEART CATH 2013-04-19 15:56:14.570 +/- HEART CATH 2013-04-19 15:59:51.150 +/- HEART CATH 2013-04-19 17:06:45.050 PCI 2013-04-19 23:00:32.133 PCI 2013-04-20 10:53:01.400 PCI 2013-04-21 12:27:59.420 PCI 2013-04-22 13:48:33.443 PCI 2013-04-23 17:26:41.247 21 The Future of Process Mining • Discover process flows from even logs (Van Der Aalst) • Discover BPMN from event logs or database tables (exploit richer data semantics) • Data mining of event logs for similar patterns (process signatures), and further discovery of process flows within pattern clusters • Process mining is the combination of data mining and business process management, and very much an active research field with tremendous potential in helping healthcare organization understand their processes. T. Topaloglou RVHS Information Management 22 Thank you ttopaloglou@rougevalley.ca