Utilizing Systems Engineering Methodologies to Enhance Clinical Decision Support Matt Johnson, Katie Schwalm, Linda Bashaw, Robert Chang, and Christopher Petrilli
Utilizing Systems Engineering Methodologies to Enhance Clinical Decision Support Matt Johnson, MS Katie Schwalm Linda Bashaw, RN Robert Chang, MD Christopher Petrilli, MD
Outline Brief overview of Michigan Medicine Systems Engineering What are we talking about? Why is this important? Methodology Who is involved? Case Studies Takeaways Questions
Michigan Medicine: Key Facts Licensed Beds (Medical & Surgical): 1,000 Emergency/Urgent Care Visits 104,000 Discharges ~49,000 Outpatient Clinic Visits: 2.1 M Surgical Procedures: 54,000 Faculty Members: 2,700 Nurses: ~5,000 Residents in Training: 1,199 Medical Students: 708
Our Team Performance Improvement Team Department of Internal Medicine Resource The content of this presentation represents Department-led projects.
What are we talking about? Systems Engineering Clinical Decision Support
Why is this important? Increased pressure on care teams complexity of patient care documentation requirements speed of decision making Cognitive burden Provider burnout Demand for care value
Methodology
Multidisciplinary Team Approach Project Management Engineers Programmers Clinicians
Critical Value of Contributors EMR Support, Programmers, Nurses, Physicians, Clerical Support, Pathology, Phlebotomy, Patient Advisors, etc. Improved Outcomes Patient and Employee Satisfaction
Systems Engineering Integration Clinical Decision Making Individualized process Significant information gathering Discrete decisions trigger care delivery Care Delivery Events occur based on orderable events in the EHR Processes occur in silos Systems level performance is not seen from provider level view without significant effort Systems Engineering Connect disconnects in EHR Identify processes that are highly variable in expectations of delivery Provide feedback to providers to enable action Develop processes that provide, redundant, reliable mechanisms for quality, care delivery
Case Study #1 Tacrolimus Blood Draws
Case Study - Overview Problem: Physicians noticed that tacrolimus blood level draws were often not being drawn at the appropriate times. Blood level draws require a critical time sequence between medication administration and actual time of blood draw. Ideal time gap is 12 ± 2 hours Patient receives evening medication dose Morning lab draw to test blood level Results provided to physician and medication may be adjusted Patient receives next medication dose
Process Mapping and Analysis Medication Ordering and Admin. Workflow Tacrolimus med. order placed by ordering provider Med. given to patient twice daily at 9:00 AM and 9:00 PM Dose admin. Documented in medical record Continue dosing times per provider oder Physician/APP Nursing Nursing Nursing Identified that only 68.5% met appropriate time range Two independent Possible workflows order change based on results No communication between processes Medication Ordering and Admin. Workflow Tacrolimus blood level order placed by ordering provider Physician/APP Order placed as time critical lab for morning draw Physician/APP Phlebotomy coordinates time critical draw to occur per schedule Physician/APP Phlebotomy draws specimen from patient Physician/APP Test is processed and results are pushed to medical record Physician/APP Results interpreted and determination is made on plan of care Physician/APP Lack of time standard for lab draw
Developed System Level Understanding Developed temporal distributions medication administration and lab draws. Foundation for simulation modeling. Draw-to-Dose n = 9,053 labs Dose-to-Dose Identified noticeable shift of early blood draws. Trough Time
Simulation Modeling of Interventions Shifting lab draw times to later in the morning improves trough times. Ideal
Model Validation All Units Trough Times (0-18 hr. Time Window) ICU Transplant (UH5C) All Units (Excl. Transplant + ICU) All Nurse Draw Units Sample Size 4004 667 787 2550 1107 Troughs within 12±2 hrs 68.50% 66.9% 79.03% 65.72% 63.86% 84% compliance with 7:30 AM time draw orders 39% compliance with 7:30 AM draw time orders Intervention Switched to pre-populated order set for 7:30 AM. Educational feedback
Decision Support Takeaway 1. System interactions between discrete, independent processes are not always understood at the care level. 2. Mapping out these disconnects, provides foundation for system level control. 3. Many interventions can be quite simple, though identifying the root cause can take significant effort.
Case Study #2 Heparin Induced Thrombocytopenia (HIT) Testing
Case Study - Overview Problem: Hematologists and Hospitalists believed that many expensive laboratory tests used to diagnose Heparin Induced Thrombocytopenia (HIT), were being ordered without any clinical value added. Suspected Condition Ideal Condition
Process Mapping and Analysis Evaluated time stamping of lab orders and subsequent result values for every test ordered in the last 12-months. 62.4% both tests are ordered at the same time. Suspected Condition 89% of the results are negative. Translates to 56% of tests providing no clinical value.
Current State and Interim Measures Test #1 and Test #2 ordered together 62% of the time. Test # 1 results and 89% of the time, no indication for test #2 Test #2 already sent to outside lab Intervention Pathology holds send out testing of confirmatory test (Test #2). When 1 st test results and is negative, a message is sent to ordering provider and Test #2 is canceled. Prevents inappropriate tests from being sent out. Actual Result: Use of the test has gone down by 75%
Integrated Ordering Decision Support Future State Pathology uses reflexive order set rules to place order for 2 nd test when appropriate. Provides built in education support in the order set. If 1 st test positive, sends next test. Prevents add. work from pathology. Prevents action when both tests are ordered. Educates provider on best practice. Prevents up to 56% of testing waste.
Decision Support Takeaway 1. Effective decision support can provide point-of-care education in addition to streamlining workflows to prevent waste or inappropriate care. 2. Clinician anecdotes regarding care delivery limitations can be an excellent* source for system level initiatives. *Caution: Important to model first before launching into initiative/project
Case Study #3 Chest Pain Pathway
Case Content Study Title - Overview Problem: Low risk (non-ami) patients coming to the ED with chest pain are not effectively risk stratified or worked up efficiently. Consequences: Variable and prolonged length of stay. Over-utilizing resources on patients who may not need additional testing.
Process Content Title Mapping and Analysis Patient arrives at ED presenting with Chest Pain ED RUN triage desk initiated 10 minute algorithm Patient transferred to intake for ECG and then to room Results and Pathway Determination Rule-out and discharge home Observe and downstream testing A current state of managing a patient presenting with chest pain has many sequenced events. Very methodical process with set expectations. Initial Troponin Ordered and Drawn Follow up Troponin Ordered and Drawn Admit Discrete-event modeling
Current Content Title State Analysis ED Arrival EKG 1 st Troponin 2 nd Troponin * * Time between 1 st and 2 nd Troponin Test
Process Content Title Level Analysis Current state time reduction improvement Time Savings 7,639 hrs. With improved advances in testing Time Savings 15,513 hrs. Intervention: Integrating a timed troponin order set. Reduce order variability by automatic process. Reduces clinician ordering burden.
New Content Testing Title Decision Support Intervention: Provide guidance into workflow on interpretation of new testing and downstream ordering. 1 st Troponin T= 0 hr. 0h 53 pg/ml 19 pg/ml < 0h < 53 pg/ml 0h 19 pg/ml 2 nd Troponin T= 2 hr. 3 rd Troponin T= 4 hr. Result Interpretation AND AND* This is Complicated! 2h > 19 pg/ml 2h 6 pg/ml 2h < 6 pg/ml 2h 6 pg/ml 2h < 6 pg/ml AND 2h < 6 pg/ml Greater opportunity for error. Ideal scenario for clinical If suspicion decision for ACS high, suggest to get a third troponin. Otherwise, stress test or CTCA. support. 4h 6 pg/ml 4h < 6 pg/ml Opportunity to provide education Rule-In in addition to decision support. Potential recent MI, CKD, LVH* Rule-In Rule-In Downstream testing and, if negative, consider other etiologies for elevated troponin. AND 2h 6 pg/ml Rule-In 2h < 19 pg/ml AND 2h < 6 pg/ml Rule-Out
Integrated Decision Support Workflow Intervention: Development of prepopulated workflows, reducing clicks and streamlining a recommended pathway. Sequenced tests are released at the same time to nursing Links provider to workflow
Decision Content TitleSupport Takeaways Evaluate the variability and process performance at all process steps of your workflow. Identify opportunities for error prevention by providing real-time feedback.
Case Study #4 Inpatient Event Communication to Oncologists
Oncologist Content Title Admit/Discharge Event Notifications Problem: Oncology physicians not aware of their patients being admitted to the hospital. Resulting in: Lack of physician collaboration on the patient s plan of care while in the hospital. Patient and provider dissatisfaction.
Oncologist ADT Event Notifications Oncologist Aware of Patient Hospital Visit N = 181 patient notifications No 35% Yes 65%
Oncologist ADT Event Notifications Distribution: days from patient s last outpatient visit to hospital encounter 65% of patients in our validation study were admitted within one month of most recent outpatient visit Algorithm notifies oncologist on the right patients that are receiving active treatment.
Notification Content Title in Physician InBasket Work Queue Delivery of notifications to standardized, accepted workflow. Real-time notifications to prevent delays in communication. Updates on status of care provides higher value content to the clinician: Admission Transfer to ICU Patient deceased Discharge order to hospice. In Basket My Messages Overdue Rx Requests Rx Requests Patient Call Encounters Message (Non-Encounter) Communications Inpatient Notifications Result Errors High-level initial information displayed: Status Patient Event Type Admission Date Discharge Date
Decision Content TitleSupport Takeaways Data collection and validation allows accurate modeling to simulate the problems that physicians experience. Define an effective and proven algorithm for identifying the right population of patients for an intervention.
Closing the Gap Providing relevant clinical information
Change Management is difficult in the hospital environment Emphasis on Patient Safety & Clinical Quality Technological Advances Demographic Changes Financial Pressures
Successful Change Management Leadership Support Integration Structure that supports the change. Culture of continuous improvement.
Successful Change Management Simplicity Communication Visual Management How & Who
Example of Simplicity & Communication Visual and concise Try to get to a 1-page limit and use multiple modes of communication. Cater and adapt communication to specific groups. Embed in EHR workflows when appropriate.
Successful Change Management Accountability Dashboards for care pathways, divisional quality metrics, and new processes. Dive down to unit, service, and provider level. Process and outcome driven dashboards.
Key Takeaways Multidisciplinary teams = essential for success Use decision support to improve disjointed processes Discrete event modeling can boost understanding of hospital operations and help identify opportunities to implement decision support. Change management is difficult especially when implementing decision support initiatives, but thorough communications plans and leadership support can breakdown barriers.
Thank you! Questions? Email Us: johmatth@med.umich.edu kschwalm@med.umich.edu
Acknowledgements Internal Medicine Quality Program (Drs. Scott Flanders, Jim Froehlich, & Maria Han) Department of Internal Medicine (Dr. John M. Carethers, Musty Habhab, & Jolena Nollar) Performance Improvement Team (Tammy Ellies & Liz Spranger)