Solving the Physician Attribution Puzzle for Length of Stay Session 61, March 6, 2018 Ingrid Carlson Wurpts, PhD, Dignity Health Angelica Chanco, MD, Dignity Health 1
Conflict of Interest Ingrid C. Wurpts, PhD and Angelica Chanco, MD Have no real or apparent conflicts of interest to report. 2
Agenda Why is provider attribution important in improving certain hospital metrics? How was the Substitutable Medical Approaches and Reusable Technology (SMART) Provider Attribution Algorithm developed? How was SMART Provider Attribution (SPA) extended to other use cases? How is SPA being used in a dashboard to improve clinical outcomes? 3
Learning Objectives List the potential problems with using the attending physician as the attributable physician for length of stay and HCAHPS physician communication questions Compare attending versus the SMART provider attribution method for attributing HCAHPS physician communication and length of stay Explain how a visual analytics dashboard can be used to monitor and improve patient outcomes such as length of stay 4
Dignity Health: Who Are We? 5 th LARGEST Non-Profit Health Care System in the nation 8.3k licensed beds 38 Hospitals 682k Patient days Ingrid C. Wurpts, PhD Data Scientist Angelica Chanco, MD Clinical Analyst 5
Provider Attribution 1. Link between encounter-level healthcare metric and responsible provider 2. Necessary for targeted interventions to improve physician communication, etc. 3. Previous approach: Attribution given to attending at discharge physician 4. Challenge: The attending at discharge may not have had any significant involvement in the patient s care 6
Why Address Provider Attribution at Dignity? Goal: Improve HCHAPS physician communication scores across the enterprise Previous interventions targeted attending physicians I barely saw that patient! How could we use data to improve attribution? Critical Success Factor: Collaborative effort between Medical Informatics and Data Science team and Care Management team 7
SMART Provider Attribution Algorithm 1. We developed a statistical algorithm that uses clinical documentation to determine provider attribution more accurately than using attending physician 2. All physicians on a patient s chart receive an attribution rank 3. Attribution can include non-physician providers (PA, NP, CNM) if desired 4. Algorithm has been validated for two use cases: HCAHPS physician communication Length of Stay (LOS) 8
SPA Algorithm Development Algorithm development Gold standard data set: clinical abstractors rated which provider(s) were most responsible for communication with the patient across 100 encounters Data mining techniques used to determine which clinical engagement variables (e.g., med orders, surgery notes) in the patient s chart could predict the abstractor s top rated provider(s) most closely Final algorithm SPA is a weighted sum of specific data fields from each patient s chart. Each provider is given a score, and provider with top score within each patient s chart is given attribution. 9
SPA vs. Attending Comparison Attending at discharge physician: PPV* = 65/100 = 65% is the probability that attending at discharge physicians were identified as primarily responsible for communication by the abstractors. 95% CI [55%, 74%] SMART Provider Attribution Algorithm: PPV = 89/100 = 89% is the probability that physicians identified as primary by SPA were identified as primarily responsible for communication by the abstractors. 95% CI [82%, 95%] *Positive Predictive Value (PPV) 10
SPA vs. Attending Comparison Originally interested in medical service line. Extended to OB and surgical to have continuity across our adult service lines SPA provides more accurate attribution than attending especially for medical and OB patients. 11 Med cases only: Algorithm PPV = 60/68 = 90% [80%, 96%] Attending PPV = 44/68 = 65% [52%, 76%] OB cases only: Algorithm PPV = 20/22 = 91% [71%, 99%] Attending PPV = 11/22 = 50% [28%, 72%] Surgical cases only: Algorithm PPV = 9/10 = 90% [56%, 100%] Attending PPV = 10/10 = 100%
Extending SPA to Another Use Case SPA was implemented into our data ecosystem Care management leaders receive a monthly report of average physician communication scores, attributed by our SPA algorithm Data consumers were interested in using SPA for other encounterlevel metrics, like length of stay (LOS) Is the provider primarily responsible for communication also primarily responsible for LOS? Collected gold standard data on LOS attribution 12
Using SPA for Length of Stay Attending at discharge physician: PPV = 27/48 = 56% is the probability that attending at discharge physicians were identified as primarily responsible for communication by the abstractors. 95% CI [41%, 71%] SMART Provider Attribution Algorithm: PPV = 37/48 = 77% is the probability that physicians identified as primary by SPA were identified as primarily responsible for communication by the abstractors. 95% CI [62%, 88%] 13
Why is it important for Dignity Health? 5 th LARGEST Non-Profit Health Care System in the nation 8.3k licensed beds 38 Hospitals 682k Patient days Even a 1% reduction in unnecessary LOS days can yield big results: $17.5M Cost savings $2,571 = an average estimate of expenses incurred by the hospital to provide a day of inpatient care in CA, NV, AZ hospitals 1999-2015 AHA Annual Survey, Copyright 2016 by Health Forum, LLC, an affiliate of the American Hospital Association. 14
SMART Provider Attribution Dashboard Dashboard Admission / Discharge / Transfer systems Electronic Medical Records Financial Analytics & Dignity Health tables CMS tables Hadoop SPA Algorithm Refresh Rate: Daily 15
Dashboard data Focus on opportunity Days variance to GMLOS by Payor Top 10 providers over GMLOS (days and patients) Top 10 MS-DRGs over GMLOS (days and patients) Top 10 Service Lines over GMLOS (days and patients) Encounter level information displayed Patient demographics, payer and provider information * GMLOS Geometric Mean Length of Stay 16
Dashboard data (continued) Trending Days over GMLOS by facility Provider days and patients above GMLOS User customizable cohorts for data discovery Filter, sort, export data within dashboard Save searches 17
SPA Dashboard Development & Rollout Feb 2017 App Prototype Testing June 2017 Pilot July 2017 Enterprise Rollout Feb 2018 Completion of Rollout Rapid development In-house development team Rapid rollout 1 region every 2-3 weeks (except holidays) 18
SPA Dashboard Development and Rollout (continued) 35 Facilities completed 220k Encounters USERS: Hospital Administration Care Coordination Hospitalist Groups Leaders Analysts Encounters with days over GMLOS Data from 10/2015 onwards 19
Action Opportunities Clinical Practice Optimization Length of Stay Committees Hospitalist Groups Physician Outlier meetings Utilization Management Sepsis Initiative Data for monitoring and trending Trend LOS for Implemented Initiatives 20
Sample Cases Case 1 Case 2 Case 3 Dr. Smith: high % of patients over LOS Workflow review reveals improvement opportunities upon coordination with care team Sepsis (DRG 871) major contributor to LOS opportunity days Initiative to improve sepsis bundle compliance Consistently excellent performance in Joint Replacement (DRG 470) Ortho service lines Best practices dissemination to other service areas 21
Iterative SPA & Targeted Intervention Improvement Model SPA App User Launch SPA App Discover Design Develop Deliver Launch Measure Maintain 22
Implementation Takeaways Challenge Dirty source data Increasing adoption of SPA application Applying information gleaned from SPA Solution Work with source systems Data in, data out ; internal cleansing Ensure facility sponsors are engaged Education and training Cultivate system best practices Refining algorithm Collecting data; scheduled for refinement 2018 23
Summary SPA helps solve the physician attribution puzzle Identified limitations of using the attending physician as the attributable physician for length of stay and HCAHPS physician communication questions Shared Dignity Health experience on how this insight led to actionable initiatives: Compared attending versus the SMART provider attribution method for attributing HCAHPS physician communication and length of stay Developed a visual analytics dashboard currently used to monitor and improve patient outcomes such as length of stay 24
Summary (continued) Provider attribution is necessary for targeted improvement of patient-level outcomes, but previous attribution methods were insufficient We used data mining to develop a new attribution method (SPA) SPA performed better than the previous method for both physician communication and LOS use cases Rapid paper-to-production deployment is possible with dedicated resources, a clear directive, and user base support Improvement is an iterative process it is important to measure and monitor, then design and implement changes 25
Questions Ingrid.Wurpts@dignityhealth.org Mel.Chanco@dignityhealth.org We welcome feedback! Please complete the online session evaluation 26