Downloaded 71 times
![Predictive Model Confidence
Why do we believe the Readmission Model?
Comparing existing models with standard C-Statistic (Area under
ROC Curve) measure of performance
– Random coin toss selection: 0.5
– State-of-art techniques(ACG): (0.70 to 0.77)[1]
– Current Allina technique: 0.861
Allina Model was found to have a precision* of ~ 0.9
*Precision is the fraction of Predicted patients that actually have a PPE. In this case, on a dataset in
which it was tested about 90% of patients predicted by the model had a PPE. Note, this is different
from sensitivity, which is the fraction of actual PPE instances that are predicted.
1 Shannon M.E. Murphy, MA, Heather K. Castro, MS, and Martha Sylvia, PhD, MBA, RN, “Predictive Modeling in Practice: Improving the
Participant Identification Process for Care Management Programs Using Condition-Specific Cut Points”, POPULATION HEALTH
MANAGEMENT, Volume 14, Number 0, 2011](https://image.slidesharecdn.com/hassession16howallinahealthusesanalyticstotransformcare-141030151954-conversion-gate01/75/How-Allina-Health-Uses-Analytics-to-Transform-Care-HAS-Session-16-23-2048.jpg)
Uploaded byHealth Catalyst
How Allina Health Uses Analytics to Transform Care - HAS Session 16
Allina Health is a large health system in Minnesota serving over 3 million patients annually. Facing rising healthcare costs and an unsustainable model, Allina transitioned to focus on higher value care through advanced analytics. Allina developed predictive models and dashboards to identify at-risk patients, coordinate care more effectively, and drive improvements in outcomes like reducing readmissions. By leveraging enterprise data and strategic use of analytics, Allina has achieved leading quality scores and cost reductions while better supporting providers to improve population health.
More Related Content
Similar to How Allina Health Uses Analytics to Transform Care - HAS Session 16
More from Health Catalyst
How Allina Health Uses Analytics to Transform Care - HAS Session 16
- 1. Session #16: HowAllina Health Uses Analytics to Transform Care Penny Ann Wheeler, MD President and Chief Clinical Officer, Allina Health
- 2. ADVANCING CARE THROUGHANALYTICS THE ALLINA HEALTH JOURNEY Penny Wheeler, M.D. President and Chief Clinical Officer September 2014
- 3. Key Questions •Who is Allina Health? • Why change? • What are the new measures of success? • What’s needed to move to higher value care? • How do we use advanced analytics to drive improvement? • What are our results thus far and lessons learned? 3
- 4.
- 5. Allina is theRegion’s Largest Health Care Organization • 13 Hospitals • 82 Clinic sites • 3 Ambulatory care centers • Pharmacy, hospice, home care, medical equipment • 26,000 employees • 5,000 physicians • 2.8 million+ clinic visits • 110,000+ inpatient hospital admissions • 1,658 staffed beds • 3.4B in revenue • 32% Twin Cities market share 5
- 6. The Imperative forChange: The Traditional Healthcare Model is Broken Representative timeline of a patient’s experiences in the U.S. health care system http://www.iom.edu/~/media/Files/Activity%20Files/Quality/LearningHealthCare/Release%20Slides.pdf
- 7. Why Change? Iffood prices had risen at medical inflation rates since the 1930s *Source: American Institute for Preventive Medicine 2009 1 dozen eggs $85.08 1 pound apples $12.97 1 pound sugar $14.53 1 roll toilet paper $25.67 1 dozen oranges $114.47 1 pound butter $108.29 1 pound bananas $17.02 1 pound bacon $129.94 1 pound beef shoulder $46.22 1 pound coffee $68.08 10 Item Total $622.27 7
- 9. All About CreatingValue… 9 Value = Good / Cost “Quality improvement is the most powerful driver of cost containment.” - Michael Porter, PhD Economics Harvard Business School
- 10. Preventable Complications UnnecessaryTreatments Inefficiency Errors Services That Add Value 40% Waste 60% Value All Services Add Value 100% Value Future Now What We Pay For… 10
- 11. Poll Question #1 In your opinion, which of the 4 categories of waste is the most important to address by the healthcare industry? a) Preventable Complications b) Unnecessary Treatments c) Inefficiency d) Errors
- 12. Four Measures ofSuccess: Allina Health 2016 Strategic Outcomes 1. Patient Care/Experience 2. Population Health 3. Patient Affordability 4. Organizational Vitality 12 Better Care/ Experience Better Health Reduce per capita costs Organizational Vitality
- 14. Triple Aim IntegrationInitiatives Quality Roadmap Goal Initiative(s) 1) Perform under payment for quality and value models Accountable care pilots • Pioneer ACO • Commercial partnerships 2) Align incentives across employed and affiliated providers Allina Integrated Medical Network 3) Give providers the data and information needed to improve outcomes Advanced analytics infrastructure including a robust Enterprise Data Warehouse (EDW) 4) Provide consistently exceptional care without waste • Primary care team model redesign • Care management/patient engagement • Clinical program optimization 5) Support transformation with new skills development Allina Advanced Training Program
- 15. Allina Health EnterpriseHealth Management Platform Transitioning Data to Actionable Information
- 16. Bridging Historical, Current,and Predictive Information Selected Health Intelligence & Delivery Tools at Allina PPR Dashboard “Potentially Preventables” Census Dashboard Enterprise Data Warehouse Reporting Workbench Retrospective Real time Predictive What happened? What is happening? What may happen? General Specific Readmissions Model Modeling of Potentially Preventable Events
- 17. Poll Question #2 For healthcare providers, on a scale of 1-5, how well do you feel you are using predictive information to address potentially preventable events? 1) No use 2) Just starting or sporadic use 3) Moderate use but increasing 4) Good use 5) Very strong use 6) Unsure or not applicable
- 18. Example: Supporting CareCoordination Predicting Unnecessary Admissions and Readmissions Challenge – Substantially reduce unnecessary admissions and readmissions Solution – Predict patients at high risk for unnecessary admissions and readmissions – Develop and use census dashboard to identify and manage patients – Prioritize care coordination and clinical interventions based on risk level – Predictive model C-statistic of 0.729 Results – Reduced readmissions for patients who received transition conferences (June 2013-June 2014) • High-risk patients: 15.8% decrease in readmissions • Moderate-high-risk patients: 5.4% decrease in readmissions
- 19. Getting the Modelto the Bedside The Census Dashboard Identifies Patient Readmit Risk Identifies Transition Conference Status Identifies Prior IP Visits in Last Week & Month
- 20. 20 Allina Results:Heart Failure 25% 20% 15% 10% 5% 0% Combined Metro Combined Metro Linear (Combined Metro) 2011 Q12011 Q22011 Q32011 Q42012 Q12012 Q22012 Q32012 Q42013 Q12013 Q22013 Q32013 Q4
- 21. RARE Campaign Graphprovided by ICSI 21
- 22. The Readmission ModelResults: How are our patients grouped? • High Risk: – 20 – 100% Readmission Risk: 7% of population • Moderate-High Risk: – 10 – 20% Readmission Risk: 19% of population • Moderate Risk: – 5 – 10% Readmission Risk: 35% of population • Low Risk: – 0 – 5% Readmission Risk: 39% of population 22 0% to 5% 5% to 10% 10% to 15% 15% to 20% 20% to 25% 25% to 35% 35% to 80% 45% 40% 35% 30% 25% 20% 15% 10% 5% Percent of Total Patients 39% 35% 13% 6% 3% 3% 1% Percent of total Readmissions 14% 31% 22% 13% 9% 7% 5% 35% 30% 25% 20% 15% 10% 5% 0% 0% Percent of Total Readmissions Percent of Total Patients Model estimated percent probability of readmission
- 23. Predictive Model Confidence Why do we believe the Readmission Model? Comparing existing models with standard C-Statistic (Area under ROC Curve) measure of performance – Random coin toss selection: 0.5 – State-of-art techniques(ACG): (0.70 to 0.77)[1] – Current Allina technique: 0.861 Allina Model was found to have a precision* of ~ 0.9 *Precision is the fraction of Predicted patients that actually have a PPE. In this case, on a dataset in which it was tested about 90% of patients predicted by the model had a PPE. Note, this is different from sensitivity, which is the fraction of actual PPE instances that are predicted. 1 Shannon M.E. Murphy, MA, Heather K. Castro, MS, and Martha Sylvia, PhD, MBA, RN, “Predictive Modeling in Practice: Improving the Participant Identification Process for Care Management Programs Using Condition-Specific Cut Points”, POPULATION HEALTH MANAGEMENT, Volume 14, Number 0, 2011
- 24. Example: Basic CostCurve for Individual $9,000 $8,000 $7,000 $6,000 $5,000 $4,000 $3,000 $2,000 $1,000 $0 with a Major Hospitalization -20 -19 -18 -17 -16 -15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Months Before and After High Cost Event Healthways Data for Diabetics with heart Failure(blue line) 24 Point of traditional payer-based care management Point of predictive intervention Green: potential cost curve with predictive intervention
- 25. Example: Supporting CohortManagement Providing Care to Patients with Diabetes Challenge – Provide superior care for Allina Health’s diabetic population Solution – Identified and stratified diabetes cohorts using registries – Identified gaps in care for diabetes patients (e.g. A1c, blood pressure management) – Provided workflow capability for care teams to manage the population through ambulatory quality dashboard Results – Highest national score for Diabetes Care Quality Measure in 2012 of all CMS Pioneer ACOs – U.S. leader in management of diabetes patients and Diabetes Optimal Care results
- 26. Supporting Cohort Management Driving Improvement through Access to Information Select by patient, clinic, provider or any combination Filter by Pioneer Shows performance of composite measure components ACO Patients
- 27. Example: Supporting Wellness& Prevention Successfully Keeping Patients Well Challenge – Avoiding future illness is core to superior population health management Solution – Established and reported on optimal care scores for individuals – Identified gaps in care and accurately connected them to care teams to close gaps in care Results – Eliminated significant gaps in wellness screening and preventative care – Allina Health has achieved some of the best ambulatory optimal care scores in the nation through a focused clinician engagement strategy using the EHMP Colon Cancer Screening Optimal Care 76.0% 71.0% 66.0% 61.0% 88.0% 86.0% 84.0% 82.0% 80.0% 78.0% 76.0% 74.0% Mammogram Optimal Care Goal = 85% Jan-11 Mar-11 May-11 Jul-11 Sep-11 Nov-11 Jan-12 Mar-12 May-12 Jul-12 Sep-12 Nov-12 Jan-13 Mar-13 May-13 Jul-13 56.0% Colon Cancer Screening Optimal Care Goal = 73% Mammogram Optimal Care
- 28. Supporting Wellness &Prevention Ambulatory Dashboard MD Name Ability to focus on a specific provider or patient population Shows performance on optimal care and component measures with patient detail, provider name and clinic
- 29. Summary This isonly just the start… Lessons Learned – Pareto analysis of population data key for determining opportunity and focus – Consistent quality drives lower cost of care • Focus on waste / “unhelpful care variation” – Use predictive modeling to focus care management resources – Strengthen the patient/primary care team relationship – Keep the patient at the center of all decisions
- 30.
- 31. Transition from Volumeto Value Planning for the inflection point Payment Type Penetration FFS Global payment Other Time 100% 50% 5% • Retain patients (keepage) • Regulatory requirements • Manage risk progression • Payment reform • Increase volume • Maximize payment • Minimize cost • Meet regulatory requirements Today Transition Tomorrow Phase Objectives • Evolve priorities based on: • Contracts • Populations • Regulatory changes
- 32. Driving Improvement toAdvance Care The Clinical Program Infrastructure Clinical Program Infrastructure Clinical /Operational Leadership Team Regional and system wide physician, administrative and clinical operations leaders needed to implement best practice Information Management Infrastructure Measurement System Staff support personnel and systems necessary to measure clinical, financial and satisfaction outcomes for key clinical processes Implementation Support Staff and systems necessary to develop, disseminate, support and maintain the clinical knowledge base necessary to implement best practice
- 33. Translating Concept toAction Selection of Key Allina Health Initiatives Allina Integrated Medical (AIM) Network – Aligns 900+ independent physicians and 1,200 Allina Health employed physicians to deliver market-leading quality and efficiency in patient care – Clinical Service Lines (CSLs) – Provide consistently exceptional and coordinated care across the continuum of care and across sites of care. CSLs are physician-led, professionally-managed and patient centered. Medicare Pioneer ACO – Member of CMS Pioneer Pilot Demonstration – Above average performance for 25 of 33 quality performance measures, including the highest performer for 3 of the measures – Held the Pioneer ACO Population to 0.8% cost growth for 2012 Northwest Metro Alliance – A multi-year collaboration between HealthPartners & Allina Health in the Northwest Twin Cities suburbs focused on the Triple Aim and a learning lab for ACOs – Since the Alliance model was implemented, medical cost increases have been below the metro average for the past two years and cost increases were less than one percent for two years in a row – Expanded access to stress tests for ED patients with chest pain and prevented 480 low-risk chest pain inpatient admissions, saving an estimated $2.16 Million in 2012
- 34. Pioneer ACO SelectedFocus Areas Area of Focus Implemented Tactics Preventable Admissions & Emergency Department Visits • Applied risk stratification to provide outreach and support to patients at risk for preventable events through Advanced Care Team or Team Care resources • Outreach to patients who have not been seen, check treatment compliance and schedule visit • Using After-Visit-Summary instructions during patient follow-up care • Develop patient-centered goals • Provide social worker support if needed • Provide support for Advanced Care Planning Preventable Readmissions • Applied predictive tool to identify patients most at risk for readmission • Prepare integrated After-Visit-Summary and provide the patient w/a Discharge ‘Packet’ • Provider transitions • Care transitions intervention • Determine and leverage role of pharmacist • Patient education • Skilled nursing facility transitions Mental Health • Care coordination for high-risk patients • Assign a Primary Care Provider to each MH patient • Eliminate delayed access • Effective management of MH resources through patient prioritization • Efficient patient transitions Late Life Supportive Care • Redesigning care so that patient’s needs are documented and that caregivers including family are able to access, understand, and comply during the course of caring for the patient End Stage Renal Disease (ESRD) • Currently in process of reviewing potential opportunities with nephrologists
- 35. Results: Allina’s ElectiveInductions < 39 Weeks (%) 35% 30% 25% 20% 15% 10% 5% 0% 2009-01 2009-04 2009-07 2009-10 2010-01 2010-04 2010-07 2010-10 2011-01 2011-04 2011-07 2011-10 2012-01 2012-04 2012-07 2012-10 2013-01 2013-04 Allina Allina 2009 Baseline Allina 2013 Goal