Managing Transfusion Medicine with Predictive Analytics March 1, 2016

Transcription

1 Managing Transfusion Medicine with Predictive Analytics March 1, 2016 Ben Cleveland, Data Scientist, UnityPoint Health Rhiannon Harms, Executive Director, Strategic Improvement and Planning

2 Conflict of Interest Ben Cleveland, MSHS Has no real or apparent conflicts of interest to report. Rhiannon Harms, CNMT, R.T.(N) Has no real or apparent conflicts of interest to report.

3 Agenda Background: Delivering Value Objective: Identify Excessive Transfusions Strategy: Risk Adjustment of Red Blood Cell (RBC) Use Execution: Patient-Level Modeling Deployment: Dyad Partnerships

4 Learning Objectives Describe a novel approach for risk-adjusting blood product resource utilization through data modeling in healthcare at the provider level Describe the value of implementing predictive analytics in the inpatient setting to reduce costs and increase quality of care Describe a multi-faceted approach to help ensure responsible use of blood products

5 An Introduction of How Benefits Were Realized for the Value of Health IT Quality Clinical Outcomes Patient and System Savings

6 UnityPoint Health

7 Delivering Value to the Patients We Serve

8 Restrictive Transfusion Practice Recent clinical trials indicate: Patients can tolerate lower hemoglobin levels than previously thought No additional adverse affects May have reduced morbidity and mortality Hebert PC, et al. N Engl J Med 1999; 340: Lacroix J, et al. N Engl J Med 2007; 356: Carson JL, et al. N Engl J Med 2011; 365:

9 The Early Interventions Used to Precede the Advanced Analytics Work Education on AABB Standards Led by Lab Directors and Pathologists Reporting of historical, descriptive utilization patterns By Hospital, APRDRG, and Attending Provider Build buy-in on clinical decision support solutions Best Practice Advisory

10 Audience Poll Your providers most common reaction to utilization data? A. But my patients are sicker. B. No way. That can t be right! C. Those aren t my patients. D. Wait, what? Tell me more!

11 Audience Poll Your providers most common reaction to utilization data? A. But my patients are sicker. B. No way. That can t be right! C. Those aren t my patients. D. Wait, what? Tell me more!

12

13

14 Risk Adjustment Risk adjustment describes methods for determining whether patient characteristics will necessitate higher utilization of medical services. -American Medical Association

15 Utilization Variance Transfusion Decision Variation Variation in Patient Factors Variation in Provider Practice

16 Objective

17 Risk Adjustment Methodology

18 Expected RBC Use Pr Transfusion Expected Units Transfusion) Too Often or Too Much x Hemoglobin Level

19 Detail Level of the Data Attending Physician Aggregate data on inpatient stay Ordering Physician Ordering Physician Ordering Physician Ordering Physician Detail data on individual transfusion decisions

20 Patient Factors Governing Transfusion Decisions Hemoglobin Levels (g/dl) Diagnostic Related Group Factors Influencing Transfusion Decision Severity of Illness Gender & Age

21

22 Modeling Methods: Logistic Regression Models the probability of an event happening given the values of different variables Can also observe what type (positive or negative) and how much of influence different variables have on the probability of an event In our case, the event we re predicting is a transfusion, and the variables we re studying are hemoglobin values, severity of illness, age, and gender

23 Logistic Regression Output

24

25 Current State Minimum Hb Level Transfusion Probability >10 Very Unlikely 8-10 Unlikely Current Threshold 7-8 Likely 6-7 Very Likely <6 Required

26 Future Threshold Minimum Hb Level Future State Transfusion Probability >10 Unnecessary 8-10 Very Unlikely Current Threshold 7-8 Unlikely 6-7 Very Likely <6 Required

27 To Recap Understanding appropriate utilization patterns is very difficult with descriptive statistics at virtually all organizational levels Risk-Adjustment at the patient-level allows you to drill into potentially inappropriate utilization at any level of your organization

28

29

30

31 Risk-adjustment indicates patients received more RBC units than expected given their case mix

32 Aggregate at any Organizational Level

33 Sharing the Results and Driving Practice Pattern Changes

34 Top percentiles show high system agreement with transfusion decisions Bottom percentiles show low system agreement with transfusion decisions

35 A Summary of How Benefits Were Realized for the Value of Health IT Quality Clinical Outcomes - Estimated 4,800 unit reduction over 2 years Patient and System Savings - Estimated $2 Million

36 Questions Ben Cleveland Rhiannon Harms