UNOS Labs: Leveraging Behavioral Science to Improve Organ Allocation. Darren Stewart, M.S., Sr. Research Scientist United Network for Organ Sharing

Transcription

1 UNOS Labs: Leveraging Behavioral Science to Improve Organ Allocation Darren Stewart, M.S., Sr. Research Scientist United Network for Organ Sharing

2 Disclosure I have no financial relationships to disclose relevant to my presentation AND My presentation does not include discussion of offlabel or investigational drugs

3 Accept or refuse: Is this enough data to decide? Deceased donor Potential recipient Age Age Cause of death Cause of renal failure 3

4 Accept or refuse: Is this enough data to decide? Deceased donor Potential recipient Age = 74 Age = 29 Cause of death Cause of renal failure 4

5 Accept or refuse: Is this enough data to decide? Deceased donor Potential recipient Age = 74 Age = 29 Cause of death Cause of renal failure Medically urgent = YES 5

6 Accept or refuse: Is this enough data to decide? Deceased donor Potential recipient Age = 48 Age = 29 Cause of death Cause of renal failure Creatinine Medically urgent = NO 6

7 Accept or refuse: Is this enough data to decide? Deceased donor Potential recipient Age = 48 Age = 29 KDPI =99% Cause of death Cause of renal failure Creatinine Medically urgent = NO 7

8 Accept or refuse: Is this enough data to decide? Renal imaging Deceased donor CIT Local vs import Potential recipient CPRA Blood type PHS? Age Age = 29 KDPI =60% Cause of death Stable/inotropes Creatinine Other labs HLA Mismatch EPTS Cause of renal failure Time on dialysis, waitlist Medically urgent = NO 8

9 Are the right decisions being made on behalf of patients? 9

10 An Effective Organ Offer System Should 1. Minimize cognitive burden (information overload) 2. Induce accept/refuse decisions that will most benefit patients 3. Help speed up decision-making 10

11 Coming attractions DonorNet Mobile 2.0 New, decisionmaking dashboard 11

12 Virtual center for innovation and research in transplantation Behavioral Science Big Data Information System Innovations Experimental Design Predictive Analytics Translational Research Optimize User Experience Decision Support Tools Machine Learning Idea Incubator Collaboration with the most innovative minds in healthcare, technology and research 12

13 Behavioral Science: Offer Simulation Studies SimNet, a mock DonorNet system

14 Behavioral Science: Offer Simulation Studies Clinical Decision-Making Dashboard Predicted survival without transplant: 5.7 years Predicted survival with this transplant: 10.2 years Predicted survival benefit? Yes (4.5 years) Predicted 1-year graft survival probability: 93% Predicted time to next KDPI<53% offer? days

15 Transplantation

16 AJT

17 Patient-specific, Net Benefit Decision Tools High KDPI benefit ECD benefit Massie, et al, AJT 2014 Merion, et al, JAMA Chow, et al, AJT 2013 PHS IR benefit 17

18 18

19 The missing piece in the dialectic around artificial intelligence and machine learning in health care is understanding the key step of separating prediction from action and recommendation. Clinicians should seek a partnership in which the machine predicts (at a demonstrably high accuracy), and the human explains and decides on action. a black box model can lead physicians to good decisions but only if they keep human intelligence in the loop, bringing the societal, clinical, and personal context. 19

20 Don t Fear Intelligent Machines work with them 20

21 Concluding Thoughts 1. Transplant acceptance decisions are complex 2. Thousands of potentially usable kidneys are discarded each year 3. UNOS Labs was created to test potential system enhancements for better, more rapid decisions 4. Predictive analytics and machine learning have the potential to improve evidence-based decision making 21

22 What Do You Think? 22

23 Interactive polling question #1 How should UNOS respond to requests to add additional data to DonorNet to support organ utilization decisions? a. Don't add it. There's too much data already. b. Just add it. The more data to help inform decisions, the better. c. Have a committee evaluate the merits of the information before potentially displaying it. d. Test the information in a lab environment first to determine how best to display it and to assess behavioral effects e. Both c) and d) 23

24 Interactive polling question #2 Would you trust well-vetted, patient-specific, clinical predictions and decision aids (e.g., expected recipient survival with this particular organ) to help your offer acceptance decision-making? a. Yes, I would rely on them heavily in my decision-making b. Yes, but I would only rely on them minimally c. No, I would not trust them nor use them to inform in my decision-making 24

25 A shout out to 25

26 Industry Expertise: Behavioral Science; Research

27 UNOS Labs Clinician Insider Group (CIG) Carl Berg, MD (Duke) Rich Formica, MD (Yale) David Klassen, MD (UNOS) Kristin Mekeel, MD (UC-San Diego) Lloyd Ratner, MD, MPH, FACS (Columbia) Bob Stratta, MD (Wake Forest)

28 UNOS Research Department Data Quality Team Brooke Chenault, CCRA Jessie Maker, MS Tammy Snead Tenisha Atobrah Penny Brock Sharon Graves Clay McDonald Jo Smith Data Scientists Harris McGehee Subhash Jaini, MS Project Leaders Alex Garza Heather Neil Data Analytics Team Eric Beeson Julia Barnhart, MS Tim Baker Jude Maghirang, MS Yulin Cheng John Beck, MSME Sherri Williams Denise Tripp Ann Harper, MPH Wes Rosson Alice Toll Aparna Sadavarte, MS Jeremy Winslow Questions? Scientists and Analysts Wida Cherikh, PhD John Rosendale, MS Darren Stewart, MS Amber Wilk, PhD Jennifer Wainright, PhD Anna Kucheryavaya, MS Read Urban, MPH Amanda Robinson, MS Rebecca Lehman, PhD Samantha Noreen, MS Gabe Vece, MSPH Chris Custer, MS Victoria Garcia, MPH Leadership Ryan Ehrensberger, PhD, FACHE David Klassen, MD Sarah Taranto Bob Carrico, PhD Cherri Taylor