Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
SlideShare a Scribd company logo

UX STRAT Online 2021 Presentation by Dr. Jofish Kaye, Anthem

•
•

These slides are for the following session presented at the UX STRAT Online 2021 Conference: "Strategy & Organization for AI & UX in Healthcare" Dr. Jofish Kaye Anthem: Senior Director of Interaction Design & AI

1 of 29
Download to read offline
Strategy & Organization for AI & UX in Healthcare Bridging AI & UX DePaul Showcase 6 November 2021 Jofish Kaye, Ph.D Senior Director of IxD & AI Anthem Health Platforms
Nokia Research Center B.S Brain & Cognitive Science MIT M.S. Media Arts & Science, Media Lab Ph.D Information Science Cornell Consulting Assistant Professor Stanford d.School, CS Yahoo Labs & Flickr Mozilla Anthem Health Platforms Co-Chair, CHI 2016 A quick introduction
Strategic Recommendations for AI/UX in Health Health UX AI
4 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
5 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
UX Recommendations 6 Health UX AI
7 Jakob Nielsen’s Ten Usability Heuristics
AI recommendations 8 Health UX AI
9 TRIPOD Checklist: Prediction Model and Validation
Health recommendations 10 Health UX AI
11 Sperl-Hillen et al (2010): Outpatient EHR-Based Diabetes Clinical Decision Support That Works: Lessons Learned From Implementing Diabetes Wizard. Diabetes Spectrum 2010 Jul; 23(3): 150-154. https://doi.org/10.2337/diaspect.23.3.150
12 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
Case study: deep learning for diabetes 13 Health UX AI
A few facts about diabetes • 34M Americans have diabetes, ~95% T2D, similar levels globally • 1/4 of all healthcare spend in the US • Diabetes is measured by a blood test, HbA1c Non-diabetic <5.6% Prediabetic 5.6%-6.5% Diabetic >6.5% • People with diabetes are at higher risk for blindness, kidney failure, heart disease, stroke, amputation and premature death. 14 https://www.cdc.gov/diabetes/basics/type2.html
Deep learning for diabetes 15 Paper under review; preprint available
16 “Figure S2. Forest plot results of NMA for clinical subgroup 1 Average treatment effect of medication combinations for insulin naïve, under 65 and CCI<=2 stratum. Treatment cohort size is annotated (sz).”
17 “Figure S2. Forest plot results of NMA for clinical subgroup 1 Average treatment effect of medication combinations for insulin naïve, under 65 and CCI<=2 stratum. Treatment cohort size is annotated (sz).”
Recommendations for AI / UX in Health 18 Health UX AI
1. Show Context 19 ✓ X
2. Show your work 20 Factors influencing the probability of hospital admission. Baseline is 12%; this patient has a 29% likelihood.
3. Respect Expertise 21 X% of doctors prescribed a beta blocker for patients like this. Learn more Data on Y0,000 patients like this suggests ACE inhibitors work Z% better than beta blockers on similar patients. Learn More ✓ ✓ High blood pressure: Recommend beta blocker. X Chart: https://www.myhealth.va.gov/mhv-portal- web/ss20190822-track-health-blood-pressure. Recommendations are entirely faked. This is not medical advice. Not that kinda doctor.
4. Predict and forecast 22 Note: This example does show uncertainty in predictions, but inaccurately shows the relative probabilities, only the span of possibilities. 0 50 100 150 200 250 300 1-Jun-16 1-Sep-16 1-Dec-16 1-Mar-17 1-Jun-17 1-Sep-17 1-Dec-17 1-Mar-18 1-Jun-18 1-Sep-18 1-Dec-18 1-Mar-19 1-Jun-19 1-Sep-19 1-Dec-19 1-Mar-20 1-Jun-20 1-Sep-20 1-Dec-20 1-Mar-21 1-Jun-21 1-Sep-21 1-Dec-21 Blood Lipids Rosuvastatin 5mg Total Serum HDL LDL Triglycerides Note: This example shows predictions, but only the highest probability.
23 So what does that mean strategically and organizationally? 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast
Epistemology: Ways of knowing 24 Health UX AI
Ways of knowing in UX 25 Health UX AI
Ways of knowing in AI 26 Health UX AI
Ways of knowing in health 27 Health UX AI
UX Strategy: Ways of knowing 28 Health UX AI
Thank you! Jofish Kaye, Ph.D Anthem Health Platforms @jofish linkedin.com/in/jofish

More Related Content

UX STRAT Online 2021 Presentation by Dr. Jofish Kaye, Anthem

  • 1. Strategy & Organization for AI & UX in Healthcare Bridging AI & UX DePaul Showcase 6 November 2021 Jofish Kaye, Ph.D Senior Director of IxD & AI Anthem Health Platforms
  • 2. Nokia Research Center B.S Brain & Cognitive Science MIT M.S. Media Arts & Science, Media Lab Ph.D Information Science Cornell Consulting Assistant Professor Stanford d.School, CS Yahoo Labs & Flickr Mozilla Anthem Health Platforms Co-Chair, CHI 2016 A quick introduction
  • 3. Strategic Recommendations for AI/UX in Health Health UX AI
  • 4. 4 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
  • 5. 5 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
  • 11. 11 Sperl-Hillen et al (2010): Outpatient EHR-Based Diabetes Clinical Decision Support That Works: Lessons Learned From Implementing Diabetes Wizard. Diabetes Spectrum 2010 Jul; 23(3): 150-154. https://doi.org/10.2337/diaspect.23.3.150
  • 12. 12 What’s the role of UX in AI & Strategy? (or: where are we going?) Recommendations in UX, AI, Health Case study: UX of AI Recommendations in Health 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast Ways of knowing in UX, AI, & Health
  • 13. Case study: deep learning for diabetes 13 Health UX AI
  • 14. A few facts about diabetes • 34M Americans have diabetes, ~95% T2D, similar levels globally • 1/4 of all healthcare spend in the US • Diabetes is measured by a blood test, HbA1c Non-diabetic <5.6% Prediabetic 5.6%-6.5% Diabetic >6.5% • People with diabetes are at higher risk for blindness, kidney failure, heart disease, stroke, amputation and premature death. 14 https://www.cdc.gov/diabetes/basics/type2.html
  • 15. Deep learning for diabetes 15 Paper under review; preprint available
  • 16. 16 “Figure S2. Forest plot results of NMA for clinical subgroup 1 Average treatment effect of medication combinations for insulin naïve, under 65 and CCI<=2 stratum. Treatment cohort size is annotated (sz).”
  • 17. 17 “Figure S2. Forest plot results of NMA for clinical subgroup 1 Average treatment effect of medication combinations for insulin naïve, under 65 and CCI<=2 stratum. Treatment cohort size is annotated (sz).”
  • 18. Recommendations for AI / UX in Health 18 Health UX AI
  • 20. 2. Show your work 20 Factors influencing the probability of hospital admission. Baseline is 12%; this patient has a 29% likelihood.
  • 21. 3. Respect Expertise 21 X% of doctors prescribed a beta blocker for patients like this. Learn more Data on Y0,000 patients like this suggests ACE inhibitors work Z% better than beta blockers on similar patients. Learn More ✓ ✓ High blood pressure: Recommend beta blocker. X Chart: https://www.myhealth.va.gov/mhv-portal- web/ss20190822-track-health-blood-pressure. Recommendations are entirely faked. This is not medical advice. Not that kinda doctor.
  • 22. 4. Predict and forecast 22 Note: This example does show uncertainty in predictions, but inaccurately shows the relative probabilities, only the span of possibilities. 0 50 100 150 200 250 300 1-Jun-16 1-Sep-16 1-Dec-16 1-Mar-17 1-Jun-17 1-Sep-17 1-Dec-17 1-Mar-18 1-Jun-18 1-Sep-18 1-Dec-18 1-Mar-19 1-Jun-19 1-Sep-19 1-Dec-19 1-Mar-20 1-Jun-20 1-Sep-20 1-Dec-20 1-Mar-21 1-Jun-21 1-Sep-21 1-Dec-21 Blood Lipids Rosuvastatin 5mg Total Serum HDL LDL Triglycerides Note: This example shows predictions, but only the highest probability.
  • 23. 23 So what does that mean strategically and organizationally? 1. Show Context 2. Show your work 3. Respect expertise 4. Predict and forecast
  • 24. Epistemology: Ways of knowing 24 Health UX AI
  • 25. Ways of knowing in UX 25 Health UX AI
  • 26. Ways of knowing in AI 26 Health UX AI
  • 27. Ways of knowing in health 27 Health UX AI
  • 28. UX Strategy: Ways of knowing 28 Health UX AI
  • 29. Thank you! Jofish Kaye, Ph.D Anthem Health Platforms @jofish linkedin.com/in/jofish