Download as PDF, PPTX
Uploaded byJoyce Lee
When Perfect Algorithms Meet Imperfect Healthcare Systems: My talk at the Machine Learning for Healthcare Meeting 2018
This document discusses the intersection of diabetes data and healthcare systems, emphasizing the barriers patients face in managing diabetes effectively, such as trapped, infrequent, and confusing data. It highlights the need for improved technology and support to enable better patient engagement and decision-making, ultimately driving innovation in diabetes care. A focus on human-centered design, patient education, and collaborative approaches is proposed to overcome the challenges and enhance health outcomes.
More Related Content
Similar to When Perfect Algorithms Meet Imperfect Healthcare Systems: My talk at the Machine Learning for Healthcare Meeting 2018
When Perfect Algorithms Meet Imperfect Healthcare Systems: My talk at the Machine Learning for Healthcare Meeting 2018
- 1. When Perfect Algorithms Meet Imperfect HealthcareSystems Joyce Lee, MD, MPH Robert Kelch Professor of Pediatrics doctorasdesigner.com @joyclee on Twitter
- 2.
- 3. www.doctorasdesigner.com medium.com/@joyclee Professor, Robert KelchProfessor of Pediatrics Clinical/Health Outcomes Research (Obesity/Diabetes) Learning Health Systems (Clinical Informatics, Quality Improvement, Participatory Design) Applications of Emerging Technologies in Healthcare (Ethnography, Analysis of Diabetes Online Maker Communities) @joyclee
- 4. @joyclee Agenda Why Diabetes DataMatters Diabetes Data Barriers Patient Innovation with Data
- 5.
- 6. @joyclee It’s an infiniteamount of math over a lifetime
- 7. It’s an infiniteamount of math over a lifetime @joyclee
- 9. Blood Glucose Meters Continuous Glucose Monitoring(CGM) Systems Insulin Pumps
- 10. @joyclee Glucose levels changeall the time Puberty Growth Activity
- 11. Glucose levels impacthealth outcomes
- 12. Management is trulypatient-driven Provider Time managing diabetes (1 hour in a year?)Patient Time Spent Managing Diabetes
- 13. Data Nirvana forCars @joyclee “The Autonomous Car”
- 14.
- 15. @joyclee “The Artificial Pancreas” Continuous GlucoseMonitor Insulin Pump Mobile Phone with Algorithm Data Nirvana for Diabetes
- 16. The Singularity! Goodbye PediatricEndocrinologist! You have been eliminated! @joyclee
- 17.
- 18. @joyclee “The Artificial Pancreas” “Closingthe Loop" Medtronic 670G DIY In Development
- 19. What are the DiabetesData Barriers? @joyclee
- 20. No data Trapped data Infrequentdata Isolated data Confusing data Biased data Judgmental data Suspicious data @joyclee
- 21.
- 22. We still giveout mostly analog tools ”It was unacceptable to me in 2002, when my son was diagnosed, to be given needles, an insulin vial, and a piece of paper.” - Jeff Brewer, Bigfoot CEO @joyclee
- 23. Blood Glucose Meters(100%) Insulin Pumps (50%) Continuous Glucose Monitoring (CGM) Systems (15%) Connected Pens (<1%) AP Systems (<1%) Device Adoption Rates (Type 1 Diabetes) @joyclee
- 24. Diabetes tools arevery expensive $1411 Starter Kit $596 Transmitter (every 3 months) $349/month for Sensors $70-80 Starter Kit $120-150/month for Sensors @joyclee
- 25. Insurance companies don’tlike to pay for things like blood glucose test strips, connected pens, and CGMs
- 26. Burnout + Stigma + Burden+ Frustration = No BG data @joyclee
- 27.
- 28.
- 29.
- 30.
- 31. People Prefer Mobileto Desktop @joyclee
- 32.
- 33. Downloads and diabetesdecision- making only happens at clinic 3-4 visits per year with 15 minutes per visit spent on data Meter/pump/CGM is collected at the clinic visit and a data download is exported as a PDF and scanned into the media tab of the chart @joyclee
- 34. Mental Health Issues 20,028 callsin 2017 @joyclee School Forms Supplies Prior Authorization Insurance Child Protection Services DME Pharmacy
- 35. Patient data reviewoutside of clinic is reactive, not proactive Reporting blood glucose numbers over the phone PDF attached to a Portal Message (page limits) Fax (email-fax) 48 hour turnaround time
- 36.
- 37.
- 38. We get upsetwhen the Color Printer is not working in clinic @joyclee
- 39.
- 40. Technology is aboutculture change ”We’re living through this time right now where technology is a Trojan Horse for change. We say technology, but we mean innovation. We say interoperability and open data, but we mean culture change.” -Susannah Fox @joyclee
- 41. Onboarding Patients tothe Patient Portal and Diabetes Data Platforms is no one’s job Not the Health IT Specialists Not the Medical Assistants Not the Educators Not the Doctors @joyclee
- 42. There is noformal patient education focused on how to download data, how to interpret it, and how to use it to adjust insulin doses “Am I supposed to still keep a logbook?” “I don’t feel comfortable making dose adjustments without first consulting with the CDE or endocrinologist. I mean, it’s his life, you know?” @joyclee
- 43. “Routine Downloaders” =Downloaded data at least once between routine clinic visits every 3 months which was four or more times in the past year (40%) “Routine Reviewers’’ = Reviewed the data at least ‘‘most of the time’’ he/she downloaded (27%) Lower A1c for Routine Reviewers: 7.8% vs. 8.6% (p=0.001) Diabetes Data Platform Use Wong et al, 2015 @joyclee
- 44. There is alack of standardization for data visualizations @joyclee
- 45. @joyclee There is alack of standardization for data visualizations
- 46.
- 47. @joyclee Are they coveringall carbs? Are they carb counting carefully? Are they inputting all the carbs / BG values into the pump? Are they bolusing before or after meals? Can I trust the data? Insulin lasts 2-3 hours There are A Lot of Unmeasured Variables
- 48.
- 49. Guilt Loss of Control Alienation Obsession Datais not always empowering Isolation @joyclee CSCW 2017, Kaziunas, Ackerman, Lindtner, Lee
- 50.
- 51. Patients and cliniciansdon’t trust a black box “I don’t know what it’s doing so how can I trust it?” “I wouldn’t give up my DIY AP” @joyclee
- 52. And that doesn’t eveninclude linking this vital diabetes data to the clinical EHR data! @joyclee “Garbage data”
- 53. Lack of User-CenteredDesign for the EHR No culture of human-centered design in Health IT Design without a strategic understanding of what metrics are needed to improve care Build without enough input from users Deploy without iteration and testing Physician Resentment/Anger @joyclee
- 54. Clinical EHR: Acombination of Microsoft Word and Pinterest @joyclee Clinicians are inputting data in unstructured format in the notes Data is being lost and/or underutilized Patient paper questionnaires and the diabetes data are scanned to PDF
- 55. @joyclee No data Trapped data Isolateddata Infrequent data Confusing data Suspicious data Garbage data PDFMediaTab
- 56. How do weaddress these Diabetes Data Barriers? @joyclee
- 57.
- 58.
- 59. @joycleeKumar RB etal J Am Med Inform Assoc. 2016 May;23(3):532-7.
- 60.
- 61. Patients, caregivers, cliniciansand researchers work together to choose care based on best evidence; together they drive discovery as natural outgrowth of patient care; and ensure innovation, quality, safety and value, all in real-time. - C3N Project @joyclee “
- 62. Design for Users Quality Improvement Implementation, Sustainability, Outcomes Human-centered Design DataOutcomes Health IT @joyclee
- 63. Aim: To decreasethe % of the population with HbA1c ≥ 9% and increase the % of the population with ≥ 0.5% HbA1c interval improvement Preference driven treatment and effective self- management Enhanced registry population management & Pre-visit planning Peer/community support Education/training to support technology use and patient viewing and problem solving with blood glucose data between visits Interventions/toolkits for addressing barriers to adherence Efficient use of technology and data to support care Access to care and regular follow-up Screening for depressionPsychosocial Support Shared decision making Partnership between engaged patients and the care team Effective use of EHR by diabetes team for population management • % of pts. testing ≥4 times/day or using CGM (6/7 days/week) • % of pts. giving 3 or more short-acting boluses/day • % of pts reviewing data between visits • % pts setting, documenting, and reviewing goals • % completed pre-visit planning • % with ≥ 4 visits per year • % of pts with annual CDE/RD/SW visit • % of pts on case mgmt. pathway • % pts screened for depression Developing a Clear Measurable Aim and a Theory of Change Care Process Measures @joyclee
- 64. Local Infrastructure tosupport an LHS Team (Director, Associate Director, Patient Partner/Advisor, Project Manager/Analyst) Patient Engagement (Patient Advisor/Advisory Board, Website/Newsletter) QI interventions (Depression Screening, High-risk Patient Recall, Portal Onboarding, Data Engagement Curriculum) Improving Data/Technology Systems @joyclee
- 65.
- 66.
- 67.
- 68. @joyclee “No one isgoing to use that tool if you can’t BOLD the text!” Rogue commas Tool for Data Input Insulin sensitivity 12AM 90 2:30 AM 110 4 AM 230* 10 AM 160 Patient Instructions Insulin sensitivity 12AM 12AM 90 2:30 AM 110 4 AM 23, 0* 10 AM 160
- 69. @joyclee MediaTab Outcomes that Matter Toolsfor Structured Data Collection Patient Reported Outcomes (Portal Questionnaires) Clinical Interface Redesign Tools for Population Management
- 70. How might our machinelearning colleagues help? @joyclee
- 71. Understand how, when,and what data is desired by different stakeholders to solve the right problems Think ecosystems not apps @joyclee
- 72. Population Level Monitoring Prioritizationof high-risk patients Tools for Seamless Data Acquisition Data Pattern Review Documentation Communication
- 73. Watch patient expertssolve problems @joyclee
- 74. Who’s the real medicalexpert? @joyclee
- 76.
- 77.
- 78.
- 79. 21. France 22. Netherlands 23.New Zealand 24. Mexico 25. Croatia 26. South Africa 27. Israel 28. Japan 29. Switzerland 30. Hungary 31. Belarus 29,000+ Int’l 11. Australia 12. Portugal 13. Denmark 14. Germany/Austria 15. Turkey 16. Russia 17. Romania 18. Korea 19. Ireland 20. Brasil 55,000+ worldwide 1. Italy 2. UK 3. Spain 4. Sweden 5. Bulgaria 6. Norway 7. Poland 8. Czech+Slovakia 9. Canada 10. Finland
- 80.
- 81.
- 82. How do youget your CGM in the Cloud? @joyclee
- 83. “The DIY ArtificialPancreas” Loop/Loopkit OpenAPS AndroidAPS @joyclee
- 84. Interoperability and choicematter Reduce the mental burden Proactive not reactive Be transparent What should I consider and why? Give access to all of the data @joyclee
- 85. The machine isnot an end. An airplane is not an end: it is a tool. Tools are created to allow you to reach greater goals, and machines should not distract from this pursuit. In fact, you should barely notice that they’re there. @joyclee “
- 86. Jacob Dwyer, AshleyGarrity, Valeria Gavrila, Emily Hirschfeld Dorene Markel, Ram Menon, Amy Ohmer, Lilia Verchichina, Michelle Wichorek, Pediatric Diabetes Team, The Nightscout Foundation, T1D Exchange Acknowledgements @joyclee Joyce Lee, MD, MPH doctorasdesigner.com [email protected] http://www.nightscout.info/ https://www.nightscoutfoundation.org/contribute-data/ AHRQ, R21HS023865 and PCORI Engagement Award (1442-UMich)