Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

OPPORTUNITY AND IMPACT OF AI IN MEDICINE AND HEALTH DELIVERY

696 views

Published on

Artificial Intelligence and Machine Learning are transforming the work of human labor. Healthcare professionals will see their work transformed and augmented with this technology, but the manner in which these changes will occur is nuanced. In this presentation, I will explore the manner in which the labor of healthcare will be transformed, review evidence to support this prediction, and remark on the changes already underway.

Published in: Health & Medicine
  • Be the first to comment

OPPORTUNITY AND IMPACT OF AI IN MEDICINE AND HEALTH DELIVERY

  1. 1. 1 Robert Mittendorff MD MBA, Partner, Norwest Venture Partners 11 January 2018 OPPORTUNITY AND IMPACT OF AI IN MEDICINE AND HEALTH DELIVERY
  2. 2. 2 Robert Mittendorff MD (2018) AI in Healthcare Speaker Introduction Robert Mittendorff, MD, MBA Partner Norwest Venture Partners Venture & growth equity investor, and emergency physician focused on companies in the digital health, healthcare IT, healthcare services, and medical device and diagnostics spaces.
  3. 3. 3 Robert Mittendorff MD (2018) AI in Healthcare Conflict of Interest Robert Mittendorff MD MBA Ownership Interest via Norwest Venture Partners in HealthCatalyst, Omada Health, iRhythm, CareCloud, ClearData, TigerText, Crossover Health, iCardiac, Qventus (fka AnalyticsMD), TalkSpace, Visitpay The Views expressed herein are my own and are not attributable to any company investment or employer
  4. 4. 4 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  5. 5. 5 Robert Mittendorff MD (2018) AI in Healthcare AI: The Three Impacts • Leverage Human Labor to Increase Capacity to Treat by 10x-100x • Allow Patients to Self Manage Their Conditions With Persuasive Software • Enable New Interventions, Like Finding Problems in Real Time and Nudging Providers or Managers with Solutions
  6. 6. 6 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  7. 7. 7 Robert Mittendorff MD (2018) AI in Healthcare THE HORSE: HOW AUTOMATION IMPROVES EFFICIENCY From Dead Horses to the Model T Source: Library of Congress website (top left and bottom left, Mittendorff analysis of auto registration data and horse and mule population in the US from 1900 to 1950 Technology Substitution: US Autos and Horses Technology and Labor Substitutes: From Horse to Car
  8. 8. 8 Robert Mittendorff MD (2018) AI in Healthcare THE STORY OF THE MILKMAN: HOW TECHNOLOGY DISRUPTS JOBS Pasteurization Refrigeration in Home Ice Man Milk Man State and Federal Laws 1971 FDA requires all milk transported interstate to be pasteurized; …bye bye Milkman Source: Library of Congress website Technology and Labor Substitutes: Milk Man & Pasteur
  9. 9. 9 Robert Mittendorff MD (2018) AI in Healthcare Source: Library of Congress website Technology and Labor : H&R Block and Turbotax
  10. 10. 10 Robert Mittendorff MD (2018) AI in Healthcare TECHNOLOGY ADOPTION ALSO LEADS TO OBSOLESCENCE (PCI VS CABG) Technology and Labor Substitutes: Surgery To Stents
  11. 11. 11 Robert Mittendorff MD (2018) AI in Healthcare THE HORSE: HOW AUTOMATION IMPROVES EFFICIENCY Source: NHDS 2009; MarketRealist 2016. Labor is 50% of a Healthcare System’s Expense And Half of Admissions are Unscheduled (from ED) Share of Hospital Admissions by Source 50% = ER Percent of Hospital Operating Expenses By Source 49% = Salary
  12. 12. 12 Robert Mittendorff MD (2018) AI in Healthcare FEWER PROVIDER ORGANIZATIONS OWN MORE PHYSICIANS THAN EVER Physician Employees 2014 2000 30% 70% Source: Kaiser Family Foundation 2015 reports and NEJM Policy Center 2016 Labor Centralization Leads to Scalability
  13. 13. 13 Robert Mittendorff MD (2018) AI in Healthcare HITECH ACT FUELED THE FIRST WAVE OF HEALTH INFORMATION DIGITIZATION 1. Clinical and operational improvements at the system level requires digitization & structuring of data 2. Once information is digitized, analytics professionals and a culture of evidence based improvement must be created 3. Once an organization has the capability to organize around data, insight, and practice change, it must be continually rewarded for improving performance EMR Adoption By US Physician Practices INSERT Source: ONC for HIT, Health IT Dashboard, https://dashboard.healthit.gov Digitization Leads to Mass Customization of Interventions
  14. 14. 14 Robert Mittendorff MD (2018) AI in Healthcare A Little Background on Other Relevant Work Source: (left) (right) Brown, E, et al. Nature Neuroscience 2004. Multiple neural spike train data analysis: state-of-the-art and future challenge. Record from the Brain Use Neural Network Model Predict Location Based on Neural Signals
  15. 15. 15 Robert Mittendorff MD (2018) AI in Healthcare Human Neural Systems and The Practice of Medicine
  16. 16. 16 Robert Mittendorff MD (2018) AI in Healthcare Human Neural Systems and The Practice of Medicine Radiology Dermatology Endocrinology Emergency Medicine Cardiac Surgery
  17. 17. 17 Robert Mittendorff MD (2018) AI in Healthcare Human Neural Systems and The Practice of Medicine Radiology Cardiac Surgery
  18. 18. 18 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  19. 19. 19 Robert Mittendorff MD (2018) AI in Healthcare AI Will Be Everywhere
  20. 20. 20 Robert Mittendorff MD (2018) AI in Healthcare MORE THAN 90 HEALTHCARE AI STARTUPS HAVE BEEN FUNDED Source: CBInsights 2016 Hunting the Blockbuster in AI
  21. 21. 21 Robert Mittendorff MD (2018) AI in Healthcare AI is a Set of Technologies – It is Not a Company
  22. 22. 22 Robert Mittendorff MD (2018) AI in Healthcare “Gives computers the ability to learn without being explicitly programmed" - Arthur Samuel Machine Learning: Finding Patterns in Data
  23. 23. 23 Robert Mittendorff MD (2018) AI in Healthcare Hard vs. Easy Problems in Machine Learning
  24. 24. 24 Robert Mittendorff MD (2018) AI in Healthcare Train: Find the patterns in the data Price 50,000 100,000 150,000 200,000 250,000 300,000 1000 2000 3000 Size Fitted line plot Test: Let's use the patterns we have found to do the prediction House size Estimated price 2000 ? 2200 ? 1450 ? Train and Test in Machine Learning Models
  25. 25. 25 Robert Mittendorff MD (2018) AI in Healthcare Should the model focus on being right 100% of the time or it ok to be wrong sometimes but find more of the ‘high priced homes’? Price 50,000 100,000 150,000 200,000 250,000 300,000 1000 2000 3000 Size Fitted line plot Precision (PPV) and Recall (Sens) in Performance
  26. 26. 26 Robert Mittendorff MD (2018) AI in Healthcare WHAT IS ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING IN A NUTSHELL? Machine Learning (ML): The use of algorithms to structure, parse, and learn from data sets (statistical or other) patterns that can be used to classify or predict in novel data sets. Supervised learning algorithms (training data is labeled) Unsupervised learning algorithms (training data in unlabeled) Semi supervised learning algorithms (training data is a mix of labelled and unlabeled) (examples include Regularization /Elastic Net, Regression and Regression Trees, Nearest Neighbor, Decision Trees, Bayesian approaches, and 50+ more) Deep Learning (a ML technique): The use of (frequently) neural networks (NN) trained on data sets to recognize patterns and features without an explicit parametric model. Neural networks can then be used on novel datasets to predict or classify. AI, ML, Deep Learning, and the Jargon
  27. 27. 27 Robert Mittendorff MD (2018) AI in Healthcare WHAT IS ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING IN A NUTSHELL? Artificial Intelligence (AI): Engineering machines to think or mimic the thinking and operations of a human. Machine Learning is frequently considered a set of technologies found in an AI. In Healthcare we like evidence to demonstrate capabilities in a “VALIDATION DATASET” that the system has not been exposed to in training. AI, ML, Deep Learning, and the Jargon
  28. 28. 28 Robert Mittendorff MD (2018) AI in Healthcare A Simple Neural Network Example
  29. 29. 29 Robert Mittendorff MD (2018) AI in Healthcare Deep Learning and (Convolutional) Neural Networks
  30. 30. 30 Robert Mittendorff MD (2018) AI in Healthcare Searle and the Chinese Room: AI and Consciousness
  31. 31. 31 Robert Mittendorff MD (2018) AI in Healthcare AI in Clinical Decision Support: Must It Explain WHY?
  32. 32. 32 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  33. 33. 33 Robert Mittendorff MD (2018) AI in Healthcare The Persuasion of AI Technology
  34. 34. 34 Robert Mittendorff MD (2018) AI in Healthcare BEHAVIOR IS A BIG PROBLEM TO SOLVE IN FLATTENING THE COST CURVE Source: McGinnis et al, New England Journal of Medicine 2002. The Digital Blockbuster is Behavior Change
  35. 35. 35 Robert Mittendorff MD (2018) AI in Healthcare AUTOMATE THE LABOR INTENSIVE, CLINICALLY VALIDATED APPROACH The Diabese Adult The “Healthy” Adult High Calorie Poorly Balanced Diet Low Activity Poor Medical “Compliance” Calorie Appropriate Diet Moderate Activity Reasonable “Compliance” ? $6,500 / Year $1,000 / Year Behavior Change Converts A to B
  36. 36. 36 Robert Mittendorff MD (2018) AI in Healthcare Intro to lorem ipsum
  37. 37. 37 Robert Mittendorff MD (2018) AI in Healthcare DIGITAL THERAPY FOR BEHAVIOR CHANGE: AS GOOD AS A DRUG? Omada Health Prevent™ Program: The Birth of a Digital Therapeutic Mass Customization for Disease Prevention; AI?
  38. 38. 38 Robert Mittendorff MD (2018) AI in Healthcare BEHAVIOR MODIFICATION CAN BE MORE POTENT THAN A DRUG 31% 58% Metformin DPP (Behavior Alone) 8% of the population developed diabetes NNT of 14 5% of the population developed diabetes NNT of 6 Source: McGinnis et al, New England Journal of Medicine 2002, CDC DPP Data. Technology and Labor Substitutes & Augmentation
  39. 39. 39 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  40. 40. 40 Robert Mittendorff MD (2018) AI in Healthcare WHAT IS ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING IN A NUTSHELL? Source: left, “Her” movie website, right, “Minority Report” movie website, all rights reserved to original publishers. AI and the Automation of Labor for Behavior Change AI/ML & Predictive & Prescriptive Analytics in Real Time From Scaling to Replacing Labor with AI
  41. 41. 41 Robert Mittendorff MD (2018) AI in Healthcare K162513: “The Arterys Software is intended to be used to support qualified cardiologist, radiologist, or other … practitioners for clinical decision-making... AI and ML Will First Come of Age As “Clinical Decision Support” to Make Humans More Efficient Source: Forbes article 1/20/17 including quote on bottom right. FDA Intended Use Claim from Arterys K162513 %10k clearance in October 2016.
  42. 42. 42 Robert Mittendorff MD (2018) AI in Healthcare K161201: “ClearRead CT™ is comprised of computer assisted reading tools designed to aid the radiologist in the detection of pulmonary nodules during review of CT examinations of the chest on an asymptomatic population. [Riverain Technologies] AI and ML Will First Come of Age As “Clinical Decision Support” to Make Humans More Efficient
  43. 43. 43 Robert Mittendorff MD (2018) AI in Healthcare Zebra: AI in Radiology
  44. 44. 44 Robert Mittendorff MD (2018) AI in Healthcare MedyMatch: AI in Acute Care Radiology
  45. 45. 45 Robert Mittendorff MD (2018) AI in Healthcare System of Action Frontline ActivationManagement Deep Dive, root cause analysis, Instant performance tracking Real-time collaboration, coordination, course correction Mission Control Situational Awareness Qventus acts as a platform for decision-making and action. It is informed by historic, real-time and predictive data. Qventus: AI enabled Air Traffic Control for Hospitals
  46. 46. 46 Robert Mittendorff MD (2018) AI in Healthcare How Qventus Works
  47. 47. 47 Robert Mittendorff MD (2018) AI in Healthcare (AIR) TRAFFIC CONTROL IN PATIENT FLOW USING AI AND MACHINE LEARNING Real Time Nudges Reduce LOS, Unnecessary Tests, and Left Without Being Seen Rates Source: AnalyticsMD data on actual customer deployments
  48. 48. 48 Robert Mittendorff MD (2018) AI in Healthcare Protenus: AI in Healthcare Privacy
  49. 49. 49 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  50. 50. 50 Robert Mittendorff MD (2018) AI in Healthcare LEVELS OF EVIDENCE REVIEW Level 1 Systematic Reviews & Randomized Controlled Trials Level 2 Cohort Studies Level 3 Case-Controlled Studies Level 4 Case Series Level 5 Case Based Reasoning or Experts The Evidence Pyramid: More is Better And Even Nicer If Randomized and Controlled
  51. 51. 51 Robert Mittendorff MD (2018) AI in Healthcare Significant Digital Health Research & AI Is Being Performed in Randomized Interventional Approach Source: Analysis of 3,791 records of clinical trials at www.clinicaltrials.gov with “mobile OR app” in content
  52. 52. 52 Robert Mittendorff MD (2018) AI in Healthcare DOES CLINICAL DECISION SUPPORT IMPROVE OUTCOMES • 2016 Clinical Decision Support in the ICU with (near) Real Time Data: 25 articles reviewed in meta-analysis of approaches of CDS in AIMS. • 2012 Computerized Clinical Decision Support for Diabetes Management: 15 studies, with several at high risk of bias. • 2012 Machine Learning and AI Source: (1) Simpao, et al. A systematic review of near real-time and point-of-care clinical decision support in anesthesia information management systems. J Clin Monit Comp 2016. (2) Jeffery, R. Diabetic Medicine, 2012. “Computerized clinical decision support systems in diabetes management may marginally improve clinical outcomes, but confidence in the evidence is low because of risk of bias, inconsistency and imprecision.” “ There is strong evidence for the inclusion of near real-time and point-of-care CDS in [Anesthesia Information Management Systems] to enhance compliance with perioperative antibiotic prophylaxis and clinical documentation…” Simpao, et al. 2016 Clinical Decision Support Improves Care
  53. 53. 53 Robert Mittendorff MD (2018) AI in Healthcare DOES CLINICAL DECISION SUPPORT IMPROVE OUTCOMES • 2017. Raju et a. Development of a Deep Learning Algorithm for Automatic Diagnosis of Diabetic Retinopathy. Precision Healthcare through Informatics 2017. • 2017, Esteva, A. Dermatologist – level classification of skin cancer with deep neural netowkrs. Nature. 2017 Bejnordi, B. Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA. Source: (1) Simpao, et al. A systematic review of near real-time and point-of-care clinical decision support in anesthesia information management systems. J Clin Monit Comp 2016. (2) Jeffery, R. Diabetic Medicine, 2012. “In this task, the CNN achieves 72.1 ± 0.9% (mean ± s.d.) overall accuracy (the average of individual inference class accuracies) and two dermatologists attain 65.56% and 66.0% accuracy.” “Approximately 35,000 images were used to train the network, which observed a sensitivity of 80.28% and a specificity of 92.29% on the validation dataset of ~53,000 images.” AI and Diagnosis: Performing Like Doctors? “The top 5 algorithms had a mean AUC that was comparable with the pathologist interpreting the slides in the absence of time constraints (mean AUC, 0.960 [range, 0.923-0.994] for the top 5 algorithms vs 0.966 [95% CI, 0.927-0.998] for the pathologist WOTC)..”
  54. 54. 54 Robert Mittendorff MD (2018) AI in Healthcare DOES ‘SMART TEXTING’ / SMS LEAD TO BEHAVIOR CHANGE: THE EVIDENCE • 2002: The first text messaging study published in health (1); “Mobile phone text messaging can help young people manage asthma” • 2014 Househ, et al.: First Meta-analysis of reviews (umbrella): 13 systematic reviews • 2015: Hall, A. et al.: 15 systematic reviews of 89 unique studies ranging from [10-5,800 patients/study] • 2015 Hall, A, et al 2015 Included Studies (abridged list of the 89): • Diabetes: 16 of 16 studies reported statistically significant effects on health outcomes or health behaviors (large variation in size & design) • Smoking Cessation: 6 of 8 studies demonstrating statistically significant behavior change or outcomes (smoking cessation, self report; 7 RCT) • Weight Loss/Physical Activity: 11 of 19 had statistically significant effects on weight and/or activity • Chronic Disease Management: 3 of 4 well designed studies from a group of 16 demonstrated statistical significance in outcome or behavior • Medication Adherence: 20 of 33 had statistically significant effects on behaviors or outcomes with asthma (3/3) and HIV (5/10) Source: (1) Neville, R, et al. BMJ 2002. (2) Househ, et al. Health Inform J. 2014 (3) Hall, A, et al. Anna Rev Public Health 2015 Mar 18; Burke, et al. AHA Scientific Statement: Current Science on Consumer Use of Mobile Health for Cardiovascular Disease Prevention. Circa 2015. “Our review found that the majority of published text-messaging interventions were effective when addressing diabetes self-management, weight loss, physical activity, smoking cessation, and medication adherence for antiretroviral therapy” - Hall, A, et al. 2015 (3) “…low to moderate research evidence exists on the benefits of SMS interventions for appointment reminders, promoting health in developing countries and preventive healthcare…” - Househ, et al. 2014 (2) AI and “Push” Reminders Work In Specific Settings
  55. 55. 55 Robert Mittendorff MD (2018) AI in Healthcare AI in Depth in Derm: Malignant vs. Benign Outperforms Source: Esteva, et al. Nature 2017. Dermatologist-level classification of skin cancer with deep neural networks
  56. 56. 56 Robert Mittendorff MD (2018) AI in Healthcare AI in Dermatology: Outperforming Dermatologists Source: Esteva, et al. Nature 2017. Dermatologist-level classification of skin cancer with deep neural networks
  57. 57. 57 Robert Mittendorff MD (2018) AI in Healthcare AI in Dermatology: The Beginning of Automation Source: Esteva, et al. Nature 2017. Dermatologist-level classification of skin cancer with deep neural networks
  58. 58. 58 Robert Mittendorff MD (2018) AI in Healthcare Agenda • The Labor of Healthcare: Leverage, Automation, and AI • AI, ML and Data Science: Terminology and Technology • Engagement and Persuasive Design: From Psychology to Digital Health • Interventional Approaches in Outcomes • State of the Evidence • Summary
  59. 59. 59 Robert Mittendorff MD (2018) AI in Healthcare AI: The Three Impacts • Leverage Human Labor to Increase Capacity to Treat by 10x-100x • Allow Patients to Self Manage Their Conditions With Persuasive Software • Enable New Interventions, Like Finding Problems in Real Time and Nudging Providers or Managers with Solutions
  60. 60. 60 Robert Mittendorff MD (2018) AI in Healthcare Questions • Robert Mittendorff MD MBA contact information: • rmittendorff@nvp.com • @doctorrem • https://www.linkedin.com/in/robertmittendorffmd

×