When Billy Beane decided to employ a recent Harvard graduate to use advanced statistical analysis to build a championship major league baseball team, he changed the game forever. While Beane’s famous early 2000s team never won a World Series, multiple 100-win seasons and a new record for the longest winning streak got the attention of teams across the MLB, all while on one of the league’s lowest payrolls. Most people know Beane’s story as it was popularized in the book—and later in the movie—Moneyball.

In healthcare, we are overdue for a “Moneyball” revolution. The shift towards value-based payment has made it clear that our system needs to do a better job generating outcomes that matter to patients — a positive health-care experience, improved health, and good quality of life. The machine learning techniques that were used to algorithmically determine a player’s value were light-years ahead of the archaic methods that had been used in baseball up to that point. Similarly, many of our conventions in delivering care come from an era when healthcare was delivered primarily by doctors and nurses with elite training whose success depended mostly on content expertise. A key component to value-based transformation in healthcare will be artificial intelligence. Without AI, medicine will never advance to a state where the totality of a patient’s data can be used to find predictive signals that will lead to enhanced treatment and population health interventions that improve outcomes.

Our guest this week is Andrew Eye, the founder and CEO of ClosedLoop.ai, the recently announced winner of the CMS Artificial Intelligence Health Outcomes Challenge. Listen and find out why Andrew and ClosedLoop are exemplars in the race to value!

Episide Bookmarks:

02:00 The Billy Beane story and how, in healthcare, we are overdue for a “Moneyball” revolution

03:00 A key component to value-based transformation in healthcare is artificial intelligence

04:00 Andrew Eye – a national leader in AI in Value-Based Care – and his company ClosedLoop.ai

06:45 Partnership with Dave DeCaprio following his work with the Human Genome Project

07:30 How Andrew’s daughter’s medical condition provided “WHY” inspiration to build a next-gen predictive analytics platform

09:20 How ClosedLoop.ai beat out the world’s leading technology and healthcare organizations to win the CMS AI Health Outcomes Challenge!

11:25 “Physician trust in AI is crucial. Algorithms never saved anybody’s life. We predict the future so that you can change it.”

12:50 Creating an open source, AI-based predictive model for predicting COVID-19 Vulnerability

13:00 Andrew discusses what it was like to submit the winning submission for the CMS AI Challenge without electricity in the Texas Snowpocalypse!

14:00 CMS’ focus on AI Explainability and how ClosedLoop was “born to win”

17:00 “Explainable AI” (XAI) versus “Black Box” machine learning algorithms

19:00 Early AI firms were reluctant to share “secret sauce” of proprietary algorithms and the impact on physician trust and external validation of bias

20:00 “We’re not building models. We are building a machine that builds models.”

20:20 “The idea that there is one algorithm that is best for every healthcare organization in the country is a total fallacy.”

21:00 “Explainability in AI is absolutely critical to helping care teams have more effective interventions in population health.”

22:30 Physician paranoia about “machines taking over” where there work will be eventually outsourced to algorithms and other artificial tools of clinical reasoning

23:45 The impact of AI on Radiology and how that scenario differs from other instances in medicine where AI is applied to population health

25:20 The opportunity to augment clinician pattern recognition with AI that goes far beyond manual chart review for surface insights

26:15 “There is going to be a point in time where patients choose a doctor based on whether or not they are using all of the available information.”

27:00 The challenges of algorithmic bias and fairness in ensuring health equity and references to recent research (article here and here)

30:30 Label choice in ML algorithms where costs are used as a proxy for health

31:30 Differentiation between algorithmic bias (based on math) and algorithmic policy (based on policy)

33:30 The inferiority complex that healthcare organizations have with “data shaming” and AI can pull predictive signals out of messy data

34:30 The Medical Home Network AI case study that focuses on Social Determinants of Health

35:30 Augmenting claims data with ADTs, Rx data, and Health Risk Assessments

37:00 “Until you squeeze all of the predictive signal from the data that you have, you shouldn’t be shopping for data that you don’t have.”

38:50 Andrew explains (in layman’s terms) the Receiver Operating Characteristic (ROC) curve used in statistical validation

39:50 Why a really accurate model for the entire population is not as important as the accuracy within the 3-5% of patients that actually drive up costs

40:45 Focusing on “percent capture” is more important that ROC curves

42:00 Building the right population health AI model by asking the right questions

42:30 Ensuring successful integration of predictive modeling in the workflow of the provider and pop health team

44:45 Feature drift that occurs in algorithms when datasets change and input values are affected

46:00 MLOps as a process of taking an experimental Machine Learning model into a production system

48:20 Andrew explains the FDA regulatory environment for AI in healthcare and its impact on health equity

51:10 Andrew shares his thoughts on Big Data futurism and the future of AI as “the next Moneyball” opportunity

52:30 The need for lowering cost to predictive models in prediction of rare diseases

53:00 How CMS is leading the charge in data liquidity with Blue Button 2.0 APIs

54:00 Organizations leading in value-based care are the ones investing in AI/ML

What happened in the comments section of the FCC's net neutrality hearing?