Presentation at AHIMA on Predictive Modeling given on September 24, 2018:
Welcome to the age of data – the revolution has arrived made possible by new sources of data available in every industry and every walk of life. No area has been untouched and we see examples from baseball to the steady stream of customized adverts and content. Big data is improving the utilization of resources, efficiency, capacity, and ultimately access to healthcare. The next frontier is predicting futures. Can we identify the patients that are sickest and mitigate their disease progression? Traditional population health programs have focused on the high utilization patients predicting their progression and applying interventions to improve their health and decrease the costs associated with treating their disease progression. But the analysis of data has its perils risking false discoveries as our unconscious bias impacts our search of these huge troves of data in a modern-day version of “I know the facts, now let’s find ‘em” to “confirm” our facts. But the reality of healthcare and increasing insights into a precision-guided healthcare system it is increasingly clear that no membership population is “average”. The interplay between even a couple of input factors and comorbidities can be very non-linear.
Predictive Modeling is Here - Dance with the Dragon of Artificial Intelligence
1. @drnic1
Predictive Modeling is Here
Dance with the Dragon of Artificial Intelligence
Nick van Terheyden, MD
Founder & CEO of Incremental Healthcare
2. @drnic1
Conflict of Interest Statement
• The Presenter is Founder and CEO of Incremental Health
• Provides Strategic Guidance and Services to Healthcare
Companies in a Range of Areas including Data Analytics,
Population Health and Personalized Healthcare, Big Data,
Artificial Intelligence
• The Presenter is a Big Fan of the Fantasy Drama: Game of
Thrones
3. @drnic1
Nick van Terheyden, MD Founder & CEO, Incremental Healthcare
AboutMe http://about.me/obiwan
Twitter http://twitter.com/drnic1
LinkedIn http://www.linkedin.com/in/nickvt
Blog http://www.IncrementalHealthcare.com/
FaceBook http://www.facebook.com/drnic1
E-Mail DrNick@IncrementalHealthcare.com
Google Voice (301) 355-0877
Where you can find me
4. @drnic1
Agenda
• Introduction
• Health Declines in the Context of Life
• Undiagnosed Conditions in our Population
• Is Data Analytics in Use Elsewhere
• The Invisible Patient
• Population Targeting
• Questions
5. @drnic1
Undiagnosed Silent Preventable Killers
• Uncontrolled hypertension leading cause of
heart disease and stroke
• 1 in 3 U.S. adults (70 Mil) have Hypertension
• 13 million U.S. adults aren’t even aware they
have Hypertension
• 84 million people suffer from cardiovascular
disease - causing about 2,200 deaths a day,
averaging one death every 40 seconds
• 8 million adults have undiagnosed diabetes and
87 million have pre-diabetes
9. @drnic1
A Word on Bias
• Reproduce the inherent biases present in the data
they are provided with
• Disproportionately creating a ‘feedback loop’ by
which the predicted outcome simply becomes a self-
fulfilling prophecy
• Models Created with the Bias of the Builders either
intentionally or unintentionally
10. @drnic1 2009 Continua Health Alliance Brigitte Piniewski, MD
0 70
Age
IllnessPre-IllnessWellness
Death
60-80% Lifestyle
Modifiable Health
13. @drnic1
Financial Risk within Populations
Known Risk
We all know this and are effectively managing it.
Members with a significant claims history; and a correspondingly high
risk classification, who will likely utilize significant healthcare resources.
Unknown Risk
We know little about this and are doing little to manage it.
Members with an insignificant claims history, with a correspondingly low
risk classification, who are likely to utilize significant healthcare resources
14. @drnic1
“Unknown” Risk is the Deal Breaker
By 2025, 49% of Americans
Will Have Chronic Diseases
Cost of treatment due to progression from risk to disease in a population of 1000 individuals
1000 individuals
Depression 75
Allergic Rhinitis 56
Hypertension 32
Heart Disease 12
Asthma 7
Type 2 Diabetes 6
Cancer 2
Annual Progression
Risk to Disease
Additional Annual
Cost due to Disease
$325,000
Source: Wu, Shin-Yi and Green, Anthony. Projection of Chronic Illness Prevalence and Cost Inflation. RAND Corporation, 10/2000
21. @drnic1
Validation Against Previous Year Experience
CAD HYPERTENSION STROKE
Prevalence
Actuary
prediction
Actual
occurrence
Prediction
2014-2015 Health Member Experience
DIABETES
Tested against national
standards and real
patient populations, and
further verified with
Banner’s own member
data
23. @drnic1
Interventional Population Health Management Overview
Targeted Intervention Population and Value
Target Population 2,243
Total Predicted Savings $13,259,153
PMPY Average Savings $5,911
Average ROI 1,078%
Population Description Size
Total 46,320
With usable data 41,068
With clinical intervention potential 28,310
High value T2D/CAD risk factor intervention potential 2,243
Medicare Advantage 2017
26. @drnic1
Massive Multiplayer Online Role Playing Games
(MMRPOG/MMO)
• $31 Billion Market
• League of Legends – 100 Million
• Heroes in World of Warcraft – 70 Million
• WoW $5 Million in Subscription Fees Every Day
• Steam Concurrent Player Peak – 13 Million
• Gaming is Voluntary and Does not Pay
• WoW Average Player 17-22 Hours per week
• WoW Total of 5.93 Million Years of Player Time
– Equivalent to the time taken for the Human species to evolve
27. @drnic1
Final Thoughts
• The BiggestRisk are the Patients We have Not Seen
• Supermarkets KnowMore Aboutour Nutrition than the Healthcare System Does
• Prioritization for the Application ofClinicalResources = Triage
• Innovationand Augmentation ≠ReplacingPeople
• Genomics and the Biome willrevolutionize our DiagnosisandTreatment of Disease –
This Includes Surgery
• We Don’t Go Online– We Live Online
28. @drnic1
Nick van Terheyden, MD Founder & CEO, Incremental Healthcare
AboutMe http://about.me/obiwan
Twitter http://twitter.com/drnic1
LinkedIn http://www.linkedin.com/in/nickvt
Blog http://www.IncrementalHealthcare.com/
FaceBook http://www.facebook.com/drnic1
E-Mail DrNick@IncrementalHealthcare.com
Google Voice (301) 355-0877
Questions
29. @drnic1
Predictive Modeling is Here
Dance with the Dragon of Artificial Intelligence
Nick van Terheyden, MD
Founder & CEO of Incremental Healthcare
Drogon - Named after Daenerys' late husband, Khal Drogo
Weather
Baseball/Money Ball, Stock Market/The Big Short
High School Drop Outs
Crime - Hotspot policing
http://www.espn.com/espnw/news-commentary/article/7577771/stats-created-moneyball
High School Graduation – Chicago Public Schools
https://dsapp.uchicago.edu/projects/education/
Data Science for Social Good (DSSG)
https://dssg.uchicago.edu/project/identifying-high-school-students-who-may-not-graduate-on-time/
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.725.9316&rep=rep1&type=pdf
Cambridge Analytica – Elections
Baseball – Moneyball - Michael Lewis
Stock Market – The Big Short Michael Lewis
Pre Crime
Manchester England - https://www.bbc.co.uk/programmes/b08msdc2
Birmingham, Snata Cruz CA - https://www.channel4.com/news/predictive-policing-tackling-crime-before-it-happens
http://www.cxotoday.com/story/using-big-data-analytics-to-predict-crime-patterns/
https://www.independent.co.uk/news/uk/home-news/police-big-data-technology-predict-crime-hotspot-mapping-rusi-report-research-minority-report-a7963706.html
But Not all predictive Analytics work out
https://www.kdnuggets.com/2018/06/football-world-cup-predictions.html
Bronn The Dragon Slayer
Bronn, who shot down one of Daenerys' dragons, Drogon, in the final scene of "Spoils of War."
10
Predictive Value of Personalized Risk Assessment to Prevent Multifactorial, Complex and Chronic Disease
International Journal of Medical Technology, ISSN: 2051-574X, Vol.23, Issue.1 1116, 2015
Application of Personalized Risk Assessment to Prevent Chronic Diseases Using Aggregate Population Based Data
International Journal of Medical Technology, ISSN: 2051-574X, Vol.23, Issue.2 1130, 2016
Utilization of Genetic Data Can Improve the Prediction of Type 2 Diabetes Incidence in a Swedish Cohort
PLoS ONE 12(7): e0180180. https://doi.org/10.1371/journal.pone.0180180, 2017