Transcript Presentation Template - Healthcare Analytics Summit

Session #23
There’s A 90% Probability That Your Son Is Pregnant:
Predicting The Future Of Predictive Analytics In Healthcare
Dale Sanders
SVP Strategy
Health Catalyst
Poll Question #1
To what degree is your organization using predictive analytics to improve care
were reduce cost?
a) We are not using any predictive analytics, that I know about
b) We are experimenting with predictive analytics in small use cases, but as yet have seen
no improvements in care or cost
c) We are using predictive analytics in a small number of use cases and the results have
been positive
d) We are using predictive analytics in a large number of use cases and the results have
been positive
e) Unsure or not applicable
2
Acknowledgements
Dr. Eric Siegel, Columbia University
Ron Gault, Aersospace Corporation
3
4
Key Themes Today
1.
Action Matters: Predictive analytics (PA) without actions and interventions are
useless
2.
Human Unpredictability: Humans behavior, like the weather, is inherently difficult to
predict with a computer
3.
Socio-Economics: Most of healthcare’s highest risk root causes lie outside the care
delivery system’s ability to intervene
4.
Missing Data: We are missing key data in healthcare, particularly clinical outcomes
data, required for accurate predictive models… so we need to leverage collective
wisdom of experts until we close the data gap
5.
Wisdom of Crowds: In the pursuit of objectivity of analytics, don’t forget the wisdom
of subjective experts sitting right next to you
6.
Social Controversy: Even with accurate PA, are we socially prepared to act? Do we
want to know? Are we intruding on people’s future?
5
Common Concepts
& Provocative Thoughts
6
Man vs. Machine 
Objective
Subjective
Man + Machine
7
Financial Industry Got It,
Long Ago
“Information about the transactions of money has
become almost as important as the money itself.”
- Walter Wriston, former chairman and CEO of Citicorp, awardee
of Presidential Medal of Freedom, 1989
• Could you cut and paste “health” for “money”?
• What if we gave healthcare away at a discount– or for free-- just so
we could collect the data for its analytic value?
• What if Health Catalyst started a healthcare delivery system so we
could collect and control the ecosystem for the downstream value of
the data?
8
The Basic Process of Predictive
Analytics
9
“Beyond math, there are no facts; only interpretations.”
- Friedrich Nietzsche
10
Challenge of Predicting Anything Human
11
Sampling Rate vs. Predictability
The sampling rate and volume of data in an experiment is directly proportional
to the predictability of the next experiment
12
Healthcare and patients
are continuous flow,
analog process and
beings
But, if we sample that
analog process enough,
we can approximately
recreate it with digital data
Thank you for the graphs, PreSonus
13
We are asking physicians and nurses to act as our
“digital samplers”… and that’s not going to work
14
The Human Data Ecosystem
15
We Are Not “Big Data” in Healthcare, Yet
16
Predictive Precision vs. Data Content
17
The Wisdom of Crowds &
Suggestive Analytics
18
The Wisdom of Crowds*
The Criteria For Designing A “Good” Crowd
Criteria
Description
Diversity of
opinion
Individual members of the group possess personal insights or facts on a topic,
even if it’s simply an unusual interpretation of data and facts on that topic
Independence
Individual members of the group form their own opinions and are not prone to
the overt and predictable influence from other members of the group
Decentralization
Knowledge on a given topic does not reside in central decision making
bodies, and important decisions can be made by members of a local,
decentralized crowd who most readily feel the consequences of those
decisions
Aggregation
There are methods and techniques for gathering and aggregating the
collective intelligence of the crowd
*--James Surowieki
19
Poll Question #2: Guess The Weight Of The Steer
c
c
2014 Southwest Washington State Fair
Charlie Brown, 8-yr old
Swiss steer; the Guessee
Dave Fenn, Owner
Levi Wallace, Guessor
20
2,767 pounds…!
21
Amazon: Predictive or Suggestive?
22
Poll Question #3
How many physicians were
working in Utah in 2010?
2012 Physician Workforce Report from the Utah Medical Education Council
23
5,596
24
Predictive Analytics
Outside Healthcare
25
US Strategic
Command,
underground
command
center…
prior to 9/11
26
Nuclear Operations
How And Where Can A Computer Help?
Reduce variability
in decision making
& improve
outcomes
Launch
prematurely?
Launch too late?
28
Desired Political-Military Outcomes
1.
Retain U.S. society as described in the Constitution
2.
Retain the ability to govern & command U.S. forces
3.
Minimize loss of U.S. lives
4.
Minimize destruction of U.S. infrastructure
5.
Achieve all of this as quickly as possible with minimal expenditure
of U.S. military resources
29
Odd Parallels
Healthcare Delivery and Nuclear Delivery
“Clinical” observations
•
Satellites and radar indicate an enemy launch
Predictive “diagnosis”
•
Are we under attack or not?
Decision making timeframe
•
< 4 minutes to first impact when enemy subs
launch from the east coast of the US
“Treatment” & intervention
•
Launch on warning or not?
30
•
•
•
•
Subjective
Objective
Assessment
Plan
31
NSA, Terrorists and Patients
The Odd Parallels of Terrorist Registries and Patient Registries
32
Predicting Terrorist Risk
Risk = P(A) × P(S|A) × C
• Probability of Attack
• Probability of Success if Attack occurs
• Consequences of Attack (dollars, lives, national psyche, etc.)
• What are the costs of intervention and mitigation?
• Do they significantly outweigh the risk?
33
Nuclear Weapons Risk Scenarios
What are the “adverse events” we were trying to predict and avoid?
•
NUCFLASH

•
Broken Arrow

•
Loss, theft, seizure, destruction of nuclear weapon
Bent Spear

•
Accidental or unexpected event, e.g., nuclear detonation or non-nuclear detonation or
burning
Empty Quiver

•
Accidental or unauthorized launch that could lead to the outbreak of war
Damage to a weapon that requires major repair, and has the potential to attract public
attention
Dull Sword

A nuclear safety deficiency that cannot be resolved by the local unit
34
“Mr. Sanders, while your 9-year tenure as an inmate has been
stellar, our analytics models predict that you are 87% likely to
become a repeat offender if you are granted parole. Therefore,
your parole is denied.”
- 2014, 80% of parole boards now use predictive analytics for case
management*
*--The Economist, “Big data can help states decide whom to release from prison”, Apr 19th 2014
35
“Evidence Based” Sentencing
20 States use predictive analytics risk
assessments to inform criminal sentencing
Thank you Sonja Star, New York Times
36
Recidivism Risk Assessment: Level of Service/Case
Management Inventory (LS/CMI)*
15 different scales feed the PA algorithm
Criminal History
Barriers to Release
Education/Employment
Case Management Plan
Family/Marital
Progress Record
Leisure/Recreation
Discharge Summary
Companions
Specific Risk/Needs Factors
Alcohol/Drug Problems
Prison Experience - Institutional Factors
Antisocial Patterns
Special Responsivity Consideration
Pro-criminal Attitude Orientation
42.2% of high risk offenders recidivate within 3 years.
*--Nov 2012, Hennepin County, MN, Department of Community Corrections and Rehabilitation
37
“Since the publishing of Lewis' book, there
has been an explosion in the use of data
analytics to identify patterns of human
behavior and experience and bring new
insights to fields of nearly every kind.”
38
eHarmony Predictions
“Heart”  of the system: Compatibility Match Processor (CMP)
• 320 profiling questions/attributes per user
• 29 dimensions of compatibility
• ~75TB
• 20M users
• 3B potential matches daily
• 60M+ queries per day, 250 attributes
Thank you, Thod Nugyen, eHarmony CTO
39
Twenty-Nine Dimensions of Compatibility
Thank you, Ryan Barker, Principal Software Engineering – Matching, eHarmony
40
41
The Good Judgment Project
•
Funded by Director of National Intelligence, brainchild of Philip Tetlock
•
Can groups of non-experts with access only to open source information,
predict world events more effectively than intelligence analysts with
access to classified information? What about “internationally recognized”
experts?
•
Since 2011: 5,000 forecasters, 1M forecasts, 250 topics

•
“…from Eurozone exits to Syrian civil war”
Non-expert forecasters are 65% better than the experts, 30-60% better
than predictive algorithms
42
Predictive Analytics
Inside Healthcare
43
True Population
TruePredictive
Population
Risk
Management
Health
Management
Very Little ACO Influence
>/=30% Waste*
100% ACO Influence
Thank you, for the diagram, Robert Wood
Johnson Foundation, 2014
Very Little ACO Influence
*Congressional Budget Office, IOM,
“Best Care at Lower Cost”, 2013
44
Socioeconomic Data Matters
Not all patients can functionally participate in a protocol
At Northwestern (2007-2009), we found that 30% of patients fell into one
or more of these categories:
• Cognitive inability
• Economic inability
• Physical inability
• Geographic inability
• Religious beliefs
• Contraindications to the protocol
• Voluntarily non-compliant
45
The key to predictive analytics in the future of healthcare
will be the ability to answer this two part question:
What’s the probability of influencing this patient’s behavior
towards our desired outcome and how much effort (cost)
will be required for that influence?
46
Example Variables: Readmission Drivers
Newborn delivery
Multiple prior admissions
High creatinine
High ammonia
High HBA1C
Weighted
Predictive
Model
Risk of
Readmission
Low Oxygen Sats
Age
Admitting physician is pulmonologist or infectious diseases
Which evidencebased Intervention?
How much will it
cost? How much will
it reduce risk?
Prior admission for CHF
traumatic stupor & coma
Prior nutritional disorders
Diabetic drugs
Thank you, Swati Abbott
47
Most Common Causes for Readmission
Robert Wood Johnson Foundation, Feb 2013
1.
Patients have no family or other caregiver at home
2.
Patients did not receive accurate discharge instructions, including
medications
3.
Patients did not understand discharge instructions
4.
Patients discharged too soon
5.
Patients referred to outpatient physicians and clinics not affiliated with
the hospital
48
What Else Are We Trying to Predict?
Common applications being marketed today
• Identifying preventable re-admissions: COPD, MI/CHF, Pneumonia, et al
• Sepsis
• Risk management of decubitus ulcers
• LOS predictions in hospital and ICU
• Cost-per-patient per inpatient stay
• Likelihood of inpatient mortality
• Likelihood of ICU admission
• Appropriateness of C-section
• Emerging: Genomic phenotyping
49
Closing Thoughts & Questions
1.
Action Matters: What is the return in investment for intervention? Are we prepared
to invest more... or say “no”… to patients who score low on predicted engagement?
2.
Human Unpredictability: The mathematical models of human behavior are
relatively immature.
3.
Socio-Economics: Can today’s healthcare ecosystem expand to make a
difference?
4.
Missing Data: Without patient outcomes, the PA models are open loop.
5.
Wisdom of Crowds: Suggestive analytics from “wise crowds” might be easier and
more reliable than predictive analytics, until our data content improves
6.
Social Controversy: How much do we want to know about the future of our health,
especially when the predictive models are uncertain?
50
Financial Industry Got It,
Long Ago
“Falling sick is not just an individual’s problem. Nations
crumble when their people are not strong. History is full
of events riddled with diseases that brought societies to
their knees.”
- Kofi Annan, former Secretary-General of the United Nations
51
Sometimes, the predictions are wrong 
Arthur Henning,
the Nate Silver of
the 1930s-1950s,
missed this one…
52
Analytic
Insights
Questions &
Answers
A
Session Feedback Survey
1.
On a scale of 1-5, how satisfied were you overall with this session?
1)
2)
3)
4)
5)
Not at all satisfied
Somewhat satisfied
Moderately satisfied
Very satisfied
Extremely satisfied
2.
What feedback or suggestions do you have?
3.
On a scale of 1-5, what level of interest would you have for additional,
continued learning on this topic (articles, webinars, collaboration, training)?
1)
2)
3)
4)
5)
No interest
Some interest
Moderate interest
Very interested
Extremely interested
54