Should healthcare be more digitized? Absolutely. But if we go about it the wrong way... or the naïve way... we will take two steps forward and three steps back. In this 90-minute webinar, Dale Sanders, President of Technology at Health Catalyst describes the right way to go about the technical digitization of healthcare so that it increases the sense of humanity during the journey. The topics Dale covers include: • The human, empathetic components of healthcare’s digitization strategy • The AI-enabled healthcare encounter in the near future • Why the current digital approach to patient engagement will never be effective • The dramatic near-term potential of bio-integrated sensors • Role of the “digitician” and patient data profiles • The technology and architecture of a modern digital platform • The role of AI vs. the role of traditional data analysis in healthcare • Reasons that home grown digital platforms will not scale, economically Most of the data that’s generated in healthcare is about administrative overhead of healthcare, not about the current state of patients’ well-being. On average, healthcare collects data about patients three times per year from which providers are expected to optimize diagnoses, treatments, predict health risks and cultivate long-term care plans. Where’s the data about patients’ health from the other 362 days per year? McKinsey ranks industries based on their Digital Quotient (DQ), which is derived from a cross product of three areas: Data Assets x Data Skills x Data Utilization. Healthcare ranks lower than all industries except mining. It’s time for healthcare to raise its digital quotient, however, it’s a delicate balance. The current “data-driven” strategy in healthcare is a train wreck, sucking the life out of clinicians’ sense of mastery, autonomy, and purpose. Healthcare’s digital strategy has largely ignored the digitization of patients’ state of health, but that’s changing, and the change will be revolutionary. Driven by bio-integrated sensors and affordable genomics, in the next five years, many patients will possess more data and AI-driven insights about their diagnosis and treatment options than healthcare systems, turning the existing dialogue with care providers on its head. It’s going to happen. Let’s make it happen the right way.

1. Raising the Digital
Quotient of Healthcare
Dale Sanders
President of Technology
Health Catalyst
August 2018
Creative Commons Copyright

2. • 15 years in military industrial complex where conflict leads to profits
• 21 years in the healthcare industrial complex where illness leads to profits
My Background
B.S. Chemistry,
biology minor
US Air Force Command,
Control,
Communications, &
Intelligence (C3l) Officer
TRW/National Security Agency
• START Treaty
• Nuclear non-proliferation
• US nuclear command & control
system threat protection
Intel Corp, Enterprise
Data Warehouse
Chief Data Guy; Regional Director
of Medical Informatics,
Intermountain Healthcare
CIO, Cayman Islands
National Health System
Technology &
Product
Development,
Health Catalyst
Reagan/Gorbachev
Summits
• Space Operations
• Nuclear Warfare
Planning and
Execution; NEACP &
Looking Glass
CIO, Chief Analytics
Guy, Northwestern
Memorial Healthcare
20181983
2

3. • Thanks to McKinsey
3
Organizational Digital Quotient = Data Assets x Data Skilled Labor x Data Usage
Digital Quotient (DQ)

4. Here’s Your Data Assets Roadmap
4

5. • Improving the soft, human side of
our digital healthcare strategy
• Attributes of a modern digital
platform
• Thoughts on AI and precision
medicine
Today’s Story
5
Once upon a time
in healthcare…

6. Our National Data Strategy is a Train Wreck
• 271 measures in QPP
• 86 related to General Internal
Medicine
• 37% invalid, 28% questionable
validity
• Highest suicide rate of any
profession
• American Psychiatric Association (APA) 2018. Abstract 1-227,
presented May 5, 2018
• >50% burnout in all specialties
We’re losing our physicians
6

7. MGMA Survey Aug 2018
• 84% of physicians participating in MIPS
• 82% consider MACRA QPP as very or
extremely burdensome
• 73% said MIPS is “a government program
that does not support their practice’s
clinical quality priorities”
7

8. A Clinician’s Life
8
Total available
time for
analytics
associated
with local
process
improvement,
creativity, and
patient care
Time required by compulsory
measures from CMS, private
payers, and professional
societies

9. How Clinicians Feel…
9

10. HHS Priorities
• End the opioid crisis
• Health insurance reform –
availability and affordability
• Drug pricing – lower costs of
prescription drugs
• Value-based care – accelerate it
10

11. To Regulate or Not?
• Generally speaking, the Federal government regulates safety and
adherence to standards that benefit the consumer
• Historically, the Federal government has not regulated quality of
products or consumer experience
• Utilities => cost, quality of service, and standards are regulated
• HHS/CMS is a hybrid … world’s largest customer of healthcare +
the world’s largest government of healthcare
Boundaries of Federal Regulation vs. Free Market
11

12. Poll Question
If you had to choose only one, where should HHS/CMS apply their
influence most?
• Quality of Care – Outcomes – 68%
• Cost of Care – 13%
• Safety of Care – Do no harm – 19%
12

13. Signs of Common Sense at HHS/CMS
• Pricing transparency… publicly available on website
• Future rules forthcoming on transparency around out-of-network costs
• Removing 18 measures from Inpatient Quality Reporting Program
• The cost of collecting the data outweighs the value; or no longer
relevant
• Removing 25 measures from IQRP that are redundant in other
programs
• Easing documentation requirements for certification, e.g. where data
is located in the EHR
• Promoting Interoperability Program (PIP)
• Flexibility and new scoring that focuses on interoperability, less on
process of care measures
IPPS and LTCH Final Rules
13

14. Proposed Rules
MACRA Quality Payment Program
Physician Fee Schedule
• Remote patient monitoring reimbursements
• Removing 34 quality reporting measures
• Add 10 measures, four of which are outcomes-based
• Consolidate PQRS, MU, VBPM
• Simplify E&M coding (how about we just get rid of it?)
• Advancing virtual care
• Medicare Part B drug payments reflect actual costs
Open for Comment until Sep 10
14

15. We’ve missed the human, softer side of
becoming a data-driven industry.
15

16. Our current “data-driven”
strategy in healthcare is
sucking the life out of
physicians’ sense of Mastery,
Autonomy, and Purpose
Mastery, Autonomy, Purpose
16

17. • Quantitative predictability is the metric of scientific precision
• Said otherwise, the progression of any body of science is measured by its
predictability
17

18. If I had a tattoo, this is what it would be…
• Find The Truth
• Tell The Truth
• Face The Truth
Humans & Their Biology are Difficult to Predict
• The truth in healthcare data is rarely “The Truth,”
but we talk to physicians as if it were
• When you communicate the truth, realize that it’s
only an approximation, and be sensitive to the
human who’s receiving the message
• Help people face the truth without feeling
threatened and over-measured
18

19. Enabling the Digital Healthcare Conversation
"I can make a health optimization recommendation for you, informed
not only by the latest clinical trials, but also by local and regional data
about patients like you; the real-world health outcomes over time of
every patient like you; and the level of your interest and ability to
engage in your own care. In turn, I can tell you within a specified range
of confidence, which treatment or health management plan is best
suited for a patient specifically like you and how much that will cost.”*
Between a physician and their patient
19
*—Inspired by the Learning Health Community
Outcomes and cost data, predictive analytics, machine learning, social determinants of health data,
recommendation engines

20. Creating a Digital Twin
1. Digitize the assets you are
trying to manage and optimize
Airplanes
Air traffic control,
baggage handling, ticketing,
maintenance,
manufacturing
Patients
Registration, scheduling,
encounters, diagnosis,
orders, billing, claims
We haven’t digitized the patient, and we’ve only digitized a clinical encounter to drop a bill
2. Digitize your production
process for managing the assets
you are trying to understand and
optimize
What’s Required to become “Digitized?”
20

21. At Best, EHRs Hold 8% of the Data We Need
• Only 20% of factors affecting health outcomes fall inside traditional
healthcare delivery
• On average, patients have 3 healthcare encounters per year
• We are missing data for the other 362 days of the year
• Healthy patients represent our ideal AI training set … but we have
no data on healthy patients
21

22. • About two-thirds of patients don’t want or cannot be “engaged”
• What they really want: When they are sick, they want to be
diagnosed and then treated safely, affordably, personally, efficiently,
and precisely
• Keep that in mind as we lay out a strategy and priorities for digital
health
My Observation About Patient Engagement
22

23. • Netherlands study
• Rate of patient requests for
a specific therapeutic or
diagnostic intervention,
1985-2014
• Significant increase in
requests by patients
• Significant increase in
compliance by GPs
Patients Owning Their Care
23
Requests for blood tests: 2x increase
Requests for urine tests: 26x increase
Requests for radiology/imaging: 2.4x increase
Requests for medication prescription: 1.2x
increase

24. Digital Accuracy Digital Sampling
We can’t possibly provide personal health or precision medicine
with only three patient data samples per year
24

25. This is my analog life
This is healthcare’s
digital view of my life
25

26. #NoMoreClicks
Physicians have become healthcare’s
analog-to-digital samplers.
That’s not a good role for human beings.
26

27. AI Needs Breadth & Depth of Data in the Domain
Rows x Columns
Records x Facts
27

28. Microns-thin, one-inch skin-pliable sensors
with integrated Bluetooth antenna, CPU,
physiologic monitors, and wireless power
28

29. • University of
California San
Diego
• 3-D, bio-
integrated,
stretchable
sensors
• EEG, EMG, ECG,
respiration, skin
temperature, eye
movement, body
motion
29

30. • 160, Ptolemy – the Earth is the center of the
Universe
• 1500s, Copernicus – the sun is the center of
the universe
• Today – the healthcare system is the center
of the patient’s data universe, creating a
monopoly of knowledge
• Tomorrow – patients will generate and
control more data about themselves than
their healthcare provider
Ending the Data Monopoly in Healthcare
30

31. • Enabled by bio-integrated sensors,
patients hold more data about
themselves than the healthcare system
• Their data is constantly being updated
and uploaded to cloud-based AI
algorithms
• Those algorithms diagnose the patient’s
condition, calculate a composite &
specific health risk scores, and
recommend options for treatment or
maintaining health
• The algorithm suggests options for a
“best fit” care provider (volumes,
outcomes, in-network) and the ability to
socially interact with other patients like
them
Future of Diagnosis and Treatment
31
• The patient engages with the care provider,
enabled with the output of the AI algorithms

32. • Different patient types have different
data profiles required for the active
management of their outcomes and
health
• I’m not talking about quality measures
• I’m talking about telemetry, diagnostics
and functional status about the state
of the patient, not the state of
healthcare processes
Rise of The Digitician and Patient Data Profiles
32
• It’s the Digitician’s job to prescribe the right
sensors and proactively collect this data for
patients in their panel, and feed the analytics
of that to the care team and patient

33. Attributes of a
Modern Digital
Platform

34. As computer scientists, we overlooked the last and
critically important layer in the technology stack…
User Interface
Application Software
Operating System
Hardware
34

35. The Evolution of Data Modeling in Analytics
Monolithic,
enterprise
data
model
Late binding
data models,
aka, schema
on read
“We know
all the use
cases, a
priori”
“We know
none of the
use cases, a
priori”
Intermediate
data models
“We know
some of the
use cases,
a priori”
• Harmonized vocabulary
• Comprehensive and
persistent agreement
about binding logic, e.g.
CMS value sets
35

36. Binding Data in the Pipeline, Not the Data Model
36

37. 7 Attributes of a Modern Digital Platform
1. Reusable clinical and business logic: Registries, value sets, and other data logic lies on top of the raw data
and can be accessed, reused, and updated through open APIs, enabling third-party application development.
2. Single data stream feeds analytics and workflow applications: Near- or real-time data streaming from
the source all the way to the expression of that data through the platform that can support transaction-level exchange of data
or analytic processing.
3. Integrates structured and unstructured data: Integrates text, images, and discrete structured data in the
same environment.
4. Closed-loop capability: The methods for expressing the knowledge in the platform, include delivering that knowledge
at the point of decision making, for example, back into the workflow of source systems, such as an EHR.
5. Microservices architecture: In addition to abstracted data logic, open microservices APIs exist for platform
operations such as authorization, identity management, data pipeline management, and DevOps telemetry. These
microservices also enable third-party applications to be built on the platform, and constant delivery of software updates, rather
than massive, major updates.
6. AI/Machine learning: Natively runs AI and machine learning models, and enables rapid development and utilization of
ML models, embedded in all applications.
7. Agnostic data lake: The platform can be deployed over the top of any healthcare data lake. The reusable forms of
logic must support different computation engines; e.g., SQL, Spark SQL, SQL on Hadoop, et al.
37

38. Data
Integration
Batch Data
Realtime Data
Mirth,
RabbitMq
Flash Data
Engine
SQL
Big Data
.NET
DOS App Cluster
Apps
Microservices
Angular, .NET, Java, Docker,
Kubernetes
Datamart Designers & Tools
SAMD, SMD, Atlas, Ops Console, Data Portal
Data & Compute Cluster
SQL Server, Hadoop, Spark,
ElasticSearch
Transactional data
store
SQL, Azure Disk
External
Data
Sources
EMR
SQL
HL7
X12
FHIR
Flat-files
XML
DOS Marketplace
Apps, Content, AI models
Flash AI
Engine
healthcare.ai
.NET, R,
Python
Azure
Azure
Data Operating System Architecture
AI Cluster
R, Python,
Spark
HTTP
FHIR
Curated Data Content
Shared Data Marts, Terminology, Measures, Populations
SQL
HTTP EMRs
External
Apps
Reports
Web
Services
SQL
Files
HL7
FHIR
38

39. Who will be the digital disrupter of healthcare?
• Employers (e.g. Amazon, Berkshire, Chase) — 27%
• Silicon Valley (e.g. Apple, Google, Microsoft) — 39%
• Current provider systems (e.g., role models like Geisinger) — 6%
• EHR vendors — 7%
• None of the above — 21%
Poll Question
39

40. Thoughts on AI and
Precision Medicine

41. Predictive Risk Fatigue
Predictions of risk, without a
plan or the ability to
intervene, are a liability to the
decision maker, not an asset
41

42. Discriminative neural networks
mimic the human pattern
recognition & classification
process… “Those are people.”
Generative Adversarial Networks
(GANs) mimic the opposite
human process… “This is what
people look like.”
42

43. Pattern Recognition
• Consortium of Japanese
researchers
• 94% accuracy detection
of small polyps
• Image analysis will and is
moving fast
• EHR and other pattern
recognition is moving
slowly
• It’s all about digital
sampling … digital density
43

44. Rules-Based Registries Have Flaws
• Lund University, Sweden
• Using k-means and hierarchical
clustering
• Five distinct subtypes of adult-
onset diabetes
• This type of work is going to
change our taxonomy of
diseases
44

45. • Mt. Sinai study
• Visualizes and explores clusters of
patients grouped together by
algorithms
• 2,551 Type 2 diabetic patients
clustered on 73 clinical variables
• A rules-based approach would not
find these subgroups
Topographical Data Analysis & Diabetes Subgroups
Li L, Cheng W-Y, Glicksberg BS, et al. Identification of type 2 diabetes
subgroups through topological analysis of patient similarity. Science translational
medicine. 2015;7(311):311ra174. doi:10.1126/scitranslmed.aaa9364.
45

46. Let the Data Tell Us Its Story
• Tsinghua University, Beijing
• Convolutional Neural Networks
(CNN) applied to EHRs
• Automatically extract semantic
information
• Perform automatic diagnosis
without artificial construction of
rules or knowledge bases
Rather than us assume we know it…
46

47. • Polygenic risk scores
• Genes that interact additively to
influence phenotypic
expression
• More genetic data => dramatic
increase in score accuracy
• “One test described last year
can guess a person’s height to
within four centimeters, on the
basis of 20,000 distinct DNA
letters in a genome.”
47

48. • Wright State University,
University of California Davis,
and Universidade Nova de
Lisboa
• eHarmony and Tinder AI
concepts applied to patient-
primary care relationship
matching
• Model homophily, trust
• Less primary care churn, better
health
48

49. Data Volume vs.
AI Model Sophistication
“Invariably, simple models and a
lot of data trump more elaborate
models based on less data.”
49
“The Unreasonable Effectiveness of Data”, March 2009,
IEEE Computer Society; Alon Halevy, Peter Norvig, and
Fernando Pereira, of Google
49

50. AI Algorithms are Commodities, Digital Platforms
and Infrastructure are Not
“…it is dangerous to think of these
quick wins as coming for free. Using
the software engineering framework
of technical debt, we find it is
common to incur massive ongoing
maintenance costs in real-world ML
systems.”
Neural Information Processing Systems (NIPS)
Advances in Neural Information Processing Systems 28 (NIPS 2015)
50

51. The machine learning code, in the black box, is a small fraction of
the ML ecosystem
This is not the land of small, niche startups or home grown systems

52. The Siren’s Temptation of Home Grown Digital Platforms
• Public cloud makes the infrastructure an
incredibly appealing and affordable
commodity
The hard part is…
• The collection, curation, and management
of data and the logic associated with that
data
• The development of APIs and applications
Remember when we were all building our
own PCs?
Plug your sailors’ ears, Odysseus 
1868, Firmin Girard

53. Healthcare Analytics Summit 18
Sept. 11-13, Salt Lake, Grand America Hotel
TOBY COSGROVE, MD
former CEO and President of
Cleveland Clinic (2004-2017),
who as a cardiac surgeon
performed more than 22,000
operations and holds 30 patents
for medical innovations
KIM GOODSELL
the actualized ‘genomified,’ quantified,
digitalized “patient of the future," her debut at
the 2014 Future of Genomic Medicine
conference made headline news
announcing— “The patient from the future,
here today”
DANIEL KRAFT, MD
a Stanford and Harvard trained physician-
scientist, inventor, entrepreneur, and
innovator, Kraft is the Founder and Chair of
Exponential Medicine, a program that
explores convergent, rapidly developing
technologies and their potential in
biomedicine and healthcare
BRENT JAMES, MD
former Chief Quality Officer at
Intermountain Healthcare - known
internationally for his work in
clinical quality improvement,
patient safety, and the
infrastructure that underlies
successful improvement efforts
PENNY WHEELER, MD
President and Chief Executive
Officer of Allina Health,
returns a second time as one
of the most popular HAS
speakers ever
MARC RANDOLF
Co-founder of Netflix, Marc will
share the Netflixed story: how a
scrappy Silicon Valley startup
brought down Blockbuster and
the lessons that could be
applicable to healthcare
JILL HOGGARD GREEN
PhD, RN, Chief Operating Officer – Mission
Health and President – Mission Hospital,
and recently named to the 2017 Becker’s
Healthcare list of the country’s top Women
Hospital and Health System Leaders to
Know
ROBERT WACHTER, MD
global leader in healthcare safety,
quality, policy, IT; Chair of the
Department of Medicine, University of
California, San Francisco; best-selling
author, “The Digital Doctor: Hope, Hype
and Harm at the Dawn of Medicine’s
Computer Age”
More highlights
4 Digital Innovators (Keynotes)
AI Showcase (10 walkabout case studies)
Digitizing the Patient Showcase (10-12 stations)
28 Educational, Case Study, and Technical Breakouts
24 Analytics Walkabout Projects
More Networking (Introducing “Brain Date”)
CME Accreditation For Clinicians
5-Star Grand America Hotel Experience
96 Total Presentations
National keynotes
Employer
Innovation
Scott
Schreeve
MD, CEO, Crossover Health
Payer
Innovation
Kevin
Sears
Executive Director of Marketing
and Network Services, Cleveland
Clinic
Biosensor
Innovation
John
Rogers
PhD, Founding Director, Center
Bio-Integrated Electronics,
Northwestern University
Pricing
Innovation
Gene
Thompson
Project Director, Health City
Cayman Islands

54. In Closing…
• Drive: Our digital strategy must enhance
Mastery, Autonomy, and Purpose
• Freud: Our data isn’t as big as we like to
think it is in healthcare
• Platform: It’s overdue in healthcare by 10
years
• Patients: The data ecosystem is shifting to
them
• Debt: AI will disrupt healthcare, no doubt
about that, but it requires massive data
infrastructure

55. Q&A

56. Thank You!