Finally…Healthcare 2.0 Data and The Age of Analytics Dale Sanders, June 2012.

Finally…Healthcare 2.0

Data and The Age of Analytics

Dale Sanders, June 2012

My Background

• Analytics and data warehousing, the constant theme• A CIO by various titles for 70% of my 29 yr career

– 15 years: US Air Force, national intelligence, consulting– 14 years: Healthcare

• Integrated Delivery System: Intermountain• Academic Medical Center: Northwestern• National Health System: Cayman Islands• Founder, Healthcare Data Warehousing Association

• Current affiliations– Mentor CIO, Cayman Islands NHS– SVP, Healthcare Quality Catalyst– Senior Research Fellow, The Advisory Board Company

Of Course We Have to Talk About It…The Supreme Court Decision

• Individual Mandate in the Affordable Care Act– The only valid constitutional debate. Everything else is just politics.

• Congress has the clear right to levy taxes– If you don’t buy insurance, you have to pay the IRS an additional “penalty”… specifically

avoided using the word “tax”• Insurance exempted from Congressional oversight as an Interstate Commerce (McCarran-

Ferguson Act 1945)– Individual Mandate would be unconstitutional as a form of Congressional regulation of

interstate commerce

• The Justice’s opinions pound this “it’s a tax but not a tax” back and forth like a tennis ball

• But…– A tax cannot be legally challenged until it is levied (Anti Injunction Act, 1793)– Individual Mandate does not take effect until 2014– I wouldn’t be surprised if it gets challenged again, as a tax

• In any case, the Act is going to increase the cost of care to taxpayers so get ready for healthcare costs to reach >20% of GDP if not amended

Agenda

• New thinking in Healthcare 2.0• A Roadmap for Healthcare Analytic Adoption• Analytics: The Technical Options• Prioritizing Analytics & Process Improvement• Predictive Analytics vs Suggestive Analytics• Watson and Big Data

5

Healthcare Billing at a Restaurant

• You wait 45 minutes for a table, even though you had a reservation.

• You tell the waiter that you’re hungry– but there’s no menu.

• The waiter returns with a meal that he thinks is appropriate for you…but he doesn’t

know how much it costs.

• You have no idea what the food is or what it costs, but you agree to eat it.

• You leave without knowing your bill.

• The restaurant sends the bill to your bank, not you.

• Your bank tells the restaurant, “Your waiter ordered the wrong thing for you. We’re not

paying for it.”

• 90 days later, the restaurant calls to tell your account is being turned over to collections.

Employers Are Taking Over

• “We know the healthcare systems that provide the best care and control their costs. We know which ones can show us their data that proves they manage quality and cost. And we know where those healthcare systems operate, geographically. Guess what criteria we consider most important, now, when making a decision to open a new plant or office?”– Paul Grundy, MD, Global Director of IBM Healthcare Transformation

The Healthcare 2.0 Analytics Equation

The Value of Your Healthcare Product =

Quality of Health

Cost of Production + Margin

$$ dividends to employers and patients

Cost of Production = Cost of Operations

The CFO in Healthcare 2.0 must be able to answer: “How much does it cost for us to produce the best health?”

Healthcare Data Maturity Stages

• Stage 1: Data Collection– Characterized by the

expanded adoption of EMRs• Stage 2: Data Sharing– Characterized by the

expanded adoption of HIEs• Stage 3: Data Analysis– Characterized by the

adoption of data warehouses

Healthcare 2.0

Healthcare 1.0

Analytic Technology Is the Easy Part

“There are a lot of companies who think they are using data…but historically that sort of data has been used to confirm and support decisions that had already been made by management, rather than learn new things and discover what the right answer is.”

The cultural change is for managers to be willing to say, ‘That’s an interesting problem, that’s an interesting question. Let’s set up an analysis to understand it; let’s set up an experiment; Show some vulnerability and say, ‘Look we are open to the data.’”

Erik Brynjolfsson Schussel Family Professor of Management

MIT Sloan School of Management

Analytic Value = Analytic Culture x Analytic Technology

Health 2.0 Analytics: Beyond Utilization Metrics

1

2

3

4

5

67

Key Message: Rapidly Adaptable Analytics

• As a C-level executive in healthcare, everything I want to improve is constrained by software that can’t adapt fast enough

• Healthcare 2.0 analytics will require a rapidly adaptable infrastructure of data and visualization tools– Data collection must be adaptable– EMRs and other

source systems– Data extraction and loading must be adaptable– Analytic data models must be adaptable– Visualization tools must fit a variety of needs

• The visualization tools and the underlying data models must be de-coupled

12

Assessing Analytic Capability

• The EMR Adoption Model from HIMSS– Imperfect, but invaluable to the industry

• What about the same concept applied to analytics?

13

Vocabulary Infrastructure established: Searchable metadata repository, core data elements linked with standardized naming and data typesLevel 1

Automated internal reporting: Key performance indicators and dashboards for hospital and clinic management, for executives, front line managers, and physicians Level 2

Automated external reporting: For financial incentives such as P4P, PQRS and MU; and accreditation/regulatory bodies such as JCAHO, ACC, STS, NRMI, HEDISLevel 3

Broad analytic deployment: A permanent integrated technical and clinical improvement teams for top 10 conditions; self-service data visualization for at least 60% of employeesLevel 5

Waste identification and elimination: Integrated patient specific costing and claims data used for identification and elimination of non-value add/non-evidence based activitiesLevel 6

Personalized patient analytics: Integration of genomic, familial, text, and patient self-reported data used for predicative modeling, preventative care and wellness managementLevel 7

Evidenced-based analytics: Patient registries for top ten conditions within the organization; Chronic condition management reports; measurement of clinical guideline usage (e.g. orders sets)Level 4

Cumulative CapabilityLevel

Healthcare Analytic Adoption Model©

Level 0 Major data sources in a single repository: Minimum EMR Level 3 data, Revenue Cycle, Financial, Costing, Supply Chain, Patient Experience integrated into a data warehouse

How Are You Going To Get There?

• How long will it take to reach Level 6?– With the right combination of technology, cultural

alignment and process improvement framework…– You can get there in 9 months, in one Clinical

Work Process area• How much will it cost?– At Northwestern, we spent about $1M per year

for three years to reach a robust, sustainable capability

Technical Options Available Today

• EMR vendors– No track record of success– Analytics are limited to the data collected in their products

• Build your own from scratch– Costly, risky… would you build your own EMR?

• Point solutions– One for JCAHO, one for physician performance, one for supply

chain, one for hospital operations, et al…– Redundant patch work of data; costly; not extensible;

enterprise wide analytics are not possible– Scarce analyst skills are spread across multiple products

Analytic Options Available Today (cont)

• Outsource and build from scratch– Consulting firms generally use this approach– Costly, risky

• Outsource and build with generic, reusable enterprise healthcare data model– IBM, Oracle, I2B2– Generic models = One size fits all = Poor adaptability– Are there any success stories?

• Outsource and build with adaptable, reusable data models, ETL, and data marts– Very few vendors in this space

WHAT’S NEXT?So… you have the Enterprise Data Warehouse technology

Quality Improvement: Focus on Process

• One of the fundamental concepts of quality improvement theory is to identify key work processes, then organize around them.

– A limited number of these processes make up the vast majority of services you provide to patients (80/20 rule). We want to prioritize this subset of key processes in our quality improvement efforts.

19

Key Process Analysis (KPA)

• 2002: Dr. David Burton, Tom Burton– Develop the KPA model while they were at Intermountain– Used the Enterprise Data Warehouse (EDW) as the

enabler• “How do we identify the care processes that offer

the greatest opportunity for quality and cost improvement?”– Normalized for apples-to-apples comparison across

clinical process families– Adjusted for severity

EDW

Case Mix Billing Data

Cost Data

APR/DRGGroupings

KPA Data Flow

Care Process Data Mart

EDW

KPA AlgorithmHighest

OpportunityCare Processes

KPA Visualization

Clinical Leadership

Detailed analysis of variation and outcomes

Analysts

Organizing Around Processes

For example…• Clinical Program: Women & Newborns– Care Process Family: Deliveries• Clinical Work Process: Vaginal & C-Section

– APR/DRG Grouper Codes (subset example here…)» 540 Cesarean» 542 Vaginal with complications» 560 Vaginal, normal delivery

Process the Data through the KPA Algorithm

1. Organize the codes into Work Processes

2. Calculate and rank by frequency (case count) and cost dollars for each clinical work process

3. Determine percentages of total cost dollars each clinical work process represents

23

Pareto Analysis In-patient Resources

24

Cumulative %

% of Total Resources Consumed for each sub-service line

Key Findings:

Number of Sub-Service Lines(e.g., Delivery, Medical Cardiology, Gastroenterology)

• 80% of all in-patient resources are represented by 23 Sub-Service Lines

23 CPMs

80%

9 CPMs

50%

• 50% of all in-patient resources are represented by 9 Sub-Service Lines

Prioritize

• Opportunity = Volume x Variation

– Removing variability in processes is the first step in process improvement and measurement

– As a general rule, standardization leads to lower cost and better outcomes

• Because we will not be able to work on all clinical work processes at once, we must have some way of prioritizing and planning our work to pursue the greatest opportunities for improvement first.

25

26

Internal Variation versus Variable Direct CostY-

Axi

s =

Inte

rnal

Var

iatio

n in

Var

iabl

e D

irect

Cos

t

Bubble Color = Clinical ProcessBubble Size = Case Count

X Axis = 2009-2011 Variable Direct Cost

1

2

3

4

Coefficient of Variation

• The Coefficient of Variation allows variation to be evaluated between data sets with different scales

Coefficient of Variation = Standard Deviation

Mean

27

Sample Data

• Admit dates 2009 – 2011

– 5.3M records/encounters

• Inpatient

– 776,895 records/encounters

• APRDRG, AdmitDate, and DischargeDate are not null

– 242,675 records/encounters

28

Case Count Pareto by Sub-Service Line

29

The Data By Clinical Area

RANK Service Line Sub-Service Line

% of Total

Running % Case Count

Work Process Count

1Women & Children's Delivery 12.2% 12.2% 29,432 2

2Women & Children's Normal Newborn 11.9% 24.1% 28,624 1

3 General Medicine Gastroenterology 7.2% 31.3% 17,302 134 Cardiac Services Medical Cardiology 6.9% 38.2% 16,739 135 General Medicine Pulmonology 6.2% 44.4% 14,904 96 Behavioral Health Psychiatry 5.1% 49.5% 12,304 107 Orthopedics Joint Replacement 3.8% 53.3% 9,077 38 General Surgery Colorectal/Lower GI 3.2% 56.5% 7,806 19 Behavioral Health Substance Abuse 2.8% 59.3% 6,845 3

10 General Medicine Nephrology 2.7% 62.1% 6,590 5

Sub-Service Line OptionsData Driven Criteria Review

Service Line Sub-Service Line Case Count Rank

Payments Rank

LOS Hours Rank

Variable Direct Cost

Rank

Variable Direct Cost

Opportunity Rank

Data Driven Criteria Results

Behavioral Health Psychiatry 6 13 2 9 1?

Women & Children's Delivery 1 1 1 1 2

?

Cardiac Services Medical Cardiology 4 5 6 3 3?

General Medicine Pulmonology 5 7 5 6 4?

Orthopedics Joint Replacement 7 2 11 2 5?

General Medicine Gastroenterology 3 6 3 4 6?

General Medicine Infectious Disease 11 8 9 8 7?

Oncology/Hematology (Medical) Oncology (Medical) 14 11 15 14 8

?

General SurgeryColorectal/Lower GI 8 3 7 7 9

?

Spine Fusion 22 9 23 11 10?

31

32

Women & Children’s: C-SectionAverage Variable Direct Cost per Case

by Provider by Severity Score

Bubble Color = APR DRG Severity ScoreBubble Size = Case Count for provider

X Axis = Average VariableDirect Cost per Case for provider

Y Ax

is =

Gro

uped

by

APR

DRG

- Se

verit

y Sc

ore

33

Women & Children’s: Vaginal DeliveryAverage Variable Direct Cost per Case

by Provider by Severity Score

Bubble Color = APR DRG Severity ScoreBubble Size = Case Count for provider

X Axis = Average VariableDirect Cost per Case for provider

Y Ax

is =

Gro

uped

by

APR

DRG

- Se

verit

y Sc

ore

34

Women & Children’s: Normal NewbornAverage Variable Direct Cost per Case

by Provider by Severity Score

Y Ax

is =

Gro

uped

by

APR

DRG

- Se

verit

y Sc

ore

Bubble Color = APR DRG Severity ScoreBubble Size = Case Count for provider

X Axis = Average VariableDirect Cost per Case for provider

35

Women & Children’s: Normal NewbornAverage LOS Hours per Case

by Provider by Severity Score

Y Ax

is =

Gro

uped

by

APR

DRG

- Se

verit

y Sc

ore

Bubble Color = APR DRG Severity ScoreBubble Size = Case Count for provider

X Axis = Average VariableDirect Cost per Case for provider

The Data Might Be Ready, But…

• When choosing a clinical process improvement area to address, consider…– Clinical leadership readiness– Can the vision for the clinical program be articulated?– External pressure and agendas

• Community, State, or Federal• Employers• Payer incentives• Donors

– Research agendas

Clinical ImpactWhat about…

There are dozens, but we are running out of time, so just a few around appendectomy…

Understanding Appendectomy LOS

Waste In Healthcare

• Don Berwick, JAMA, April 2012• Annual waste (2011 figures)– Failures of care delivery: $102B-$154B– Failures of coordinated care: $25B-$45B– Overtreatment: $158B-$226B– Administrative complexity: $107B-$389B– Pricing failures: $84B-$178B– Fraud and abuse: $82B-$272B

% of Appendectomy Patients Receiving Evidence Based Tests

Evidence-Based Antibiotic Use

Predictive Analytics, Watson, Big Data

What about…

The Problem with Predictive Analytics

• Predicting healthcare comes in two flavors– Too easy

• BMI 29, smoker, sedentary = multiple chronic diseases

• 65 yrs, living alone, post-CABG = re-admission

– Too hard: BMI 21, active, 39 yrs, non-smoker = stroke

• BRCA1 and BRCA2 mutations = 60% chance of breast cancer– Very difficult personally to take action…what

would you do?– Over 150 known genetic markers for risk that we

largely ignore

Predicting vs. Acting

• Even if we can predict, we have major obstacles against proactive mitigation & intervention– Culturally– Operationally– Behaviorally

• “We knew the [9/11] scenario was a risk, and some airlines had already put storm doors on their cockpits to mitigate hijackers, but we just didn’t push it [the mitigator] hard

enough.”

Suggestive Analytics©

• Surround the decision making environment with suggestions, based on analytic data–Much easier than predicting

• Worth reading– “Nudge: Improving Decisions About

Health, Wealth, and Happiness”

Suggestive Analytics

The Antibiotic Assistant

Antibiotic Protocol

Dosage Route Interval Predicted Efficacy

Average Cost/Patient

Option 1 500mg IV Q12 98% $7,256

Option 2 300mg IV Q24 96% $1,236

Option 3 40mg IV Q6 90% $1,759

47

• An EMR clinical decision support tool developed at Intermountain Healthcare

The Antibiotic Assistant Impact

• Complications declined 50%

• Avg # doses declined from 19 -> 5.3

• The replicable and bigger story– Antibiotic cost per treated patient: $123 -> $52

– By simply displaying the cost to physicians

48

Watson

• First, a little background–National Security Agency–Natural Language Processing and Text

Mining

• Watson is revolutionary. – It’s the first thing in my IT career that really

excited me… everything else has been incremental or variations of the same flavor

Watson’s Technology

• Apache – Unstructured Information Management

Architecture (UIMA)– Hadoop– Java, C++

• Lexicals and ontologies–  DBPedia, WordNet, and Yago 

• IBM Content Analytics with Enterprise Search

• 90 IBM Power 750 servers enclosed in 10 racks

• 16 Terabytes of memory• A 2,880 processor core• Linux based• Estimated to have cost $1B - $2B

What is Watson?

• Near-word associations coupled with semantic mapping and zillions of sources of knowledge… digitized books, encyclopedias, news feeds, magazines, blogs, Wikipedia, etc.– Equivalent to approximately 240 million

pages, in memory• Jeopardy answer– “A famous red coiffed clown or just any

incompetent fool”• Watson’s correct answer– “Who is Bozo?”

• Watson searched its indexes for near-word associations, recognized that Bozo was the most common word in the indexes that was missing from the question

Watson’s Problem With Healthcare

• Watson’s training set for Jeopardy was a HUGE collection of human wisdom, academic and otherwise, stretching back 1,000 years– Wikipedia; digitized books, magazines, newspapers,

journal

• What’s the training set for healthcare wisdom? – A few decades of clinical trials and journals? – Claims processing data from a dysfunctional

healthcare system? – No outcomes data to speak of…– Progress notes? Radiology reports?

Pathology reports?• Watson is not going to impact healthcare in

the near term like many hope it will…but it’s still very cool

Big Data Technology• Big Data technology was built to process

web log, semi- and non-structured tagged text on a massive basis for Google, Amazon. eBay, Facebook, etc.– Our text data in healthcare is not massive and is not

tagged– We (healthcare) are small fry data compared to Silicon

Valley … we can solve our analytic needs with less complex technology

• Completely different business processes and information context– Googles and Facebooks of the world are collecting and

analyzing data about their business processes that are completely different in content and structure than anything in our current healthcare environment  

• The skills required for big data are equally big and more rare than platinum

• No major impact on healthcare other than “gee whiz” for at least 4-6 years.

Questions, Thoughts, or Challenges?

[email protected]@hqcatalyst.com Cell/Text: 970-403-6090LinkedIn: http://www.linkedin.com/in/dalersanders