turning big data insights into action through advanced analytics

© 2014 IBM Corporation

C203: Big Data and Healthcare: Key Technologies and Strategies

Turning Big Data Insights into Action through Advanced Analytics

DATA SUMMIT 2014

Craig Rhinehart

Director, IBM Smarter Care Strategy and

Market Development

© 2014 IBM Corporation 2

C203: Big Data and Healthcare: Key Technologies and Strategies

2:00 pm - 2:45 pm

Turning Big Data Insights into Action through Advanced Analytics

•Healthcare is being disrupted by the collision of a wave of innovation. The emergence of Big Data, cloud computing, analytics, and social media and increasing wellness awareness through mobile devices are revolutionizing patient-centered connected care. This session discusses the new technologies and approaches that offer promise in terms of better data understanding for organizations and individuals.

Hashtag: #BigDataNY

Twitter: @CraigRhinehart, @DBTADataSummit

Craig Rhinehart Blog: http://craigrhinehart.com


Please note

IBM’s statements regarding its plans, directions, and intent are subject to change or withdrawal without notice at IBM’s sole discretion.

Information regarding potential future products is intended to outline our general product direction and it should not be relied on in making a purchasing decision.

The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion.

Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon many factors, including considerations such as the amount of multiprogramming in the user’s job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve results similar to those stated here.


• Time Machine: A Decade of Reversal: An Analysis of 146 Contradicted Medical Practices

• Case Study: Seton Healthcare• Case Study: University of North Carolina School of

Medicine• Introduction to Smarter Care• Questions and Discussion

4

Topics


Who Do You Trust?

A Decade of Reversal:

An Analysis of 146 Contradicted Medical Practices

“40.2% reversed the original standard

of care … and only 38.0% reaffirmed

the original standard of care”

“Amputations or Analytics …”Craig Rhinehart Blog


(Billion) Wasted on missed opportunities along with unnecessary, error-prone and inefficiently delivered services 3

$585B

US physicians still reliant on paper based medical records systems 4

45%

Hours each day an average family physician spends on direct patient care 3

8.0

Hours required to meet the patient care guidelines each day 3

21.7

1 World Health Statistics 2011 from World Health Organization2 The World Health Report 2000 – Health Systems: Improving Performance from World Health Organization3 Best Care at Lower Cost: The Path to Continuously Learning Health Care in America from Institute of Medicine / National Academy of Sciences4 Physician Adoption of Electronic Health Record Systems: United States, 2011 from National Center for Health Statistics

US rank in Healthcare spending 1

1st

US rank in quality of care delivered 2

37th

80%90%

of the world’s data is unstructuredof the world’s data was created in the last two years

An Ocean of Unused Data

Healthcare Transformation: A Work in Progress


Unstructured data is messy but filled with key

medical facts

Medications, diseases, symptoms, non-symptoms,

lab measurements, social history, family history …

and much more


20% of people generate80% of costs

Health care spending

Health status

Healthy Low Risk

High RiskAt RiskActive

Disease

EarlyClinical

Symptoms

Disease and cost-of-care progression

Time

Early intervention opportunities identification

Early intervention opportunities identification

70% of US deaths from chronic diseases


Seton Healthcare FamilyReducing CHF readmission to improve care

Business ChallengeSeton Healthcare strives to reduce the occurrence of high cost Congestive Heart Failure (CHF) readmissions by proactively identifying patients likely to be readmitted on an emergent basis.

What’s Smart?IBM Content and Predictive Analytics for Healthcare solution will help to better target and understand high-risk CHF patients for care management programs by:

Smarter Business Outcomes• Seton will be able to proactively target care management

and reduce re-admission of CHF patients.• Teaming unstructured content with predictive analytics,

Seton will be able to identify patients likely for re-admission and introduce early interventions to reduce cost, mortality rates, and improved patient quality of life.

IBM solution• IBM Content and

Predictive Analytics for Healthcare

• IBM Cognos Business Intelligence

• IBM BAO solution services

• Utilizing natural language processing to extract key elements from unstructured History and Physical, Discharge Summaries, Echocardiogram Reports, and Consult Notes

• Leveraging predictive models that have demonstrated high positive predictive value against extracted elements of structured and unstructured data

• Providing an interface through which providers can intuitively navigate, interpret and take action

“IBM Content and Predictive Analytics for Healthcare uses the same type of natural language processing as IBM Watson, enabling us to leverage information in new ways not possible before. We can access an integrated view of relevant clinical and operational information to drive more informed decision making and optimize patient and operational outcomes.”

Charles J. Barnett, FACHE, President/Chief Executive Officer, Seton Healthcare Family

Featured on


The Data We Thought Would Be Useful … Wasn’t

• Structured data not available, not accurate enough, without the unstructured data - which was more trustworthy

What We Thought Was Causing 30 Day Readmissions … Wasn’t

• 113 possible candidate predictors expanded and changed after mining the data for hidden insights

New Hidden Indicators Emerged … Readmissions is a Highly Predictive Model

• 18 accurate indicators or predictors (see next slide)

Predictor Analysis % EncountersStructured Data

% Encounters Unstructured Data

Ejection Fraction (LVEF) 2% 74%

Smoking Indicator 35%(65% Accurate)

81%(95% Accurate)

Living Arrangements <1% 73%(100% Accurate)

Drug and Alcohol Abuse 16% 81%

Assisted Living 0% 13%

What Really Causes Readmissions at Seton

Key Findings from Seton’s Data

97% at 80th percentile

49% at 20th percentile


What Really Causes Readmissions at Seton

Top 18 Indicators

1. Jugular Venous Distention Indicator

2. Paid by Medicaid Indicator3. Immunity Disorder Disease Indicator4. Cardiac Rehab Admit Diagnosis with CHF Indicator5. Lack of Emotion Support Indicator6. Self COPD Moderate Limit Health History Indicator7. With Genitourinary System and Endocrine Disorders8. Heart Failure History9. High BNP Indicator10. Low Hemoglobin Indicator11. Low Sodium Level Indicator12. Assisted Living13. High Cholesterol History14. Presence of Blood Diseases in Diagnosis History15. High Blood Pressure Health History16. Self Alcohol / Drug Use Indicator17. Heart Attack History18. Heart Disease History

New Insights Uncovered by Combining Content and Predictive Analytics

• Top indicator JVDI not on the original list of 113 - as well as several others

• Assisted Living and Drug and Alcohol Abuse emerged as key predictors - only found in unstructured data

• LVEF and Smoking are significant indicators of CHF but not readmissions

• A combination of actionable and non-actionable factors cause readmissions


The Impact of Readmissions at Seton

CHF Patient X – What Happened? Admit / Readmission

30-Day Readmission

98% 98% 96% 95% 96% 100%

Apr-18-2009 May-12-2009 May-20-2009 Oct-11-2009 Nov-24-2009 Dec-20-2009

8 days24 days 144 days 44 days 26 days

Individual Patient Data at Each Encounter (Patient X @ Dec 20, 2009)

Patient X was hospitalized 6 times over an 8 month period. The same basic information was available at each encounter and Patient X’s readmission prediction score never dropped below 95% (out of possible 100%)

Patient Population Monitoring Clinical and Operational Data


Overview

• Improve collection of Physician Quality Reporting System (PQRS) measures

• Prepare for meaningful use (Stage 3) and improve test result follow-up

• Avoid Centers for Medicare and Medicaid Services (CMS) readmission penalties


Physician Quality Reporting System (PQRS) measures

PQRS is a reporting program that uses a combination of incentive payments and payment adjustments to promote reporting of quality information by eligible professionals (EP)

• Services furnished for Medicare Part B

PQRS measures• Breast cancer screening (mammograms)• Colorectal cancer screening• Pneumovax


Problem

Mechanism for capturing procedures performed at outside institutions is unreliable:• Mammogram• Colo-rectal cancer (CRC) screening• Pneumovax

PQRS measures are typically captured using structured Electronic Medical Record (EMR) data


Health maintenance in electronic medical record


Free-text note in patient’s EMR

Not captured in structured data


Physician Quality Reporting System (PQRS) measures

UNC Healthcare utilized IBM Content Analytics (ICA) to analyze free-text clinical documents to improve the accuracy of our 2012 PQRS measures

We used ICA to analyze clinical notes for those patients whose structured EMR data did not indicate compliance with PQRS measure requirements

• Mammogram, CRC screening, Pneumovax


Results

142

155

236

49

27

247

Mammogram(297)

CRC screening(285)

Pneumovax(271)NLP Capture of PQRS

Measure

0

50

100

150

200

350

300

250

Nu

mb

er

of

pa

tie

nts

ou

t o

f c

om

pli

an

ce

(s

tru

ctu

red

da

ta)


Results

60

53

55

61

Mammogram CRC screening Pneumovax

Structured data

48

50

52

54

56

64

62

58

Pe

rce

nt

of

pa

tie

nts

me

eti

ng

re

qu

ire

me

nts

60

Structured data + NLP

53

63


Follow-up of abnormal test results documented in free-text reports

Mammograms | Colonoscopies | Pap tests | Chest x-rays


Breast cancer and mammograms

• Second deadliest cancer in women

• Screening mammograms can reduce mortality

28% of women requiring short-term follow-up for abnormal mammography results do not receive recommended care


Mammography reports


Performance of NLP for accurately identifying mammography results

Characteristics Value

Precision 98%

Recall 100%

F Measure 99%

• Precision = positive predictive value If NLP reports a result, the probability that the result is correct

• Recall = sensitivityif there is a result in the mammography report, the probability that NLP identifies and accurately classifies the result

• F-measure = summary measure of NLP performance

Revised NLP model has 100% performance for all characteristics


Structured Data is Not Enough Unstructured data significantly increases the

richness and accuracy of analysis and decision making … including paper / faxes

Today’s Care Guidelines Only Get You So Far Not granular enough to deliver on the promise of

personalized medicine with data driven insights 1, 2

Manual Processes and Traditional Workflow Approaches Don’t Work Process complexity increases with disease

complexity … changing conditions require process adaptability 3

Knowledge and Guidelines

Data Driven Insights

1. Dijun Luo, Fie Wang, Jimeng Sun, Marianthi Markatou, Jianying Hu,Shahram Ebadollahi, SOR: ScalableOrthogonal Regression for Low-Redundancy Feature Selection and its Healthcare Applications. SDM’12

2. Jimeng Sun, Jianying Hu, Dijun Luo, Marianthi Markatou, Fei Wang, Shahram Edabollahi, Steven E. Steinhubl, Zahra Daar, Walter F. Stewart. Combining Knowledge and Data Driven Insights for Identifying Risk Factors using Electronic Health Records. Under submission at AMIA’12

3. Blind Surgeon Metaphor Problem - W.M.P. van der Aalst, M. Weske, and D. Grünbauer. Case Handling: A New Paradigm for Business Process Support. Data and Knowledge Engineering, 53(2):129-162, 2005

1. Dijun Luo, Fie Wang, Jimeng Sun, Marianthi Markatou, Jianying Hu,Shahram Ebadollahi, SOR: ScalableOrthogonal Regression for Low-Redundancy Feature Selection and its Healthcare Applications. SDM’12

2. Jimeng Sun, Jianying Hu, Dijun Luo, Marianthi Markatou, Fei Wang, Shahram Edabollahi, Steven E. Steinhubl, Zahra Daar, Walter F. Stewart. Combining Knowledge and Data Driven Insights for Identifying Risk Factors using Electronic Health Records. Under submission at AMIA’12

3. Blind Surgeon Metaphor Problem - W.M.P. van der Aalst, M. Weske, and D. Grünbauer. Case Handling: A New Paradigm for Business Process Support. Data and Knowledge Engineering, 53(2):129-162, 2005

What Have We Learned So Far?


A Smarter approach to delivering better outcomes

• Build longitudinal “data driven” evidence based population insights

• Uncover hidden intervention opportunities

• Proactively deliver accountable and personalized care in a patient centered model

• Collaborate across caregivers to focus on high cost, high need patients

• Prevent at-risk patients from progressing to high cost, high need

• Build longitudinal “data driven” evidence based population insights

• Uncover hidden intervention opportunities

• Proactively deliver accountable and personalized care in a patient centered model

• Collaborate across caregivers to focus on high cost, high need patients

• Prevent at-risk patients from progressing to high cost, high need

IBM Strategy to Support Care Coordination:


The path forward

IBM Smarter Care uncovers valuable insights into lifestyle choices, social determinants, and clinical factors enabling holistic and individualized care to optimize outcomes and lower costs

Social

determinants such as where one is born,

grows, lives, works and ages have direct

impact on an individual’s overall health and

well being

Lifestyle

choices have direct impact on an

individual’s mental and physical wellness

Lifestyle

choices have direct impact on an

individual’s mental and physical wellness

Clinical

factors such as specific medical symptoms,

history, medications, diagnoses, etc are

indicators of an individual’s health

© 2014 IBM Corporation 28© 2013 IBM Corporation28

Find out more about IBM Smarter Care

http://www.ibm.com/smarterplanet/us/en/smarter_care/overview/

Visit my blog or follow me on Twitter

http://craigrhinehart.com

@CraigRhinehart

Craig RhinehartDirector, IBM Smarter Care Strategy and Market [email protected]

Thank You

turning big data insights into action through advanced analytics

Data & Analytics