Transforming Health Care Through Big Data Science Innovation

Transforming Health Care Through

Big Data Science Innovations

Georgia Tourassi, PhD

2015 ORAU Annual Meeting Oak Ridge, TN March 5, 2015

A Nation in Crisis

2

Predictions about US Health in 2030…

3

Projected Popula0on: 365+ million


4


More than 72 million elderly


5




6


300% increase in healthcare costs

$9 trillion/yr


1.  What are the challenges of healthcare delivery?

2.  What is the value of Big Data for healthcare transformation?

3.  What are the challenges with Big Health Data?

7

Outline

Healthcare Challenges

Be#er delivery

Be#er outcomes

Fewer dispari5es

Lower cost

8

$ 2.3 Trillion / yr

The Practice of Medicine is a SCIENCE

“The fundamental problem with the quality of American medicine is that we have failed to view delivery of health care as a science. …That’s a mistake, a huge mistake.”

9

Peter Provonost, MD Professor, Anesthesiology and Cri0cal Care Medicine, and Surgery Professor Health Policy & Management Johns Hopkins University

Transform healthcare through effective use of information

10

Big Data

Effec5ve use of IT

Proac5ve care

Quality metrics Lower Costs

Government regula5ons

Where is the big health data coming from?

•  Typical EHR size –  ~1MB healthy young, no images –  40 MB middle-‐aged w/ health issues,

no images –  3-‐5 GB w/ health issues and images

•  Es0mated size of current US digi0zed data

–  600 petabytes

•  Es0mated different medical devices

–  1.5 million

•  2015: global health data, 20 Exabytes

11

Clinical Data Medical History

Imaging Data Expression arrays

Personal Genomics

Telehealth / sensor data

Health Claims Social Media

Healthcare data is growing by 15 Petabytes a day in the US. Up to 80% of healthcare data is currently unstructured.

Where is the big health data coming from?

12

Are we ready for healthcare data integration?

13

But what about value?

“Big Data are data whose scale, diversity and complexity require new architecture, techniques, algorithms, and analy0cs to manage it and extract value and hidden knowledge from it” IMIA working group on “Data Mining and Big Data Analy0cs” From R. Bellazi, IMIA Yearbook of Medical Informa0cs 2014

14

The Maslow Pyramid of Big Data Needs

15

Stage 4: Wisdom

Stage 3: Knowledge Genera5on

Stage 2: Informa5on Retrieval

Stage 1: Data Collec5on & Storage

Decisions

Predic0ons Visualiza0on Repor0ng

Sta0s0cs Analy0cs Querying

ETL Data Fusion Data Integra0on

The Big Data Value Proposition

•  Health care efficiency via beier access to pa0ent data (par0cularly for addressing health dispari0es)

•  Earlier disease detec0on via real-‐0me analysis of mul0modality pa0ent data (e.g. mHealth, ICU)

•  Personalized pa0ent care •  Popula0on health management by con0nuously aggrega0ng

and analyzing public health data •  Fraud, waste, and abuse detection

16

McKinsey Global Ins0tute – May 2011: Big Data: The next fron0er for innova0on, compe00on and produc0vity

What are the computing challenges with Big Health Data?

17

Real 0me access to data

Data integra0on

Extreme scalability

Radical flexibility

Storage infrastructure

Pa0ent privacy

Data governance

MUST: Holistic View of the Lifecycle of Data-Intensive Discovery

Can we scale up in all three aspects of data-driven discovery?

Opera5onal Flow

Descrip5ve Analy5cs

Diagnos5c Analy5cs

Predic5ve Analy5cs

Prescrip5ve Analy5cs

Concept adapted from Gartner

Analy5cal Flow

History •  Formed in 2013 to integrate ORNL’s data-‐

driven, data-‐intensive biomedical research programs.

•  HDSI members include biomedical researchers, system architects, data scien0sts, computer scien0sts, IT services, HPC opera0on experts

Vision •  Accelerate data-‐driven biomedical discoveries

and healthcare delivery advancement Mission •  Develop innova5ve, scalable, and robust

technologies for organizing, integra0ng, and analyzing complex data at scale

Priori5es: •  Deliver methodological and applied scien0fic

innova0ons, informa0cs tools, and compu0ng infrastructure to enable effec0ve use of data for individual and public benefit.

•  Advance a broad range of sponsor and health policy priori0es while serving as a neutral en5ty.

•  Build health data science community capacity via a User Facility for collabora0ve engagement and targeted educa0on and training.

Health Data Sciences Institute at ORNLAdvancing the Utility of Data to Achieve Better Health Outcomes at Lower Cost

Innovate

Incubate

Accelerate

Research focus areas

Health Informa5cs Computa0on

for personalized solu0ons

Mul0-‐modality data analy0cs

Computer-‐aided decision support

Informa0ve visualiza0on

Health Economics and Policy

Computa0on for transla0onal impact

in healthcare management and policy

Detec0on and early predic0on of misuse, underuse, or overuse of health services

Healthcare system M&S: Analysis of pa0ent,

provider, and health system

interac0ons

Popula5on Health Dynamics Collec0on, analysis,

and modeling of data for disease and human health behavior surveillance

Social media analy0cs to study disease spread, health behaviors,

informa0on dissemina0on

Digital epidemiology

Analysis of heterogeneous unstructured big health data

Key contribu5on: A scalable, interac0ve analy0c and visualiza0on plasorm for exascale data

Data analysis

Exploratory data analysis

Confirmatory analysis

Evidence gathering

Visual analysis

CMS: Fraud detection

Descrip5ve Analy5cs

Diagnos5c Analy5cs

Predic5ve Analy5cs

Prescrip5ve Analy5cs

Descriptive Analytics: What happened ?

Diagnostic Analytics: Why did it happen ?

Predictive Analytics: What will happen ?

Prescriptive Analytics: How to stop it from happening ?

Individual Provider Payer

Health state

Demographics

Cogni5ve characteris5cs

Behavioral traits

News Media

Social Media

Individuals

Diffusion processes •  Thresholds •  Beliefs •  Attudes •  Preferences •  Behaviors

Group communica5on processes, social networks

Decision to seek treatment

Disease Progression

Health outcomes

Care Progression

Diagnosis, Treatment selec5on

Decision to seek insurance

Claims

Pa5ent Compliance

Payment

Diagnos5c models

Care Progression models

Type

Specialty

Quality of Care

Cogni5ve characteris5cs

Provider Organiza0ons

(ACO)

Payment models

Regional varia5on

Professional Organiza0ons

Providers

Diffusion processes •  Treatment paierns •  Base rates

Deduc5bles

CoPays

Extent of coverage

Modeling and simulating healthcare delivery futures at scale

Digital epidemiology: Environmental Risk Factors

Smart Crawler

WWW Profiles

Match

NLP

Detect Iden0ty: Name, Job, Loca0on

Survivor’s LinkedIn Profile

Gender: First Name

Age: High school or College Year

Lifeline: Educa0on and Work History

Crawled 5,000,000 Pages

37,500 PaFent Stories 12,500 IdenFFes Found

1,200 LinkedIn Profiles Matched

Link EPA data Compare aggregate environmental exposure profiles between cancer and controls

Summary

25

Gordon Gekko, Wall Street (1987) •  The most valuable commodity I know of is informaFon (data).

Lao Tzu •  To aQain knowledge, add things every day. To aQain wisdom, subtract things every day.

Ronald Coase (Nobel Prize in Economics, 1991) •  Torture the data and nature will always confess.

Thank you

26

Georgia Tourassi,

PhD

Director,

Health

Data Sciences Institute

Oak

Ridge

National Laborato [email protected]

Georgia Tourassi, PhD Director, Health Data Sciences Ins5tute

Oak Ridge Na5onal Laboratory [email protected]

Transforming Health Care Through Big Data Science Innovation

Data & Analytics

big health data

global health data

data mining

value of big data

big data analy0cs

healthcare data integration

public health data fraud

us health