strata big data feb. 26, 2013 let the data decide: predictive analytics in healthcare

44
Strata Big Data Strata Big Data Feb. Feb. 26, 2013 26, 2013 Let the Data Decide: Let the Data Decide: Predictive Analytics in Healthcare Predictive Analytics in Healthcare Eugene Kolker Eugene Kolker [email protected] , , [email protected]

Upload: cyrah

Post on 20-Jan-2016

22 views

Category:

Documents


0 download

DESCRIPTION

Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare Eugene Kolker [email protected] , [email protected]. Outline. Appetizer: Introduction 2. Main Course: Prioritized Improvements (US News & World Report Metrics) - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Strata Big DataStrata Big Data Feb. 26, 2013Feb. 26, 2013

Let the Data Decide:Let the Data Decide:Predictive Analytics in HealthcarePredictive Analytics in Healthcare

Eugene KolkerEugene [email protected],,

[email protected]

Page 2: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

1. Appetizer: Introduction

2. Main Course: Prioritized Improvements(US News & World Report Metrics)

3. Second Course: Personnel & Reduction of Waste (Nurses’ Turnover Trends)

4. Dessert: General Observations

Thanks to Edd Dumbill, Alistair Croll,O’Reilly Media, the Organizers, & You!

OutlineOutline

Page 3: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

1. Why Should You Care?1. Why Should You Care?

DataData

KnowledgeKnowledge

ActionAction

BenefitsBenefits

Our motto: Our motto: Accelerating Accelerating and optimizing your work and optimizing your work through intuitive, reliable through intuitive, reliable and powerful analyticsand powerful analytics

Page 4: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

What is Seattle Children’s?What is Seattle Children’s?

Seattle Children’s: Seattle Children’s: Hospital – Research – FoundationHospital – Research – Foundation

SCH: Non-profit, network, tertiary, 100 y.o.

$0.8 Bln/yr, 5,000 FTEs, 350 beds (plans: 600)

SCH covers 5 States (WWAMI region), 0-21 y.o.

6th on USNWR ranking of Ch. Hospitals

5th on Federal ranking of Ch. Research Institutes

Page 5: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Chief Data Officer Chief Data Officer @ SC

DirectorDirector, Bioinformatics & High-throughput Analysis LabBioinformatics & High-throughput Analysis Lab,

SC Research Institute

Affiliate Professor Affiliate Professor @ Depts of Biomedical Informatics & Medical Education and Pediatrics, University of Washington

MS in Applied Mathematics & Computer Sciences,

PhD in Structural Molecular Biology (Bioinformatics) +

Business School

Executive and Founding Executive and Founding EditorEditor of: of:

““OMICSOMICS A Journal of Integrative Biology A Journal of Integrative Biology” and “Big DataBig Data””

Who is EK? Who is EK?

Page 6: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

In today’s data-driven age, healthcare is transitioning In today’s data-driven age, healthcare is transitioning

from opinion-based decisions from opinion-based decisions to informed decisions to informed decisions

based on data and analytics. based on data and analytics.

Analyzing the data reveals trends and knowledge Analyzing the data reveals trends and knowledge

that may that may run contrary to our assumptions run contrary to our assumptions causing causing

a a shift in ultimate decisions shift in ultimate decisions that in turn will that in turn will better better

serve both patients and healthcare enterprisesserve both patients and healthcare enterprises . .

AbstractAbstract

Page 7: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Helmet laws“helmet laws are associated with a 13% reduction in bicycle-related

head injuries, a 9% reduction in non-head bicycle-related injuries, and

an 11% increase in all types of injuries from the wheeled sports.“

Buying habits“conservatives like established national brands—and are significantly

less likely to try new items”

Sober vs. intoxicated eye witnesses“Intoxicated eyewitnesses are no less reliable than sober ones, and

neither is very good at picking kidnappers out of a lineup”

WSJ Data news: Last weekend, Feb. 16-17WSJ Data news: Last weekend, Feb. 16-17

Page 8: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Volume, Veracity, Velocity, Variety, and Value

Banking/Marketing/IT: Volume, Velocity, and Value

Healthcare/Life Sciences:Healthcare/Life Sciences:Veracity, Variety, and ValueVeracity, Variety, and Value

5 Vs of Big Data5 Vs of Big Data

Page 9: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

This talk illustrates This talk illustrates our collaborative work our collaborative work

with key stakeholders, including with key stakeholders, including

executive leadershipexecutive leadership, and describes , and describes

a few a few representative, data-driven, and representative, data-driven, and

cost-effective projectscost-effective projects..

Abstract, Cont.Abstract, Cont.

Page 10: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

US News & World Report US News & World Report (USNWR) Metrics(USNWR) Metrics

Sponsors:

David Fisher, SVP, Medical Director andTom Hansen, CEO

Objective: to prioritize enterprise-wide

improvements based on USNWR Metrics (utilized as

Hospital & Departmental Metrics)

2. 2. Prioritized Improvements

Page 11: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Prioritized Improvements, Cont.Three key recommendations:

1. Focus on care and outcomes for A1 and A2Departments (medical service lines)

2. A1 and A2 Department-specific Marketing3. Implement 1-Day Immunization Reporting

This work is described in Kolker E. & Kolker E., Chief Data Officer in Healthcare: Predictive Analytics

Transforms Data to Knowledge to Action,

In: Chief Data Officer: Enterprise Data Solution for Business Challenges, MIT Press, 2013, in

press.

Since 2007, SC moved from 11th to 6th rank

Page 12: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

USNWR 2012 Honor Roll

Rank Hospital Points Departments

1 Boston Children's Hospital 20 10

1 Children's Hospital of Philadelphia 20 10

3 Cincinnati Children's Hospital Medical Center 19 10

4 Texas Children's Hospital, Houston 13 8

5 Children's Hospital Los Angeles 6 5

6 Seattle Children's Hospital 5 4

7 Nationwide Children's Hospital, Columbus, Ohio 4 3

7 Children's Hospital Colorado, Denver 4 3

9 Children's Hospital of Pittsburgh of UPMC 3 3

9 Johns Hopkins Children's Center, Baltimore 3 3

9 Ann and Robert H. Lurie Children's Hospital of Chicago

3 3

9 St. Louis Children’s Hospital- Washington University

3 3

Page 13: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Model for C1 department

Reconstructed

model based on

provided data

Empirically

determined

transformation

applied to data

Page 14: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

FY12-F11 Changes

Department 2012 rank 2011 rank Change Points

B3 7 8 +1 1

B1 11 15 +4 0

C2 19 20 +1 0

C1 14 19 +5 0

B2 17 22 +5 0

C4 4 2 -2 2

A1 8 10 +2 1

C3 22 17 -5 0

A2 11 11 0 0

B4 5 7 +2 1

Overall 6 7 +1 5

Page 15: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Prioritizing Improvements

Department Percent score increase needed for goal

Percent score increase needed for stretch goal

A1* + 0.8%

A2* 1.4% 7.8%

B1** 4.2% 8.4%

B2** 4.4% 12.0%

B3** + 4.4%

B4** + 5.3%

C1 8.8% 24.6%

C2 10.7% 18.3%

C3 17.1% 26.3%

C4 + +

*First and **Second priority improvements

Page 16: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Guiding Improvements

Categories broken down for each departmentCalculated as:

maximum possible increaseneeded increase

Need total of 100 points in a column

to reach goalStill, reputation has major influence in every

department, however, there are numerous important factors to be improved

Page 17: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

A1 Department

Category Points forgoal

Points forstretch goal

Reputation -- 2,438

Nurse-patient ratio -- 377

X1 management -- 222

Clinic volume -- 102

X2 treatment volume -- 96

Commitment to best practices -- 89

Surgical volume -- 87

Overall infection prevention -- 46

Specialized clinics and programs -- 38

Advanced clinical services -- 32

Subspecialist availability -- 24

Page 18: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

A2 Department

Category Points forgoal

Points for stretch goal

Reputation 1,177 213

Preventing deaths of Y1 patients 272 49

Success in reducing ICU infections 181 32

Y2 management 160 29

Y3 management 149 27

Y4 management 136 24

Nurse-patient ratio 110 19

Y5 management 93 17

Patient volume 70 12

Overall infection prevention 67 12

Nonsurgical procedure volume 59 10

Commitment to best practices 29 5

Advanced clinical services 13 2

Page 19: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Recurring Categories

Category Number of departments

Reputation 9

Advanced clinical services 9

Nurse-patient ratio 9

Overall infection prevention 8

Commitment to best practices 8

Patient volume 6

Surgical volume 5

Success in reducing ICU infections 5

Page 20: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

2. Implement 1-Day Immunization Reporting

All categories have both general & departmental measurements

1. Advanced clinical services

- Services and programs organized around a particular

diagnosis, disease, need, or age group

2. Overall infection prevention

- Hospital commitment to reducing infection risk

(tracking infections, immunization reporting, etc.)

3. Commitment to best practices

- Includes participating in conferences, safety procedure

guidelines, database tracking, etc.

4. Success in reducing ICU infections

- Rates of infection in ICUs

Page 21: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Nurses’ Turnover TrendsNurses’ Turnover Trends

Sponsors: Lisa Brandenburg, Hospital President

andSteven Hurwitz, VP, HR

ASK: something is happening with nurses WHAT? WHY? and HOW to deal with it?

3. Personnel + Reduction of Waste3. Personnel + Reduction of Waste

Page 22: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Nurses’ Turnover TrendsNurses’ Turnover Trends

Findings:

1. Termination rates higher after Magnet Status

2. After Magnet Status more experienced nurses leaving more often

3. Overall termination decreases with experience, especially Involuntary termination

Page 23: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Nurses’ Turnover Trends, Cont.Nurses’ Turnover Trends, Cont.

4. Termination higher for VPs A & E, lower for others

5. Higher termination for nurses living in Seattle

6. No difference in termination for night versus day shifts

Page 24: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

MethodsMethods

For this initial look, we broke down nurses’ turnover into 3 categories: Active, Involuntary, and Voluntary terminations.

We initially looked at differences in age, gender, years since hired, whether they had been rehired, department, ethnicity, and FTE.

We have added comparisons on reporting VP, Seattle residency status, and shift.

We also compared 2 time periods:Before and After Magnet Status

Page 25: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Methods, Cont.Methods, Cont.

For all comparisons except time period, odds ratios (with

95% Confidence Interval) were calculated for each variable:Odds = P(termination)/P(active)Odds Ratio (OR)=odds(Male)/odds(Female), e.g.

Hence, an OR = 1 implies no difference in termination rates, OR > 1, Males (or whatever category) has higher termination rate, OR < 1 lower termination rate

Analyses were done unadjusted as well as with

an adjustment for age and adjustments for age and

experience (years since hire).

Page 26: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Conclusion 6: SHIFT (Night Conclusion 6: SHIFT (Night vs.vs. Day) Day)

Unadjusted Age Adj. Age and Exp. Adj.

INV TERMVOL TERM

Od

ds R

atio

0.2

0.5

1.0

2.0

5.0

10

.0

Termination looks higher on night shift, but the difference gone after adjusting for age and experience.

Page 27: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Conclusion 5:Conclusion 5:ZIP in Seattle (Seattle ZIP in Seattle (Seattle vs.vs. Other areas) Other areas)

Unadjusted Age Adj. Age and Exp. Adj.

INV TERMVOL TERM

Od

ds R

atio

12

5

Termination higher for nurses living in Seattle.

Page 28: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Conclusion 4: VPs (Conclusion 4: VPs (vs. vs. other VPs)other VPs)

Involuntary TerminationInvoluntary Termination

Involuntary termination higher for A and E. Note – OR = 1 for D (Unadj. and Age Adj.).

Unadjusted Age Adj. Age and Exp. Adj.

0.1

0.2

0.5

1.0

2.0

5.0

10

.0

A

ED

CB

Page 29: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

29

Conclusion 2: Before and After Magnet StatusConclusion 2: Before and After Magnet Status

Experience of Terminated NursesExperience of Terminated Nurses

Experience of terminated nurses is higher After Magnet Status (Age Adj.).

Mean Experience

Involuntary Termination

Voluntary Termination

Before Magnet 0.14 0.63

After Magnet 0.8 1.5

Page 30: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

3. Three Follow-up Actions3. Three Follow-up Actions

1. Discussions with (experienced) nurses

2. Bringing external consultant in-house

(psychology, sociology, nursing)

3. Hiring re-adjustments

Page 31: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Bottom Line for 2. & 3. Bottom Line for 2. & 3.

Do you want to:

A: Improve the health of your patients

B: Cut huge amounts of waste

C: Increase your rankings?

How about all three?How about all three?

Page 32: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Working Together

4. General Observations4. General Observations

Page 33: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

EXA_3: Improve Care + Cost Savings

Summary:

1. Medically Complex Patients (2+ Chronic Diseases), 80-20 rule

2. Question: Number of Medications? Answer: 5+

3. Extremely complicated model with simple Q&A

Sponsor: Mark Del Beccarro, VP, Medical Affairs

Page 34: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

EXA_4: Model of Seattle Downtown

Champions: Champions: Blake Nordstrom, Matt Griffin, Blake Nordstrom, Matt Griffin, Jim Hendricks + DSAJim Hendricks + DSA

• An An index of Downtown vitality index of Downtown vitality which examineswhich examines four categories: four categories: Live, Work, Shop, Live, Work, Shop, andand PlayPlay

• Enables comparison of Downtown across time Enables comparison of Downtown across time

• 20052005 is baseline with score of 100 is baseline with score of 100

Page 35: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

50

75

100

125

150

2005 2006 2007 2008 2009

Inflation Adjusted Data

Live

Work

Shop

Play

IS

Vitality Index: Integrated Score

Page 36: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

IS

100

Work

Shop

Live

Play

IS

100%

Dashboard: Integrated Score (Inflation Adj.)

Live

75%

WorkIS

Play

Shop

Page 37: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

EXA_5: DELSA Global, EXA_5: DELSA Global, delsaglobal.orgdelsaglobal.org

Data-Enabled Life Sciences Alliance Data-Enabled Life Sciences Alliance (DELSA Global)(DELSA Global) DataData

KnowledgeKnowledge

ActionAction

BenefitsBenefits

Page 38: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

“To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: s/he may be able to say what the experiment died of.”

Ronald FisherRonald Fisher, Cambridge U, 1938

Our motto: Accelerating and optimizing your work through intuitive, reliable and powerful analytics

Big data, Predictive analytics, Computational modeling:

From Data through Knowledge & Action to Outcomes & Benefits

4. Bottom Line4. Bottom Line

Page 39: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Thanks to Team:Thanks to Team:Roger HigdonRoger Higdon Natali KolkerNatali Kolker Bill Bill

Broomall Broomall Winn Haynes Winn Haynes Chris MossChris MossBeth Stewart Beth Stewart Greg YandlGreg Yandl Imre JankoImre Janko Andrew LoweAndrew LoweLarissa Stanberry Larissa Stanberry Maggie Lackey Maggie Lackey Randy Salomon Randy SalomonChris HowardChris Howard SkylarSkylar JohnsonJohnson Nate Anderson Nate Anderson Courtney MacNealy-KochCourtney MacNealy-Koch Gerald van BelleGerald van BelleVural Ozdemir Vural Ozdemir Matthias HebrokMatthias Hebrok Corinna Gries Corinna Gries Biaoyang LinBiaoyang Lin Todd Smith Todd Smith Geoffrey Fox Geoffrey Fox

Peter ArzbergerPeter Arzberger Dan AtkinsDan Atkins Deborah Deborah ElvinsElvins

Rob Arnold Rob Arnold Jack FarisJack Faris Evelyne & Ben Evelyne & Ben KolkerKolker

David Fisher, Lisa Branderburg, Kelly Wallace, Skip Smith, Wes David Fisher, Lisa Branderburg, Kelly Wallace, Skip Smith, Wes Wright, Mark Del Beccarro, Sandy Meltzer, Steven Hurwitz, Bruder Wright, Mark Del Beccarro, Sandy Meltzer, Steven Hurwitz, Bruder Stapleton, Peter Tarczy-Hornoch, Troy McGuire, Judy Dougherty, Lee Stapleton, Peter Tarczy-Hornoch, Troy McGuire, Judy Dougherty, Lee HunstmanHunstman

Jim HendricksJim Hendricks Tom Tom HansenHansen

Support:Support: NSF, NIH, SCRI, Robert McMillen Foundation, NSF, NIH, SCRI, Robert McMillen Foundation, Gordon and Betty Moore FoundationGordon and Betty Moore Foundation

Page 40: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Contact EK: [email protected]@yahoo.com or [email protected]@seattlechildrens.org

For more info: kolkerlab.orgkolkerlab.org and delsaglobal.orgdelsaglobal.org

Thank You!

Any questions?

Page 41: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Additional SlidesAdditional Slides

Page 42: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Radom representation of data today:

REGULATION (WSJ Feb. 15, Week in Ideas: Daniel Akst)

Helmet HeadwindAmerican kids need more exercise, but are helmet laws making them ride their bicycles less?Two economists say that could be the case. Helmet laws, they found, are associated not only with fewer bike-related head injuries for children but also with fewer non-head biking injuries. More than 20 states have laws requiring bike helmets, with various age limits, as do localities."For 5-19 year olds," the researchers write, "we find the helmet laws are associated with a 13% reduction in bicycle head injuries, but the laws are also associated with a 9% reduction in non-head bicycle related injuries and an 11% increase in all types of injuries from the wheeled sports."The increase in injuries from other wheeled sports suggests young riders might be shifting to skateboards and roller skates instead of bicycling.

"Effects of Bicycle Helmet Laws on Children's Injuries," Pinka Chatterji and Sara Markowitz, National Bureau of Economic Research Working Paper 18773 (February)

American kids need more exercise, but are helmet laws

making them ride their bicycles less?

Page 43: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Radom representation of data today:

MARKETING (WSJ Feb. 15, Week in Ideas: Daniel Akst)

Buying ConservativelyBringing a new product to market? You'll have a harder time in conservative parts of the country, a paper implies.A trio of business professors studied six years of supermarket purchases in counties covering nearly half the U.S. population and found that, when it comes to groceries, conservatives like established national brands—and are significantly less likely to try new items."These tendencies," the researchers wrote, "correspond with other psychological traits associated with a conservative ideology, such as preference for tradition and the status quo, avoidance of ambiguity and uncertainty, and skepticism about new experiences."Conservative ideology was measured in the study by Republican voting behavior and religiosity. In counties high on both measures, generic products fared worse and new products had lower penetration.

"Ideology and Brand Consumption," Romana Khan, Kanishka Misra and Vishal Singh, Psychological Science (Feb. 4)

Page 44: Strata Big Data Feb. 26, 2013 Let the Data Decide: Predictive Analytics in Healthcare

Radom representation of data today:

CRIMINAL JUSTICE (WSJ Feb. 15, Week in Ideas: Daniel Akst)

Unreliable, Sober or NotIntoxicated eyewitnesses are no less reliable than sober ones—but neither is very good at picking culprits out of a lineup.Researchers in Sweden gave screwdrivers to two groups of presumably eager volunteers with the aim of a 0.04 blood alcohol concentration in one, and 0.07 in the other—both above the 0.02 Swedish limit for driving but below the 0.08 level that is standard in the U.S.Then the participants, along with an alcohol-free control group, were shown a staged kidnapping on video. A week later the volunteers were asked to pick the kidnappers out of a lineup.All three groups of participants performed about the same—better than chance but poorly nonetheless. The poor showing was in keeping with prior studies.

"Do Sober Eyewitnesses Outperform Alcohol Intoxicated Eyewitnesses in a Lineup?" Angelica Hagsand and four other authors, The European Journal of Psychology Applied to Legal Context (January)

Intoxicated eyewitnesses are no less reliable than sober ones—but neither is

very good at picking culprits out of a lineup.