3750 ldr 102813 webcast3 · 5 presented by: pamela peele, phd, is the chief analytics officer of...

35
1 2 Copyright Information Copyright © 2013 HealthLeaders Media The “Predictive Analytics: Reducing Readmissions and Hospitalizations” webcast materials package is published by HealthLeaders Media, a division of HCPro, Inc. For more information, please contact us at: 75 Sylvan Street, Suite A-101, Danvers, MA 01923. Attendance at the webcast is restricted to employees, consultants, and members of the medical staff of the Licensee. The webcast materials are intended solely for use in conjunction with the associated HealthLeaders Media webcast. The Licensee may make copies of these materials for internal use by attendees of the webcast only. All such copies must bear the following legend: Dissemination of any information in these materials or the webcast to any party other than the Licensee or its employees is strictly prohibited. In our materials, we strive to provide our audience with useful and timely information. The live webcast will follow the enclosed agenda. Occasionally, our speakers will refer to the enclosed materials. We have noticed that non- HealthLeaders Media webcast materials often follow the speakers’ presentations bullet-by-bullet and page-by- page. However, because our presentations are less rigid and rely more on speaker interaction, we do not include each speaker’s entire presentation. The enclosed materials contain helpful resources, forms, crosswalks, policies, charts, and graphs. We hope that you will find this information useful in the future. Although every precaution has been taken in the preparation of these materials, the publisher and speaker assume no responsibility for errors or omissions, or for damages resulting from the use of the information contained herein. Advice given is general, and attendees and readers of the materials should consult professional counsel for specific legal, ethical, or clinical questions. HealthLeaders Media a division of HCPro, Inc., is not affiliated in any way with The Joint Commission, which owns the JCAHO and Joint Commission trademarks; the Accreditation Council for Graduate Medical Education, which owns the ACGME trademark; or the Accreditation Association for Ambulatory Health Care (AAAHC).

Upload: others

Post on 31-Aug-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

1

2

Copyright Information

Copyright © 2013 HealthLeaders Media

The “Predictive Analytics: Reducing Readmissions and Hospitalizations” webcast materials package is published by HealthLeaders Media, a division of HCPro, Inc. For more information, please contact us at: 75 Sylvan Street, Suite A-101, Danvers, MA 01923.

Attendance at the webcast is restricted to employees, consultants, and members of the medical staff of the Licensee. The webcast materials are intended solely for use in conjunction with the associated HealthLeadersMedia webcast. The Licensee may make copies of these materials for internal use by attendees of the webcast only. All such copies must bear the following legend: Dissemination of any information in these materials or the webcast to any party other than the Licensee or its employees is strictly prohibited.

In our materials, we strive to provide our audience with useful and timely information. The live webcast will follow the enclosed agenda. Occasionally, our speakers will refer to the enclosed materials. We have noticed that non-HealthLeaders Media webcast materials often follow the speakers’ presentations bullet-by-bullet and page-by-page. However, because our presentations are less rigid and rely more on speaker interaction, we do not include each speaker’s entire presentation. The enclosed materials contain helpful resources, forms, crosswalks, policies, charts, and graphs. We hope that you will find this information useful in the future.

Although every precaution has been taken in the preparation of these materials, the publisher and speaker assume no responsibility for errors or omissions, or for damages resulting from the use of the information contained herein. Advice given is general, and attendees and readers of the materials should consult professional counsel for specific legal, ethical, or clinical questions.

HealthLeaders Media a division of HCPro, Inc., is not affiliated in any way with The Joint Commission, which owns the JCAHO and Joint Commission trademarks; the Accreditation Council for Graduate Medical Education, which owns the ACGME trademark; or the Accreditation Association for Ambulatory Health Care (AAAHC).

Page 2: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

3We will begin shortly!

If you are not hearing music or you areexperiencing any technical difficulties,

please contact our help desk at 888-364-8804.

4

Page 3: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

5

Presented by:

Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience along with 12 years of academic research experience to her position as the leader of healthcare analytics at the Health Plan. She is responsible for data analytic activities, economic modeling, predictive modeling, and statistical analysis. Her work focuses on the application of economic and statistical models to improve the health and welfare of UPMC members. She currently holds faculty appointments in Health Policy & Management and in Psychiatry in the School of Medicine at the University of Pittsburgh. She is core faculty at the Center for Research on Health Care at the University of Pittsburgh Medical Center and an elected fellow of the Centre for Interuniversity Research and Analysis on Organizations in Montreal, Canada.

6

Presented by:

Christine VanZandbergen, MPH, MS, PA-C,is the associate clinical information officer at

Penn Medicine, a $4.3 billion healthcare provider organization consisting of over 2,000 physicians providing services to the Hospital of the University of Pennsylvania, Penn Presbyterian Medical Center, Pennsylvania Hospital and the health system network that serves the city of Philadelphia, the surrounding five-county area and parts of southern New Jersey. She joined Penn Medicine in June 2002 as a Cardiac Surgery Physician Assistant and is currently responsible for oversight of all clinical decision support tools from the ideation process, through pilot, and deployment across the health system in a multi-system EHR. She continues her clinical practice in cardiac surgery at Pennsylvania Hospital.

Page 4: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

7

BUILDING A LEARNING ORGANIZATION FOR POPULATION HEALTH MANAGEMENT

Pamela Peele, PhD

Chief Analytics Officer

UPMC Insurance Services Division

August 2013

8

Levels of Analytics Framework

Standard ReportsWhat happened?

AlertsWhat actions are needed?

Query DrilldownWhat exactly is the problem?

Ad hoc ReportsHow many, how often, where?

Statistical AnalysisWhy is this happening?

OptimizationWhat’s the best that can happen?

Predictive ModelingWhat will happen next?

ForecastingWhat if these trends continue?

Degree of Intelligence

Com

petit

ive

Adva

ntag

e

From Tom Farre, “The Analytical Competitor”, in Analytics: The Art and Science of Better, ComputerWorld Technology Briefing.

UPMC HP: 2009

UPMC HP: 2006

Page 5: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

9

Staff – 2006

Business Analyst (30)

Accounting

• Excel

• Access

• Crystal Reports

10

Current Staff

Clinical Program

Evaluation (5)

Epidemiology

Biostatistics

Health Services Research

Strategic Business Analysis

(6)

Finance

Economics

Policy

Statistics

Database & Data

Quality (7)

Finance

Economics

Policy

Statistics

Modeling (3)

Physics

Mathematics

Biomedical Engineering

Statistics

Operations (5)

Economics

Industrial Engineering Operations

Communications

Statistics

Page 6: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

11

Staff Skills and Backgrounds

• Industry Knowledge

• Data visualization skills

• Data ECTL (extraction, cleaning, transformation, loading) skills

• Statistics

• Health Services Research

• Data Mining

• Financial modeling & evaluation

• Presentation, writing, and communication skills

• Formally trained but NOT blinded by their training

– Challenge deeply held beliefs

12

Tools

• Database: SQL, Toad

• Statistics: SAS, STATISTICA, STATA, R

• Data Mining: STATISTICA, R

• Text Mining: STATISTICA

• Modeling & Simulation: MATLAB, Mathematica, Vensim, GEPHI

• GIS: ArcGIS

Page 7: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

13

Integrated Data to Support Clinical Management Population Health Strategy and Clinical Support

• Many types of disparate data available

• Medical Claims

• Behavioral Health Claims

• Pharmacy Claims (allows medication possession ratio MPR)

• Workers’ Compensation Claims

• Short-Term Disability

• Absenteeism Data From Time Cards

• On-Site Biometric Screening Results

• Health Risk Assessments – (self-reported)

• Care Management Assessments/Phone interaction

• Enrollment & Demographic Data

• Lab Values

GOAL: develop centralized registry of member clinical presentation and lifestyle profiles for clinical analysis

14

Identifying Health Conditions by SEPARATE Data Source

Page 8: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

15

1,596 1,994 2,197 2,344

4,086 5,698 5,698 6,774

4,324 6,588 6,588 7,658982 2,715 2,715 2,715

2,200 6,366 7,597 7,597

2,738 2,738 2,738 2,738

0 1,442 5,721 6,119

132 132 8,593 8,878

11,795 16,036 21,005 21,913

Identifying Health Conditions by AGGREGATING Data Source

16

Stratification Data Flow

HealthPlaNET

Page 9: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

17

PREDICTIVE ANALYTICS AND READMISSION RISK

Using the electronic health record to automatically identify patients at risk for 30-day readmission

Christine VanZandbergen, MPH, PA-C

Associate Clinical Information Officer

Penn Medicine

18

The University of Pennsylvania Health System

• Four acute care hospitals:

– Hospital of the University of Pennsylvania

– Penn Presbyterian Medical Center

– Pennsylvania Hospital

– The Chester County Hospital

– 1,740 acute care beds

– 80,020 acute care admissions in 2012

– 73 ACGME-accredited training programs

Page 10: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

Transforming Data Into Value

Phase 3 – Precision MedicineData

Inform

ation

Knowledge

Phase 2 – “Meaningful Use”

Phase 1 – Foundation

YEAR

VALU

E

2004 2006 2008 2010 2012 2014 2016 2018 2020

• Implement Common Inpatient EMR Applications

• Implement Common Ambulatory Applications

• Support Standard Processes Across Penn Medicine

• Aggregate Patient Data in Centralized Data Warehouse

• Development of Dashboards and Alerts

• Integration With Perelman School of Medicine IS

• Development of Early Warning Surveillance System

• High Performance Computing Center

• Research Data Warehouse 

• Implement Integrated EMR

• Integrate Phenotype and Genomic Data

• Establish Enterprisewide IS Governance Structure

• Solidify Enterprise IS Infrastructure

• Extend Mobility Capabilities 

19

20

Key Elements: Operational Predictive Analytics

Operational ReadinessOperational Readiness

Analytics CapabilityAnalytics Capability

Data Capture

Data Capture

• Advocacy• C-suite• Front line

• Infrastructure

• Advocacy• C-suite• Front line

• Infrastructure

• Predictable• Versatile• Predictable• Versatile

• Accurate• Reliable• Available

• Accurate• Reliable• Available

Page 11: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

21

Penn Strategy

22

Background: National Readmissions

Source: “After Hospitalization: A Dartmouth Atlas Report on Post-Acute Care for Medicare beneficiaries” September 2011.

Page 12: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

23

Purpose

• To develop simple predictive model for 30-day readmission

• Implement an automated alert in our hospital’s electronic health record that identifies on admission patients at high risk for 30-day readmission

24

Methods: Systematic Review

Page 13: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

25

Method: Predictors From Review

• Healthcare resource utilization

– Length of stay, number of prior admissions, and previous ED visits

– Studies have not consistently identified threshold values for these predictors

• Patient characteristics

– Comorbidities, living alone, discharged to home, and payer

– Evidence is mixed regarding other factors, including age and gender

26

Analysis: Prediction Rule

Rule Sensitivity PPV % of Total Hospital

Visits Flagged

Looking 12 months back from Index HospitalizationPH>0 and PE>0 30 14 11

PH> 1 25 19 7

PH>1 and PE>0 20 19 5

PH>1 or PE>1 33 13 13

Looking six months back from Index Hospitalization

PH>0 36 14 13

PH>1 18 24 4

PH = prior hospitalization PE = prior ED visit

Page 14: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

27

Analysis: Final Prediction Rule

Rule Sensitivity PPV % of Total Hospital

Visits Flagged

Looking 12 months back from Index HospitalizationPH>0 and PE>0 30 14 11

PH> 1 25 19 7

PH>1 and PE>0 20 19 5

PH>1 or PE>1 33 13 13

Looking six months back from Index Hospitalization

PH>0 36 14 13

PH>1 18 24 4

Patients with >1 inpatient admission in the 12 months preceding the current inpatient stay

28

Implementation:Readmission Risk Flag

Page 15: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

29

Implementation:Readmission Risk Flag

30

Results: 21 months After implementation

42% 85%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Readmission No Readmission

First 21 Months After Implementation

Readmit Flag

No Flag

Sensitivity Specificity

Page 16: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

31

22% 93%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Readmission No Readmission

First 21 Months after Implementation

Readmit Flag

No Flag

PPV NPV

Results: 21 months After implementation

32

Readmission Rates Entity One

-1%

1%

3%

5%

7%

9%

11%

13%

15%

2010

-07

2010

-08

2010

-09

2010

-10

2010

-11

2010

-12

2011

-01

2011

-02

2011

-03

2011

-04

2011

-05

2011

-06

2011

-07

2011

-08

2011

-09

2011

-10

2011

-11

2011

-12

2012

-01

2012

-02

2012

-03

2012

-04

2012

-05

2012

-06

2012

-07

2012

-08

2012

-09

2012

-10

2012

-11

2012

-12

2013

-01

2013

-02

2013

-03

2013

-04

2013

-05

2013

-06

% E

nco

un

ters

UPHS 7 & 30 day unscheduled readmission, FY11 - FY13

30 Day 07 Day

Implementation

Page 17: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

33

Readmission Rates Entity One

-1%

1%

3%

5%

7%

9%

11%

13%

15%

2010

-07

2010

-08

2010

-09

2010

-10

2010

-11

2010

-12

2011

-01

2011

-02

2011

-03

2011

-04

2011

-05

2011

-06

2011

-07

2011

-08

2011

-09

2011

-10

2011

-11

2011

-12

2012

-01

2012

-02

2012

-03

2012

-04

2012

-05

2012

-06

2012

-07

2012

-08

2012

-09

2012

-10

2012

-11

2012

-12

2013

-01

2013

-02

2013

-03

2013

-04

2013

-05

2013

-06

% E

nco

un

ters

Entity 1: 7 & 30 day unscheduled readmission, FY11 -FY13

30 Day 07 Day

34

-1%

1%

3%

5%

7%

9%

11%

13%

15%

2010

-07

2010

-08

2010

-09

2010

-10

2010

-11

2010

-12

2011

-01

2011

-02

2011

-03

2011

-04

2011

-05

2011

-06

2011

-07

2011

-08

2011

-09

2011

-10

2011

-11

2011

-12

2012

-01

2012

-02

2012

-03

2012

-04

2012

-05

2012

-06

2012

-07

2012

-08

2012

-09

2012

-10

2012

-11

2012

-12

2013

-01

2013

-02

2013

-03

2013

-04

2013

-05

2013

-06

% E

nco

un

ters

Entity 2: 7 & 30 day unscheduled readmission, FY11 -FY13

30 Day 07 Day

Readmission Rates Entity Two

Page 18: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

35

Readmission Rates Entity Three

-1%

1%

3%

5%

7%

9%

11%

13%

15%

2010

-07

2010

-08

2010

-09

2010

-10

2010

-11

2010

-12

2011

-01

2011

-02

2011

-03

2011

-04

2011

-05

2011

-06

2011

-07

2011

-08

2011

-09

2011

-10

2011

-11

2011

-12

2012

-01

2012

-02

2012

-03

2012

-04

2012

-05

2012

-06

2012

-07

2012

-08

2012

-09

2012

-10

2012

-11

2012

-12

2013

-01

2013

-02

2013

-03

2013

-04

2013

-05

2013

-06

% E

nco

un

ters

Entity 3: 7 & 30 day unscheduled readmission, FY11 -FY13

30 Day 07 Day

36

Implementation

5%

6%

7%

8%

9%

10%

11%

12%

13%

14%

15%

2010

-07

2010

-08

2010

-09

2010

-10

2010

-11

2010

-12

2011

-01

2011

-02

2011

-03

2011

-04

2011

-05

2011

-06

2011

-07

2011

-08

2011

-09

2011

-10

2011

-11

2011

-12

2012

-01

2012

-02

2012

-03

2012

-04

2012

-05

2012

-06

2012

-07

2012

-08

2012

-09

2012

-10

2012

-11

2012

-12

2013

-01

2013

-02

2013

-03

2013

-04

2013

-05

2013

-06

HUP PAH PMC

UPHS Entities: 30 Day Unscheduled Readmissions

Page 19: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

37

30-Day Readmission Rates

38

30-Day Readmission Rates

0%

5%

10%

15%

20%

25%

30%

En

cou

nte

rs (

%)

UPHS Readmission rate per Risk Flag group

Risk Flag Group: 30 Day Readmission Rate No Risk Flag Group: 30 Day Readmission Rate

Page 20: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

39

Conclusion

• Proof of concept

• Real-time, “actionable” data

• Integrates into provider workflow

• Available to all care providers

• Unclear impact on outcomes

40

Limitations andFuture Directions

• Needs further prospective evaluation

• Improve sensitivity and specificity

• How is the information used?

• Impact on clinical care and outcomes

Page 21: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

41

Future Directions:Predictive Analytics

Data Warehouse

Data Warehouse

Operational System

Operational System

Data Mining

Predictive

Model

Predictive

Model

Implementation

Information

Implementation

Information

EvaluationEvaluation

Operational System

Operational System

Predictive Goal

Predictive Goal

42

Don’t wait until the end of the program to submit your question!

Submit now by using the Q & A boxlocated on your screen. Type in your question.

Click the Icon to send.

1

2 3

Page 22: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

43

BUILDING A LEARNING ORGANIZATION FOR POPULATION HEALTH MANAGEMENT

Pamela Peele, PhD

Chief Analytics Officer

UPMC Insurance Services Division

August 2013

44

CMS Penalties

Fiscal Year 2013 2014 2015

Targeted Conditions AMI, HF, PN AMI, HF, PN AMI, HF, PN, COPD, CABG, PCI, VascularProcedures

Maximum Penalty 1% 2% 3%

44

CMS adjusts for age, sex, comorbid diseases, and past medical history. Each hospital is evaluated with respect to the national average.

All-cause 30-day readmits are also included as a HEDIS quality measure in CMS’ star program for health plans

Page 23: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

45

Geographic Variationin Readmit Rates

45

46

Modeling Issues

• Data is usually “local”

• Data is usually a “snapshot”

• Data is fine-grained

– 13,000 ICD-9 codes

• Will increase to 68,000 with ICD-10

– Example ICD-10: V9107XA - Burn due to water-skis on fire

• Can roll-up using Clinical Classification Software (CCS)

– 9,600 CPT4 codes

• Can roll-up using Berenson Eggers Type of Service

• Target (30-day readmit) is relatively small

• Readmission rates are complicated by the presence of comorbidities

• Most models do not focus on actionability46

Page 24: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

47

Days to Readmit as aProxy for Actionability

47

Example: 10% of all Anemia discharges are readmitted within 10 days

4848

6: Operations on the respiratory system7: Operations on the cardiovascular system16: Miscellaneous diagnostic and therapeutic procedures

Page 25: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

49

Readmission Models

• Most models focus on:

– A specific population: Medicare, Medicaid, etc.

– A specific disease: Heart Failure, COPD, …

– A single hospital

• 18% of all 30-day readmits occur at a different hospital

49

50

Identification of At-Risk Patients

Allaudeen, Schnipper, et. al. “Inability of Providers to Predict Unplanned Readmissions,” J Gen Intern Med., 2011 Jul.

• 159 discharges

• Age ≥ 65

• Evaluated readmission predictions from: Attending Physician, Resident Physician, Intern Physician, Case Manager, Nurse

– None of the predictions differed from chance

50

Page 26: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

51

Readmission Models

Kansagara, Englander, et. al. “Risk Prediction Models for Hospital Readmission: A Systematic Review,” JAMA, 2011 Oct.

• Reviewed 26 models

– “Most current readmission risk models perform poorly”

• Table 4 contains complete list of predictor variables for all used in at least one model

– Most common:

• Specific medical diagnosis or CCI (24)

• Age (19)

• Sex (15)

• Prior admissions (14)

• Alcohol or substance use (11)

• Mental health comorbidity (9)

51

52

Model Types

• Models derived from retrospective administrative data

– May be used for comparison of hospitals

• Models derived from real-time data

– EMR

– Augmenting administrative data with clinical data does not meaningfully improve an HF readmission model: Hammill, Curtis et. al., 2011

• Models derived from survey or chart review data

• Almost all models are based upon a logit model

– Exceptions: Neumann “Measuring Performance in Health Care: …”, 2004; Amalakuhan et. al. “A Prediction Model for COPD Readmissions…,” 2011; Lee, “Selecting the Best Prediction Model for Readmission,” 2012.

Page 27: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

53

Notable Models

Hasan, Meltzer, et. al., “Hospital Readmission in General Medicine Patients: A Prediction Model,” 2009 Dec. (online); J Gen Intern Med., 2010 Mar.

• General Population

• c-statistic = 0.61

• Insurance status

• Marital status

• Has regular physician

• CCI

• SF12 health questionnaire

• Number of prior admissions

• Length of stay

54

Notable Models

van Walraven, Dhalla, et. al. “Derivation and validation of an index to predict early death or unplanned readmission after discharge from hospital to the community,” Canadian Medical Association Journal, 2010 Apr.

• General Population

• c-statistic = 0.684

Length of stay

Acuity (emergent admission)

Comorbidity (CCI)

Emergency department visits

54

Page 28: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

55

Notable Models

Amarasingham, Moore, et al., “An Automated Model to Identify Heart Failure Patients at Risk for 30-Day Readmission or Death Using Electronic Medical Record Data,” Med Care, 2010 Nov.

• Tabak mortality score

• History of depression/anxiety

• Single

• Male

• Number of address changes

• Medicare patient

• In lowest socioeconomic quintile

• History of cocaine use

• History of missed clinic visits

• Number of prior admissions

56

Notable Models

Donzé, Aujesky, et. al, “Potentially Avoidable 30-Day Hospital Readmissions in Medical Patients,” JAMA Intern Med., 2013 Apr.

• Target = Avoidable Readmission (36.6% of all readmissions)

• c-statistic = 0.71

Hemoglobin at discharge

Oncology

Sodium Level at discharge

Procedure during index admission

Index Type of admission (elective vs. non-elective)

Admissions during previous 12 months

Length of stay56

Page 29: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

57

El Camino Hospital Case Study

• Predictors of readmission:

– Age

– Discharge location

– Is the patient’s PCP identified in the medical record?

– Diagnosis for

• CHF

• PN

• Stroke

• Sepsis

• Renal failure

• Resulted in 25% reduction in readmits

57

http://www.cio-chime.org/chime/press/CaseStudy/ElCamino_Case_Study.pdf

58

Allina Health Model

• General Population

• c-statistic = 0.73

• 30 Variables

– Clinical Data

– Demographic Data

– LOS

– Prior Admits

.

.

.

58

http://www.sourcemediaconferences.com/HCS13/Hauptshare.pdf

Page 30: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

59

UPMC Readmission Model

• 1.5 years of discharges to home

– 50,600 in training set

– 21,700 in validation set

• Claims-based model

• Boosted Decision Tree

– Predictions are generated prior to the index admission

– Predictive Positive Value = 28%

• Base Readmit Rate = 14.8%

– Sensitivity = 57%

• Model captures over half of all readmissions

– AUC = 0.6859

60

0.990.880.770.660.550.440.330.22

500

400

300

200

100

0

Probability

Fre

qu

en

cy

0.70.5

Distribution of Probability for ReadmissionFY09 Acute Inpatient Discharges (All LOB)

n = 38,840

Most impactable opportunity to

prevent readmission

Most impactable opportunity to

prevent readmission

Discharge Advocate: Risk Models Identify Readmission “Sweet Spot”

60

At discharge, patients lacked competency in their own conditions and care: • 37% able to state the purpose of all their medications• 14% knew their medication’s common side effects• 42% able to state their diagnosis

Low Risk of ReadmissionLow Risk of

ReadmissionLess impactable despite

high readmission riskLess impactable despite

high readmission risk

Single Acute Episodes

Early/Mid Stage Chronic Disease

End StageChronic Disease

Page 31: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

61

UPMC Model Performance

Condition Sensitivity Specificity

DM with complications 74.6% 60.2%

Sickle Cell 74.4% 53.8%

Hypovolemia 71.6% 70.5%

Complications of surgical or medical care

65.2% 63.2%

Respiratory infection 59.2% 73.1%

Obstructive chronic bronchitis 69.7% 59.8%

Asthma 59.3% 74.2%

Spondylosis 24.0% 91.0%

Osteoarthritis 16.2% 93.8%

All 54.5% 74.4%

61

62

Sensitivity by WeekPost Discharge

62

Sensitivity increases after two weeks

Page 32: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

63

Days to Readmit as Proxyfor “Actionability”

MDC Description Readmit Rate

Days to Readmit –10th Percentile

1 Diseases of the Nervous System 12.9% 20*

4 Diseases of the Respiratory System 15.7% 17

5 Diseases of the Circulatory System 15.2% 16

6 Diseases of the Digestive System 14.2% 17

7 Diseases of the Hepatobiliary System & Pancreas 18.2% 12

10 Endocrine, Nutritional & Metabolic Diseases 11.7% 24

11 Diseases of the Kidney & Urinary Tract 17.0% 14

16 Diseases of the Blood & Blood-forming Organs & Immunological Disorders

27.3% 9

63

* 10% of all MDC 1 discharges are readmitted within 20 daysResults based on 211,198 discharges to home

64

Effect of Planned Visiton Readmits

64

Page 33: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

65

65

Questions & Answers

How to submit a question:

1. Go to the Q & A box located on your screen.

2. Type in your question.

3. Click the Icon to send.

1

2 3

66

Submit a question:

1. Go to the Q & A box located on your screen.2. Type in your question.3. Click the Icon to send.

Questions & Answers

Christine VanZandbergen, MPH, MS, PA-CAssociate Clinical Information Officer

Penn MedicinePhiladelphia, Pa.

Pamela Peele, PhDChief Analytics Officer

UPMC Health PlanPittsburgh, Pa.

Page 34: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

67

Thank you for attending!

Be sure to register forHealthLeaders Media’s next live program:

Telemedicine: A Strategic Tool to Grow Revenue and Drive Efficiency

November 19, 2013 at 1:00 p.m. Eastern

http://www.hcmarketplace.com/prod-11623

68

This concludes today’s program.

Please do not close your browser.When the presentation ends you will be automatically redirected

to the post-event survey.

Page 35: 3750 LDR 102813 Webcast3 · 5 Presented by: Pamela Peele, PhD, is the chief analytics officer of the UPMC Insurance Services Division. Peele brings 13 years of patient care experience

Cer

tifica

te o

f Atte

ndan

ce

atte

nded

Eliz

abet

h Pe

ters

enV

ice

Pres

iden

t H

CPr

o, In

c.

a 90

-min

ute

web

cast

on

Oct

ober

28,

201

3

“Pre

dict

ive

Anal

ytic

s:

Redu

cing

Rea

dmis

sion

s an

d H

ospi

taliz

atio

ns”