ward - predictive modeling - boston - 2011-08-02...2011/08/02 · a comprehensive analytic...

A Comprehensive Analytic Framework for Accountable Care and Care Management

The World Congress 2nd Annual Leadership Summit on Predictive AnalyticsBoston, MA

Richard E. Ward, MD, MBAReward Health Sciences, Inc.

August 2, 2011

Health SciencesHealth SciencesREWARDREWARD

Copyrighted 2011, Reward Health Sciences, Inc.1

Outline

• How can ACOs reduce cost• IT investment priorities• Using analytic models

– Population Management – Provider– ACO Financial Models


Whatever we call it….

• Accountable Care Organization• Patient‐Centered Medical Home• Clinical Integration• Population Management• Value Based Health Care• Managed Care 2.0

Copyrighted 2011, Reward Health Sciences, Inc.

Providers• Taking responsibility for cost

and quality of care for a defined population of patients

• Working as a team• Sharing some gains and

bearing some risk

ENABLING INVESTMENTSNEW STRUCTURES

NEW INFORMATION TECHNOLOGYNEW ANALYTIC CAPABILITIES

3

Clinician Workstation‐ Results‐ Profiles‐ To Do List‐ Guidelines

Sources of Cost Savings for ACOs


Cost Impact

Reduce Use of Low Value Services of Specialists and

Facilities

PCP Referral Influence

Reduce Rate of Avoidable Clinical

Events

Patient Self‐Management Support

Care Coordination

Reduce Resources Per Clinical ServiceLean

Reduce Duplication of Services

Clinical Decision Support

Health Information Exchange

Provider Consolidation increasing Market Power

Increase Price per Clinical Service or

Episode

Delivery System Transformation

Patient‐Centered Medical Home

and

Accountable Care Organization

and

Meaningful Use of Health Information

Technology

4

Health Information Technology


Accountable CarePatient CenteredPopulationProcess

Guidelines & ProtocolsMeasures

Going PaperlessClinical Data Accessibility, Efficiency, SecurityOld Vision

New Vision

5

Health Information Technology


Process

DataOld Vision

New Vision

6

Benchmarks

Goals

Quality & Cost

PerformanceAnalysisLiterature

Expert Opinion

BestPractices

Data

Outcomes

Process

Feedback

IncentivesProtocols

Guide

lines Implementation

HealthCare

Care‐Delivery

Care‐Planning


Systems to Enable Process Transformation

HealthCare

Care‐Delivery

Care‐Planning


Leverage Workflow Automation / Business Process Mgmt Technology used in other industries

TightlyIntegrated

Care Planning ToolsPatient CenteredProblem Oriented

SmartPopulation

Care ProcessManagement Tools

Physician controlledMeasurableCoordination

8

Unstructured Passively Structured

• Free text• Dictated and Transcribed• Dictated and voice‐

recognized• Document Images• Optical Character

Recognition

• Drawings• Clinical Images• Sounds

• Text‐to‐code logic• Commands to include text blocks in notes

• Loose XML messages

Actively Structured

• Registry• Questionnaire• Form‐based Template Charting

• Problem‐oriented clinical documentation templates

• Rigorous XML messages

Enables:• Reminders and alerts• Performance measures• Comparative effectiveness

Health Information


Analytic Data Repository

RawVersioned

Data

Source Systems

Reports &Reporting

Applications

out

Analytic Data Repository Framework to Support ACOs

Scheduling

Admit, Discharge,Transfer (ADT)

Billing

MedicationAdministration

Operating Room

Credentialing

Etc.

in

Data Derivation Engines & Services

Disease ID Risk ScoresGaps in Care Episodes of Care

Clinical Data Repository

Cubes & Other Summary

Data Structures

Care Relationships

Specialty / Peers

ReferralRelationships Etc.

Derived data

Analyzable Data

• Normalized• Documented• With derived

entities and attributes



MODELSREPORTS &MEASURES

vs.

Looking back Looking ahead

11

12

Intervention Design

Cause‐Effect Model

includes

Process Model

includes

Intervention Model

informsinformsEvaluation

Plan informs

ProcessMgmtSystem

Configuration

informs

Clinical ProgramOperations

orchestrates

Activity Datacreates

enablesextrapolation of

Calculated ActualOutcomes

toProjected Outcomes

For AlternativeIntervention Designs

enablescalculation of

supports assumptions of

confirms plausibility of

Effect Measurement

informs

informs


ChronicConditionsWellness

Concerns& Symptoms

Acute Conditions

ElectiveSurgical Conditions

Complex Catastrophic Conditions

Continuum of Patient Needs

Using Models for Care Management


14

Is Care Management Effective?• Are drugs effective?• Is a scalpel effective?

• Which population?• What point in time?• What intervention?• What outcomes of interest?• What time horizon?• What evidence threshold?

It depends


15

TARGETEDHOLISTICCompeting Intervention Design Philosophies

• Easier to design• Respects professionalism• Addresses patient complexity• Difficult to evaluate

Many “triggers”

General Assessment

Multi‐Issue Care Plan

Intervention Periodas Coach Evolves Goalsand Revised Care Plan

• Consistent intervention process enables process improvement

• Targeting protocol can be applied to comparison population for evaluation

Targeting of PatientsBased on Objective CriteriaBased on Opportunity to

Benefit from aparticular intervention

Outreach Protocol

Intervention Protocol


16

Using Intervention Models to Explore Alternative Interventions

Care Transition Nurse On Site

Care Transition Nurse On Phone

Identified Population/Spend $100 $100

Patients Identified in when still in hospital

$100 $48

Target Rate$100 $41

Reach and Engagement Rate

Effectiveness Rate in avoiding need for readmission

$65 $13

Total Gross Savings $20 $2

100% 48%

100% 86%

65% 32%

30% 15%

IllustrativeCopyrighted 2011, Reward Health Sciences, Inc.

17

Intervention ModelsAssumptions

Epidemiology‐‐‐‐‐‐‐‐Effectiveness‐‐‐‐‐‐‐‐Economic‐‐‐‐‐‐‐‐Preferences‐‐‐‐‐‐‐‐‐‐

Optimistic Best Pessimistic‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐

Calculations Results

Can be used to determine:• Optimal targeting threshold• Program dynamics (ramp up)• Uncertainty (ranges)• Geographic critical massCopyrighted 2011, Reward Health Sciences, Inc.

18

Illustrative


19

Illustrative


20

Number of IP admissions per 1000 members identified with CHF, by percentile of risk score

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

0102030405060708090100

Percentile of Symmetry risk score

IP A

dmit

Rat

e pe

r 100

0

Predicted rate per 1000

Overall IP Rate

Illustrative

Threshold

Y N

0% 10% 20% 30% 40% 50% 100%Target RateCopyrighted 2011, Reward Health Sciences, Inc.

21

Diabetes Disease Management

(1,000,000)

-

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

0% 5% 10%

15%

20%

25%

30%

35%

40%

45%

50%

55%

60%

65%

70%

75%

80%

85%

90%

95%

100%

Finding Target Penetration that Yields Max Net Savings:Maximizing Beneficial Impact for Members for the Amount Spent

Gross Savings

Cost

Net Savings

Dollars

41%

Fixed Cost

Illustrative

Target Penetration Rate (as % of Diabetes population)


22


(1,000,000)

-

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

0% 5% 10%

15%

20%

25%

30%

35%

40%

45%

50%

55%

60%

65%

70%

75%

80%

85%

90%

95%

100%


Gross Savings

Cost

Net Savings

Dollars


41%

Fixed Cost

Threshold

Y N

Illustrative


23


(1,000,000)

-

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

0% 5% 10%

15%

20%

25%

30%

35%

40%

45%

50%

55%

60%

65%

70%

75%

80%

85%

90%

95%

100%


Gross Savings

Cost

Net Savings

Dollars

41%

Fixed Cost

Illustrative



24

Highest ROI Does Not Yield Maximum Net Savings or Maximum Penetration Rate for Member Impact


(1,000,000)

-

1,000,000

2,000,000

3,000,000

4,000,000

5,000,0000% 5% 10%

15%

20%

25%

30%

35%

40%

45%

50%

55%

60%

65%

70%

75%

80%

85%

90%

95%

100%

-

0.50

1.00

1.50

2.00

2.50

Gross Savings

Cost

Net Savings

ROIDollars

41%18%

Fixed Cost

ROI*

Increasing the target penetration rate from 18% to 41% leads to a lower ROI, but the net savings increases by 24%.

Illustrative



25

Chronic Disease Management

(0.15)

(0.10)

(0.05)

-

0.05

0.10

0.15

0.20

0.25

- 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45

Variable Cost PMPM

Net

Sav

ings

PM

PM IHDCHFDiabetesCOPDAsthma

47% of Ischemic Heart Disease

87% of CongestiveHeart Failure

41% of Diabetes

34% ofCOPD 20% of Asthma

Max Net Savings Signature Illustrative


26

Does global opportunity score / stratification make sense with targeted interventions?

Intervention Proxy for Return

Care Transition ProgramFor Patients Admitted to Hospital

Probability of Being Re‐AdmittedWithin 30 days of Discharge to Home

Nurse Advice about Pros and ConsOf Spine Surgery

Probability of Getting Back SurgeryIn Next Year

Nurse Coaching to IncreaseChronic Condition Self‐ManagementMotivation and Effectiveness

Probability of Being Admitted toHospital in Next Yearfor Chronic Disease


27

Dynamic Models

• Thinking like an accountant analyzing accounts receivable


28

Model of One Engagement Cohort for a Program Component

Total Savings

$ per 100 Members

Intervention

Engage

Outreach

121110987654321

Months Since Member was Targeted

Capital/Operating Cost, Benefit Savings

Total $ Impact

$ per 100 Members

Targeting Volume

Target

Development Cost

121110987654321

Months Since Started Program Development


Model of Program Component, Rolled‐Out “Go” Decision

Total Portfolio $

Etc.

Program B

Program A

AprMarFebJanDecNovOctSepAugJulyJunMayAprMarFebJan

20122011


Model of Overall Portfolio of Clinical Programs, in Calendar Time


29

-$5M

$0M

$5M

$10M

$15M

$20M

Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Benefit Cost SavingsOperational CostsInvestment CostsQuarterly Economic Impact

Dynamic Models

Quarterly Economic Impact

2009 2010 2011 2012 2013 2014

Case Management

Illustrative


30

Dynamic Models


2009 2010 2011 2012 2013 2014

ILLUSTRATION

Chronic Condition Management

‐$1.0M

‐$0.5M

$0.0M

$0.5M

$1.0M

$1.5M

$2.0M

$2.5M

$3.0M

$3.5M

Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

Benefit Cost Savings

Operational Costs

Investment Costs



31

Dynamic Model of Entire WCM Solution


2009 2010 2011 2012 2013 2014

ILLUSTRATION

‐$3M

‐$2M

‐$1M

$0M

$1M

$2M

$3M

$4M

$5M

$6M

Benefit Cost Savings

Operational Costs

Investment Costs & Core Costs

Net Quarterly Impact


32

Analyzing UncertaintyUsing Monte Carlo Simulation

Assumptions

Calculations

90% Interval of Uncertainty


33

Chronic Condition Management—Sensitivity Analysis

2014 Cumulative Net Savings Frequency Distribution 2014 Cumulative Net Savings Variable Sensitivity

Contribution to Variance

Illustrative

1%

1%

1%

1%

2%

17%

71%

7%

0% 20% 40% 60% 80%

Other

Average Length of RegularEngagement Phone Calls (min)

MA PPO Annual Inflation (ProgramCosts) Growth Rate

MA PPO Annual Medical SpendGrowth Rate Above Inflation

Double Counting Assumption

Engagement Rate (% of reachedmembers engaged)

Member Reach Rate (% of targetedmembers reached)

Total Spend Reduction for EngagedMembers

0

200

400

600

800

1,000

1,200

‐$10M $4M $18M $32M $46M $60M

Freq

uency


34

Example of “Hurricane Diagram” WCM Solution Cumulative Net Savings

90%Confidence

Note: Based on a Monte Carlo analysis with 10,000 trials, and triangular distributions on 72 input variables for entire portfolio

Range of Outcomes—Cumulative Portfolio Net Savings

ILLUSTRATION

‐$20M

$M

$20M

$40M

$60M

$80M

$100M

$120M

$140M

Jun‐10 Dec‐10 Jun‐11 Dec‐11 Jun‐12 Dec‐12 Jun‐13 Dec‐13 Jun‐14 Dec‐14


35

InghamInghamInghamInghamInghamInghamInghamInghamIngham

KalamazooKalamazooKalamazooKalamazooKalamazooKalamazooKalamazooKalamazooKalamazoo

KentKentKentKentKentKentKentKentKent

OaklandOaklandOaklandOaklandOaklandOaklandOaklandOaklandOakland

WashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenaw WayneWayneWayneWayneWayneWayneWayneWayneWayne

Modeling Geographically‐Sensitive Interventions

County # Facilities

# Nurse Case Mgrs

Annual Net Savings

Engaged LocallyOakland County 85 10 $ xWayne County 91 8 $ xKent County 22 4 $ xWashtenaw

County 19 3 $ x

Ingham County 7 2 $ xKalamazoo

County 12 2 $ x

Engaged TelephonicallyAll Other

Counties 364 $ x

= Counties targeted locally

ILLUSTRATIVEIn‐HospitalDischarge Planning


36

WashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenawWashtenaw

KentKentKentKentKentKentKentKentKentOttawaOttawaOttawaOttawaOttawaOttawaOttawaOttawaOttawa

MacombMacombMacombMacombMacombMacombMacombMacombMacombOaklandOaklandOaklandOaklandOaklandOaklandOaklandOaklandOakland

WayneWayneWayneWayneWayneWayneWayneWayneWayne

Modeling Geographically‐Sensitive Interventions

Top Counties –ILLUSTRATIVE SNF/LTC Spend

County Members SNF-LTC Spend % of SNF Spend

Wayne County 1,281 $ x 10-15%

Oakland County 1,470 $ x 10-15%

Kent County 1,292 $ x 5-10%

Macomb County 778 $ x 5-10%

Washtenaw County 546 $ x 5-10%

Ottawa County 478 $ x 2-5%

Kalamazoo County 330 $ x 2-5%

Ingham County 298 $ x 2-5%

Genesee County 183 $ x 1-2%

Muskegon County 195 $ x 1-2%

Livingston County 166 $ x 1-2%

Jackson County 246 $ x 1-2%

St. Clair County 196 $ x 1-2%

Calhoun County 122 $ x 1-2%

Grand Traverse County 164 $ x 1-2%

Berrien County 153 $ x 1-2%

Saginaw County 185 $ x 1-2%

Eaton County 167 $ x 1-2%

Bay County 97 $ x 1-2%

Allegan County 139 $ x 1-2%

Monroe County 117 $ x 1-2%

Wexford County 52 $ x 1-2%

$ x$ x$ x$ x$ x$ x

Total SNF/LTC Spend

Nursing HomeCare Coordination

Illustrative


37

Projected benefit cost savingsAnnual savings by initiative category

$5,006

$6,281

$7,530

$8,472

$0

$1,000

$2,000

$3,000

$4,000

$5,000

$6,000

$7,000

$8,000

$9,000

2010 2011 2012 2013

Ben

efit

cost

sav

ings

($k)

Service utilization ConditionClinical IT Core clinical processNew group Planned

Projected benefit cost savingsAnnual savings by initiative category as % of total benefit cost

0.23% 0.24% 0.24% 0.24%

0.18%

0.23% 0.25% 0.25%

0.25% 0.25% 0.25% 0.25%

0.35%

0.40%0.43% 0.44%

0.00% 0.00% 0.00% 0.00%0.0%

0.1%

0.1%

0.2%

0.2%

0.3%

0.3%

0.4%

0.4%

0.5%

0.5%

2010 2011 2012 2013

Ben

efit

cost

sav

ings

(%)

Service utilization ConditionClinical IT Core clinical processNew group Planned

Modeling for Provider‐facing Clinical ProgramsSavings for Customer X for 41 Initiatives in the BCBSM Physician Group Incentive

Program

ILLUSTRATIVECopyrighted 2011, Reward Health Sciences, Inc.


MODELSREPORTS &MEASURES vs.

Looking back Looking ahead

38

CareOffered

CareReceived

PhysiologicEffect

Mortality Reduced,FunctionImproved

CostsSaved orIncurred

PopulationHealthier

PremiumLowered

Process Outcome

• Simpler to define, concrete • Less expensive• Fewer confounding variables• Less measurement variation• Faster improvement cycle• Less problem with turnover

• Measures based on ultimate goals• More intuitive to consumers• Avoids “micro‐management”• Promotes innovation

RE Ward (9/96)

Types of Measures

ReminderSystem

DocumentedStandards

QACommittee

Structure


CareOffered

CareReceived

PhysiologicEffect

Mortality Reduced,FunctionImproved

CostsSaved orIncurred

PopulationHealthier

PremiumLowered

Process Outcome

Meaningful Use of

Certified HIT components

RE Ward (9/96)

Types of Incentives

ReminderSystem

DocumentedStandards

QACommittee

Structure

Quality ofCare Metrics

ACOMedicare Gain

Sharing


41

Measurement of Outcomes

• Can only measure events that did not happen by comparison

• Two basic types of comparison groups:–Pre‐Post–Concurrent


42

• Formal analysis uses more rigorous methods to deal with potential confounding variables and assess confidence interval.

• Iterative process requires methods expertise; impractical to do over and over for monthly reporting.

• Outcome measure defined so as to be able to define the denominator population symmetrically for intervention and comparison group.

• Comparison could be historical or concurrent.

• Objective is to track actual results to determine if expected results are achieved.

Periodic RetrospectiveProgram Evaluation

ConcurrentOutcomes Monitoring

The Levels of Effect Measurement


43

• Comparison group is not truly comparable

• Noise > Signal

• Noise = “common cause” or “random” variation in people and their response to disease and treatment

BIASVARIATION

The Two Key Challenges to Measurement


44

• Risk adjustment does not help.

TightEligibilityCriteria

More ConsistentIntervention

(“Lab Conditions”)

Reduce Variation Increase sample size (“Power”)

Methods to Address Variation


45

Addressing Bias

Top Five ROI Bias Issues

1. Regression to the Mean2. Biased Secular Trend Adjustment3. Once‐chronic‐always‐chronic “migration bias”4. Risk Factor Switcharoo5. Volunteer Bias with “I did my best” control for

confounding


46

6,533

3,450

-1,0002,0003,0004,0005,0006,0007,0008,0009,000

10,000

Pre Intervention (3 months) Post Intervention (3 month)

Aver

age

Cost

Per

Cas

e (P

MPM

)Regression to the Mean

47.1%Reduction!

$3,083SavingsPer Case!

Medicare Advantage Cases referred between April 2007 ‐ Dec 2008. n=11,768

Case Management – Cost per Case before and after referral

Illustrative


47

Regression to the Mean



*Post date ranges in relation to 5‐days after targeting.

-

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

Pre61-90

Pre31-60

Pre0-30

Post'0-30

Post31-60

Post61-90

Post91-120

Post121-150

Post151-180

Post181-210

Days in Relation to Targeting for Case Management*

Cos

t Per

Mem

ber P

er M

onth

Engaged

Illustrative


48

Regression to the Mean



*Post date ranges in relation to 5‐days after targeting.

Engaged

Not Engaged-

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

Pre61-90

Pre31-60

Pre0-30

Post'0-30

Post31-60

Post61-90

Post91-120

Post121-150

Post151-180

Post181-210

Days in Relation to Targeting for Case Management*

Cos

t Per

Mem

ber P

er M

onth

Illustrative


49

Solution = Outcomes Monitoring with “Re‐qualification”Regression to the Mean

0

50

100

150

200

250

300

350M

on 1

Mon

2

Mon

3

Mon

4

Mon

5

Mon

6

Mon

7

Mon

8

Mon

9

Mon

10

Mon

11

Mon

12

Mon

13

Mon

14

Mon

15

Mon

16

Mon

17

Mon

18

Mon

19

Mon

20

Mon

21

Mon

22

Mon

23

Mon

24

Mon

25

Mon

26

Mon

27

Mon

28

Pre-Intervention Actual

Pre-Intervention Trend

Illustrative


50

0

50

100

150

200

250

300

350M

on 1

Mon

2

Mon

3

Mon

4

Mon

5

Mon

6

Mon

7

Mon

8

Mon

9

Mon

10

Mon

11

Mon

12

Mon

13

Mon

14

Mon

15

Mon

16

Mon

17

Mon

18

Mon

19

Mon

20

Mon

21

Mon

22

Mon

23

Mon

24

Mon

25

Mon

26

Mon

27

Mon

28



Expected Post-Intervention Trend


Ramp‐Up Intervention Steady State

Illustrative


51

0

50

100

150

200

250

300

350M

on 1

Mon

2

Mon

3

Mon

4

Mon

5

Mon

6

Mon

7

Mon

8

Mon

9

Mon

10

Mon

11

Mon

12

Mon

13

Mon

14

Mon

15

Mon

16

Mon

17

Mon

18

Mon

19

Mon

20

Mon

21

Mon

22

Mon

23

Mon

24

Mon

25

Mon

26

Mon

27

Mon

28




Post-Intevention Actual



Illustrative


52

0

50

100

150

200

250

300

350M

on 1

Mon

2

Mon

3

Mon

4

Mon

5

Mon

6

Mon

7

Mon

8

Mon

9

Mon

10

Mon

11

Mon

12

Mon

13

Mon

14

Mon

15

Mon

16

Mon

17

Mon

18

Mon

19

Mon

20

Mon

21

Mon

22

Mon

23

Mon

24

Mon

25

Mon

26

Mon

27

Mon

28




Post-Intevention Actual



Illustrative


53

Applying Outcomes Monitoring to aVendor‐delivered Disease Mgmt Program


54

Using Statistical Models


55

Key Conclusions

• Plans and Providers must prepare to share risk• Systems should capture Actively Structured Data• Cause‐Effect models should be developed to support

intervention design, prospective outcomes estimates and evaluation plan

• Intervention Models should be used to prospectively estimate outcomes of clinical programs and to determine optimal targeting

• Engagement Cohort method should be used to model the dynamics of program ramp‐up and ramp‐down.

• Monte Carlo analysis should be used to assess uncertainty • Pre‐Post Analysis without “requalification” is analytic

malpractice


Thank You!


QuestionsContact Info:

Richard E. Ward, MD, [email protected]

519‐817‐8300

56

ward - predictive modeling - boston - 2011-08-02...2011/08/02 · a comprehensive analytic...

Documents