using predictive modeling to target interventions

Using Predictive Modeling Using Predictive Modeling to Target Interventionsto Target Interventions

Barry P. Chaiken, MD, MPHBarry P. Chaiken, MD, MPHChief Medical OfficerChief Medical Officer

ABQAURP ABQAURP -- PSOSPSOS

2

OverviewOverview

Cost and Quality Trends Cost and Quality Trends

Disease Management and ModelingDisease Management and Modeling

Predictive Modeling FundamentalsPredictive Modeling Fundamentals

Accuracy of Models Accuracy of Models -- Case StudyCase Study

3

02468

1012141618

1991 1993 1995 1997 1999 2001 2003*

Change Per Capita In Health Care Change Per Capita In Health Care Spending and GDPSpending and GDP

Source: B. Strunk and P. Ginsburg, “Tracking Health Care Costs: Trends Stabilize But Remain High in 2002,” Health Affairs (Web Exclusive June 11, 2003); B. Strunk and P. Ginsburg, Tracking Health Care Costs: Trends Slow in First Half of 2003, Center for Studying Health System Change, December 2003.

Percent

Health Care Spending

GDP

* Data for January through June 2003, compared with corresponding months in 2002

8.5%

2.9%

4

0

3

6

9

12

15

18

1985 1988 1991 1994 1997 2000 2003*

Growth in Per Enrollee Premiums and BenefitsGrowth in Per Enrollee Premiums and Benefits

Source: Heffler et al., “Health Spending Projections for 2002-2012,” Health Affairs (Web Exclusive February 7, 2003) for 1985–2001; Employer Health Benefits 2003 Annual Survey, The Kaiser Family Foundation and Health Research and Educational Trust, September 2003 for 2002–2003.

Percent

Premiums per enrollee

Benefits per enrollee

* Data for growth between Spring 2002 and Spring 2003

5

Drivers of Care ManagementDrivers of Care Management

50% 50% -- preventive carepreventive care

30% 30% -- lack recommended acute carelack recommended acute care

40% 40% -- lack recommended chronic carelack recommended chronic care

30% 30% -- receive contraindicated acute carereceive contraindicated acute care

20% 20% -- receive contraindicated chronic carereceive contraindicated chronic care

6

Recommended Care and Quality VariesRecommended Care and Quality Varies

Source: McGlynn et al., “The Quality of Health Care Delivered to Adults in the United States,” The New England Journal of Medicine (June 26, 2003): 2635–2645.

55

68 6554

49 45

0

20

40

60

80

Overall CAD Hypertension Asthma Hyperlipidemia Diabetes

Percent Receiving Recommended Care

7

0%10%20%30%40%50%60%70%80%90%

100%

U.S. Population Health Expenditures

Health Care Costs Concentrated in Sick FewHealth Care Costs Concentrated in Sick Few

1%5%

10%

55%

69%

27%

Source: AC Monheit, “Persistence in Health Expenditures in the Short Run: Prevalence and Consequences,” Medical Care 41, supplement 7 (2003): III53–III64.

50%

97%

$27,914

$7,995

$4,115

$351

Expenditure Threshold (1997 Dollars)

8

OverviewOverview





9

DM DefinedDM Defined

“Knowledge“Knowledge--based process intended to based process intended to improve continuously the value of health care improve continuously the value of health care delivery from the perspectives of those who delivery from the perspectives of those who receive, purchase, provide, supply and receive, purchase, provide, supply and evaluate it.”evaluate it.”

----James B. Couch, MDJames B. Couch, MD

10

DM Criteria SpecificDM Criteria Specific

High dollar and volumeHigh dollar and volume

Preventable complicationsPreventable complications

Short time frame for resultsShort time frame for results

Treatment variabilityTreatment variability

Extensive patient nonExtensive patient non--compliancecompliance

Practical guidelinesPractical guidelines

Measurable quality metricsMeasurable quality metrics

11

DM ProcessesDM Processes

Identify patientsIdentify patients

Develop therapeutic programsDevelop therapeutic programs

Improve outcomesImprove outcomes

Achieve acceptable cost levelsAchieve acceptable cost levels

Provide evidence based careProvide evidence based care

12

Spectrum of CareSpectrum of Care

PatientsProvider

Home Care

InstitutionsLong Term

CareNursing Homes

HospitalsCenters ofExcellence

SpecialistOutpatient

Clinics

PCPAllied HealthProfessionals

Self-Directed SecondaryPrimary Long TermCareTertiary

Disease Management Lower Costs Higher Costs

13

Risk Measurement PyramidRisk Measurement Pyramid

CaseMgmt.

NeedsAssessment

QualityImprovement

Payment/Finance

PracticeResource

Mgmt.

DiseaseMgmt.

HighDiseaseBurden

Single HighImpact Disease

Users

Users & Non-Users

Population Segment Source: Weiner JP. Presentation at National BCBS Association meeting, Chicago, 1/30/03.

14

OverviewOverview





15

Predictive Modeling DefinedPredictive Modeling Defined

Use of clinical information available for all Use of clinical information available for all members of a population to predict future members of a population to predict future healthcare needs, overall or for specific healthcare needs, overall or for specific types of services.types of services.

Source: CB Forrest, “Population-based Predictive Modeling Using ACGs: Application to Disease and Case Management”, International ACG User’s Conference, November 11, 2003

16

Reasons for Predictive ModelingReasons for Predictive Modeling

Existing high utilization by the fewExisting high utilization by the few

Uses available data to identify highUses available data to identify high--riskrisk

Allows intervention early in disease cycleAllows intervention early in disease cycle

Enhances case and disease managementEnhances case and disease management

17

Predictive Modeling FocusPredictive Modeling Focus

Case management targetingCase management targetingIdentify persons for care programsIdentify persons for care programs

Disease management risk stratificationDisease management risk stratificationIntensity tiersIntensity tiers

Financial forecastingFinancial forecastingActuarial riskActuarial risk


18

What Predictive Modeling DoesWhat Predictive Modeling Does

Stratify membersStratify membersEnhance impact of interventionsEnhance impact of interventions

Probabilistic identification of high Probabilistic identification of high utilizersutilizersAssign risk scoresAssign risk scores

–– Describe comparative severity of illnessDescribe comparative severity of illnessIdentify members not receiving proper careIdentify members not receiving proper careHighlight inconsistency of careHighlight inconsistency of careProspectively identify adverse eventsProspectively identify adverse eventsAllow focused interventionsAllow focused interventions

Maximize benefits of disease managementMaximize benefits of disease managementDiscover inefficient careDiscover inefficient care

19

Additional Uses of ModelingAdditional Uses of Modeling

Influence adoption of best practicesInfluence adoption of best practices

Track effectiveness of interventionsTrack effectiveness of interventions

Establish pay for performanceEstablish pay for performance

Set more accurate premiumsSet more accurate premiums

Develop contracts with providersDevelop contracts with providersActuarialActuarial

Help plan network compositionHelp plan network compositionBased on member needsBased on member needs

20

Target PopulationsTarget Populations

Risk stratifyRisk stratifySubpopulationSubpopulation

Risk factorsRisk factors

Identify most likely to benefitIdentify most likely to benefit

Develop specific, targeted interventionsDevelop specific, targeted interventionsProbabilities for certain outcomesProbabilities for certain outcomes

Practice guidelinesPractice guidelines

Practice standardizationPractice standardization

–– Decrease variationDecrease variation

21

Modeling Utilizes Available DataModeling Utilizes Available Data

Traditional Traditional –– demographic datademographic dataAge, sex, occupation, prior costsAge, sex, occupation, prior costs

Clinical data from claimsClinical data from claims

Pharmacy dataPharmacy data

Diagnostic testsDiagnostic tests

Health risk appraisalsHealth risk appraisals

QuestionnairesQuestionnaires

22

Components of ModelingComponents of Modeling

Entire population dataEntire population dataBaseline assessment used to predict futureBaseline assessment used to predict future

Risk assessment periodRisk assessment periodStatic: year 1 predicts year 2Static: year 1 predicts year 2Rolling: assign score every periodRolling: assign score every period

Outcomes of interestOutcomes of interestChanges in health statusChanges in health statusCostly healthcare eventsCostly healthcare eventsOverall healthcare expendituresOverall healthcare expenditures


23

Statistics in Predictive ModelingStatistics in Predictive Modeling

Screening TestTP

(True Positive)

FN(False Negative)

TN(True Negative)

FP(False Positive)

Specificity

Sensitivity

Predicative Pos. Value

Predicative Neg. Value

TN/FP + TN

TN/FN + TN

TP/TP+FN

TP/TP + FP

+

+ -

-

24

Sensitivity Versus Predictive ValueSensitivity Versus Predictive Value

51%51%25%25%Top 10 %Top 10 %

36%36%36%36%Top 5%Top 5%

14%14%69%69%Top 1%Top 1%

% All True % All True Cases IdentifiedCases Identified

% True Cases % True Cases Among GroupAmong GroupScore CutScore Cut--PointPoint

Predictive Pos. Value

Sensitivity

25

First Generation ModelingFirst Generation Modeling

Utilize demographic dataUtilize demographic dataAge, sex, diagnosesAge, sex, diagnoses

Rely upon historical financial dataRely upon historical financial data

Predict riskPredict risk

26

Second Generation ModelingSecond Generation Modeling

Utilize first generation data sourcesUtilize first generation data sources

Incorporate second generation sourcesIncorporate second generation sourcesPharmacy dataPharmacy data

Lab dataLab data

Test dataTest data

Predictions based on risk adjustmentPredictions based on risk adjustmentDCGsDCGs, , ACGsACGs, , ETGsETGs

27

Third Generation ModelingThird Generation Modeling

Utilize first and second generation sourcesUtilize first and second generation sourcesIncorporate other sources and modelsIncorporate other sources and models

Health risk appraisals, questionnairesHealth risk appraisals, questionnairesSurveysSurveysRegional variabilityRegional variabilityACGsACGs, , DCGsDCGs, , ETGsETGs

Model the modelsModel the modelsChoose the best modeling of models for resultsChoose the best modeling of models for resultsLearn from previous data modelingLearn from previous data modeling

28

Perform dataPerform data ccleanupleanup

Split data into 2 yearsSplit data into 2 years

Use Year1 data to predict Year2 costUse Year1 data to predict Year2 cost

Identify modeling clusters and select best driversIdentify modeling clusters and select best driversDiseases, enrollment groups, product line Diseases, enrollment groups, product line

Episodes of Care (ETGs)

MEDai’s Clinical

Groupings

Patient Characteristics•Age•Gender•Enrollment•Insurance type

Provider Specialty

Selected Resource Measures

Timing & Frequency of

Services

Risk Score

DrugsEpisodes of Care (ETGs)

MEDai’s Clinical

Groupings

Patient Characteristics•Age•Gender•Enrollment•Insurance type

Provider Specialty

Selected Resource Measures

Timing & Frequency of

Services

Risk Score

Drugs

Third Generation ProcessThird Generation Process

Source: MEDai Inc.

29

Select model for optimum training of each clusterSelect model for optimum training of each clusterLinear & Nonlinear / Regression, Neural Networks….,Linear & Nonlinear / Regression, Neural Networks….,

Modeling of ModelsModeling of Models

Source: MEDai Inc.

30

OverviewOverview





31

0

5000

10000

15000

20000

25000

30000

35000

40000

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100

Fu

ture

Do

llars

Predicted

Actual

Accuracy of Third Generation ModelAccuracy of Third Generation Model

• Predicted Costs By Memberversus

• Actual Costs Experienced that Year

Each data point represents a single group of members within a range of predicted costs from the lowest predicted group to the highest predicted group (100 groups each with 1900 members)

Source: MEDai Inc.

32

Diabetes Model AccuracyDiabetes Model Accuracy

95 members per observation95 members per observation

0

10000

20000

30000

40000

50000

60000

70000

80000

90000

0 20000 40000 60000 80000 100000 120000 140000

Predicted $

Ac

tua

l $

Source: MEDai Inc.

33

0

5000

10000

15000

20000

25000

30000

35000

40000

0 10000 20000 30000 40000 50000 60000 70000

Predicted $

Ac

tua

l $

Depression Model AccuracyDepression Model Accuracy

105 members per observation105 members per observation Source: MEDai Inc.

34

HypertensionHypertension Model AccuracyModel Accuracy

230 members per observation230 members per observation

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

0 10000 20000 30000 40000 50000 60000 70000

avg predicted $

av

g a

ctu

al $

Source: MEDai Inc.

35

Underwriting Impact for Employer GroupsUnderwriting Impact for Employer Groups

$6 $6 $124 $124 $118 $118 36 36 KK

($36)($36)$168 $168 $204 $204 3939JJ

$5 $5 $135 $135 $131 $131 4141II

$15 $15 $126 $126 $111 $111 4444HH

($36)($36)$134 $134 $170 $170 4848GG

($32)($32)$113 $113 $145 $145 5151FF

($21)($21)$93 $93 $114 $114 5555EE

($43)($43)$124 $124 $167 $167 6161DD

$28 $28 $149 $149 $121 $121 7878CC

($12)($12)$139 $139 $151 $151 7878BB

$6 $6 $119 $119 $114 $114 8080AA

DifferenceDifferenceInternal Actuary Internal Actuary Premium SPMPMPremium SPMPM

MEDai forecast MEDai forecast $PMPM$PMPMMembersMembers

Sample of Sample of Employer Employer GroupGroup

• Health plan XYZ compared their premiums for a sample of employer groups using actuary vs. MEDai’s. • The actuarial model underestimated on the majority of small employer groups in comparison to MEDai.• This creates substantial losses, since 80% of the employers are small-group. The MEDai forecasting provides a

savings opportunity that approximated $11 million for 100,000 lives.• The client states that “in the final quarter of 2001, the actual cost for these groups shows clearunderestimation by the internal actuary forecasting.” Source: MEDai Inc.

38

Successful Predictive ModelingSuccessful Predictive Modeling

Identify clear goalsIdentify clear goalsModels fit some better than othersModels fit some better than others–– Actuarial versus care managementActuarial versus care management

Assess available data inputsAssess available data inputsDemographic, claims, pharmacy, lab valuesDemographic, claims, pharmacy, lab values

Secure a product championSecure a product championKey to any successful implementationKey to any successful implementation

Apply effective change managementApply effective change managementAdjustment of approach to care managementAdjustment of approach to care management

39

ReferencesReferences

MonheitMonheit AC, Persistence in health expenditures in the short run: AC, Persistence in health expenditures in the short run: Prevalence and consequences, Medical Care 41, supplement 7 Prevalence and consequences, Medical Care 41, supplement 7 (2003): III53(2003): III53––III64.III64.StrunkStrunk B, Ginsburg P, Tracking health care costs: Trends stabilize B, Ginsburg P, Tracking health care costs: Trends stabilize but remain high in 2002, Health Affairs (Web Exclusive June 11, but remain high in 2002, Health Affairs (Web Exclusive June 11, 2003)2003)StrunkStrunk B, Ginsburg P, Tracking health care costs: Trends slow in B, Ginsburg P, Tracking health care costs: Trends slow in first half of 2003, Center for Studying Health System Change, first half of 2003, Center for Studying Health System Change, December 2003.December 2003.HefflerHeffler et al., Health spending projections for 2002et al., Health spending projections for 2002--2012, Health 2012, Health Affairs (Web Exclusive February 7, 2003) for 1985Affairs (Web Exclusive February 7, 2003) for 1985––2001.2001.HefflerHeffler et al., Employer health benefits 2003 annual survey, The et al., Employer health benefits 2003 annual survey, The Kaiser Family Foundation and Health Research and Educational Kaiser Family Foundation and Health Research and Educational Trust, September 2003 for 2002Trust, September 2003 for 2002––2003. 2003. McGlynnMcGlynn et al., et al., ““The Quality of Health Care Delivered to Adults in The Quality of Health Care Delivered to Adults in the United States,the United States,”” The New England Journal of Medicine (June 26, The New England Journal of Medicine (June 26, 2003): 26352003): 2635––26452645

40

ReferencesReferencesWeiner JP, Weiner JP, ““Predictive modeling and risk measurement: Predictive modeling and risk measurement: Paradigms, potential and pitfalls.Paradigms, potential and pitfalls.”” Presented at a symposium Presented at a symposium on on ““Predictive ModelingPredictive Modeling”” sponsored by the National Blue sponsored by the National Blue Cross/Blue Shield Association, Chicago, January 30, 2003.Cross/Blue Shield Association, Chicago, January 30, 2003.Forrest CB, Forrest CB, ““PopulationPopulation--based predictive modeling using based predictive modeling using ACGsACGs: : Application to disease and case management,Application to disease and case management,”” International International ACG UserACG User’’s Conference, November 11, 2003.s Conference, November 11, 2003.LeGrowLeGrow G, Metzger J, G, Metzger J, EE--Disease ManagementDisease Management, First Consulting , First Consulting Group, Group, ©© 2001, California HealthCare Foundation2001, California HealthCare FoundationCouch JB, Couch JB, The Physician’s Guide to Disease ManagementThe Physician’s Guide to Disease Management. © . © 1997, Aspen Publishing, Inc., Gaithersburg, MD1997, Aspen Publishing, Inc., Gaithersburg, MDKongstevdtKongstevdt PR, PR, The Managed Health Care Handbook, 3The Managed Health Care Handbook, 3rdrd. ©. ©1996, Aspen Publishers, Inc., Gaithersburg, MD1996, Aspen Publishers, Inc., Gaithersburg, MDwww.medai.comwww.medai.com

using predictive modeling to target interventions

Documents