WIXLIB

Using Predictive Models to Move MedicaidProviders Toward Best Practices and Reduce WastePredictive Modeling SummitGary Redding, Vice President and Practice Leader, Thomson ReutersMichelle McAllister, Consulting Manager, Thomson Reuters

AGENDA Thomson Reuters Using Predictive Models to Move MedicaidProviders Toward Best Practices and ReduceWaste 2009 Thomson Reuters Questions and Discussion2HEALTHCARE PAYER SOLUTIONS

Thomson ReutersGary Redding, Vice President and Practice Leader, Thomson Reuters

2009 Thomson ReutersOUR FOOTPRINT4HEALTHCARE PAYER SOLUTIONS

Knowledge To ActThomson Reuters is the world’s leading source ofintelligent information for business and professionals.FINANCIALLEGALHEALTHCARE & SCIENCE Integrated decision support solutions for improvingclinical and business performanceTAX &ACCOUNTINGMEDIA Insights for better healthcare touching 150 millionlives 2009 Thomson Reuters We are known for our: Independence Objectivity Reliability5HEALTHCARE PAYER SOLUTIONS

UNPARALLELED EXPERTISE ACROSSTHE SPECTRUM OF HEALTHCAREGE, FedEx,GM, Ford,200 Boeing,EmployersCorning, DowMany U.S.hospitals100 HealthPlans3,000 HospitalsFederalGovernmentCMS, AHRQ,CDC, SAMHSA,VA, DOD11 state employers28 Medicaidagencies8 HHS agenciesAll major U.S.pharmacompanies 2009 Thomson ReutersCIGNA,Paramount,Emblem,Humana,NumerousBCBS plansStateGovernmentPharma6HEALTHCARE PAYER SOLUTIONS

A Trusted Partner to Government and edpac 2009 Thomson ReutersResearch& Non-ProfitStateGovernment28 Medicaid agencies11 state employersAARPAcademyHealthBridges to ExcellenceCalifornia Cooperative Healthcare Reporting initiativeIHA (Integrated Healthcare Association)Pacific Business Group on HealthMidwest Health Initiative (St. Louis)Robert Wood Johnson FoundationOthers77HEALTHCARE PAYER SOLUTIONS

Medicaid RMSALGALATX 2009 Thomson ReutersFL8HEALTHCARE PAYER SOLUTIONS

OUR VALUEPROVIDING INTELLIGENT INFORMATIONTO MAKE BETTER DECISIONS FASTERDECISION SUPPORT 2009 Thomson ReutersClinical and ManagementReporting, Modeling,AnalysisMANAGING DATACONSULTING ANALYTICS400 Sources,SAS 70 CertifiedSubject Matter Experts,Account TeamMEASURING PERFORMANCEBENCHMARKING & NORMSQuality Measurement, P4P,Physician Measurement3,000 Measures, 69 Million Lives,PULSE SurveyREFERENTIAL DATAENGAGING CONSUMERSConsumer Advantage, InformedEnrollmentClinical, Drug, Disease, Poison,Toxicology9HEALTHCARE PAYER SOLUTIONS

We Help Our Customers . . . Analyze state and national trends Profile provider performance Monitor population health Fight Fraud, Abuse, Waste and Overpayment Improve clinical performance and outcomes Forecast, monitor and manage healthcare costs Oversee vendor contracts Support care management Formulate payment policy Launch cost control strategies 2009 Thomson Reuters Support prevention programs Answer any healthcare programquestionHEALTHCARE PAYER SOLUTIONS

Using Predictive Models to Move MedicaidProviders Toward Best Practices and Reduce WasteMichelle McAllister, Consulting Manager, Thomson Reuters

Overview Context– What is Predictive Modeling?– History of Predictive Modeling How it Works– Predictive Modeling Basics– A Useful Approach to Predictive Modeling and Targeting 2009 Thomson Reuters Case Studies: Applications in Moving ProvidersToward Best Practices and Reduce Waste12HEALTHCARE PAYER SOLUTIONS

Context

What is Predictive Modeling? Use of demographic, diagnostic, and utilizationinformation, with analytic models to predict:– Beneficiaries who will be high-risk/high-cost in the future– Future costs– Future utilization– Influence Best Practices– Help Reduce Waste 2009 Thomson Reuters in order to better intervene, manage risk, ensurequality and set rates14HEALTHCARE PAYER SOLUTIONS

History of Predictive Modeling Research beganon health-basedrisk models Health Insurance Planof California (HIPC)implemented ahomegrown riskadjustment model 2009 Thomson Reuters State of WashingtonEmployees and Stateof Colorado’s Medicaidprogram used riskadjustment forpayments15 1st Society of Balanced Budget Act –Actuaries studymandated risk-adjustedpayments forcomparing severaldifferent risk modelsMedicare Choice foryear 2000 (Principaland risk adjusters.Inpatient Dx Model)HEALTHCARE PAYER SOLUTIONS

History of Predictive Modeling (cont.) 2009 Thomson Reuters 2nd Society ofActuaries studycomparing severaldifferent riskmodels and riskadjusters. DCGschosen best inclass. CMS selects an ‘allencounter model’16 3rd Society ofActuaries studycomparing severaldifferent risk modelsand risk adjusters.DCGs chosen best inclass. 12 modelstested.HEALTHCARE PAYER SOLUTIONS

History of Predictive Modeling (cont.)Individual Level R-squared60.00%DCG All-EncounterACGsCDPS50.00%CRGsIngenix ERG* R squaredExplains predictive value.A higher R squaredequates to a better degreeof predictability.DCG RxGroups40.00%Ingenix PRG30.00%20.00% 2009 Thomson Reuters10.00%0.00%ConcurrentProspectiveSociety of Actuaries, “A Comparative Analysis of Claims-based Methods of Health Risk Assessment for Commercial Populations,” 2007HEALTHCARE PAYER SOLUTIONS

How it works

Predictive Modeling Basics Most models are designed to require 1 year’s worthof enrollment and medical and/or pharmacy data. Purpose Varies– Predictive Modeling in Pricing Identify difference in cost between people and groups to priceaccordingly– Predictive Modeling in Care Management 2009 Thomson Reuters Resource use and intensity is the focus19HEALTHCARE PAYER SOLUTIONS

DCGs Thomson Reuters currentlyutilizes Diagnostic Cost Groups(DCGs)– DCGs are a population-basedclassification and riskadjustment methodologyDCG Models in Advantage SuitePopulationGroupMedicareAll - encounterMedicaidAll - encounter– Developed and licensed byVerisk Health Inc.– Selected by CMS for theMedicare Choice ProgramVariantAll – encounterCommercialRx Groups –Rx/InpatientRx Groups – RxOnly 2009 Thomson ReutersEtc.HEALTHCARE PAYER SOLUTIONS

The DCG models work by: Classifying raw administrative (medical and/orpharmacy claims as well as eligibility and aidcategory) data into coherent clinical groupings Applying clinically valid hierarchies and interactionsto create an aggregated, empirically valid patientscore at the individual beneficiary level 2009 Thomson Reuters Correlating the scores with the cost of the healthburden carried by the beneficiary Aggregating individual scores by groups of interestcreates group-level predictive results specific tomany Medicare/Medicaid applications21HEALTHCARE PAYER SOLUTIONS

Medical Episode Grouper (MEG)What is an episode? An episode is all care for a course of treatment of a disease– Primary and secondary diagnosis codes grouped into one of 560disease categories– Stratified by severity of illness– Disease categories and clinical criteria specified by physicians atJefferson Medical College Includes inpatient, outpatient and drug claims– Not every claim is assigned an episode Also includes admissions 2009 Thomson Reuters Patients may have more than one episode at any given time22HEALTHCARE PAYER SOLUTIONS

2009 Thomson ReutersMEG providesclinically relevantgrouping23HEALTHCARE PAYER SOLUTIONS

MEG: Grouping MethodologyAppropriatelygroup claimsinto episodesaccording todiseasecategory andrelative timebetweenservices Classification - Diagnosis codes from healthcare claims and otheradministrative data are grouped into one of the over 550 DiseaseStaging disease categories and severity stages. Beginning and Ending Episodes - Clean periods unique to eachdisease category are used to group each claim into an episode. Inclusion Logic - Less specific episode groups occurring in closeproximity to specific episodes are combined with the specificepisodes, e.g. “other gastrointestinal or abdominal symptoms" and"appendicitis". Drug Data - Mappings of National Drug Codes (NDCs) to episodegroups enables pharmacy claims to be grouped to relevant episodes. Lookback Procedure - Lab and diagnostic imaging claims precedingan episode are examined to determine whether they should becombined with the episode. 2009 Thomson 4RxOfficeVisitThe completeepisode ranges intime between the labtest and the finaloffice visit.HEALTHCARE PAYER SOLUTIONS

MEG—Putting it All TogetherEpisode an e 2009 Thomson ReutersVisitDRUG TRANSACTION RMLOOKUP 1-1595-11-1525HEALTHCARE PAYER SOLUTIONS

Medical Episode Grouper (MEG)What is the analytic value of using episodes? Attributes all costs to diseases– Particularly prescription drugs Finds a physician that is accountable for treatment– Managing Physician Identifies the physician who guided treatment usually providingmultiple E & M visits Use to identify the decision-maker in a patient’s care– Primary Physician Identifies the physician with the highest costs Use to evaluate if specific episode costs 2009 Thomson Reuters Measures and compares the costs of treating disease– Which providers and plans are most efficient– Both price and resource consumption26HEALTHCARE PAYER SOLUTIONS

MEG: Potential Drug Substitutions – ACEI vs. ARBQuintile 1Quintile 512%41%59%88% 2009 Thomson ReutersACEIARBACEIARB ARB has no generic substitute and is more costly than therapeutically similar ACEI. Quintile 5 patients are more likely to receive an ARB.27HEALTHCARE PAYER SOLUTIONS

MEG:Variation and Cost of Antithrombotic Treatment People prescribe/take antithrombotics for hypertension to lower risk ofischemic event – “it makes sense.” Aspirin/antiplatelet is recommended for secondary prevention of ischemicevent. However:– Warfarin not recommended.– Antiplatelet therapy not recommended for primary prevention, as risk ofhemorrhage is greater than risk reduction of ischemic event. 2009 Thomson ReutersCost of Antithrombotic Drugs for Primary PreventionQuintile 1 3,686Quintile 5 213,962Lip GYH, Felmeden DC. Antiplatelet agents and anticoagulants for hypertension. CochraneDatabase of Systematic Reviews 2004, Issue 3. Art. No.: CD003186. DOI:10.1002/14651858.CD003186.pub228HEALTHCARE PAYER SOLUTIONS

Episodes and Patient-Level Adjustment Diagnostic Cost Groups (DCGs)– Risk adjustment methodology used to predictcurrent or future patient costs, e.g. relative riskscore (RRS)– Unit of analysis – the patient– Based on all prior or current year claims to identifypatient-level complexity/comorbidities 2009 Thomson Reuters Together, MEG and DCGs provide a completepicture of a patient29HEALTHCARE PAYER SOLUTIONS

Disease Staging Framework Initially developed under contract to NCHSR withongoing private development by Medstat/Jeffersonteams Software converts a stream of claims into clinicallyhomogeneous groups Takes over 15,000 ICD-9-CM codes to 560 diseasecategories Independent of setting or treatment 2009 Thomson Reuters Etiology assigned to each category Severity stratification based on robust clinicalcriteria30HEALTHCARE PAYER SOLUTIONS

A Useful Approach to Predictive Modeling andTargeting: Risk Adjusted EpisodesWhat is Risk Adjusted Episodes? An improved method of benchmark adjustment in Decision Analystthat combines the disease and severity strength of MEG with thepopulation-based risk adjustment power of the DCG grouper Basing risk adjustment on average episode costs ignores significant riskfactors which physicians take into account when treating a patient.– Age and gender– Comorbidities Basing risk adjustment on a single illness burden score (for the patient)ignores vital information about diseases and severity levels. 2009 Thomson Reuters– Risk adjustment using illness burden accounts for the risk of a population across allconditions, but is not enough to determine expected costs for a specific condition. Combining episodes and illness-burden methodologies addresses theshortcomings of each.31HEALTHCARE PAYER SOLUTIONS