Predictive Analytics And Accountable Care Organizations

Predictive Analyticsand Accountable CareOrganizationsApplication of Modelsfor ManagingPopulations andProvider NetworksRon RussellDecember 5, 2012

The Healthcare ChallengeWhat is Contributing to the Rising Trend in Spending? 3 of every 4 spent in healthcare is on chronic disease*Diabetics who are not compliant with taking their insulin, not visiting a physician, and notmanaging their weight will eventually end up in the ER or be hospitalizedAlmost 18% of Medicare patients are readmitted within 30 days of discharge**This outcome is often related to uncoordinated post-hospitalization care or a lack of focus onprevention and quality outcomes Diagnostic imaging from CAT scan and MRI contribute 26.5 billion inunnecessary use of health services***Fee-for-service payment systems incent providers to do more with too little focus on qualityof care or appropriate utilization*CDC**CMS***McKinsey Global Institute 2012 Verisk Health, Inc. All Rights Reserved2

Accountability Reduce cost Improve quality Value based reimbursement Shared risk arrangements Bundled payments 2012 Verisk Health, Inc. All Rights Reserved3

Successful population health management requires theeffective application of predictive modeling andanalytics to patients and providers 2012 Verisk Health, Inc. All Rights Reserved4

Data Driven StrategyPopulation Health ManagementIdentify patients with high-risk score AND at risk of expensive medical servicesPractice Variation ManagementSupport providers with higher than expected utilization with care managementservices 2012 Verisk Health, Inc. All Rights Reserved5

Choosing a Predictive ModelAge and gender only explain 3-5% of variationPredictive models achieve up to 27% of variation at the individual and level and provide ahighly accurate cost projection at the population group levelToday predictive models are tailored to payer type, a wide range of outcomes (cost, events,payment), and prediction periodsPrediction Period Concurrent predictive models (retrospective) measure the risk of population groups foran historic period, based on all of the conditions present in the period Prospective models predict risk for a future period based on the demographic andclinical mix of the population in the baseline periodDiagnosis only or Diagnosis with prior cost and utilizationOutcome Predicted Medical plus pharmacy cost Total risk Medical cost only Pharmacy cost only Primary Care funding Likelihood of hospitalization or Emergency Room encounters Range of utilization services 2012 Verisk Health, Inc. All Rights Reserved6

Population HealthManagement7 2012 Verisk Health, Inc. All Rights Reserved

Selecting Members for CareManagement:Clinical Interventions with an Impact Assess the health status of the population Identify the group of individuals at high risk of future utilization or poor healthoutcomes Focus on the subset of people that case managers believe they can impact througha defined intervention 2012 Verisk Health, Inc. All Rights Reserved8

Allocate Resources to HighestRisk Patients Based on Predicted CostHigh Cost Case IdentificationGenerate a list of predicted high cost individualsThe Care Manager at the practice site can analyze the patient roster each month and provide a casereview for each PCPFocused on medication compliance, gaps in care, and regular office visits in an effort to avoid urgentcare services in the ER, avoidable hospitalizations, and complicationsIncorporating conditions and priorutilization into predictions willimprove the predictive accuracy ofcost projections. These patients arehigh priority for care management 2012 Verisk Health, Inc. All Rights Reserved9

Disease Registry of the MemberPopulationClinical ConditionProspectiveRisk ScoreRate /1,000TotalCurrentNewPlan** ry Artery Disease4.002,8122,1292820.314Cerebrovascular hronic Renal Failure9.4364246444.63.6Cost PMPYClinical ConditionER Visits/1,000Plan** NormHypertension 8,187 9,911341325146138Hyperlipidemia 6,641 7,9432262228584 10,189 12,672380372192177 8,777598533150152Coronary Artery Disease 16,012 23,555626641391407Cerebrovascular Disease 22,832 33,288935964587580COPD 18,007 28,209756860486494Chronic Renal Failure 32,240 43,605861801615587DiabetesAsthma 7,129Plan** NormAdm/1,000Plan** NormNOTE: the benchmark norm includes 11 Million lives and is adjusted forage, gender & region of the country 2012 Verisk Health, Inc. All Rights Reserved10

Tools to Assess Care StrategiesStratification of the Diabetic population by comorbidityDiabetes by ComorbidityProspectiveRisk ScoreTotalCurrentRate/1,000PMPYERADMAll Diabetics3.017,2915,49052.7 10,189380192Hypertention4.022,2381,82516.2 14,256487277Complicated Hypertension6.932782252 22,632740470Uncomplicated Hypertension3.601,9601,60014.2 13,051451250Hyperlipidemia3.261,3151,1359.5 11,011322165Back Pain4.141,1679408.4 13,601501258CAD5.677585935.5 22,005751531Cancer6.384363273.2 26,73553541311.482351781.7 38,987975733Cerebrovascular Disease7.962311731.7 32,7481,283914COPD8.701851301.3 34,4691,05385711.891831291.3 52,4761,6221,339Asthma4.921751421.3 19,614852441Atrial Fibrillation6.891741371.3 32,658882785Major Depression5.471731341.3 21,422789519Chronic Renal FailureCHF 2012 Verisk Health, Inc. All Rights Reserved11

Quantify OpportunitiesStratification of the Diabetic population by comorbidityDiabetes by ComorbidityProspectiveRisk ScoreTotalCurrentRate/1,000PMPYERADMAll Diabetics3.017,2915,49052.7 10,189380192Hypertention4.022,2381,82516.2 14,256487277Complicated Hypertension6.932782252 22,632740470Uncomplicated Hypertension3.601,9601,60014.2 13,051451250Hyperlipidemia3.261,3151,1359.5 11,011322165Back Pain4.141,1679408.4 13,601501258CAD5.677585935.5 22,005751531Cancer6.384363273.2 26,73553541311.482351781.7 38,987975733Cerebrovascular Disease7.962311731.7 32,7481,283914COPD8.701851301.3 34,4691,05385711.891831291.3 52,4761,6221,339Asthma4.921751421.3 19,614852441Atrial Fibrillation6.891741371.3 32,658882785Major Depression5.471731341.3 21,422789519Chronic Renal FailureCHF 2012 Verisk Health, Inc. All Rights Reserved12

Target Individuals for CareCoordinationCohort: Diabetes Hypertension Obesity without Antihyperlipidemic Rx 20 Members 2012 Verisk Health, Inc. All Rights Reserved13

Ambulatory Care Sensitive Admissions1. ACS admissions are potentially avoidable admissions given timely andappropriate ambulatory care2. Their prevalence is influenced by age and the clinical condition mix ofthe population3. Higher than risk expected rates may be indicative of restricted accessto primary care services, delayed attention to a medical need, or poorpatient compliance4. Variation in ACS rates can lead us to a rich set of opportunities forcare improvement and cost savingsACS Admissions Provide a Window onAmbulatory Care Management Opportunities 2012 Verisk Health, Inc. All Rights Reserved14

ACS ConditionsConsensus around key factors that will reduce ACS admissions Improve access to primary care servicesCreate clinical care teamsEngage patients to foster complianceStrengthen social support systemsThere is no “one” right rate The prevalence of chronic conditions will influence the likelihood of anACS admissionACS Conditions can be separated into Acute and ChronicSelected ConditionsTop 5 Chronic Conditions of interest include: AnginaAsthmaCongestive Heart FailureCOPDDiabetes 2012 Verisk Health, Inc. All Rights Reserved15

Seeking ACS AdmissionsUsing a likelihood of hospitalization (LOH) model, we analyzed the ability of themodel to correctly identify individuals at risk of hospitalization in generalOf the individuals listed in the top 0.5% at risk, 37% of them were admitted tothe hospital in a 6 month periodOur assumption was that a significant proportion of admissions were potentiallyavoidable so we used ACS criteria to test our hypothesis 2012 Verisk Health, Inc. All Rights Reserved16

Targeted Individuals Incur ACSAdmissions that are PotentiallyAvoidablePerformance statistics for six-month validation studyACS AdmissionsIndividuals withACS AdmACS AdmissionsTotal AllowedDays IncurredAvg CostALOSACS 0.5% LOH List192224 1.4 M1,010 6,3764.5Cohort .05 ‐ 5% LOH List310337 2.3 M1,335 6,8024.0Cohort 6 ‐ 20% LOH List520540 2.4 M1,679 4,3523.1Reaching out to the top 0.5% of members proactively holds the potential ofreducing admissions and improving outcomes through high qualityambulatory care 2012 Verisk Health, Inc. All Rights Reserved17

List of ACS Admissions from the StudySample ACS conditions in six-month period top 0.5% CohortTOP VOLUME ACS ADMISSIONSIndividuals# ADMBronchitis & Asthma3138Gastroenteritis26Chest PainDaysAvg CostALOS 172,634125 4,5433.327 83,243132 3,0834.91823 132,38172 5,7563.1Pneumonia1717 85,29482 5,0174.8Diabetes1620 98,25178 4,9133.9COPD1315 67,04359 4,4703.9Seizures1215 52,73835 3,5162.3Cellulitis77 30,88636 4,4125.1 2012 Verisk Health, Inc. All Rights ReservedTotal Allowed18

Accurate Identification of IndividualsMost Likely to Seek Care in the EDLikelihood of Emergency Department Encounter: Performance statistics for 6month Validation studyMember Cohort onLikelihood List LOEDTop 0.5% on LOED List0.5 to 5% 2012 Verisk Health, Inc. All Rights ,0316,253PositivePredictiveValue82%55%Visit Count fortheseIndividuals Rate per Person4,13112,7224.02.019

Analysis of Emergency RoomEncountersTop 10 Diagnosis in ER for 6 month Validation studyTop 0.5% Members: LOED Model has 82% PPVDescription Primary Diagnosis ED VisitAcute Upper Respiratory InfectionFever, No other symptomsOtitis Media, No other symptomsViral Infection, No other symptomsAcute PharyngitisCoughNoninfectious GastroenteritisVomiting AlonePneumonia, OrganismHeadacheTotal Top Ten CausesTOTAL ED VISITS 2012 Verisk Health, Inc. All Rights 51246Pct All ,03420

Bringing It All Into FocusData driven analytic toolsallow for understandingrisk and making decisionsabout where to focusresourcesClinical ConditionsHigh cost casemanagementRisk ofHospitalizationFuture CostPrograms for unstableclinical conditionsComplex casemanagement at theepicenter of risk 2012 Verisk Health, Inc. All Rights ReservedRisk of Emergency Visit21

Practice VariationManagement22 2012 Verisk Health, Inc. All Rights Reserved

Building a Strong Primary CareFoundation Primary Care Activity Level model can be used for prospectivepayment Risk score reflects the resource intensity for direct primary services andfor care coordination services Individual scores varied from 16 times average for the top 0.5% of thephysician panel to a low of 1/10th the average for the lower 30% of c/2012/08000 2012 Verisk Health, Inc. All Rights Reserved23

Building a Strong Primary CareFoundation Service utilization models estimate the comparative rate ofhospitalization, readmissions, ACS admissions, ER, specialtyphysician services, diagnostic testing, outpatient imaging, andpharmacy cost 2012 Verisk Health, Inc. All Rights Reserved24

Building a Strong Primary CareFoundation 2012 Verisk Health, Inc. All Rights Reserved25

Commercially Insured Populations There is not one “right” rate for ACS admissions Benchmarks are useful Risk adjustment is essentialAdmissionCategoryVerisk HealthBenchmarkCommercialClientExampleClient TieredNetworkProviderVariationfrom 9.021.2 2012 Verisk Health, Inc. All Rights ReservedThe annualACS rate is36% higherthan risk1.3626

Comparative Performance ProfilesPlotting two quality measures for seven large provider groups: care gap index andACS admission ratio of actual-to-expected rateACS Admission Rate IndexMore thanexpectedMore thanexpectedCare Gap Index 2012 Verisk Health, Inc. All Rights Reserved27

Comparative Performance ProfilesTwo of the Capitol Group providers with large patient panels have a high caregap index and higher than expected ACS Admission rateACS Adm 50/1,000ACS Index 1.97ACS Adm 32/1,000ACS Index 1.58 2012 Verisk Health, Inc. All Rights Reserved28