Predictive Modeling and Risk Adjustment
Applications in the U.S. and I t
A li ti      i th U S d International M k t
                                 ti l Markets
                Dr. Rong Yi, Vice President, Consulting
  Dr. Thomas Zahn, General Manager, DxCG Gesundheitsanalytik GmbH
             World Health Congress, European Conference
                         Brussels, May, 2009
Overview

    Risk Adjustment and predictive models in the
             j             p
    big picture
    How are models developed
    Applications in healthcare financing,
    management, delivery and reforms




 ©2009 Verisk Health, Inc.                         2
Setting up the Stage -
Healthcare Dichotomy

                            Healthcare




                  Financing          Delivery




                  Feasibility
                                   Cost, Quality,
                     andd
                                   and Efficacy
                 sustainability
Healthcare Financing Challenges

     For all decision-makers from governments down
       to individual consumers:
       How much?
       How many years?
       Whom to cover?
       What services to offer?
       …

       Need to understand and predict acc ratel
                nderstand             accurately:
       • Risk selection
       • Aging and disease burden
          g g
       • Benefit design and outcomes
Healthcare Delivery Challenges

    Resource planning
    Health d
    H lth and wellness management
                 ll             t
     •   Population
     •                           g
         Disease and case management
     •   Wellness and productivity
     •   Patient education and outreach
     •   …

    Need to:
     • Stratify the population and prioritize management efforts
     • Measure outcomes (cost, quality and efficacy)
     • …
Risk adjustment and predictive modeling
as decision support tools
              pp
How are Models Built?
         Claims Data +
    Clinical Classification +
Econometric/Statistical Modeling
Data used as input

     Enrollment information:
     • age, sex, eligible months, basis of eligibility (e.g., disabled)
     Claims information
     •   Diagnosis
     •   Procedures
         P      d
     •   Pharmacy
     •   Long term care
     •   Spending – timing, categories, patterns
          p        g       g,     g     ,p
     •   Utilization – hospital, ED, specialty

     Not everything is used for prediction!
     • D
       Depends on client needs, available d t model’s i t d d
              d      li t     d     il bl data,   d l’ intended
       use, and the tradeoff between easy of use and added
       predictive accuracy
Risk Prediction using Medical
Diagnosis only

                       John Smith
                       Age: 45
                       Sex: M


                          Hypertension
                           essential hypertension

                          Type II Diabetes Mellitus
                           type II diabetes w/ renal manifestation

                          Congestive Heart Failure
                             g
                           hypertension heart disease, w/ heart failure
   6.35x sicker than      Drug/Alcohol Dependence
       average             alcohol dependence


                       Relative Risk Score: 6.35
Clinical Classification Systems
Example – DCG/HCC Diag Grouping
 Why Grouping?
         Distill massive amount of data to create useful analytic units
                                                             y

 Most Common Classification Systems:
      DCG/HCC - Grouping of Diagnosis Codes
      RxGroups - Grouping of Drug Codes
      BETOS - Grouping of Procedure Codes

 Example: The DCG/HCC Classification System

                ICD-9 or ICD-10 Diagnosis Codes


                         DxGroups (DxGs)
                           (784 groups)

                        Hierarchical C diti
                        Hi     hi l Condition
                         Categories (HCCs)
                            (184 groups)
What Can be Predicted?
… almost anything as long as data supports
           y g          g           pp

         Individual level prediction. Can be g p up by
                          p                  grouped p y
         age/gender, medical conditions, geography, benefit
         design, etc.
         Predicts within a year, across years and for multiple
                            y ,          y                 p
         years,
         Examples of predicted outcomes:
     •      Total healthcare cost or a subset, e.g., drug cost
                                                    g      g
     •      Distribution of risk, e.g., probability of cost more than
            $100,000
     •      Healthcare utilization, e.g., future hospitalization, avoidable
            ER visits, total length of stay use of advanced imaging tests
                visits                  stay,
     •      Disease progression and spending persistence
     •      Duration of injury
Risk adjustment and predictive modeling
as decision support tools
              pp
Bring All Together
                              Operations                               Marketing
                                                   Health Plans


            Revenue             Work Comp                              Government 
           Optimization          Insurers                               Agencies         Underwriting




      Employer              Employers
                                                                             Provider           Medical 
      Reporting                                                               Groups          Management




                                                                                Financial 
                Pharmacy Audit          Pharmacy              Consultants       Analysis




                                                                                                                 13
 © Verisk HealthCare 2008                                                                                  10/16/2008
Applications of Risk Adjustment and
         Predictive Modeling
         P di ti M d li
Healthcare Reform Example
Massachusetts Universal Healthcare

        2006 Massachusetts landmark legislation
                                         g
        Procurement of healthcare for the previously
        uninsured
    •      Capitation based on age/sex/geography/benefit design
              p                   g       g g p y            g
    •      Structured bids to incentivize lower bidders
        Challenge: risk selection. Healthier members cluster in
        lower-priced p
              p      plans
           Need to establish a fairer and more sustainable procurement
           process
        Solution:
    •      Fairer – account for disease burden
    •      Sustainable – prospective risk adjustment to ensure financial
           stability
Risk-based Capitation Illustration

                                 Systemwide
                            Monthly Payment/Capita
                                     $420




      Health Plan A              Health Plan B               Health Plan C
 Relative Risk Score 1.16   Relative Risk Score 0.61   Relative Risk Score 1.52
  Budget: $420 x 1.16         Budget $420 x 0.61         Budget: $420 x 1.52
          = $488                     = $256                     = $640


       Implementation:
       • Budget Neutral
       • Further Adjustments on benefit design, geographic factor,
       new/partial enrollees
           / ti l      ll
       • Quarterly adjustments to smooth cash flow
For Health Plans

 Care and disease management identification
                          g
 and stratification
 Understand changes in risk over time –
 actuarial and underwriting
  • Aging and increased medical comorbidities
  • Disease progression
 Reinsurance and stop loss arrangements
 Compare providers fairly, adjusting for
 differences in health status
 diff        i h lth t t
                                         ©2005 by DxCG®, Inc.
For Self-Insured Employers

 Pay health plans based on resource needs
    y        p
 rather than risk selection
 Compare health plans fairly (accounting for
 health status)
 Negotiate rate reductions from inefficient
 plans




                                     ©2005 by DxCG®, Inc.
For Providers

 Integrate with medical informatics:
  • Identify at-risk patients for the right care (high
    likelihood of hospitalization, avoidable ER use, etc.)
  • Identify care gaps and improve quality
  • Evidence-Based Medicine
 Compare quality and efficiency to their peers
  • Referral and prescription patterns
  • Use of advanced imaging tests
                         g g
 Pay-for-performance
 Patient-centered Medical Home

                                                ©2005 by DxCG®, Inc.

World Health Congress 2009 Europe Market Insight

  • 1.
    Predictive Modeling andRisk Adjustment Applications in the U.S. and I t A li ti i th U S d International M k t ti l Markets Dr. Rong Yi, Vice President, Consulting Dr. Thomas Zahn, General Manager, DxCG Gesundheitsanalytik GmbH World Health Congress, European Conference Brussels, May, 2009
  • 2.
    Overview Risk Adjustment and predictive models in the j p big picture How are models developed Applications in healthcare financing, management, delivery and reforms ©2009 Verisk Health, Inc. 2
  • 3.
    Setting up theStage - Healthcare Dichotomy Healthcare Financing Delivery Feasibility Cost, Quality, andd and Efficacy sustainability
  • 4.
    Healthcare Financing Challenges For all decision-makers from governments down to individual consumers: How much? How many years? Whom to cover? What services to offer? … Need to understand and predict acc ratel nderstand accurately: • Risk selection • Aging and disease burden g g • Benefit design and outcomes
  • 5.
    Healthcare Delivery Challenges Resource planning Health d H lth and wellness management ll t • Population • g Disease and case management • Wellness and productivity • Patient education and outreach • … Need to: • Stratify the population and prioritize management efforts • Measure outcomes (cost, quality and efficacy) • …
  • 6.
    Risk adjustment andpredictive modeling as decision support tools pp
  • 7.
    How are ModelsBuilt? Claims Data + Clinical Classification + Econometric/Statistical Modeling
  • 8.
    Data used asinput Enrollment information: • age, sex, eligible months, basis of eligibility (e.g., disabled) Claims information • Diagnosis • Procedures P d • Pharmacy • Long term care • Spending – timing, categories, patterns p g g, g ,p • Utilization – hospital, ED, specialty Not everything is used for prediction! • D Depends on client needs, available d t model’s i t d d d li t d il bl data, d l’ intended use, and the tradeoff between easy of use and added predictive accuracy
  • 9.
    Risk Prediction usingMedical Diagnosis only John Smith Age: 45 Sex: M Hypertension essential hypertension Type II Diabetes Mellitus type II diabetes w/ renal manifestation Congestive Heart Failure g hypertension heart disease, w/ heart failure 6.35x sicker than Drug/Alcohol Dependence average alcohol dependence Relative Risk Score: 6.35
  • 10.
    Clinical Classification Systems Example– DCG/HCC Diag Grouping Why Grouping? Distill massive amount of data to create useful analytic units y Most Common Classification Systems: DCG/HCC - Grouping of Diagnosis Codes RxGroups - Grouping of Drug Codes BETOS - Grouping of Procedure Codes Example: The DCG/HCC Classification System ICD-9 or ICD-10 Diagnosis Codes DxGroups (DxGs) (784 groups) Hierarchical C diti Hi hi l Condition Categories (HCCs) (184 groups)
  • 11.
    What Can bePredicted? … almost anything as long as data supports y g g pp Individual level prediction. Can be g p up by p grouped p y age/gender, medical conditions, geography, benefit design, etc. Predicts within a year, across years and for multiple y , y p years, Examples of predicted outcomes: • Total healthcare cost or a subset, e.g., drug cost g g • Distribution of risk, e.g., probability of cost more than $100,000 • Healthcare utilization, e.g., future hospitalization, avoidable ER visits, total length of stay use of advanced imaging tests visits stay, • Disease progression and spending persistence • Duration of injury
  • 12.
    Risk adjustment andpredictive modeling as decision support tools pp
  • 13.
    Bring All Together Operations Marketing Health Plans Revenue  Work Comp  Government  Optimization Insurers Agencies Underwriting Employer  Employers Provider  Medical  Reporting Groups Management Financial  Pharmacy Audit  Pharmacy Consultants Analysis 13 © Verisk HealthCare 2008 10/16/2008
  • 14.
    Applications of RiskAdjustment and Predictive Modeling P di ti M d li
  • 15.
    Healthcare Reform Example MassachusettsUniversal Healthcare 2006 Massachusetts landmark legislation g Procurement of healthcare for the previously uninsured • Capitation based on age/sex/geography/benefit design p g g g p y g • Structured bids to incentivize lower bidders Challenge: risk selection. Healthier members cluster in lower-priced p p plans Need to establish a fairer and more sustainable procurement process Solution: • Fairer – account for disease burden • Sustainable – prospective risk adjustment to ensure financial stability
  • 16.
    Risk-based Capitation Illustration Systemwide Monthly Payment/Capita $420 Health Plan A Health Plan B Health Plan C Relative Risk Score 1.16 Relative Risk Score 0.61 Relative Risk Score 1.52 Budget: $420 x 1.16 Budget $420 x 0.61 Budget: $420 x 1.52 = $488 = $256 = $640 Implementation: • Budget Neutral • Further Adjustments on benefit design, geographic factor, new/partial enrollees / ti l ll • Quarterly adjustments to smooth cash flow
  • 17.
    For Health Plans Care and disease management identification g and stratification Understand changes in risk over time – actuarial and underwriting • Aging and increased medical comorbidities • Disease progression Reinsurance and stop loss arrangements Compare providers fairly, adjusting for differences in health status diff i h lth t t ©2005 by DxCG®, Inc.
  • 18.
    For Self-Insured Employers Pay health plans based on resource needs y p rather than risk selection Compare health plans fairly (accounting for health status) Negotiate rate reductions from inefficient plans ©2005 by DxCG®, Inc.
  • 19.
    For Providers Integratewith medical informatics: • Identify at-risk patients for the right care (high likelihood of hospitalization, avoidable ER use, etc.) • Identify care gaps and improve quality • Evidence-Based Medicine Compare quality and efficiency to their peers • Referral and prescription patterns • Use of advanced imaging tests g g Pay-for-performance Patient-centered Medical Home ©2005 by DxCG®, Inc.