Health-Based Risk Assessment - Changes in Health Care ... · Health Insurance and Risk Health...

Health-BasedRisk Assessment:Risk-Adjusted Payments and Beyond

By Kathryn E. Martin, Deborah L. Rogal, and Sharon B. ArnoldJanuary 2004

AcademyHealth is the national program office for HCFO, an initiative of The Robert Wood Johnson Foundation.

A number of payers across the country,including Medicaid managed careprograms, the Medicare+Choice program,some state employee programs, and a fewprivate purchasing cooperatives, haveadjusted their payments based on healthstatus. Researchers generally agree thatmost health-based risk-assessment tools,which are used to determine how much apayment should be adjusted, providecomparable levels of predictability, and avariety of tools are in use today. As payershave gained more experience with thesetools and the concept of health-based riskassessment, both the concept and toolshave become accepted more widely.

There are differences, however, amonglevels of implementation in variousmarkets. One persistent question is whyrisk adjusted payments are more commonin publicly financed programs than in theemployer-based insurance markets. Thisquestion is particularly important given theever-increasing fragmentation of the privateinsurance market, as employers opt for“consumer-directed” and other high-cost-sharing plans. In addition to concernsabout risk pool fragmentation, there is agrowing understanding, highlighted by theInstitute of Medicine’s 2001 report Crossingthe Quality Chasm, that quality of care mustbe improved. Risk-assessment toolsoriginally developed to modify payment andexpand access are now also being used toimprove quality.

To better understand the benefits andshortcomings of risk assessment andadjustment as vehicles for maintainingviable risk pools and guiding medicalmanagement toward high-quality careand access to insurance coverage, The

Robert Wood Johnson Foundation’sChanges in Health Care Financing andOrganization (HCFO) program and theKaiser Permanente Institute for HealthPolicy brought together risk adjustmentexperts, public and private purchasers,and representatives from health plans to discuss their experiences. Theyaddressed the roles that risk assessmentand adjustment have played in keepingrisk pools intact. Attendees also exploredthe role of risk assessment in helpinghealth plans and providers improvequality of care through a variety ofinnovative uses such as predictivemodeling, high-cost care identification,provider profiling, and identification ofpatients for disease managementprograms. They also considered thebarriers to more widespread use of risk-assessment tools for adjusting paymentsand other innovative uses.

The purpose of this report is to explain theunderlying concepts and tools critical tothose considering a variety of riskassessment applications. It summarizes theexperiences to date in three market sectors:Medicare, Medicaid, and the employer-based market. It also discusses the benefitsand shortcomings of risk assessment andadjustment; highlights the differencesamong risk-assessment tools, theirappropriateness for a particular use orpopulation, and the data required to usethem; and describes how risk-assessmenttools are being used for both payment andnon-payment applications. It is our hopethat this report will serve as a primer forthose considering adopting health-basedrisk assessment, as well as an update onpurchasers’ experiences thus far.

Foreword

Anne K. Gauthier, Program Director Deborah L. Rogal, Deputy Director

1

Health Insurance and Risk Health insurance is designed totransfer individuals’ financial risk forneeded health care services to a thirdparty, such as a health plan. Inexchange for a premium, the healthplan pays for or provides a defined setof necessary health care services, evenif the costs of those services exceedsthe premiums paid. The premium isbased in part on the anticipated cost ofthe health care needs of those insured.

Risk selection exists wheneverindividuals can select among healthinsurance products offered by morethan one health plan. Adverse selectionoccurs when a disproportionately highnumber of people at increased medicalrisk enroll in a particular health plan.At times, risk selection may be randomor it can result from differences inbenefit design or provider contractingstrategies. It also can occur whenhealth plans use strategies to attracthealthier enrollees.

If plans must spend more money on anindividual (because he or she requiresmore services or more expensiveservices) than the payment they receive,they will lose money. If that is true for alarge proportion of enrolled individuals,the adverse selection can endanger theplan’s financial viability. As a result,health plans have an incentive to enrollhealthier people. Therefore, they may bein a position to compete not only on thebasis of quality and efficiency,1 but alsoaccording to whom they enroll. Intheory, if payments to plans are adjustedto reflect the health of their enrollees,plans will not have a financial incentiveto recruit healthier enrollees.

How Risk Adjustment Works The goal of risk adjustment is to payhealth plans according to the health riskof plans’ enrollees, reducing the potentialnegative financial consequences of

insuring individuals at high risk formedical expenditures. The health risk ofenrollees can be quantified using avariety of risk-assessment tools. (Seepage 3.) In a competitive privateinsurance market, risk-adjustmentmechanisms are designed based on theassumption that the total resourcesspread among plans in a particularmarket are sufficient, but inappropriatelydistributed. If all payers and purchasersare willing to participate in a risk-adjusted system, premium income canbe reallocated to more accurately reflectthe distribution of health risk amongplans. An adequate risk-adjustmentmechanism provides incentives forhealth plans to compete solely on thebasis of quality and efficiency.

Many tools for assessing risk have beendeveloped over the last decade. They are tailored to different populations,require different data, and can be usedfor adjusting risk prospectively orconcurrently.

Risk-adjusted payments have beenimplemented in varying degrees in arange of settings in both private andpublic health insurance markets overthe last decade. For example, they havebeen adopted by Medicaid managedcare programs, the Medicare+Choiceprogram, some state employeeprograms, a small number of privatepurchasing cooperatives, and a fewlarge employers. Public payers also usehealth-based risk-assessment tools toset initial managed care rates, as wellas to move money among plansaccording to the risk pool they attract.

Experts continue to debate whetherrisk adjustment will be implementedmore widely in the private insurancemarket. Some argue that, if anincreased number of employers offerconsumer-directed health plans orother plans with more cost-sharing

Introduction to Risk Assessment and Risk Adjustment

If plans must spend moremoney on an individual(because he or she requiresmore services or moreexpensive services) than thepayment they receive, theywill lose money.

2

(which may be more attractive toyounger, healthier individuals who donot have significant—and thusexpensive—health care needs), theemployer risk pool will becomeincreasingly fragmented, thus raisingthe need for risk adjustment. On theother hand, more employers areoffering only a single health plan,which reduces the need for formal riskadjustment.

Although risk-assessment tools weredeveloped to be incorporated into riskadjustment mechanisms to modifypayment and expand access, somehealth plans are now using them toimprove quality and control costs. Keyplayers in health care markets have agrowing understanding that quality ofcare must be improved, as highlightedby the Institute of Medicine’s 2001report, Crossing the Quality Chasm.2

Innovative uses of risk-assessmenttools include managing patient care(e.g., high-cost case identification),adjusting payments from plans toproviders, profiling physicians on thebasis of quality or productivity, andpaying physicians using specificquality measures. Risk-assessmenttools also are being used for predictivemodeling to aid health plans in someof their routine business functions,such as underwriting and renewalrating.

This report examines the tools availablefor assessing risk, the data needed forthem, and the current use of risk-adjusted payments in three purchasermarket sectors: Medicare, Medicaid,and the employer-based market.

Although risk-assessmenttools were developed to

be incorporated into risk-adjustment mechanisms

to modify payment andexpand access, somehealth plans are now

using them to improvequality and control costs.

3

OverviewOnce purchasers decide to adjustpayments based on health risk, they mustmake a number of decisions, including:

◆ Whether to use a risk-assessment toolthat explains the current year’s cost orpredicts next year’s costs;

◆ Which data to use to measure anindividual’s health risk; and

◆ How to collect and manipulate thosedata to predict risk.

The mechanism by which purchasers cananalyze data is called a risk-assessment“tool.” Risk-assessment tools arestructured to measure risk, either byestimating it for the future or on acurrent-year basis. Prospective tools usedata on an individual from a previous yearto estimate that person’s future expenses.Often, these tools are used to set andadjust payment. Another option is to use

a concurrent tool, which draws on healthstatus data collected in the current year toexplain expenses in the same period.

Concurrent tools typically are used forprofiling providers, although they also canbe used for payment. Research shows thatconcurrent models are more accuratemeasures of health care expenses than areprospective models. This is primarilybecause concurrent tools capture more ofthe costs of actual utilization during ayear, while prospective tools only makepredictions of future utilization.

Concurrent models also call into question how to time payments and datacollection. Each of these approachescreates different incentives. The decisionabout the type of tool to use should bebased on the application.

There are many tools available that can beused for either payment or non-payment.

Risk-Assessment Tools

Predictive Power ability to accurately explain the variation in the expenses of a given population

Underlying Logic link to daily clinical practice and whether it is clinically meaningful to providers

Incentives the behavior encouraged among providers and health plans in the short and long term

Resistance to“Gaming”

the degree to which providers and plans cannot manipulate the tool to their benefit, including an ability to verify and/or audit the results

Data Availability accessibility of the data upon which the tool is based, including the completeness, quality, and timeliness of the data

Transparency ability of stakeholders to understand the basis and operation of the tool

Simplicity how easy it is to implement and use

Reliability how stable the risk scores are over time and with data from differenthealth plans

Cost monetary and non-monetary expense of the tool and of acquiring data

Criterion Definition

Table 1: Criteria for Assessing Risk-Assessment Tools

4

Therefore, purchasers must evaluatewhich tool is best for their purposes.Goals often differ and include controllingcosts (by managing selection issues) orcreating incentives for plans to accepthigh-risk populations in order toencourage plans to compete on the basisof quality and efficiency.

Tool selection is a complicated task. Eachtool has strengths and weaknesses. Inselecting a tool, a purchaser is making atradeoff among several criteria, which aresummarized in Table 1.

If used for non-payment purposes, suchas medical management, underwriting,or provider profiling, a tool also may be evaluated on the basis of its ability to predict risk for particularsubpopulations and its overallappropriateness for a given use.

Most often, the quality of a risk-assessment tool is measured according toits ability to accurately explain variation inthe expenditures of a given individual orpopulation. The predictive power isquantified as a statistical measure knownas the R2, which is the proportion of totalvariation that can be explained by theindependent variables. The closer the R2

value is to 1, the better the predictivepower of a tool. For example, a tool withan R2 value of 0.05 explains only 5 percentof the variation in future health care costsfor a given population, whereas a tool withan R2 value of 0.25 explains 25 percent ofthe variation.

Although there are debates about howuseful the R2 statistic is in selecting a tool(largely because it is easily influenced bypeople with very low or very high healthrisk and there is difficulty estimating high-cost enrollees), it is the most popular andmost easily understood measure thatallows purchasers to compare tools’predictive power. At a minimum, R2 is auseful starting place because it is thequickest way to compare predictive abilityacross tools. Other measures gauge a tool’s

effectiveness at the population level—andcan assess groups of different sizes andrisk composition—which may be morerelevant than the R2 in some cases.

Another way to determine the adequacyof a tool is by using something called apredictive ratio. This ratio compares thetotal predicted costs with the actual costsamong people for whom cost estimatesare likely to be biased due to some sharedcharacteristic, such as Medicaid status orthe presence of diabetes. Predictive ratiosrange from less than 1 to greater than 1.Numbers less than 1 show that the modelis under-predicting the costs of the group,while numbers greater than 1 show thatthe model is over-predicting costs.Models with more diagnoses tend to dobetter on predictive ratios, even whenpredicting the costs of groups defined bya disease not in the model, because manydiseases may be correlated withdiagnoses that are included in the model.

The risk-assessment tools available topurchasers can be divided into threecategories based on the data they use:demographic information; self-reportedhealth status; and administrative data.

A fourth category—data mining—can use any of the data in the first three categories. Because it is relativelynew, data mining is discussed in more detail than other techniques. Each of the categories vary in how they rank according to the criteriadescribed in Table 1.

Demographic InformationTraditionally, payments have been riskadjusted based on demographic factors,such as age, sex, family status, andlocation of residence. Mechanisms thatpredict costs based on these demographicdata are the baseline against which othermechanisms are measured. Studies haveshown that risk-assessment tools usingdemographic information alone accountfor no more than 2 percent of health

Most often, the quality of a risk-assessment tool

is measured according toits ability to accuratelyexplain variation in theexpenditures of a given

individual or population.

5

care costs.3,4 Despite their relatively lowpredictive power, demographic-basedadjustments are easy and relativelyinexpensive to administer, and they donot produce incentives for providers tochange treatment or coding tomaximize risk scores.5 In addition, theyare transparent, reliable, and simple toverify, audit, and understand.

Self-Reported Health StatusSome purchasers have consideredadjusting payments based on the healthstatus of an individual or population asmeasured through surveys of enrollees.This method of risk assessment has notbeen adopted widely. At one time, theRAND Self-Reported Health StatusSurvey was examined for its potentialin forecasting future health care costs.The 36-item self-administered surveyinstrument was designed to capture thefull range of health characteristics.6

Self-reported health status has betterpredictive power than demographicinformation alone, but only marginally.It explains between 3 and 5 percent ofthe variation in future costs.7

Other advantages of self-reportedhealth status are that it has a solidunderlying logic, it is transparent andsimple to understand, and it does notcreate incentives for inefficiency. Onthe other hand, this method ofassessment is associated withsignificant costs. Currently, mostpurchasers do not gather self-reportedhealth information, so initial datacollection costs would likely beconsiderable. Broadly surveyingenrollees is also expensive, as well astime consuming.

Finally, self-reported health status issubject to “gaming,” with incentives forproviders to report higher levels ofhealth care need.

Administrative DataA variety of administrative data sourcescan be used to estimate medical risk,including prior expenditures, diagnosisdata (e.g., encounter data, inpatient oroutpatient claims data), or prescriptiondrug data. Research indicates thatmodels using any of these data sourceshave much higher predictive powerthan demographic data alone.

Estimating future costs based on priorutilization and expenditures has morepredictive power than many of the otheravailable tools. Some evidence suggeststhat prior expenditures can account for between 25 and 45 percent of thevariation in future health care costs.8

Adjusting payments based on priorexpenditures is easy to implement andmanage, and has a sound logic.Moreover, this approach is transparentand simple to comprehend and use.

Using prediction tools with priorexpenditure data has some drawbacks,however. For example, it creates someundesirable incentives for providers andhealth plans. If plans will be paid morebecause their enrollees have used moreservices, then they have an incentive toprovide more services instead of supplyingefficient care. In addition, risk-adjustmentsystems that rely on prior expenditurescan be more difficult to implement thanthose that use diagnosis data.

Because of the advantages of using priorexpenditure data, purchasers mightconsider beginning their health-basedpayment program with prior utilizationand expenditure data and eventuallytransition to another model to avoid thelonger-term negative incentives.

Predicting future health risk and costsfrom diagnoses has been adoptedwidely because these models alignpayments with risk while encouragingefficient provision of care.9 Expertsagree that diagnosis-based methods

Adjusting payments based onprior expenditures is easy toimplement and manage, and hasa sound logic. Moreover, thisapproach is transparent andsimple to comprehend and use.

6

have better predictive power thandemographic data (estimates rangebetween 10 and 14 percent forcommercial populations,10 and 19 to 22percent for a Medicaid-specific tool11).The underlying logic of these models isstrong because diagnoses are highlycorrelated with current and futureexpenditures. Diagnosis-based modelsare not as heavily influenced byutilization levels as are models that useprior expenditure data. With diagnosisdata, purchasers can make informedchoices about their medicalmanagement strategies.

Although information on diagnoses arecollected and reported routinely, it is oftendifficult to acquire complete andappropriate data. Purchasers rely onhealth plans and providers to submitdiagnosis information on which they canbase their adjustments. If the submitteddata are not complete, are of questionablequality, or if physicians do not codeaccurately, adjusting payments based onthat information could do more harmthan good. In some instances, purchasersmay not have been collecting diagnosisdata from plans, and therefore do nothave any information on which to basetheir adjustment. In that case, dataacquisition is not only costly but timeconsuming.12 The data challengesassociated with diagnosis-based riskadjustment are discussed in more detailin the following section.

Compared with using an approach thatdraws on demographic information,diagnosis-based models are not astransparent to all stakeholders. Theadjusted payments are sometimesdifficult to understand for those withlittle or no clinical background.Moreover, the consensus-buildingprocess that is required to educatestakeholders on the complexity of thesemodels is often long and contentious.

When these models were developed, theywere expensive to purchase because of

extensive development, clinical input,and programming costs. For the mostpart, however, the prices of software havedecreased significantly, so the acquisitioncost is no longer a barrier.

Pharmacy-Based ModelsRecently, questions have been raisedabout whether the addition ofprescription drug data is preferable todiagnosis-based adjustments. Overall,prescription-drug models have about thesame predictive ability as diagnosismodels. In one study of the predictivepower of risk adjusters, the researchersfound that pharmacy-based modelsperformed at a level similar to diagnosis-based models when the assessment wasapplied prospectively, but, when themodels were applied concurrently, thediagnostic models outperformed theprescription drug tools.13

Pharmacy data also are similar tomedical claims-based diagnostic data in that they can provide medicalmanagement information similar to thatprovided by diagnoses. In addition,prescription drugs are predictive offuture expenditures withoutencouraging greater utilization of otherservices. However, with a pharmacy-based model, plans or providers mayincrease the number of prescriptionsthey dispense to raise their risk scoresin order to demonstrate higher riskenrollees or patients. This does notpromote the efficient provision of care.

Pharmacy-based models have thefollowing advantages over diagnosis-based tools:

◆ Purchasers often have morecomplete prescription drug dataavailable in an electronic format,making it significantly easier toaccess;

◆ Prescription drug data do not need to be obtained from physicians, thus eliminating a difficult data-collection step;

Compared with using anapproach that draws on

demographic information,diagnosis-based models are

not as transparent to allstakeholders. The adjusted

payments are sometimesdifficult to understand

for those with little or noclinical background.

7

◆ People with chronic conditions may nothave encounters with the health caresystem during the data-collectionwindow that would generate diagnoses,but they generally refill prescriptionsregularly;

◆ Prescription drug data are recordedmore similarly across plans than isinformation about diagnosis, leading toless variation in reporting; and

◆ Pharmacy data are available morequickly than diagnosis data andgenerally are less expensive to collectand validate.14

Although pharmacy data are available morequickly than diagnosis data, the rapidlychanging pharmaceutical industry requiresthat pharmacy-based models be updatedmore often than diagnosis-based models.15

The primary weakness of pharmacy-basedmodels is that they are slightly less precisethan diagnosis-based models, in part dueto the vague relationship betweenconditions and pharmaceuticals.

Prescribing a drug to someone is a stepremoved from diagnosing him or her with a specific condition. For example, some conditions do not have a singlerecommended pharmacological treatment.Thus, pharmacy data may not be helpful.Alternatively, providers can treat the samecondition with different medications, andsome drugs are used to treat multipleconditions. Models may not link thepharmacy data back to the appropriatecondition or may assume an inaccurateunderlying health status.

Data MiningData mining is the science of findinginteresting and novel patterns in largeblocks of data. Data-mining techniques usesophisticated computer programs to siftthrough health plans’ data to identifypatterns that can predict the future healthcare expenditures and use of a givenpopulation. Data-mining mechanismspredict future costs based on patterns of

information, whether complete orincomplete. They may be better able thantraditional regression methods to modelnon-linear dynamics that occur within apopulation. Health plans can use thesetools to better understand the effects ofsubtle changes in health status that mayhave an impact on future costs. The coregoal of any of these is to model therelationship between data measuring acertain health indicator in a population andthe group’s usage of services and costs.

Currently, some are experimenting withneural networks, one type of pattern-recognition tool for assessing health risk.Unlike traditional regression models,neural networks and other pattern-recognition tools search through data in anon-linear way to recognize linkagesamong different combinations of data.

Data-mining methodologies do not have amedical or causal theory supporting theiruse. In other words, they are not designedto look for specific diagnoses, conditions,prescriptions, or other indicators of futurehigh risk. Rather, they seek out patterns,which may not be predictable and could betemporary or unique to a specific context.One of data-mining tools’ greatestcontributions is the possibility of improvedpredictive accuracy through “featureextraction.” These methodologies have theability to convert raw data into a set ofattributes or features that can describe theprevious year’s trends and occurrences.

Data-mining techniques are applied widelyin other industries. For example, advertisersmay use past purchasing behavior todetermine which mail-order catalogs anindividual should receive. However, thesemethodologies have not been adopted asbroadly in health care. Experts believe it isdifficult to apply data-mining results inhealth care because the tools do notnecessarily have a theoretical basis specificto health claims.

The use of data-mining mechanisms to assess risk raises policy questions about

8

what health plans and purchasersshould do with the results that thesetools provide. It seems unlikely thatpurchasers would want to adjustpayments based on these models.Without an adequate causalexplanation for the findings generatedthrough data mining, purchaserscannot fully understand the medical or policy implications that would be associated with adjusting riskaccording to their results.

As decision makers become more familiarwith the capabilities of data-miningmethodologies, they may become morecomfortable using information generatedin this way. For example, data miningcould be helpful for purposes other thanpayment, such as identifying high-costcases within a health plan.

Each type of risk assessment tool hasadvantages and drawbacks. In general,diagnosis-based, traditional-regressionmodels have been adopted more widelythan the other categories of tools foradjusting payments. Their ability toprovide balance among predictiveaccuracy, desired incentives, andadministrative burden with whichpurchasers can be satisfied contribute totheir popularity.

Applying Tools Beyond PaymentOnce plans and purchasers collectinformation for risk-assessment, theycan use it for purposes other thanadjusting payments. Risk-assessmenttools may be useful in improving healthcare efficiency, for example. They canalso help health plans to underwritemore effectively, identify candidates fordisease management, adjust paymentsfrom plans to providers, profilephysicians on the basis of quality or

productivity, and pay providers usingspecific efficiency measures. Mostoften, risk-assessment tools serve as afirst level of information on whichhealth plans can act, rather than as afinal decision tool.

Randall K. Spoeri, Ph.D., vice presidentof medical and quality informatics at HIPHealth Plan of New York, says that heand his colleagues use their physicianprofiling program to “engage providers ina substantive and useful dialogue, whichwe hope ultimately will improve thequality of care we provide.” Although anincreasing number of plans are usingrisk assessment in creative ways, littleresearch has been done to evaluate theimpact of these innovations, althoughthere has been some assessment of theprovider profiling methodologies using avariety of risk-assessment tools.

J. William Thomas, Ph.D., andcolleagues at the University ofMichigan evaluated the accuracy ofseven primary care provider profilingmethodologies, including theimplications of the differences inaccuracy for assessment of physicianperformance. They found that onaverage, profiling systems agree on 54percent to 58 percent of physiciansidentified as least efficient and 42percent to 53 percent of those identifiedas most efficient. They also found thatthe profiling systems vary significantlyin terms of the information they gather,and how they classify patients.16

Thomas is now examining whetherrisk-adjustment methodologies used togenerate reliable profiles in a primarycare setting can be extended tospecialists, given the unique factors thatarise in profiling specialty physicians.

9

Data for Risk Assessment

One of the most significant barriers toimplementing a diagnosis-basedtraditional regression health-based risk-assessment tool is the collection of data.This is in part because of the effortrequired to submit large amounts ofdata and in part due to the sensitivity ofdiagnostic models to the quality of dataused. Health plans often fear that theirdata are wrong or incomplete, eitherbecause they capitate their providers and do not require them to submitinformation, or because they truncate the data from their providers. Thus, the data may indicate that they havehealthier enrollees and cause theirrevenues to decline. Unless health plans have had reason to edit their data,they have no assurance that theinformation is accurate.

In order to make health-based payments,plans must be willing to make anongoing commitment to submitting data about their enrollees’ diagnoses. For some plans and purchasers,submission of diagnoses will require littleadditional effort. Others will have tomake a significant investment to developthe necessary infrastructure to meet datarequirements.17 Purchasers can work with plans to increase their comfort withthe selected tool and the data used inadjustments in a number of ways— bycreating risk corridors, adjusting only aportion of the payment until plans gainconfidence in the quality of their data, orby offering trial runs on data where thereare no payment implications. Theseintermediate steps can go a long waytoward increasing a plan’s commitmentto submitting complete, high-quality dataand to health-based payments overall.

As they work toward successful datasubmission, purchasers and healthplans should realize that there is nosuch thing as perfect data for health-based risk adjustment. All stakeholdersneed to work with each other toproduce an acceptable data set.

Purchasers often push plans to submita multitude of data elements, but plansmay only be able to supply limitedinformation. One way that plans andpurchasers can compromise is toreduce the number of required fields toonly those that are absolutely necessaryin order to conduct the assessment andadjustment.18 According to a studyevaluating the implementation of

HIPAA and Data Collection for Risk AssessmentIn an era in which the privacy rule within the Health Insurance Portability andAccountability Act (HIPAA) has raised consciousness about dataconfidentiality, any data collection effort needs to comply with the standardslaid out in that legislation. Although many assume that HIPAA hinders orprohibits data collection for risk adjustment purposes, the “business-associate” provisions of the rule may still permit it. By establishing businessassociate contracts between the health plan and the purchaser and the healthplan and any consultants involved in the process, the health plan can shareprotected health information with the other parties.1

The privacy rule allows providers and health plans covered by HIPAA to discloseprotected health information to people and businesses that help them carry outtheir health care activities and functions. Under HIPAA, business associates arethose entities that undertake specific functions, activities, and services.

Although risk assessment and adjustment appear to fall into these categories,some health plans still may be concerned about submitting their data to anexternal entity for these purposes. To ameliorate these concerns, benefitconsultants that specialize in health-based risk assessment have used minimumdata sets that encrypt member identifiers, limited geographic information,created age bands (instead of including an individual’s actual age), andeliminated social security numbers from records.

Permissible functions and activities include: ◆ Claims processing or administration;◆ Data analysis, processing,

or administration;◆ Utilization review;◆ Quality assurance;◆ Billing;◆ Benefit management;◆ Practice management; and,◆ Repricing.

Allowable services include:◆ Legal; ◆ Actuarial; ◆ Accounting;◆ Consulting;◆ Data aggregation;◆ Management;◆ Administrative;◆ Accreditation; ◆ Financial;◆ Actuarial services;◆ Accounting services; and,◆ Consulting.2

1 U.S. Department of Health and Human Services Office of Civil Rights Guidance Explaining Significant Aspects of the Privacy Rule, December 4, 2002, Business Associates section. See www.hhs.gov/ocr/hipaa/guidelines/businessassociates.pdf.

2 ibid.

10

health-based risk adjustment conductedat the Park Nicollet Institute, datacollection and processing obstacles canbe overcome with commitment andresources from purchasers andparticipating health plans.19

Fortunately, the experiences ofpurchasers who have adopted health-based risk adjustment suggest thatcollecting the necessary encounter data is feasible.20 For example,Medicaid programs have been able to acquire robust data through acommitment to health-based riskadjustment and an investment in adata-collection infrastructure. Theimplementation of health-basedpayment in Medicare+Choice also has helped move public programs along way in terms of data collection.

The request for data issued by theCenters for Medicare and MedicaidServices (CMS) in March 2002 causedplans to refine their data-collectioninfrastructure to comply with the new system. To gain the support ofhealth plans, CMS reduced thenumber of ICD-9 codes used for riskadjustment. Although some plans had to make a significant financialinvestment to get to that point, theynow are well-positioned to use thatinformation in other ways and,potentially, for other purchasers.

David Knutson, director of healthsystems studies at the Park NicolletInstitute, and his colleagues found that

plans may take up to three years todevelop and implement a data-collection system that providescomplete information. In addition, inthe rare cases when purchasers haveinvested in providing technicalassistance to plans to improve theirdata-collection efforts, purchasers havebeen successful in acquiring completeencounter data.21 The health-based riskassessments that have beenimplemented thus far were madepossible through data sets that are notperfect, but good enough. As health-based risk adjustment becomes morecommon and data requirements aremade clear, it will become easier toidentify good data.

Another goal of data collection forhealth-based payments is for healthplans to generate similar enough data to compare and measure healthstatus across plans.22 Increaseduniformity would improve the precision of risk adjusters. With moreconsistent data, purchasers can be moreconfident that they are adjusting forlegitimate health status differences,rather than for differences in the qualityof the data that plans can provide. Somehave argued that standards should beestablished about how to collectinformation and what to do with data ofpoor quality. It would be much easier forpurchasers to adopt health-based riskadjustment with standardizationbecause it would be easier for plans toprovide the necessary data.

Some have argued thatstandards should be

established about howto collect information

and what to do withdata of poor quality.

11

State Medicaid programs have been thelead purchasers to implement health-based risk adjustment, preceding effortsby both employers and Medicare. Thislargely is because health plans servingMedicaid enrollees are more likely toenroll higher risk enrollees. Withouthealth-based payments, the financialviability of the health plans that serve thechronically ill and disabled could be injeopardy.23 By paying plans according tothe health risk of their enrollees, statescan allocate expenditures, withoutincreasing them, in a way that reflects theplans’ costs of providing care tobeneficiaries. This, in turn, can increase

plans’ willingness to serve the Medicaidpopulation.24,25 More than in othermarkets, there is a large health-based risk-assessment infrastructure in place amongMedicaid programs.

Medicaid programs, including those in Colorado, Maryland, and Oregon, began implementing health-based riskadjustment in the late 1990s. Delaware,Michigan, New Jersey, Minnesota,Tennessee, and Utah also have incorporatedhealth-based risk assessment into theirMedicaid programs in some way, and morethan 20 other states have considered using

MedicaidProgram

ImplementationYear

Tool Prospective/Concurrent

Populations Covered ParticipatingPlans

CoveredLives

% of paymentbased on HBRA

Individual/Plan*

Utah 2002 CDPS Concurrent SSI/Disabled, TANF 4 100,000 100 Plan

Tennessee 2002 CDPS Concurrent SSI/Disabled, TANF,and Medicaid/expansion populationsexcept dual eligibles

8 1.3 milliona 100 Plan

New Jersey 1999 CDPS Prospective Aged, blind, disabledwithout Medicare,including SSI

5 42,600 100 Individual

Minnesota 2000 ACGs Concurrent TANF and medicallyneedy

7 173,000b 30% (2002),50% (2003)

Plan

Oregon 1998 CDPS Concurrent Non-Medicaredisabled, generalassistance, OregonHealth Plan forAdults and Couples

15 100,000 15 Plan

Maryland 1997 ACGs Prospective SSI/Disabled, TANF,state expansionprogram

6 410,000 100 Individual

Colorado 1998 CDPS Prospective SSI/Disabled, TANF, and a fewother specializedpopulations

5 111,000 100 Plan

Table 2: Health-Based Risk Adjustment as of 2002 – 2003.

* “Individual” indicates that the payment adjustment was based on risk assessment of individuals within a plan. “Plan” indicates that the payment was adjusted based on the health risk of a health plan as a whole.

a This number was projected to decrease in 2003 because of stricter eligibility criteria, which would lead to lower enrollment.b This number was expected to increase in 2003.

Risk Adjustment in Medicaid

12

risk adjustment or are interested in learningmore.26 In addition, some larger states,including Texas, Pennsylvania, California, and New York, are considering using riskadjustment or are close to implementing it.Table 2 illustrates how some states haveimplemented health based risk adjustment as of 2002 – 2003. According to Knutson,“There is evidence that health-based riskadjustment will continue to expand forhealth plan payment in state Medicaidmanaged care programs and probably forprovider payments in primary care casemanagement programs.”

Medicaid-Specific ToolsBecause the Medicaid population is differentfrom commercial populations in terms of theirhealth status (people in Medicaid generally aresicker and have more chronic conditions),researchers have developed tools specificallyfor assessing the health risk of Medicaidenrollees. All state Medicaid programs thathave adopted health-based risk adjustmenthave used some version of these tools.

Richard Kronick, Ph.D., and colleagues at the University of California, San Diego,have developed the Chronic Illness andDisability Payment System, which builds onthe Disability Payment System. These toolsgroup diagnoses according to chronic anddisabling diseases. The same researchersalso have been developing a pharmacy-basedtool called Medicaid Rx, which may be moreappropriate for the Medicaid populationthan other pharmacy-based tools.

At Johns Hopkins University, JonathanWeiner, Dr.P.H., and colleagues have released a new version of theirAmbulatory Care Group (ACG) tool that istargeted to Medicaid programs. ACG-Medicaid is a streamlined version of earlierACG models; It is designed specifically forrate setting and capitation in Medicaid.27

ACGs use ambulatory and inpatient claimsto predict expenditures.28 Both of these toolsare available publicly at no cost to stateMedicaid agencies.

Lessons States Have LearnedAccording to a survey conducted by the University of Maryland, BaltimoreCounty’s (UMBC) Center for HealthProgram Development and Management,states that have implemented health-basedpayments believe that doing so createdincentives for plans to focus on providinghigh-quality care rather than on selectionstrategies. At a forum sponsored byUMBC, states noted that this has improvedthe overall efficiency of the programs.29

Separately, an evaluation of Maryland’sMedicaid Managed Care program completedin 2002 showed that, overall, health-basedrisk adjustments have become part of thefabric of the state’s Medicaid program. Riskadjustment helped to ensure that the dollarsfollow patients, and discussions betweenplans and the state about payment issuesfocus heavily on the level of payment, nothow those funds are allocated.30

Some states found it useful to beginadjusting payments by using priorexpenditures, especially in the first year ortwo, before undertaking the longer-termapproach of adjusting payment by diagnoses.Adjusting by diagnoses has the advantage of encouraging plans to undertake activitiesthat will attract people with varying levels ofchronic illness and disability.31

Another crucial component of successfulimplementation of health-based riskadjustment in Medicaid is consensusbuilding among stakeholders about boththe goals of health-based payments andthe mechanics of achieving the new system. Stakeholders who should be involved include health plans,beneficiaries and their advocates,providers, state legislators, and other stateofficials.32 States have found it useful tofocus on well-defined diagnoses to avoidconflicts between payers and plans aboutreporting requirements.33

Some states found it useful tobegin adjusting payments

by using prior expenditures,especially in the first year

or two, before undertaking thelonger-term approach of

adjusting payment by diagnoses.

13

In the 1990s, there was some evidencethat Medicare+Choice (M+C) healthplans, which were paid an averagepayment rate, were enrolling a healthiermix of enrollees than was fee-for-serviceMedicare. To address this, CMS (whichwas called the Health Care FinancingAdministration at that time) began toexplore options for a system that wouldprovide more accurate payments.34

The Balanced Budget Act of 1997 (BBA)expanded managed care options forMedicare beneficiaries under the M+Cprogram. It mandated the implementationof health-based risk adjustment by January2000. In addition to creating an expandedset of options for beneficiaries, there wasthe possibility that the BBA legislationcould introduce biased selection by type ofplan (e.g., a higher proportion of sickindividuals could have chosen the PPOoption). Risk adjustment ensured moreaccurate payments for all types of plans.

In selecting the risk adjustmentmethodology, CMS was primarilyconcerned with ensuring reliability andfinding a tool that would be reasonablyinexpensive to implement—both forCMS and plans. The agency also soughtto find an approach that had simplicity inits underlying logic, was resistant togaming, and had good predictive power.

CMS chose to implement a prospective,health-based risk adjuster in two phases.In the first phase, which began in January2000, diagnoses from inpatient stays wereused to adjust payments using thePrinciple Inpatient Diagnostic Cost-Group(PIP-DCG) Risk Adjustment model,developed by Ash and colleagues.35 Morerecently, diagnoses from ambulatorysettings have been incorporated into thepayment model using a version of theHierarchical Condition Category (HCC)risk adjustment model.36 Implementationof a risk-adjusted payment using healthdata from ambulatory settings isscheduled to begin in January 2004. It is

being phased in over a four-year period: 30percent in 2004; 50 percent in 2005; 75percent in 2006; and 100 percent in 2007.

CMS uses risk adjusters that incorporatedemographic status and diagnosticinformation into a risk-adjustment factor,which is then multiplied by a basepayment amount. A risk-adjustmentfactor is calculated on a preliminary basisfor each beneficiary at the beginning ofthe year based on data from the prioryear; the factor is updated after the year isover to incorporate any additionaldiagnoses from the prior year that havebeen received. The factors are based ondiagnoses identified in the prior year only,and are not updated during the paymentyear for diagnoses that are identifiedduring the payment year.

In theory, the models are additive. In other words, additional payment isassociated with each added diagnosis.However, some diagnoses are combined ina hierarchy, and some diagnoses within ahierarchy may not change payment.

Challenges to Implementing Risk-Adjusted PaymentAmong the biggest challenges toimplementing health-based riskadjustment in Medicare have been thecost and administrative difficultiesassociated with data collection. Prior tothe implementation of risk-adjustedpayments, health plans were notrequired to report diagnosticinformation to CMS. Many plans didnot require such information from theirproviders, and providers had difficultyrecording, collecting, and submittingsuch data. Plans had to modifycontracts with providers and developthe capacity to collect this data and passit along to CMS on a timely basis. For amore detailed discussion about datachallenges, see page 9.

Risk Adjustment in Medicare

In selecting the risk-adjustmentmethodology, CMS wasprimarily concerned withensuring reliability and findinga tool that would be reasonablyinexpensive to implement—both for CMS and plans.

14

Early instructions from CMS called for thecollection of all encounter-level data byplans from inpatient and ambulatorysettings. After health plans expressedconcern about the volume of encountersand the level of burden being imposed onthem, CMS revised its requirements. Plansnow are required to provide a simplifiedformat, which includes the date, type of provider, beneficiary identifier, anddiagnosis code (only for diagnoses includedin the payment model) for each enrolledbeneficiary. Diagnoses must bedocumented in a medical record.

Some stakeholders have raised concernsthat the diagnostic data collected by planswould not be entirely accurate and, to theextent that important diagnoses aremissing, would result in reduced paymentsto plans. Although Medicare physicianpayment policies under fee-for-servicerequired that diagnostic information beincluded on the claim, such data often werenot used in payment and so may not havebeen accurate. To address this issue, CMShas developed educational materials thatplans may give to their provider networks toincrease the accuracy of diagnostic coding.

Currently, CMS uses diagnostic datacollected from health plans for paymentpurposes only. With the plannedcollection of encounter data, CMS hadconsidered using the data forrecalibrating the risk-adjustment modeland for quality-assessment purposes.The current data, however, do notsupport those efforts.

Lessons LearnedAn important lesson from the Medicareexperience is that it is possible to initiatelarge-scale data collection and use it toadjust payment, although it may bedifficult for smaller players with lessmarket share to achieve this. By phasing inthe types of data collected and the impactof the payment change, Medicare decisionmakers helped overcome plans’ fears thatinadequate payments would be made onthe basis of incomplete data. In addition,

the extensive education and technicalassistance provided by CMS helped allplans, especially smaller ones, to meet therequirements.

In establishing the data-submissionrequirements, CMS had to carefullyweigh its need for data and the potentialburden that collecting such informationwould impose on plans. Complicatingthe process was the extensive variationboth by plan and within some plans inhow they contracted with their providersand what data they were collecting.Changing systems to comply with newdata requirements was costly for manyplans and took time to implement.

CMS needed to minimize the potential foradjustments based on the data collectionmethodology, rather than on the underlyinghealth status of beneficiaries. In the end,CMS chose to allow flexibility in datacollection. Plans can submit encounter dataif they were already collecting it, but canalso choose to submit a minimum set ofdiagnostic data instead.

One question for decision makers toconsider is the extent to which otherpayers could capitalize on the data-collection infrastructure created forMedicare. If other payers choose a risk-adjustment model using inpatient andambulatory diagnoses, providers andpayers may be able to take advantage ofsignificant economies of scale byproviding the same types of data as arecurrently required for Medicare.

The choices made during the developmentof a risk-adjustment model for Medicarewere constrained, however, by thestructure of the Medicare benefit. Forexample, it was not possible to design arisk adjuster that draws on drug databecause Medicare to date has not coveredoutpatient pharmaceuticals. Other payers,especially those that have a comprehensivedrug benefit, may make different choicesabout a risk-adjustment model and thedata required to support it.

Some stakeholders haveraised concerns that the

diagnostic data collected byplans would not be entirelyaccurate and, to the extent

that important diagnoses aremissing, would result in

reduced payments to plans.

15

Employers37 face many of the same issuesof adverse selection as Medicaid andMedicare programs when they offerproducts from multiple health plans. Theyhave been far slower to adopt formal riskadjustment, however. By 1998, only 774,919(approximately 1.3 percent) of the60,599,934 enrollees with employer-basedhealth insurance coverage were in plansthat use formal demographic or health-based risk adjustment.38 Recent figuresindicate that between 100 and 150employers use risk assessment as a tool for better purchasing, particularly for premium negotiation. About 350employers use some form of demographicrisk assessment, and 50 use prescriptiondrug data to assess risk. Fewer than fivelarge employers are using risk assessmentto drive efficiency through payment.39

Some posit that, as state Medicaidprograms and Medicare adopt health-basedpayment methodologies, the employer-based insurance market will follow.40

Economists also observe that the differentconstraints and markets that private andpublic payers face, and the different goalsthey set, might lead to variation in adoptionand tool selection. For most employers, theprimary goal of formal risk adjustment isto maintain freedom of choice of healthplans for their employees.41 In some cases,if health plans perceive that they will attractthe higher-risk employees, they have notoffered coverage to employers offeringmultiple plans.

Glazer and McGuire,42,43 Frank andRosenthal,44 and Ellis45 suggest thatemployers have mechanisms other thanformal risk adjustment to address adverseselection, which is why it has not beenadopted as widely in this market. Typically,these other mechanisms are not availableto Medicare or Medicaid programs,resulting in their increased reliance onformal risk adjustment. Specifically, Glazerand McGuire note that open enrollmentprovisions, premium negotiations, andrestrictions on employees’ choices ofhealth plans are mechanisms “superior toformal risk adjustment for dealing withproblems caused by adverse selection.”

Frank and Rosenthal note that threefactors reduce the value of riskadjustment in the employer market fromthe perspective of the plan:

◆ Only a small segment of privatelyinsured individuals are offered achoice of competing health plans;

◆ Health plans share risk with payers,providers, and reinsurers; and

◆ De facto experience rating occursduring the premium negotiationprocess and management of coverageseems to substitute for formal riskadjustment.

Those who work closely with employershave seen a recent surge in interest informal risk adjustment, supporting Ellis’hypothesis46 that changes in the healthcare market may change demand andaccelerate future adoption.

Hewitt Associates notes that the currentdilemma for employers is how to controlcosts while offering employees morechoices in health care. In advisingemployers, they posit that risk adjustmentmay offer the “right combination ofemployee-driven choice and fully insuredcost control.”47 Over the next three years,they hope to have risk adjustmentimplemented for 100 percent of the livescovered by their clients’ plans.

Overall, they have found that employersare willing to invest in more complexdata sets, as long as the benefits ofcollecting and using the data outweighthe costs. Further education ofemployers can help them understand:

◆ what risk adjustment tools areavailable and the feasibility ofimplementing them;

◆ how payments will be affected; and

◆ how formal risk adjustment mightstabilize their risk pools, whichbenefits the market in the long-term,even if individual employers are“winners” and “losers” over time.48

Risk Adjustment in the Employer Market

Some posit that, as stateMedicaid programs andMedicare adopt health-basedpayment methodologies, the employer-based insurancemarket will follow.

16

The results from recent surveysconducted by researchers at the ParkNicollet Institute indicate that riskadjustment is a feasible tool and animportant component of the solutionto risk segmentation in the employermarket. A survey of purchasers whohad implemented health-based riskadjustment for payment found that

most purchasers agree that it hascontributed to fairer payment, and thatit has been worth the effort. Healthplans were also surveyed, and themajority agreed that risk-adjustedpayments did not require the collectionof unnecessary and costly data andreduced concerns about the negativefinancial impact of adverse selection.49

Efforts to implement risk-adjustedpayments in public programs have beensuccessful, in part because of substantialwork by all stakeholders to come toagreement on the goals of riskadjustment and the mechanics of thesystems. Many risk-adjustment systemshave been implemented gradually,starting with adjustment based on priorexpenditures or with a limited set ofdata. This has helped by reducing thefinancial impact of risk adjustment inthe early stages and allowing all partiesto get used to new data-collectionsystems over time.

There are many different risk-assessmenttools and risk-adjustment mechanismscurrently in use. Prior to choosing anapproach to assessing risk or adjustingpayments, participants need to take intoaccount the factors specific to their

environment, including the benefitpackage, data capabilities of providers, andcharacteristics of the population. It isunlikely that one common methodologywill emerge in light of the various healthcare environments and infrastructures inplace, and the high costs that would beassociated with making such a transition.

Further, it is not clear that onemethodology would be appropriate for allsettings and all uses, particularly for non-payment uses. Given the many parties thatare using diagnosis-based risk-assessmentmodels, however, in the future decisionmakers may be able to take advantage ofeconomies of scale by using data alreadybeing collected for other programs, such asMedicaid and Medicare.

Conclusion

17

1 Beeuwkes Buntin, M.J. and J.P. Newhouse.“Employer Purchasing Coalitions and MedicaidExperiences with Risk Adjustment.”Commonwealth Fund Issue Brief, May 1998.

2 Institute of Medicine of the NationalAcademies, Crossing the Quality Chasm: A NewHealth System for the 21st Century, March 1, 2001.

3 Kronick, R. and T. Dreyfus. “The Challenge ofRisk Adjustment for People with Disabilities:Health-Based Payment for MedicaidPrograms.” Center for Health Care StrategiesWorking Paper, November 1997.

4 Rice, C. and P. Nakahata. “Pharmacy-BasedRisk Adjustment: A Guidebook for States.”Center for Health Care Strategies WorkingPaper, September 2001.

5 Cumming, R.B. et al. A Comparative Analysis ofClaims-based Methods of Health Risk Assessmentfor Commercial Populations. Society of ActuariesMonograph, May 2002.

6 Lee, C. and D. Rogal. Risk Adjustment: A Key toChanging Incentives in the Health InsuranceMarket. Changes in Health Care Financing andOrganization Special Report, AcademyHealth,March 1997.

7 Kronick, R. and T. Drefus. “The Challenge ofRisk Adjustment for People with Disabilities:Health-Based Payment for Medicaid Programs.”Center for Health Care Strategies WorkingPaper, November 1997.

8 ibid.

9 ibid.


11 Kronick, R. and T. Drefus. “The Challenge ofRisk Adjustment for People with Disabilities:Health-Based Payment for Medicaid Programs.”Center for Health Care Strategies WorkingPaper, November 1997.

12 Gilmer, T. et al. “The Medicaid Rx Model:Pharmacy-Based Risk Adjustment for PublicPrograms,” Medical Care, Vol. 39, No. 11, pp.1188–1202.


14 ibid.

15 Gilmer, T. et al. “The Medicaid Rx Model:Pharmacy-Based Risk Adjustment for PublicPrograms,” Medical Care, Vol. 39, No. 11, pp.1188–1202.

16 “Major Differences Exist in Physician ProfilingSystems,” HCFO News & Progress, March 2002,pp. 4–5.

17 Kronick, R. et al. “Improving Health-BasedPayment for Medicaid Beneficiaries: CDPS,”Health Care Financing Review, Vol. 21, No. 3,Spring 2000, pp. 29–64.

18 Halpern, R. et al. “Implementing Health-basedRisk Adjustment: Lessons from the Field.” ParkNicollet Institute Working Paper.

19 ibid.


21 Halpern, R. et al. “Implementing Health-basedRisk Adjustment: Lessons from the Field.” ParkNicollet Institute Working Paper.


23 Rice, C. and P. Nakahata. “Pharmacy-BasedRisk Adjustment: A Guidebook for States.”Center for Health Care Strategies WorkingPaper, September 2001.

24 Kronick, R. and T. Drefus. “The Challenge ofRisk Adjustment for People with Disabilities:Health-Based Payment for MedicaidPrograms.” Center for Health Care StrategiesWorking Paper, November 1997.

25 Madden, C. et al. Risk Distribution and RiskAssessment Among Enrollees in WashingtonState’s Medicaid SSI Population. Final Report tothe Center for Health Care Strategies, July1998.

26 Knutson, D. “Health-Based Risk Adjustment:Payment and Other Innovative Uses,”background report prepared for an invitationalmeeting, Health-Based Risk Adjustment:Payment and Other Innovative Uses, sponsoredby The Robert Wood Johnson Foundation’sChanges in Health Care Financing andOrganization (HCFO) initiative and the KaiserPermanente Institute for Health Policy, May21–23, 2003.

Endnotes

18

27 Johns Hopkins ACG Web site,www.acg.jhsph.edu.

28 The Faces of Medicaid, Financing Chapter,published by the Center for Health CareStrategies.


30 Personal conversation with John Kaelin.

31 Kronick, R. and T. Drefus. “The Challenge ofRisk Adjustment for People with Disabilities:Health-Based Payment for MedicaidPrograms.” Center for Health Care StrategiesWorking Paper, November 1997.

32 ibid.


34 Riley, G. et al. “Health Status of MedicareEnrollees in HMOs and Fee for Service in1994,” Health Care Financing Review, Vol. 17,No. 4, Summer 1996, pp. 65–75.

35 Pope, G.C. et al. “Principal InpatientDiagnostic Cost Group Model for MedicareRisk Adjustment,” Health Care FinancingReview, Vol. 21, No. 3, Spring 2000, pp. 93–118.

36 Ash, A.S. et al. “Using Diagnoses to DescribePopulations and Predict Costs,” Health CareFinancing Review, Vol. 21, No. 3, Spring 2000,pp. 7–28.

37 For purposes of this report, the employer-basedhealth insurance market is understood toinclude private employers who purchase healthcoverage directly or self-insure, employers whoobtain insurance through purchasingcoalitions, and states who purchase healthinsurance coverage for their employees.

38 Keenan, P. et al. “The Prevalence of FormalRisk Adjustment in Health Plan Purchasing,”Inquiry, Vol. 38, pp. 249–251.

39 Personal communication with Jack Bruner, co-leader, Health Management Practice; HewittAssociates; Washington, DC; May 2003.

40 Rogal, D. and A.K. Gauthier, “Are Health-BasedPayments a Feasible Tool for Addressing RiskSegmentation?” Inquiry, Vol. 35, Summer 1998,p. 117.

41 Ibid., p. 118.

42 Glazer, J. and T. McGuire, “Why Don’t PrivateEmployers Use Risk Adjustment? ConferenceOverview,” Inquiry, Vol. 38, Fall 2001, pp. 242–4.

43 Glazer J. and T. McGuire, “Private EmployersDon’t Need Formal Risk Adjustment,” Inquiry,Fall 2001, Vol. 38, pp. 260–9.

44 Frank, R.G. and M.B. Rosenthal, ”Health Plansand Selection: Formal Risk Adjustment vs.Market Design and Contracts,” Inquiry, Vol. 38,Fall 2001, pp. 290–8.

45 Ellis, R.B. “Formal Risk Adjustment by PrivateEmployers,” Inquiry, Fall 2001, Vol. 38, pp. 299–309.

46 ibid.

47 Hewitt Associates, LLC, “Premium RiskAdjustment: A Practical Guide to AdjustingPremiums In a Managed ChoiceEnvironment.” White Paper, May 2003.

48 Personal communication with Jack Bruner, co-leader, Health Management Practice; HewittAssociates; Washington, DC; May 2003.


19

Kathryn E. Martin was a senior associate with AcademyHealth at the time this report was written.

Deborah L. Rogal is deputy director of the HCFO program and a director at AcademyHealth.

Sharon B. Arnold, Ph.D., is a director at AcademyHealth.

About the Authors

This report was prepared by AcademyHealth under The Robert Wood JohnsonFoundation’s Changes in Health Care Financing and Organization (HCFO) program.The program supports research, policy analysis, demonstration and evaluationprojects that provide policy leaders with timely information on health care policy,financing, and market developments. www.hcfo.net

About the Program

AcademyHealth is the professional home for health services researchers, policyanalysts, and practitioners, and a leading non-partisan resource for the best in healthresearch and policy. AcademyHealth promotes the use of objective research andanalysis to inform health policy and practice. www.academyhealth.org

About AcademyHealth

Special Thanks to:

The Kaiser Permanente Institute for Health Policy for co-sponsoring the workingmeeting, “Health-Based Risk Adjustment: Payment and Other Innovative Uses ,” withthe HCFO program on May 21 – 23, 2003. In particular, we thank Robert Crane,Murray Ross, and Laura Tollen for their assistance in developing and conducting themeeting, as well as for their assistance in preparing this report.

The following individuals also provided valuable input to discussions and in reviewingdrafts of this report: Robert Berenson, Bruce Bowen, Melinda Beeuwkes Buntin, JohnBertko, Jack Bruner, Daniel Dunn, David Frazzini, Emma Hoo, Sean Creighton,Jinnet Fowles, Sandra Hunt, Teresa Keane-North, Barbara Johnson, David Knutson,Patricia Salber, Randall Spoeri, Jason Wiley, J. William Thomas, Jonathan Weiner, andSteve Wright.

Editorial and Production Staff

Designer: Ed BrownProduction Manager: LeAnne DeFrancescoEditor: Christina Folz

1801 K Street, NWSuite 701-L

Washington, DC 20006

tel: 202.292.6700fax: 202.292.6800

e-mail: [email protected]: www.hcfo.net

Health-Based Risk Assessment - Changes in Health Care ... · Health Insurance and Risk Health...

Documents

Transcript of Health-Based Risk Assessment - Changes in Health Care ... · Health Insurance and Risk Health...