Basic Benefits and Clinical Guidelinesrmkaplan.bol.ucla.edu/Robert_M._Kaplan/1992_Publications_files/01… · Robert M. Kaplan A recent issue of the Journal of the American Medical

Basic Benefits andClinical Guidelines

EDITED BY

David C. Hadorn

wrrH A FOREWORD BYRichard D. Lamm

6Technical Considerations

Robert M. Kaplan

Copyright © 1992 by Westview Press, Inc.

Westview PressBOULDER* SAN FRANCISCO• OXFORD

6Technical Considerations

Robert M. Kaplan

A recent issue of the Journal of the American Medical Association (May15, 1991) presented 13 different proposals for national health insurance.

Most of the proposals emphasized cost reduction and improved access tocare. In order to accomplish these goals, many of the proposalsrecognized the need for clinical guidelines. Development of guidelinesmight allow society to meet its obligation to provide basic health carewithout the burden of paying for expensive and unnecessary care. ThePresident's Commission for the Study of Ethical Problems in Medicine

(1983), in its report "Securing Access to Health Care," pointed out that theterm "medical need" is vaguely defined. To some, it may imply provisionof any service that someone wants. As a result, there have beenproposals to provide objective standards that define basic benefits. The

Model Proposal represents one approach to the development of a basicbenefit package. The purpose of this chapter is to evaluate the technical

aspects of the proposal, particularly in light of an alternative approachknown as the General Health Policy Model.

The Model Proposal replies to the concern of the President'sCommission by defining necessary care as services that producesignificant net health benefit. Net health benefit is defined as the

"life-expectancy and quality-of-life improvements expected fromintervention minus any expected negative consequences." Although theproposal does not emphasize costs, it may lead to cost containmentthrough the elimination of unnecessary services. I wish to focus on thedefinition of health outcome and net health benefit. The feasibility of ageneral conceptualization of these outcomes is explored in the first part

60 Robert Kaplan

of the chapter. Then we depart from the proposal with the considerationof prioritized reimbursement for services.

In reviewing the Model Proposal, there are at least three places inwhich human judgment is required. The first is in the evaluation of

evidence about outcome. The second place that judgment is required isin the evaluation of the desirability of health states. Third, judgment isrequired in the estimation of treatment costs. Each of these issues will be

addressed in the following sections. Some reviews of health policymodels confuse these three types of judgment. It is important toemphasize that these judgments are different and are obtained fromdifferent types of experts. For example, judgments of treatmenteffectiveness are typically obtained from medical experts. Preferences forhealth states are best obtained from members of the community or frommedical patients. Cost data might be obtained from actuaries or healtheconomists.

Model Proposal's Approach to Incorporationof Outcome Evidence

One of the most difficult problems in creating guidelines is inestimating the effects of treatment. This is the first form of judgmentrequired to create a policy model. The Model Proposal uses panels toevaluate which aspects of treatment produce net health benefit.However, the emphasis is on judging the components of the clinicaldecision (i.e., should a certain drug be given) rather than on thedefinition of outcome. In order to estimate the incremental benefit oftreatment, it is essential that the risks and benefits of all treatments beevaluated in the common measurement units. In a later section we will

consider the methods for estimating these benefits in different studies.

At least three different approaches have been used to estimate healthbenefits. The first approach involves collection of prospective data usingsystematic medical experiments. For example, patients might berandomly assigned to treatment or to the control groups and thenfollowed over the course of time. Clearly, the randomized clinical trialprovides the best evidence for treatment effectiveness. Unfortunately,there are relatively few clinical trials that have gathered health outcome(quality of life) data using systematic experimental designs. Thus, for thevast majority of medical and surgical treatments, we do not have datafrom systematic experimental studies.

Further, there are problems with many experimental studies. For

example, the external validity or generalizability of the results is notalways optimal in randomized studies. It is becoming increasingly

Technical Considerations 61

apparent that patients selected for randomized clinical trials may not berepresentative of patients cared for under usual circumstances. Theinclusion criteria for many clinical trials often require that more than 90%of those screened for participation are excluded. Further, participants inclinical trials often are not representative of the general population interms of age, sex, and ethnicity. A second concern about randomized

clinical trials is that they are very expensive. It is extremely unlikely thatwe will have data from systematic clinical trials in all areas of medicineand surgery. Pharmaceutical companies and the National Institutes ofHealth are now beginning to include health-related quality of lifemeasures in their clinical trials. Nevertheless, they are focusing on asmall fraction of the potential therapeutic interventions and they often donot include the appropriate measures required for systematic outcomesassessment.

A final concern about clinical trials is that these medical experimentsoften take years to complete. Indeed, the best medical experiment wouldbe one that follows patients long enough to estimate the full profile oftreatment benefits and side effects. The difficulty is that we arewitnessing rapid change in medical options. Therapeutic approaches arechanging rapidly. By the time one approach is thoroughly evaluated ina clinical trial, a new therapeutic approach has taken its place. Thus,results from clinical trials are sometimes obsolete by the time theinvestigation is completed.

The second method for estimating treatment effects is to have expertsperform systematic reviews of the literature. For example, Weinstein andStason (1976a) performed many detailed analyses on the effects ofscreening and treatment for hypertension. Their analyses carefullyconsidered the medical literature and used the informed judgment ofexperts. Ultimately, they evaluated a variety of different public healthapproaches to the problem of high blood pressure treatment and control.Such analyses make use of data from clinical trials, observational studies,

and computer simulation. Today, a substantial number of such analyseshave been performed. For example, investigators have considered the

cost/utility of pneumococcal vaccine for the elderly (Office of TechnologyAssessment 1979), coronary artery bypass surgery (Weinstein and Stason1983), total hip replacement (Liang et al. 1990), and medical managementof rheumatoid arthritis (Thompson et al. 1988). Although these analyseswere performed by different analysts, they used a comparablemethodology. Thus, it is possible to construct league tables acrossstudies so that the cost/utilities of different interventions can be

compared.

The third approach is to obtain judgment from panels of experts. This

resembles the "science court" endorsed in the model proposal. The panel

62 RobertKaplan

members may not necessarily be cost/utility analysts. Instead, panelmembers are typically experts who render judgment on the likely benefitof various treatments. For example, Bush, Patrick, and Chen (1973)estimated the cost/utility of PKU screening for the state of New Yorkusing Delphi methods to obtain consensus from a panel of experts aboutthe most likely outcomes of PKU given that the disease was not detected.The experts used their experience and knowledge of the literature tocreate simulations of outcome under various scenarios. This processpermitted the estimation of outcomes in the absence of any trueexperiment. A similar process was used by the Oregon Health ServicesCommission which used experts to estimate the health benefits of 709condition-treatment pairs.

The committee approach has the advantage of allowing the estimationof many health benefits in a relatively short period of time. It is the onlyapproach that makes the implementation of guidelines feasible in the nearterm. A disadvantage of this approach is that physicians mayoverestimate the benefits associated with services that they offer. As hasbeen suggested (Mosteller 1990), providers tend to be overly optimisticabout the efficacy of their interventions. Indeed, Sachs, Smith andChalmers (1982) demonstrated that the likelihood that a treatment will bedemonstrated effective declines as the source of evidence about treatmentefficacy becomes more rigorous.

Clinical judgment tends to provide the most optimistic view of likelytreatment benefits. One of the concerns about the use of clinicaljudgment is the fear that some physicians will apply "gaming" strategiesin order to obtain reimbursement for their specialty. For instance, thereis clearly an incentive to provide optimistic estimates of treatment benefit.In a system such as the one proposed in Oregon, specialties that arepessimistic about their treatment efficacy risk putting themselves out ofbusiness. The Model Proposal calls for one or more panels of physiciansand non-physicians to make these evaluations. However, the presence ofnon-physicians does not assure a solution to the problem. In the absenceof evidence; non-physicians are in an even poorer position to challengeclinical opinion.

Greely (Chapter 8) endorses the use of expert committees forestablishing the efficacy of treatments in a basic benefits package. Hesuggested that there be multiple panels and that the panels be privatelyfunded. In endorsing panels, Greely implicitly accepts the fallibility ofthe process. However, the Greely proposal is an interesting one becausecompetition among panels creates checks upon the system and mayexpose physician gaming. I believe that the use of panels, as envisionedin the Model Proposal, is the only realistic solution for the near term.

TechnicalCons/derat/ons 63

The notionof a panel"grandjury"seems reasonableas long as itis

recognized that the estimate of treatment effectiveness is an iterativeprocess. We must expect that judgments of treatment efficacy willcontinually be revised as new data become available. This is not onlygood policy, it is good science. Panels should weight evidence by itsquality. For example, randomized clinical trials should be given the mostcareful consideration in the judgment of treatment effectiveness. Panelistmust also take into consideration biases associated with evaluation

methods. For example, we know that treatments that have beenevaluated in randomized clinical trials often appear less potent then thesame treatments evaluated in non randomized studies (Sachs et al. 1982).

Thus, well informed panels will need to consider these issues so that theydo not create biases against those treatments that have been evaluated inthe most systematic studies.

Future Directions: Comparisons Across Diagnosis

Practice guidelines require formal decision-making models.Mathematical models of decision making are now being proposed in avariety of health care systems. For example, these models have beensuggested for use in European, Australian and American health caresystems. There is a growing recognition that health care resources arevery limited. The Model Proposal uses formal algorithms to estimatewhich services produce benefit. The assumption is that enough moneywould be saved by elimination of unnecessary services to create accessto basic services for all people. However, the identification of effectiveand efficacious services might not solve the problem. Future iterationsof the proposal will need to consider comparisons across diagnoses andto recognize the possibility of prioritized funding. For example, we mayneed to invest in some services, but not others. In the United Kingdom,the national health service has acknowledged the need to prioritize

competing demands on their very limited budgets. Oregon has alsoproposed prioritized funding. Yet prioritization schemes make littlesense without some consideration of outcome.

The most important challenge in developing a formal model fornet-benefit and a model of resource allocation is in defining a commonunit of health benefit. Typically, the value of each specific interventionin health care is determined by considering a measure specific to theintervention or the disease process. For example, treatments for

hypertension are evaluated in terms of blood pressure while those fordiabetes are evaluated by blood glucose. Yet it is difficult to compare therelative value of investing in blood glucose versus blood pressure

64 Robert Kaplan

reduction. Traditional public health measures, such as life expectancy,are usually too crude to allow appropriate prioritization. However, webelieve a general model of health outcome is both feasible and practical.Although the Model Proposal mentions net benefit, it does nothing tooperationalize this construct. We believe the definition of outcome iscentral to the process. Our approach to this problem is described next.

A General Health Policy Model

In order to understand health outcomes, it is necessary to build a

comprehensive theoretical model of health status. This model includesseveral components. The major aspects of the model include mortality(death) and morbidity (health-related quality of life). In several papers,we have suggested that diseases and disabilities are important for tworeasons. First, illness may cause the life expectancy to be shortened.Second, illness may make life less desirable at times prior to death(health-related quality of life) (Kaplan and Anderson 1988, Kaplan andAnderson 1990).

Over the last two decades, a group of investigators at the Universityof California, San Diego, has developed a General Health Policy Model(GHPM). Central to the general health policy model is a generalconceptualization of health status. The model separates components ofhealth status into distinct components. These are life expectancy(mortality), functioning and symptoms (morbidity), preference forobserved functional states (utility) and duration of stay in health states

(prognosis).

Mortality

A model of health outcomes necessarily includes mortality. Indeed,many public health statistics focus exclusively on mortality throughestimations of crude mortality rates, age-adjusted mortality rates, andinfant mortality rates. Death is an important outcome that must be

included in any comprehensive conceptualization of health.

Morbidity

In addition to death, behavioral dysfunction is also an importantoutcome. The General Health Policy Model considers functioning inthree areas: mobility, physical activity, and social activity. Descriptionsof the measures of these aspects of function are given in many different

TechnicalConsidera_ons 65

publications (Kaplan and Anderson 1988, Kaplan and Anderson 1990).Most public health indicators are relatively insensitive to variationstoward the well end of the continuum. Measures of infant mortality, togive an extreme example, ignore all individuals capable of reading thisarticle since they have lived beyond one year following their births (weassume that no infants are reading the article). Disability measures oftenignore those in relatively well states. For example, the RAND HealthInsurance Experiment reported that about 80% of the general populationreports no dysfunction when they are interviewed. Thus, they wouldestimate that 80% of the population is well. In our work (Kaplan andAnderson 1990) we ask about symptoms or problems in addition tobehavioral dysfunction. In these studies, only about 12%of the generalpopulation reports no symptoms during a particular week. In otherwords, health symptoms are a very common aspect of the humanexperience. Some might argue that symptoms are unimportant becausethey are subjective and unobservable. However, symptoms are highlycorrelated with the demand for medical services, expenditures on healthcare, and motivations to expend personal resources for altered lifestyles.Thus, we feel that the quantification of symptoms is very important.

Utility (Relative Importance)

Given that various components of morbidity and mortality can betabulated, it is important to consider the relative importance of variouscomponents. For example, it is possible to develop measures that detectvery minor symptoms. Yet, because these symptoms are measurable doesnot necessarily mean they are important. A patient may experience sideeffects but be willing to tolerate them because the side effects are lessimportant than the probable benefit that would be obtained if themedication is consumed. Not all outcomes are equally important. Atreatment in which 20 of 100 patients die is not equivalent to one inwhich 20 of 100 patients develop nausea even though statistical testsmight come up with the same chi square value for the two treatments.

An important component of the General Health Policy Model is thatit attempts to scale the various health outcomes according to their relativeimportance. In the preceding example, the relative importance of dyingmay be weighted more than developing nausea. Weighting isaccomplished by rating all states on a continuum ranging from 0 (fordead) to 1.0 (for optimum functioning). These ratings are typicallyprovided by independent judges who are representative of the generalpopulation. Using this system it is possible to express the relativeimportance of states in relation to the life-death continuum. A point

66 Robert Kaplan

halfway on the scale (0.5) is regarded as halfway between optimumfunction and death. The weighting system has been described in several

different publications (Kaplan 1982, Kaplan et al. 1976, Kaplan et al. 1978,Kaplan et al. 1979). Controversies relevant to the measurement of theseweights will be addressed in a later section.

Prognosis

Another dimension of health status is the duration of a condition. A

headache that lasts one hour is not equivalent to a headache that lastsone month. A cough that lasts three days is not equivalent to a coughthat lasts three years. In considering the severity of illness, duration ofthe problem is central. As basic as this concept is, most contemporarymodels of health outcome measurement completely disregard theduration component. In the General Health Policy Model, the termprognosis refers to the probability of transition among health states overthe course of time. In addition to consideration of duration of problems,the model considers points at which the problem begins. For example,a person may have no symptoms or dysfunctions currently but may havea high probability of these problems in the future. The prognosiscomponent of the model takes these transitions into consideration andapplies a discount rate for events that Occur in the future. The Quality ofWell-being Scale (QWB) is a method for estimating some components ofthe general model. The QWB questionnaire categorizes individualsaccording to functioning and symptoms. Other components of the modelare obtained from other data sources (Kaplan and Anderson 1990).

A mathematical model integrates components of the model to expressoutcomes in a common measurement unit. Using information on currentfunctioning and duration, it is possible to express the health outcomes interms of equivalents of well-years of life, or as some have describedthem, Quality-Adjusted Life Years (QALYs). Consider, for example, aperson who is in an objective state of functioning that is rated bycommunity peers as 0.5 on a 0 to 1.0 scale. If the person remains in thestate for one year, he or she has lost the equivalent of 1/2 of one year oflife. So, for example, a person limited in activities who requires a caneor walker to get around the community might be hypothetically at 0.50.Over the course of an entire year, he or she would lose the equivalent of

one year of life. A person who has the flu may also get 0.50, but theillness might only last three days. Thus, the total loss in well-years mightbe 3/365 X 0.50 = 0.004 well-years.

A similar approach has been taken by a group of investigators in the

European community. This group includes researchers and analysts at

Technical Considerations 67

the Health Economics Research Group at Brunell University, the

Academy of Finland, The National Public Health Institute, the Universityof Helsinki, the Swedish Institute for Health Economics, the National

Institute of Public Health of Norway, the Department of AcademicPsychiatry, Middlesex, University of London, Erasmus University inRotterdam-the Netherlands, the Dutch Institute of Medical TechnologyAssessment, the Departments of Economics, Law, and Public Health andSocial Medicine of the University of Rotterdam, and the Centre for HealthEconomics at the University of York-UK.

The purpose of the EuroQol group is to develop standardizednon-disease specific outcome measures similar to the General Health

Policy Model. Conceptually the approaches are very similar. However,the exact methodologies differ. In particular, the EuroQol ratingclassifications do not include symptoms but do include pain and anxiety.Further, the method for obtaining utilities is different. Nevertheless, it

is of interest that a conceptually similar approach is emerging throughoutthe westernized world (Euroquol Group 1990).

How the General Health Policy Model Differs fromTraditional Conceptualizations

The two major differences between the GHPM and other approachesto health outcome measurement are: 1) the attempt to express benefitsand consequences of health in a common unit known as the well-year orquality-adjusted life year, and 2) emphasis on area under the curve rather

than point in time measurement. In the following sections, I will arguethat the general approach to health outcome is, intuitively, what patientsand consumers use as a guide. Their physicians may be more directedby a less comprehensive model that considers only a component of healthoutcome. For example, health care providers might focus on bloodpressure. Focusing on a blood pressure might allow the provider todisregard all of the other effects blood pressure management has uponhealth outcome. Consumers must integrate various sources ofinformation in their decision process. Intuitively, they are directedtoward maximization of health outcomes. However, sometimes these

decisions become overwhelming and the use of a formal model may aidtheir decision process.

A basic objective for most people is to function without symptoms aslong as possible. Clearly, early death contradicts this objective. Illnessand disability during the interval between birth and death also reduces

the total potential health status during a lifetime. The general nature ofthe health policy model leads to some different conclusions than more

68 Robert Kaplan

traditional medical approaches. For example, the traditional medicalmodel focuses on specific diseases and on pathophysiology.Characteristics of illness are quantified according to blood chemistry orin relation to problems in a specific organ system. Often, the focus ondisease-specific outcome measures leads to different conclusions thanthose evaluated using a more general outcome measure.

For example, studies on the reduction of blood cholesterol havedemonstrated reductions in deaths due to coronary heart disease.However, the same studies have failed to demonstrate reductions in total

deaths from all causes combined (Lipid Research Clinics CoronaryPrevention Trial Results 1984). All studies in the published literature in

which patients are assigned to cholesterol lowering through diet ormedication, or to a control group, have revealed that reductions incardiovascular mortality for those in the cholesterol lowering group arecompensated for by increases in mortality from other causes (Kaplan1984, Kaplan 1985). A meta-analysis of these studies has demonstratedthat the effect size for increase in deaths from non-illness causes (i.e.

accidents, murders, etc.) is larger than the effect size for reduction incardiovascular deaths (Mauldoon 1990).

Similar results have been reported for reductions in cardiovasculardeaths attributable to taking aspirin. The disease specific approachfocuses on deaths due to myocardial infarction because there is abiological model to describe why aspirin use should reduce heart attacks.Yet, in a controlled experiment in which physician subjects wererandomly assigned to take aspirin or placebo, there was no difference intotal deaths between the two groups (Kaplan 1989). Aspirin may reducethe chances of dying from a myocardial infarction, but it does not reducethe chance of dying (Steering Committee of the Physicians' Health StudyResearch Group 1988, Kaplan 1991a). The traditional, diagnosis specific,medical model argues that there is a benefit of aspirin because it reducesheart attack, but general health policy models argue that there is nobenefit of aspirin because there is no change in the chances of dying fromall other causes (Steering Committee of the Physicians' Health Study

Research Group 1989).This same line of reasoning applies to many other areas of health care.

Many treatments produce benefits for a specific outcome, but induce sideeffects that are often neglected in the analysis. Estimates of the benefitsof surgery must take into consideration the fact that surgery causesdysfunction through wounds that must heal prior to any realization ofthe treatment benefits. Further, surgeries often create complications. Thegeneral approach to health status assessment attempts to gain a global

picture of the net treatment benefits, taking into consideration bothtreatment benefits, side effects, and estimates of their relative importance.

Technical Considerations 69

Although the Model Proposal generally accepts this approach, somediscrepancies require discussion. For example, clinical guidelines mightemphasize the reduction of serum cholesterol. However, generalsimulations fail to show that cholesterol reduction reduces total mortality.

Further, there are technical problems in the implementation of the Model.We clearly need methods to assure the best estimates of health effect,compliance to guidelines, utilities, and cost. We will return to theseissues in a later section of the chapter.

Model Proposal andthe Incorporation of Preferences

The concept of net benefit implies that side effects and benefits havebeen balanced against each other. The Model Proposal suggests that thisshould be done but does not say how it should be done. This secondarea level of human judgment is another place where the Model Proposaldeparts from the General Health Policy Model. In this section we willreview the rationale for using preference judgments to estimate net effectsand then we will consider some of the controversies surrounding the useof preferences. The Model Proposal does not explicitly include preferencedata. The concept embraced by the proposal is that the panels wouldexpress their preferences in the evaluation of evidence. However, theissues in preference measurement can be complex. Eventualimplementation of the model will need to face the valuation problem.

Choices between alternatives in health care necessarily involvepreference judgment. For example, the inclusion of some services in abasic benefits package and the exclusion of others is an exercise in value,choice, or preference. There are many levels at which preference is

expressed in the health care decision process. For example, an older manmay decide to cope with the symptoms of urinary retention in order toavoid the ordeal and risks of prostate surgery, A physician may orderexpensive tests in order to insure against the very low probability that arare condition will be missed. Or, an administrator may decide to putresources into prevention for large numbers of people, instead ofdevoting the same resources to organ transplants for a smaller number.In each case, it would be argued that there is a net-health benefit.

In the General Health Policy Model, preferences are used to expressthe relative importance of various health outcomes. There is a subjectiveor qualitative component to health outcome. Whether we prefer aheadache or an upset stomach caused by its remedy is a value judgment.

Not all symptoms are of equal importance. Most patients would prefera mild itch to vomiting. Yet, providing a model of how well treatments

70 Robert Kaplan

work or a model that compares or ranks treatments implicitly includesthese judgments. Models require a precise numerical expression of thispreference. The General Health Policy Model explicitly includes apreference component to represent these issues. Our group measuresthese preferences using a simple rating scale method. Judges considerdescriptions of function and symptoms at a particular point in time.Using a 0 (for death) to 10 (for optimal function) scale, they rate thedesirability of states. These judgments do not require any informationabout treatment effectiveness or prognosis.

In contrast to the Model Proposal, the General Health Policy Modelincorporates preferences from random samples of the general population.It is recognized that administrators ultimately choose between alternativeprograms. The General Health Policy Model asserts that thesepreferences should represent the general public and not those of specificadministrators. Yet there is considerable debate about technical aspectsof preference assessment. Some of the debate has to do with whosepreferences are considered. Another issue concerns the specific methodused to obtain the preferences.

Whose Preferences?

Criticisms on the use of preferences abound. The naive critic typicallyassumes that preferences differ. For example, in most areas of preferenceassessment, it is easy to identify differences between different groups ordifferent individuals. It might be argued that judgments about net healthbenefits for white Anglo men should not be applied to Hispanic menwho may give different weight to some symptoms. We all have differentpreferences for movies, clothing or political candidates. Naive criticsassume that these same differences must extend to health states. Thus,

it is assumed, that the entire analysis will be highly dependent upon theparticular group that provided the preference data.

In Oregon, where an attempt was made to implement a general healthpolicy model, critics declared the whole process meaningless because theprogram was aimed at Medicaid recipients when the preferences cameform both Medicaid recipients and non recipients (Daniels 1991). Otheranalysts have suggested that preference weights from the generalpopulation cannot be applied to any particular patient group. Rather,patient preferences from every individual group must be obtained.

The difference between instrumental and terminal preferences isimportant to understanding this debate: The difference betweeninstrumental and terminal preference is analogous to the difference

between a means and an ends. Instrumental preferences describe the

TechnicalConsiderations 71

means by which various assets are attained. For instance, socialists andcapitalists hold different instrumental values with regard to the methodsfor achieving a fully functioning society. Different individuals may havedifferent preferences for how they would like to achieve happiness andevidence suggests that social and demographic groups vary considerablyon instrumental values.

Terminal values are the ends, or general states of being thatindividuals seek to achieve. The classic Rokeach study of values(Rokeach 1973) demonstrated that there is very little variability amongsocial groups for terminal preferences. Within health states, there is lessreason to believe that different social or ethnic groups will have differentpreferences for health outcomes. All groups agree that it is better to livefree of pain than to experience pain. Freedom from disability isuniversally preferred over disability states. It is often suggested thatindividuals with particular disabilities have adapted to them. However,when asked, those with disabilities would prefer not having them. Ifdisability states were preferred to non-disability states there would be nomotivation to develop treatments to help those with problems causing thedisabilities.

Although critics commonly assume substantial variability inpreferences, the evidence for differential preference is weak, at best. Ourinitial study demonstrated some significant, but very small differences,between social and ethnic groups on preferences (Kaplan et al. 1978).Studies have found little evidence for preference difference betweenpatients and the general population. For example, Balaban andcolleagues (1986) compared preference weights obtained from arthritispatients with those obtained from the general population in San Diego.They found remarkable correspondence. Nerenz and colleagues (1990)performed a similar study with cancer patients. Again, they found thatpreference weights for these patients and the cognitive strategies used toevaluate these descriptions were remarkably similar to those from thegeneral population.

There are very few differences by location. Patrick et al. (1985) foundessentially no differences between preferences on another health statusmeasure between those who !ive in the UK and those who live in Seattle.

We have compared residents of the Navaho nation living in rural Arizonawith the general population in San Diego and found few differences(Kaplan 1991b). The weights obtained by the Oregon Health ServicesCommission used a different scaling mefllodology and different wordingin the case descriptions. Nevertheless, differences between San Diegocitizens evaluated in the mid-70"s and Oregon citizens evaluated in the1990's were remarkably small (Kaplan et al. 1991). A similar scalingmethodology was used by the EuroQol Group in a series of European

72 Robert Kaplan

communities. The data from those studies suggest that differences inpreference among the European community are remarkably small andnon significant. We have used EuroQol scenarios and estimatedapproximate San Diego preferences for these cases. The results suggestthat preferences are remarkably similar. We do recognize that there isconsiderable variability in estimating preferences for a particular case(Mulley 1989). However, averaged across individuals, the meanpreference for different cases in different groups is remarkably similar.

We should not leave the impression that there are never any meandifferences in preference. For example, the original Kaplan et al. study(1978) did identify some significant differences between social groups.Further, the Oregon Health Services Commission did identify small, butobservable, preference differences among those who had previouslyexperienced a condition and those who had not. However, thesedifferences were typically very small. Further analysis will be requiredin order to determine whether these small differences effect the

conclusions of various analyses.

Measurement Method

In addition to the issue of whose preferences are obtained, we must

also consider how preferences are measured. Economists andpsychologists differ on their preferred approach to preferencemeasurement. Economists favor approaches based on expected utility

theory. The axioms of choice (von Neumann and Morganstern 1941)depend upon certain assumptions about gambling or trade-off. Thus,economists only acknowledge utility assessment methods that formallyconsider economic trades (Torrance 1986). The advantage of thesemethods is that they clearly are linked to economic theory. However,there are also some important disadvantages. For example, Kahnemanand Tversky (1983) have shown empirically that many of the assumptionsthat underlie economic measurements of choice are incorrect. Human

information processors do poorly at integrating complex probabilityinformation when making decisions that involve risk. Further, economicanalysis assumes that choices accurately correspond to the way rationalhumans put information together. A substantial literature fromexperimental psychology questions these assumptions. In particular,Anderson (1979) has presented substantial evidence suggesting thatmethods commonly used for economic analysis do not represent theunderlying true preference continuum. It seems that economists andpsychologists have ignored one another's literatures.

Technical Considerations 73

In summary, we have considered the preference issue in some detailand we believe that preferences can be explicitly considered in theformulation of a basic benefits package. A variety of studies haveevaluated the generalizability (Kaplan et al. 1978), the validity, and thereliability of the preference measures (Kaplan 1982, Kaplan et al 1979).Methodological studies have tested some of the specific concerns aboutrating scale methods (Kaplan et al. 1991, Kaplan and Ernest 1983). We

have little evidence that preferences differ across groups or that the smallobserved differences reported by others have much influence upondecisions in policy analysis. Having considered the evidence, we stillbelieve that rating scales provide an appropriate method for utilityassessment. However, we encourage continuing evaluations of theseproblems since they are expected to be important in the development ofnewer approaches to policy analysis.

Model Proposal and theIncorporation of Cost Information

A third level of judgment that deserves consideration is theimportance of cost in the development of a basic benefits package. TheModel Proposal differs from others in its explicit statement that costsshould not be directly considered. Hadorn (1991a -- see also Chapter 11)cites the inclusion of a cost component as the rationale for modifying theoriginal proposal offered by the Oregon Health Services Commission.

Hadorn described the "Rule of Rescue" in which the possibility of savinga severely ill person overrides cost considerations. Eddy (1990e) andothers argue that cost must be considered because there are few resources

and those available must be used equitably.There are several levels at which these issues need to be considered.

The most optimistic view is that practice guidelines that emphasize netbenefit will alleviate the need to consider cost. This might happen ifhealth care services are limited to those interventions and tests that

produce a known benefit. Estimates of the proportion of health servicesthat produce no benefit vary dramatically. For example, Brook and Lohr(1986) suggest that 30-50% of health services produce no noticeablebenefit. If this were true, and these services were not reimbursed, there

would be enough savings to fund all beneficial services independent oftheir price. The cost utility ratio is largely driven by the denominator(utility or benefit).

Unfortunately, it may be naive to think that health care costs could be

dramatically reduced if selected services were not reimbursed. Thedifficulty is that physicians will attempt to replace non-reimbursed

74 Robert Kaplan

services with other interventions. Ultimately, the amount of money thatcould be saved remains an empirical question. However, the potentialsavings are substantial. Suppose, for example, that costs could be cut by20%. Given an estimated 700 billion dollar expenditure in 1991, thesavings would be $140,000,000. A rea!location of 140 billion dollars mayprovide the funding base to allow reimbursement for all effectiveservices.

It is more likely that changes in reimbursement will not totally fix the

problem. Under these circumstances, considerations of costs areunavoidable. The Model Proposal does allow for cost/utilitycomparisons within equally efficacious treatments. For example, recentevidence suggests that TPA and streptokinase are equally effective atdissolving blood clots in the coronary arteries. Yet TPA may be 10 timesas expensive. Freedberg (1991) evaluated prophylaxes for PCPpneumonia among patients with AIDS. He found that one alternative,dapsone, produces an equal benefit for a small fraction of the cost of themethod most commonly used in the United States. When resources arelimited and we can gain equal effectiveness, costs must be considered.

Another reason that cost must be considered is that explicitconsideration of the cost/utility ratio may eventually effect the pricing ofservices. A policy that funds any effective treatment does little to controlcosts. The Oregon Health Services Commission has recommended theimplementation of resource based relative value scaling (RBRVS) as amethod for determining how much Medicaid should pay for variousservices. It has been extremely difficult to determine how prices formedical services became established in the first place.

Consider the example of cataract extraction with lens replacement. In

Michigan, this surgery costs about $2500. The ophthalmologists whoperform this 20 minute surgery argue that the high fee represents thedifficulty of the procedure, their training, and the risks. However, justacross the lake in Windsor, Ontario, ophthalmologists with identical

training perform the same procedure for about $700. Indeed, many of theCanadian surgeons gained their training in the United States and viceversus. Most likely, cataract surgery could be performed profitably at afee lower then the current prevailing rate. Explicit consideration of thecost/utility ratio produces competition between services. When efficacycannot be bettered, prices can be altered in order to make the servicemore competitive.

Finally, and most importantly, is the issue of cost consideration indefining basic benefits. The Model Proposal suggests that all treatmentsthat produce benefits should be included in a basic package. Anymethod that allows prioritization may exclude some services because theyare too expensive. The Model Proposal emphasizes that no helpful

TechnicalConsiderations 75

service should be denied. The difficultyis that resources are limited.When we failtoprioHtizetheservicesavailable,resourcesmay becomeovercommitted.

Insteadofrationingservices,thealternativeistorationpeople.Someindividualsbecome excludedfromthesystemand receiveno careatall.Thealternativesaretoincreasefundingortoallowthisimplicitrationing.Politicalopposition to raising taxes or to increasing health careexpendituresare very significant.Althoughneitheralternativeisattractive,prioritizationof servicesby costutilityratioseems moreattractivethantheotherbecauseitprovidesthemosthealthforthemostpeople.

One finalconcernabouttheuseofcostutilityanalysisisthatsomevery beneficialservicesmight be denied. In otherwords, ithascommonly beenarguedthatcost/utilityprioritizationschemeswillleavesome individualsunderserved.Certainspecificservicesmay onlybeavailabletothewealthyor thewell-insured.This,conceivably,couldhappenwhen a lifesavingserviceisextremelyexpensive.Althoughthisisan importantethicalconcern,we have seenverylittleevidencethatcost/utilityanalysiswould causethistype of problem. To date,prioritizationschemessuggestthatonlyservicesthathaveessentiallynohealthbenefitare eliminated.For example,itemsin the Oregonprioritizationlistthatdo poorlyincludeheroiccareforchildrenbornwithoutbrains(withnochanceofsurvival)orveryadvancedlifesupportforextremelow birthinfantwho alsohaveno chanceofsurviving.Itisconceivablethatthecost/utilityschemewillultimatelyresultindenialofpaymentforsome servicethatmay be ofbenefit.However,itisalsoimportantto recognizethatwe currentlyhave a systemthatdeniesvaluableservicestoverylargenumbersofpeople.The advantageofthegeneralmodelisthatitattemptstominimizethesuffering.

Summary and Conclusions

Formal models provideways of thinkingabout alternativesfordefiningbasicbenefitpackages.Thesemodelsareimpreciseand onlyintheinfancystageofconceptualdevelopment.However,themodelsdohelprefocustherationaleforresourceallocation.The essentialissuesintheallocationmodelaretodeterminewhat anepisodeoftreatmentcostsand what benefitsthetreatmentprovides.Usingthesemodels,we canseektoobtainthegreatestgood atthelowestcost.AlthoughtheModelProposaldoesnotconsidercosts,we believethatcostsmust eventuallybe considered.

Previous attempts at costcontainment have focused on cost restrictionwithoutconsiderationofbenefit.Yetcostcontainmentmakes no sense

76 RobertKaplan

without consideration of the purpose of providing care in the first place.The Model Proposal takes the opposite position by focusing on benefitand discounting cost. Estimates of outcome must be a component of anymodel. We prefer general models that include both cost and benefit.

Our General Health Policy Model is generally consistent with theModel Proposal, although there is greater emphasis on a cost componentand it is more explicit in the use of preference data. In order to useeither model we need considerably more data. In particular, we need tocreate better models for estimating medical effectiveness, we need tocarefully evaluate the measurement of preferences, and we need todetermine the impact of any proposal upon health care costs. However,the panel "grand jury" in the Model Proposal provides a good beginningfor estimating effectiveness. Preference measurement and cost estimationstrategies are available and should be incorporated into future versionsof the basic benefits proposed. Although imperfect, reasonable tools cannow be applied.

Once again, we must emphasize the development of guidelines andhealth policy models should be considered an iterative process. Analyseswill identify weaknesses in the data. These should stimulate new studiesand redevelopment of the models. The models themselves are alsoundergoing redevelopment. However, models and analyses do currentlyexist and it is not premature to begin applying these in the resourceallocation process: