Advertisement

Health Technology Assessment With Diminishing Returns to Health: The Generalized Risk-Adjusted Cost-Effectiveness (GRACE) Approach

  • Darius N. Lakdawalla
    Correspondence
    Correspondence: Darius N. Lakdawalla, PhD, Schaeffer Center for Health Policy and Economics, University of Southern California, 635 Downey Way, VPD 414K, Los Angeles, CA 90089-3333.
    Affiliations
    Quintiles Professor of Pharmaceutical Development and Regulatory Innovation, School of Pharmacy, Price School of Public Policy, Leonard Schaeffer Center for Health Policy and Economics, University of Southern California, Los Angeles, CA, USA

    National Bureau of Economic Research, Cambridge, MA, USA
    Search for articles by this author
  • Charles E. Phelps
    Affiliations
    University Professor and Provost Emeritus, University of Rochester, Rochester, NY, USA
    Search for articles by this author
Open ArchivePublished:January 12, 2021DOI:https://doi.org/10.1016/j.jval.2020.10.003

      Abstract

      Objectives

      Cost-effectiveness analysis (CEA) embeds an assumption at odds with most economic analysis–that of constant returns to health in the creation of happiness (utility). We aim to reconcile it with the bulk of economic theory.

      Methods

      We generalize the traditional CEA approach, allow diminishing returns to health, and align CEA with the rest of the health economics literature.

      Results

      This simple change has far-reaching implications for the practice of CEA. First, optimal cost-effectiveness thresholds should systematically rise for more severe diseases and fall for milder ones. We provide formulae for estimating how these thresholds vary with health-related quality of life (QoL) in the sick state. Practitioners can also use our approach to account for treatment outcome uncertainty. Holding average benefits fixed, risk-averse consumers value interventions more when they reduce outcome uncertainty (‘insurance value’) and/or when they provide a chance at positively skewed outcomes (‘value of hope’). Finally, we provide a coherent way to combine improvements in QoL and life expectancy (LE) when people have diminishing returns to QoL.

      Conclusion

      This new approach obviates the need for increasingly prevalent and ad hoc exceptions to CEA for end-of-life care, rare disease, and very severe disease (eg, cancer). Our methods also show that the value of improving QoL for disabled people is greater than for comparable non-disabled people, thus resolving an ongoing and mathematically legitimate objection to CEA raised by advocates for disabled people. Our Generalized Risk-Adjusted Cost-Effectiveness (GRACE) approach helps align HTA practice with realistic preferences for health and risk.

      Keywords

      Introduction

      Cost-effectiveness analysis (CEA) is widely used to evaluate new medical technologies—for example, by the UK’s National Institute for Health and Care Excellence or by the Institute for Clinical and Economic Review. Standard methods calculate the average increase in treatment cost per average quality-adjusted life-year (QALY) gained, also known as the incremental cost-effectiveness ratio (ICER). In this standard approach, an intervention improves economic welfare if ICERΔCostsΔQALYS<K, where K is the cost-per-QALY decision threshold adopted by the decision-making authority.
      Existing theory implies that decision thresholds should not vary with disease.
      • Garber A.M.
      • Phelps C.E.
      The economic foundations of cost-effectiveness analysis.
      The Institute for Clinical and Economic Review recommends using a range of thresholds, now set from $50 000 to $200 000 per QALY, depending on disease characteristics.
      Institute for Clinical and Economic Review
      2020-2023 Value Assessment Framework: Response to Public Comments, January 31, 2020.
      The National Institute for Health and Care Excellence uses an official threshold of K = ₤20 000 to ₤30 000 per QALY, perhaps operationally going as low as ₤13 000 per QALY.
      • Claxton K.
      • Soares M.
      • Rice N.
      • et al.
      Methods for Estimation of the NICE Cost-Effectiveness Threshold—Final Report.
      However, the National Institute for Health and Care Excellence regularly makes exceptions for rare diseases (up to ₤300 000, based on the extent of health improvement) and end-of-life care (up to ₤50 000) and in 2011 established a separate “cancer fund” for new cancer drugs, most of which did not meet the required ₤30 000/QALY limit.
      • Paulden M.
      Recent amendments to NICE’s value-based assessment of health technologies: implicitly inequitable?.
      In parallel, researchers have raised concern that traditional CEA discriminates against the severely ill or disabled.
      • Nord E.
      • Pinto L.K.
      • Richardon J.
      • et al.
      Incorporating societal concerns for fairness in numerical valuations of health programmes.
      ,
      • Basu A.
      • Carlson J.
      • Veenstra D.
      Health years in total: a new health objective function for cost-effectiveness analysis.
      The U.S. Affordable Care Act forbids using CEA that discriminates against persons with disabilities, both by the Patient-Centered Outcomes Research Institute and in determining Medicare coverage and reimbursement. To address this concern, the Institute for Clinical and Economic Review now calculates the equal value of life-years gained in parallel with standard CEA analyses,
      Institute for Clinical and Economic Review
      2020-2023 Value Assessment Framework, January 31, 2020.
      and other departures from CEA have been proposed as ad hoc ways to repair this problem.
      • Basu A.
      • Carlson J.
      • Veenstra D.
      Health years in total: a new health objective function for cost-effectiveness analysis.
      These exceptions, exclusions, and prohibitions call for deeper examination of CEA’s theoretical foundations. In a new analysis, we develop a generalization of standard CEA methods that resolves many of these issues. We begin with one of the simplest ideas in economics—that of diminishing returns. For example, an additional $10 000 is worth more when base income is $50 000 than when it is $100 000. This diminishing returns assumption is already embedded into the way CEA values nonhealth consumption gains (in a period utility function where the other argument is health-related quality of life). However, the existing CEA framework imposes the restriction that returns to health-related quality of life (QoL) never diminish.
      • Garber A.M.
      • Phelps C.E.
      The economic foundations of cost-effectiveness analysis.
      Our new study relaxes this restriction, resulting in major changes to the proper conduct of CEA. This leads to our Generalized Risk-Adjusted Cost-Effectiveness (GRACE) framework.
      The Generalized Risk-Adjusted Cost-Effectiveness framework justifies the longstanding hypothesis that illness severity should affect the per-unit value of health improvements. We also show why QoL gains for persons with disabilities are more valuable than for comparable nondisabled persons, how to assess the value of reducing variable treatment outcomes (“value of insurance”), and why risk-averse consumers may still place value on risky outcome distributions skewed toward more favorable effects (“value of hope”).
      • Lakdawalla D.N.
      • Doshi J.A.
      • Garrison Jr., L.P.
      • Phelps C.E.
      • Basu A.
      • Danzon P.M.
      Defining elements of value in health care-a health economics approach: an ISPOR Special Task Force Report [3].
      Finally, GRACE directly responds to a recent ISPOR task force urging development of methods to “augment” CEA by widening the scope of cost-effectiveness models.
      • Garrison Jr., L.P.
      • Neumann P.J.
      • Willke R.J.
      • et al.
      A health economics approach to US value assessment frameworks: summary and recommendations of the ISPOR Special Task Force Report [7].

      Methods

      Background

      In standard CEA frameworks, the willingness to pay (WTP) for health equals the marginal utility of health divided by the marginal utility of consumption. Willingness to pay per QALY (defined as K) satisfies K=CωC, where C is annual nonhealth consumption and ωC is the rate at which utility changes with income.
      • Phelps C.E.
      A new method for determining the optimal willingness to pay in cost-effectiveness analysis.
      The more consumption-related utility sacrificed by diverting spending to healthcare, the larger is ωC and the smaller the CEA threshold, and conversely. Recent estimates put ωc in the neighborhood of 0.3 to 0.5, making K about 2 to 3 times annual nonhealth consumption.
      • Phelps C.E.
      A new method for determining the optimal willingness to pay in cost-effectiveness analysis.
      ,
      • Phelps C.E.
      • Cinatl C.
      A generalized method for estimating optimal willingness to pay thresholds for health technology assessment.
      The World Bank estimates that 2019 GDP per capita in the United States was $65 118, implying (with 17% of GDP spent on healthcare) that C$55000. For the United States, this translates to about $110 000 to $165 000 per QALY, within the range recommended by the Institute for Clinical and Economic Review,
      Institute for Clinical and Economic Review
      2020-2023 Value Assessment Framework: Response to Public Comments, January 31, 2020.
      several prominent clinical organizations,
      • Anderson J.L.
      • Heidenreich P.A.
      • Barrnett P.G.
      • et al.
      ACC/AHA statement on cost/value methodology in clinical practice guidelines and performance measures: a report of the American College of Cardiology/American Heart Association Task Force on Performance Measures and task Force on Practice Guidelines.
      and previously by the World Health Organization.
      • Tan-Torres Edjerer R.
      • Baltussen T.
      • Hutubessy A.
      • et al.
      Making Choices in Health: WHO Guide to Cost-Effectiveness Analysis.

      Introducing Diminishing Returns to Health

      Our model employs a scalar health index, defined on a [0,1] interval. This could be a QALY or an alternative QoL index defined over the same interval (eg, one constructed using multicriteria decision analysis methods
      • Phelps C.E.
      • Madhavan G.
      Using multicriteria approaches to assess the value of health care.
      ) that has not been converted to utilities (eg, by use of standard gamble techniques). Multicriteria decision analysis combines different dimensions of health into a single index by introducing importance weights for each component of health.
      Under GRACE, WTP for health gains generalize from K=C[1ωC] to KωHR=C[ωHωC]R. We define 2 new parameters, R and ωH. The traditional framework restricts both to equal 1.0. What are these 2 new multipliers, ωH and R, and why might they differ from one?
      Formally, ωH is analogous to ωc. It describes how utility (happiness) changes with health-related QoL. The ratio [ωHωC] extends the traditional multiplier [1ωC] by adding the parallel concept of diminishing returns to health (ωH). When returns to QoL do not diminish, happiness continues to rise proportionately with QoL, and ωH=1. However, with positive but diminishing returns to health, 0<ωH<1, so [ωHωC]< [1ωC]. Other things equal, this reduces the traditional WTP threshold K.
      Next consider R, the disease severity ratio. R adjusts cost-effectiveness thresholds for disease severity. Quantitatively, R is the ratio between the marginal utility of health in the sick state and the marginal utility when healthy. The traditional model imposes the restriction that R = 1. This is essentially correct for the mildest diseases, but it understates the WTP threshold—sometimes considerably—for highly severe illnesses.
      Variation in R results from diminishing returns to health. If you are very sick, you derive more incremental value from a fixed QoL gain than if you are mildly sick. Greater severity means larger values of R and—consistent with prior empirical research
      • Nord E.
      Severity of illness versus expected benefit in societal evaluation of healthcare interventions.
      —higher WTP for any given health improvement.
      R1 for trivially minor illnesses, but rapidly rises with illness severity. The rate of increase depends on how rapidly diminishing returns set in. Long-established results in economics imply that consumers experience diminishing returns if and only if they exhibit risk aversion.
      • Cather D.A.
      A gentle introduction to risk aversion and utility theory.
      Therefore, the rate at which R rises with disease severity also depends on relative risk aversion over QoL (rH),which we define as analogous to relative risk aversion in nonhealth consumption, (rC). The GRACE framework demonstrates how to estimate R using the relative QALY loss from a disease and the degree of relative risk aversion in QoL. Later, we present some numbers to make this result concrete.
      We know of no estimates of rH currently, so estimates of this parameter are needed; we suggest estimation approaches below. In the interim, we suggest a benchmark assumption of rH=rC1, with appropriate sensitivity analysis around it. Because of the mathematical relationship between diminishing returns and risk aversion, one can show under this assumption that ωH=ωC. In this special case, the WTP threshold becomes CR. Thus, for example, if annual average nonhealth consumption were $55 000, the threshold becomes $55 000 ×R. Under GRACE, WTP for health increases with illness severity. We call this the risk aversion and severity-adjusted WTP, or RASA-WTP.

      Measuring Risk-Adjusted Health Gains

      Recall that diminishing returns to QoL imply consumers will be averse to risky QoL outcomes.
      • Cather D.A.
      A gentle introduction to risk aversion and utility theory.
      This requires accounting for risky treatment outcomes. To meet this need, we incorporate the utility cost of risky outcomes into the standard QALY measure. We produce a new, more general measure of QoL improvement that we call the risk-adjusted QALY, or RA-QALY. Conveniently, the RA-QALY can be expressed as the standard average QALY gain multiplied by a single mathematically defined value, ε, that combines statistical measures of treatment variability with consumer attitudes toward QoL risk. As discussed further below, our study provides details on how to estimate ε from data already gathered in standard randomized, controlled trials or technology assessments, and from consumer risk preference parameters. If a new treatment and its comparison therapy have the same risk parameters, then ε = 1, but if the new treatment reduces overall risk, then ε > 1, and conversely.
      The GRACE framework shows that reducing variance in health outcomes adds value to a treatment—akin to “the insurance value” identified in earlier literature.
      • Lakdawalla D.N.
      • Malani A.
      • Reif J.
      The insurance value of medical innovation.
      In principle, this includes both the value of physical risk reduction and the value of financial risk reduction implicit in insurance value. With complete health insurance, this reduces to the value of physical risk reduction alone.
      • Lakdawalla D.N.
      • Malani A.
      • Reif J.
      The insurance value of medical innovation.
      Allowing for physical and financial risk reduction together complicates the notation, but it is a straightforward extension of our approach.
      The GRACE framework further implies that increasing positive skewness in treatment outcomes provides value to consumers who are risk-averse and “prudent” in their risk preferences. Both risk aversion and prudence seem to characterize real-world consumer risk preferences.
      • Kimball M.S.
      Precautionary saving in the small and in the large.
      The idea that patients will prefer treatments that provide a chance for a large positive benefit has been called the “value of hope,” which has been measured in patients with cancer.
      • Lakdawalla D.N.
      • Romley J.A.
      • Sanchez Y.
      • Macleal J.R.
      • Penrod J.R.
      • Philipson T.J.
      How cancer patients value hope and the implications for cost-effectiveness assessments of high-cost cancer therapies.
      ,

      Shafrin J, Schwartz TT, Okoro T, Romley JA. Patient versus physician valuation of durable survival gains: Implications for value assessments. Value Health. 20(2):217-223.

      The RA-QALY properly incorporates these risk preferences. It penalizes treatments with more variance in outcomes but rewards those with a high degree of positive skewness—that is, hope.

      Distinguishing Health and Utility

      According to The Handbook on Cost-Effectiveness Analysis (2nd edition, page 52), “Most CEA analysts consider QALYs as measures of health.”
      • Neumann P.J.
      • Sanders G.D.
      • Russell L.B.
      • Siegel J.E.
      • Ganiats T.G.
      Cost Effectiveness in Health and Medicine.
      It further states that “For those who aspire to connect QALYs to utility theory, a number of important issues must be addressed.” Indeed, these issues have created confusion in the practice of CEA.
      Several CEA theorists have recognized that patients may have risk preferences over QALYs, even though these are not explicitly measured (cf.
      • Buckingham K.
      • Devlin N.
      A theoretical framework for TTO valuations of health.
      ). The current solution relies on the result from expected utility theory that consumers remain risk-neutral on the level of utility, even if risk-averse over levels of consumption and health. To implement it, analysts convert measures of health into measures of utility using standard gamble, time trade-off, or visual analogue scale methods. From this perspective, the ICER is correctly interpreted as incremental costs per unit gain in health-related utility. In principle, this allows for risk aversion over QoL.
      Unfortunately, one cannot reconcile this approach with a cost-effectiveness threshold that remains fixed when disease severity varies. If ICERs measure costs per utility gain, then cost-effectiveness thresholds reflect WTP for a gain in health-related utility. Standard economic analysis implies that the WTP for gains in utility varies with the level of health. Briefly, WTP for gains in utility is the inverse of the marginal utility of consumption (cf.
      • Garber A.M.
      • Phelps C.E.
      The economic foundations of cost-effectiveness analysis.
      ). Cost-effectiveness analysis presumes that utility is the product of consumption-related utility and health-related utility; therefore, the marginal utility of consumption will vary with baseline health. Thus, while health-related utility measures remain valid under QoL risk aversion, cost-effectiveness thresholds that fail to vary with health become invalid.
      The GRACE framework solves this problem by using measures of health, not utility, as inputs and then by explicitly accounting for how risk aversion in health affects valuation. Using the mean, variance, and skewness of distributions of health outcomes, we create a Taylor series approximation to any sufficiently differentiable utility function to estimate the expected utility of the health outcome.
      Generalized Risk-Adjusted Cost-Effectiveness framework practitioners have several existing options for health measurement. One is the widely used EQ-5 measure. It assesses health in 5 domains (mobility, self-care, usual activities, pain/suffering, and anxiety/depression), each rated on a 5-point severity scale (none, slight, moderate, severe, extreme). Each of these is a measure of health. They can be combined into a single scalar measure of health by assessing the relative importance of each domain—for example, using multicriteria decision analysis models (cf.
      • Phelps C.E.
      • Madhavan G.
      Using multicriteria approaches to assess the value of health care.
      ). Another option is the HUI-3 model, which uses 8 domains but is otherwise similar. However, the HUI-3 is often converted into utilities (eg, with standard gamble methods), an unnecessary step in GRACE.

      Dealing With Disability

      As discussed earlier, many people object to CEA’s valuation of health improvement for people with disabilities. If permanent disability reduces people’s life expectancy, then standard CEA methods say that improving their QoL is not as valuable, because there are fewer remaining life-years over which to enjoy the improvement. Standard CEA also says that improving life expectancy is not as valuable, because people with disabilities start with lower baseline QoL. These issues are exacerbated by the phenomenon of adaptation. People with a particular disability often rate it as less costly than would nondisabled persons (cf.
      • Groot W.
      Adaptation and scale of reference bias in self-assessments of quality of life.
      ). Thus, health ratings based on general population surveys may understate QoL for disabled people.
      The GRACE framework mitigates these concerns, because it implies that the value of improving QoL of permanently disabled people is greater than for otherwise comparable nondisabled people. Due to diminishing returns, greater preexisting disability implies greater per-QALY value in improving QoL. In some cases, analysts may wish to consider differences in nonhealth consumption levels between people with and without disability. Here, one should account for baseline consumption and any disability insurance payments. Higher consumption levels translate into higher WTP for health, and vice versa.

      Changes in Life Expectancy

      Medical technologies affect both QoL and life expectancy (LE). Many interventions extend LE, including cancer therapies, cardiac drugs, stents, various surgeries, vaccines against contagious diseases, or smoking cessation support. Some of these come with risks of reduced QoL (eg, chemotherapy for cancers). Others, such as monoclonal antibody therapies for some diseases (eg, lupus, Crohn’s disease, psoriasis, ulcerative colitis), increase risks of serious infections and even death, with the hope of increasing QoL. Our methods show how to combine the upside and downside risks to LE and in QoL in a unified way.
      Since returns to QoL diminish, the “exchange rate” between gains in LE and gains in QoL will vary with disease severity. For example, a person in a highly disabled but long-lived state might be willing to give up more LE in exchange for QoL improvements than a less disabled person. Consequently, GRACE shows that disabled persons would have different preferences about this exchange rate than otherwise similar nondisabled persons. Greater permanent disability leads to stronger preferences for QoL improvement over LE improvement. This does not represent bias against disabled persons in terms of extending their LE, but rather the plausible reality that a given gain in QoL is worth more to a person in a disabled state. As noted before, our model heightens the value of improving QoL for disabled persons, which should translate directly (in value-based healthcare finance systems) into stronger incentives for improving QoL among those with permanent disabilities. Many such treatments, of course, will also extend LE for people with the same disability.

      Results and Implementation

      Within current CEA methods, a marginal change in QALYs is defined as (ΔS)Q+S(ΔQ), where S is baseline LE, ΔS the change in LE, Q baseline QoL, and ΔQ the change in QoL. ΔC is the incremental cost of the new technology. In this traditional framework, a technology with incremental cost, ΔCost, is welfare-improving if:
      ΔCost(ΔS)Q+S(ΔQ)K
      (1)


      The GRACE framework produces a similar expression, but with a few additional parameters. We define the generalized risk-adjusted QALY (GRA-QALY) as (ΔS)δ+S(ΔQ)ε. Here, δ reflects the QoL units a consumer would give up in exchange for 1 more life-year. In the conventional framework without diminishing returns to QoL, this equals the baseline QoL level, Q. With diminishing returns, this restriction evaporates. Separately, as noted previously, ε reflects the change in value associated with uncertain treatment outcomes. Treatments with high outcome variance are worth less to risk-averse consumers, so ε < 1. In contrast, treatments with high positive skewness produce “hope” to consumers who are “prudently” risk-averse,
      • Kimball M.S.
      Precautionary saving in the small and in the large.
      so that ε > 1.
      In the GRACE framework, technologies should be adopted if and only if:
      ΔCost(ΔS)δ+S(ΔQ)εKωHR
      (2)


      or, in an acronym-rich formulation, (ΔCost)/(GRA-QALY) ≤ RASA-WTP. When ωH=1 (constant returns to QoL), ε = 1 (risk neutrality in health or non-risky QoL outcomes), and δ = Q, this collapses to the traditional formula for cost-effectiveness.

      Generalized risk-adjusted QALY

      The GRA-QALY incorporates 2 new parameters that can be readily estimated. First, δ is the marginal rate of substitution between LE and QoL. It can be estimated via time trade-off survey methods applied to each disease of interest. For example, consider a new treatment for a disease with a baseline QALY level of 0.8. The time trade-off survey would ask respondents how many years of expected survival (LE) they would give up within this disease state in exchange for restoring ideal QoL. For this disease, the estimate would be δ=0.2LE willing to forego. Once this parameter is estimated for a disease state, it can be reused for all diseases with the same baseline severity. This can also be achieved, of course, using discrete choice experiment methods.
      The term ε can be estimated using: (1) relative risk preferences over QoL, and (2) statistical moments characterizing the distribution of QALY gains. Relative risk preferences measure the degree of consumer risk aversion over risky QoL outcomes. Relative risk preferences can be used over the full range of diseases and treatments, without estimating them anew each time.
      A number of approaches to estimating relative risk preference are available. One is a structured series of discrete choice experiment questions that elicit not only relative risk-aversion, but also higher-order risk parameters, following the methods of Noussair et al.
      • Noussair C.N.
      • Trautmann S.T.
      • Van de Kuilen G.
      Higher order risk attitudes, demographics, and financial decisions.
      Another might use direct measures of happiness
      • Easterlin R.A.
      The economics of happiness.
      and relate them to income and QoL measures for the same subjects, using measures such as the EQ5
      EuroQol Group
      EuroQol: a new facility for the measurement of health-related quality of life.
      to measure health levels. Properly done, this could provide estimates of the relevant measures of rC,rH,ωC,and ωH from the same population.
      Finally, to characterize the distribution of QALY gains, analysts need to estimate: average QALYs in the treated and untreated states, the variance of QALYs in the treated and untreated states, and the skewness of QALYs in the treated and untreated states. The second and third pairs of parameters are not typically reported but are simple to calculate from cost-effectiveness models or other studies estimating QALYs gained. Again, we assume that QALYs continue to be used as a single summary measure of QoL. Our model admits a range of approaches, so long as they are applied uniformly across a wide spectrum of disease conditions.

      Risk- and severity-adjusted willingness to pay (RASA-WTP)

      Conveniently, once relative risk preference parameters are estimated, KωH can be calculated directly. This leaves the question of how to quantify the disease severity ratio, R. Table 1 shows how R varies with relative risk aversion over QoL, rH, and disease severity, measured as the percent loss in QoL, . The Table 1 values of the risk-aversion parameter, rH, span from rH=0, the assumed value in the traditional CEA model, to rH=1.25. As context, most modern estimates of rC fall between 0.7 and 1.0.
      • Phelps C.E.
      A new method for determining the optimal willingness to pay in cost-effectiveness analysis.
      ,
      • Noussair C.N.
      • Trautmann S.T.
      • Van de Kuilen G.
      Higher order risk attitudes, demographics, and financial decisions.
      ,
      • Chetty R.
      A new method of estimating risk aversion.
      The QoL loss, , ranges from 0 (no health loss) to 0.9 (90% reduction from perfect health). If =0.1, then QoL = 0.9 on a [0,1] scale. If =0.5, QoL = 0.5 on the same scale, and so on.
      Table 1R multipliers for various relative risk aversion and health loss values.
      Relative Health Loss()Relative Risk Aversion in Health(rH)
      00.250.50.7511.25
      0111111
      0.111.031.051.081.111.13
      0.311.091.21.311.431.56
      0.511.191.411.6222.38
      0.711.351.832.473.334.5
      0.911.783.155.611017.7
      Note. This is a condensed version of the table in [, Table 2].
      The R multiplier is somewhat sensitive to rH, particularly for the most severe illnesses. For =0.9, if rH=1.25, the R value rises to 17.7, and for rH=0.25it falls to 1.78—a 10-fold difference. This emphasizes the importance of acquiring more precise estimates of rH. The sensitivity to rH falls as falls.
      To improve familiarity with these ideas, Table 2 shows some estimated QoL levels for a series of disease and disability conditions, all taken from the Tufts Cost-Effectiveness Analysis Registry.
      Center for the Evaluation of Value and Risk in Health. Cost-Effectiveness Analysis (CEA) Registry.
      Table 2Severity of illness measures for representative illnesses and disabilities.
      Relative Disease SeverityRepresentative Diseases and Conditions
      0.0 to 0.1Peptic ulcer, stress urinary incontinence; benign prostatic hyperplasia
      1.1 to 0.2Grave’s disease (hyperthyroidism), sleep apnea, hypertension (uncomplicated)
      0.2 to 0.3Hypercholesterolemia (familial), end-stage knee osteoarthritis, peripheral arterial disease
      0.3 to 0.5Type 1 diabetes; acute lung injury; moderate to severe rheumatoid arthritis; relapsing remitting multiple sclerosis
      0.5 to 0.7Transient ischemic attack and carotid stenosis; traumatic brain injury; nursing home resident at risk for pressure ulcers; secondary progressive multiple sclerosis
      0.7 to 1.0Alzheimer’s disease; metastatic colon cancer; acute pulmonary embolism
      Note. All values derived from the Tufts Center for the Evaluation of Value and Risk. This table is taken from [, Table 3].
      Combining these 2 tables (and the assumption that risk preferences in health and consumption are approximately identical), we come to a clear conclusion: Current CEA methods overvalue treatments for mild illnesses (eg, peptic ulcer, benign prostatic hypertrophy, urinary incontinence) and undervalue treatments for highly severe illnesses (eg, Alzheimer’s disease, metastatic cancers, acute pulmonary embolism, pressure ulcers in nursing homes). We could be overpaying by a factor of 2-3 for mild illnesses and underpaying for severe illnesses by factors of 4-5 or more. The exact valuations will hinge on better estimates of risk preferences in consumption and health. For the sake of concreteness, suppose that rH1, and average annual nonhealth consumption is $55000. In this special case, the cost-effectiveness threshold is $55000R. Mild illnesses would require a threshold of $55 000, even though traditional CEA would often ascribe thresholds of 2 to 3 times average annual consumption. Highly severe illnesses would call for thresholds of up to $600 000 per QALY, even holding average annual consumption fixed.
      To ease estimation of R, researchers can build a comprehensive table of illness severity—that is, values of —for various health conditions. This table can be built up over time, similar to ways in which the Diagnosis-Related Group (DRG) system has improved calibration of severity to compensate hospitals properly that have different blends of case-mix severity. As with the DRG system, this should be done by neutral parties who have no specific financial interest in the outcome. The Tufts registry
      Center for the Evaluation of Value and Risk in Health. Cost-Effectiveness Analysis (CEA) Registry.
      provides one starting point for such efforts.

      Conclusions

      As health payers increasingly turn to CEA for value assessment, it becomes even more important to assure that it reflect the preferences of real people. Current models run an important risk by not considering the consequences of diminishing returns and risk aversion over health. Continuing to assume that the incremental value of health is invariant to severity of illness endangers the foundations of CEA. The combination of the diminishing returns and severity of illness adjustments suggests that we are probably overvaluing treatments of low-severity illnesses (possibly by a factor of 2 or more) and undervaluing treatments of very high-severity conditions (possibly by a factor of 5 or more). Following other economic literature,
      • Noussair C.N.
      • Trautmann S.T.
      • Van de Kuilen G.
      Higher order risk attitudes, demographics, and financial decisions.
      ,
      • Chetty R.
      A new method of estimating risk aversion.
      we suggest initially assuming that risk-related preference parameters (ε,R,δ,ωH) are constant across the population, but more refined estimates can allow these to vary across subgroups.
      Value assessments drive reimbursement, which signals to innovators where best to apply their efforts. Our analysis shows that value hinges strongly on untreated illness severity, a factor that current reimbursement processes ignore. This distorts incentives for innovation, tilting toward treating relatively mild illnesses, whereas the greatest value comes from treating the most severe illnesses.
      Our model points toward better ways to resolve concerns regarding how standard CEA discriminates against disabled people. We show that improving the QoL of disabled persons is worth more than for a comparable nondisabled person. It also shows why disabled persons’ preferences likely will tilt toward technologies that improve QoL (vs life extension).
      The GRACE framework shows how to generalize traditional CEA models to incorporate the effects of diminishing returns to health improvements as severity of illness increases. This creates cost-effectiveness thresholds (stated as multipliers of consumption) that incorporate risk preferences both in consumption and in QoL and that increase with severity of illness. Our model also incorporates measures of treatment outcome uncertainty, valuing interventions more when they not only improve average outcomes but also reduce the uncertainty surrounding those outcomes. Finally, it provides a coherent way to combine improvements in QoL and LE in this more generalized framework.
      Using this approach will lead to more sensible reimbursement of medical treatments, will improve the well-being of the most vulnerable members of our society—those with severe acute illnesses and chronic disabilities—and will rationalize financial signals sent to innovators of medical technologies as to where best to apply their efforts toward successful innovation.
      Author Contributions: Concept and design: Lakdawalla, Phelps
      Acquisition of data: Lakdawalla
      Analysis and interpretation of data: Lakdawalla, Phelps
      Drafting of the manuscript: Lakdawalla, Phelps
      Critical revision of the paper for important intellectual content: Lakdawalla, Phelps
      Statistical analysis: Phelps
      Obtaining funding: Lakdawalla
      Administrative, technical, or logistic support: Lakdawalla
      Conflict of Interest Disclosures: Dr Lakdawalla reported holding equity in Precision Medicine Group outside the submitted work; and personal fees from GRAIL, Pfizer, Novartis, Amgen, Otsuka, and Genentech outside the submitted work. Dr Phelps reported receiving personal fees from Audentes Therapeutics, Merck Sharp & Dohme, and Pfizer Pharmaceuticals outside the submitted work.
      Funding/Support: The National Institute on Aging ( 1R01AG062277-01 ) provided funding for this research.
      Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
      Acknowledgment: The authors thank Hanh Nguyen, MA, for providing valued technical assistance.

      References

        • Garber A.M.
        • Phelps C.E.
        The economic foundations of cost-effectiveness analysis.
        J Health Econ. 1997; 16: 1-31
        • Institute for Clinical and Economic Review
        2020-2023 Value Assessment Framework: Response to Public Comments, January 31, 2020.
        • Claxton K.
        • Soares M.
        • Rice N.
        • et al.
        Methods for Estimation of the NICE Cost-Effectiveness Threshold—Final Report.
        Center for Health Economics Research Paper 81, York, England2013
        • Paulden M.
        Recent amendments to NICE’s value-based assessment of health technologies: implicitly inequitable?.
        Expert Rev Pharmacoecon Outcomes Res. 2017; 17: 239-242
        • Nord E.
        • Pinto L.K.
        • Richardon J.
        • et al.
        Incorporating societal concerns for fairness in numerical valuations of health programmes.
        Health Econ. 1999; 8: 25-39
        • Basu A.
        • Carlson J.
        • Veenstra D.
        Health years in total: a new health objective function for cost-effectiveness analysis.
        Value Health. 2020; 23: 96-103
        • Institute for Clinical and Economic Review
        2020-2023 Value Assessment Framework, January 31, 2020.
      1. omitted.

        • Lakdawalla D.N.
        • Doshi J.A.
        • Garrison Jr., L.P.
        • Phelps C.E.
        • Basu A.
        • Danzon P.M.
        Defining elements of value in health care-a health economics approach: an ISPOR Special Task Force Report [3].
        Value Health. 2018; 21: 131-139
        • Garrison Jr., L.P.
        • Neumann P.J.
        • Willke R.J.
        • et al.
        A health economics approach to US value assessment frameworks: summary and recommendations of the ISPOR Special Task Force Report [7].
        Value Health. 2018; 21: 161-165
        • Phelps C.E.
        A new method for determining the optimal willingness to pay in cost-effectiveness analysis.
        Value Health. 2019; 22: 785-791
        • Phelps C.E.
        • Cinatl C.
        A generalized method for estimating optimal willingness to pay thresholds for health technology assessment.
        University of Rochester, 2020
        • Anderson J.L.
        • Heidenreich P.A.
        • Barrnett P.G.
        • et al.
        ACC/AHA statement on cost/value methodology in clinical practice guidelines and performance measures: a report of the American College of Cardiology/American Heart Association Task Force on Performance Measures and task Force on Practice Guidelines.
        J Am Coll Cardiol. 2014; 63: 2305-2321
        • Tan-Torres Edjerer R.
        • Baltussen T.
        • Hutubessy A.
        • et al.
        Making Choices in Health: WHO Guide to Cost-Effectiveness Analysis.
        World Health Organization, Geneva2003
        • Phelps C.E.
        • Madhavan G.
        Using multicriteria approaches to assess the value of health care.
        Value Health. 2017; 20: 251-255
        • Nord E.
        Severity of illness versus expected benefit in societal evaluation of healthcare interventions.
        Expert Rev Pharmacoecon Outcomes Res. 2001; 1: 85-92
        • Cather D.A.
        A gentle introduction to risk aversion and utility theory.
        Risk Manage Insur Rev. 2010; 13: 127-145
        • Lakdawalla D.N.
        • Malani A.
        • Reif J.
        The insurance value of medical innovation.
        J Public Econ. 2017; 145: 94-102
        • Kimball M.S.
        Precautionary saving in the small and in the large.
        Econometrica. 1990; 58: 53-73
        • Lakdawalla D.N.
        • Romley J.A.
        • Sanchez Y.
        • Macleal J.R.
        • Penrod J.R.
        • Philipson T.J.
        How cancer patients value hope and the implications for cost-effectiveness assessments of high-cost cancer therapies.
        Health Affairs. 2012; 31: 676-682
      2. Shafrin J, Schwartz TT, Okoro T, Romley JA. Patient versus physician valuation of durable survival gains: Implications for value assessments. Value Health. 20(2):217-223.

        • Neumann P.J.
        • Sanders G.D.
        • Russell L.B.
        • Siegel J.E.
        • Ganiats T.G.
        Cost Effectiveness in Health and Medicine.
        Oxford University Press, New York2017 (2nd ed)
        • Buckingham K.
        • Devlin N.
        A theoretical framework for TTO valuations of health.
        Health Econ. 2006; 15: 1149-1154
        • Groot W.
        Adaptation and scale of reference bias in self-assessments of quality of life.
        J Health Econ. 2000; 19: 403-420
        • Noussair C.N.
        • Trautmann S.T.
        • Van de Kuilen G.
        Higher order risk attitudes, demographics, and financial decisions.
        Rev Econ Stud. 2013; 81: 325-355
        • Easterlin R.A.
        The economics of happiness.
        Daedalus. 2004; 133: 26-33
        • EuroQol Group
        EuroQol: a new facility for the measurement of health-related quality of life.
        Health Policy. 1990; 16: 199-208
        • Chetty R.
        A new method of estimating risk aversion.
        Amer Econ Rev. 2006; 96: 1821-1834
      3. Center for the Evaluation of Value and Risk in Health. Cost-Effectiveness Analysis (CEA) Registry.