Valuing the Societal Impact of Medicines and Other Health Technologies: A User Guide to Current Best Practices

2.1 Generalized Cost-Effectiveness Analysis (GCEA) Value Flower

There have been multiple initiatives characterizing the GCEA value flower (Lakdawalla et al. 2018; Review 2023c; Sanders et al. 2016a). We categorize GCEA’s petals into four groups: (i) uncertainty, (ii) dynamics, (iii) beneficiary, and (iv) additional elements (Figure 1). The elements of conventional CEA – including the use of conventional quality-adjusted life-years (QALYs) and the direct treatment costs and potential cost offsets –are captured by the core of the GCEA value flower; the GCEA flower with none of the current petals (i.e. just the pistil) would represent conventional CEA (Figures 1 and 2). Ideally, effectiveness would be measured in a patient-centered matter and could include both general health gains – which could be measured in QALYs – and potentially disease-specific measures of effectiveness which may be more or less difficult to convert into QALYs. Conventional healthcare costs and effectiveness as core value elements, however, are not discussed in this paper as our focus is expanding beyond conventional CEA.

2.1.1 Uncertainty

This category includes three petals recognizing the values associated with uncertainty of patient’s future health outcomes borne by the stochastic natures of the treatment effectiveness and the development of disease (Table 1). Outcome certainty represents the value that risk-averse or risk-preferring patients place on the distribution of health outcomes independent of the expected value (value of reducing risk in health outcomes) as well as the potential for highly favorable outcomes enabled by treatment (value of hope).^[1] Disease risk reduction represents potential health benefits to non-patients who might gain peace-of-mind that a treatment exists should they become sick (insurance value). There is another petal in this category not directly related to uncertainty on future health outcomes: the value of knowing represents the value of improved diagnostics, which facilitates more informed treatment decisions by reducing the uncertainty surrounding a patient’s current health status. For each of these petals, the value beyond what would be accounted for in conventional CEA is largely driven by the degree of patient risk-aversion.

Table 1:

Generalized cost effectiveness analysis (GCEA) implementation petal checklist.

1. Approaches appearing in blue are the preferred methods of GCEA. GRACE = generalized and risk-adjusted cost-effectiveness; EP: Two-parameter expo-power utility function; COGS = cost of goods sold; DCEA = distributional cost-effectiveness analysis. *Suggested parameters from literature that can be used in the estimation of a given value component or method.

2.2 GCEA Versus Previous Value Inventories

GCEA differs from previous efforts to make value assessment more comprehensive (Table 2). The Second Panel emphasized the importance of both the healthcare and societal perspectives (Sanders et al. 2016b). It also introduced an “impact inventory” that featured impacts on labor productivity and caregivers, which is also considered by the ISPOR value inventory (Lakdawalla et al. 2018; Sanders et al. 2016a). In GCEA, we capture both of these broader elements, but distinguish between the productivity costs experienced by patients and caregivers through our separate productivity and family and caregiver spillover components, where the former only refers to productivity impacts on the patient and the latter concerns costs, utility impacts and productivity impacts to caregivers. Both the Second Panel and ISPOR value flower (through its “Net Costs” petal) recommend considering a wide range of costs, some of which appear within the GCEA as part of the dynamic net health system costs and direct non-medical costs petals.

Beyond what is provided in the Second Panel, GCEA also incorporates all novel value components that are part of the ISPOR value flower. More specifically, GCEA accounts for insurance value as part of the disease risk reduction component and combines ISPOR’s value of hope and reduction in uncertainty into the single outcome certainty value element. These new terms are used to better distinguish between both of these value components and because these components can now be fit into the broader generalized and risk-adjusted cost effectiveness (GRACE) methodology; within GRACE, disease risk reduction considers the ex ante health risks a healthy individual experiences whereas outcome certainty considers the variability on future health outcomes when already having the disease.

Lastly, GCEA makes further modifications and additions beyond the ISPOR value flower. Unlike the ISPOR value flower, while GCEA does not include distinct petals for QALY gains and disease severity, GCEA accounts for both elements in its patient-centered health improvements component. However, this petal recommends using generalized risk-adjusted QALYs (GRA-QALYs) rather than conventional QALYs or other alternatives of QALYs such as equal value life-years gained (evLYG). Moreover, GCEA introduced novel petals, including a societal discount rate, dynamic disease prevalence, and the value of knowing.

While in theory, all petals should be estimated for all intervention, in practice the feasibility of estimating the petals and their impact on overall treatment value estimates may vary depending on the application. In practice, there will be multiple ‘versions’ of the GCEA value flower and value assessments will likely fall somewhere between conventional CEA (i.e. just the stem or pistil), GCEA with only readily estimable petals included, and full GCEA (Figure 2). A treatment’s indication, treatment modality, and data available to researchers likely will influence the degree to which GCEA petals are included in any value assessment. The Second Panel on Cost Effectiveness in Health and Medicine noted, “cost-effectiveness analysis is not by itself a sufficient decision-making standard and that it does not capture all relevant concerns”(Sanders et al. 2016b); while GCEA moves us towards a more comprehensive view of value, researchers should still consider issues surrounding data accuracy, value comprehensiveness, and other limitations when conducting real-world decision-making.

Standard patient-derived QALYs and direct healthcare costs are accounted for in the center of the flower and therefore do not themselves have petals.

3.1 Outcome Certainty

As pointed out in the ISPOR Value Flower, risk aversion and prudence of patients play a significant role in decision making process of treatment choice(Lakdawalla et al. 2018). Therefore, the value patients place on treatments depends on their health outcome distribution and not just on mean values.

Risk-averse patients value treatments that reduce the variance of their future health (Lakdawalla and Phelps 2021). A survey of cancer patients seeking a second opinion at diagnosis identified risk in treatment outcomes and prognosis as a source of disutility (Hillen et al. 2017). Previous research on the underlying microeconomic theory of healthcare valuation demonstrates that smaller risk is a non-trivial component of treatment value to patients (Lakdawalla and Phelps 2022, 2021). To aid in understanding, consider two treatments for glaucoma: a standard pair of eyeglasses and a new surgical intervention. Assume eyeglasses improve every patient’s vision by 10 % (i.e. with certainty). The surgery, however, confers perfect vision to many patients but leaves some patients blind. Assume that, on average across all patients, surgery improves vision by 10 % despite the discrete outcome space. Conventional CEA would value these treatments equally because the average health improvement is 10 % for both, hence implying that patients are indifferent between the two options regardless of the differences in risk. However, a risk-averse patient may prefer the certain vision improvement offered by glasses to the range of possible outcomes attending surgery.

Rather than comparing average outcomes across treatments, a more complete analysis considering uncertainty in outcomes would compare certainty equivalents across treatments. A certainty equivalent is the guaranteed health benefit that a risk-averse or risk-loving consumer regards as equally preferable to a treatment’s distribution of potential benefits. A simple monetary gambling example is included in the appendix to illustrate this distinction (Appendix Conceptual Example: Outcome Certainty).

Figure 3 illustrates the difference between a certainty equivalent and an average health benefit. The blue and red curves represent the probability density functions for the health outcomes (measured by QoL) of the standard of care (SoC) and a treatment, respectively. Compared to SoC, the treatment has a greater average health benefit. Not only is the red treatment superior on average, but its variance is smaller (the red distribution is narrower than the blue distribution). Because of the reduced variance, the incremental certainty equivalent (ΔCE_QoL) exceeds the incremental average quality-of-life gain (Δμ _QoL), as patients benefit not only from the treatment’s increased average QoL but also from reduced risk (reduced variance).

Figure 3:

Conceptual representation of outcome certainty and certainty equivalence.

When considering treatments for severe, often terminal, illness with poor quality of life in the sick state, patients may place additional value on treatments that offer highly favorable, “tail of the curve” health outcomes. In the same example of the two glaucoma treatments, for a patient who has almost lost their sight, even if they are risk averse, they may be willing to try the surgery to get perfect vision rather than improving their vision 10 % from the current poor vision. A survey of patients with metastatic melanoma and metastatic breast cancer found that 77 % of patients preferred treatments with a possibility of highly favorable outcomes, with some patients willing to pay over $90,000 for such a treatment, despite the average health benefit being no better than the standard of care(Lakdawalla et al. 2012). A survey of patients with non-small cell lung cancer found that patients were willing to exchange a year or more of mean survival in exchange for the chance of durable, tail-of-the curve survival gains with both upside and downside survival risk(Shafrin et al. 2017). In Figure 3, if the treatment outcome distribution is right skewed, then ΔCE_QoL increases.

One key methodological advance for incorporating outcome certainty is GRACE’s ability to calculate the value of health gains while incorporating both preferences for reduced variance and increased positive skewness in the distribution of health outcomes through generalized risk-adjusted quality-adjusted life years (GRA-QALY). Rather than directly comparing the average health gains, GRA-QALYs instead reflect certainty equivalents (Lakdawalla and Phelps 2021, 2022). For a risk neutral patient, the GRA-QALY and the QALY are equivalent, as they seek to maximize the expected value of their future health due to treatment. However, for a risk-averse patient as an example, the gains in GRA-QALY can be more than those in conventional QALYs. Section 3.4.1 illustrates calculation of the GRA-QALY.

3.2 Disease Risk Reduction

Whereas conventional CEA focuses only on a health technology’s benefit to patients, individuals who do not currently have a given disease benefit from medical innovations that reduce their downside risk in the case that they are diagnosed with the disease in the future. The conventional CEA framework typically assesses the economic value of treatment from the perspective of the patients as the monetized health benefits net of treatment costs. However, the conventional framework overlooks the value of the treatment to the individuals who are not yet diagnosed with, but at risk of, the disease. These values include i) better expected health outcomes in the future and ii) reduced risk of the future health outcomes. The former value is represented by the expectation of the net treatment value of risk-neutral individuals. In addition, risk-averse individuals consider another value (the latter one) generated by the reduction in a utility loss from the uncertainty of having the disease, which is proportional to the net treatment value.

Specifically, for healthy risk-averse individuals, getting a disease or not in the future is a perceived risk and thus its impact on current utility is positively associated with how much their future quality of life could be negatively impacted by the disease. An introduction of a new treatment does not reduce the risk of contracting the disease itself but rather offers societal value by alleviating the loss of the quality of life incurred by the disease. Indeed, survey based methods have revealed that this risk reduction provides patients and non-patients alike with substantial value (McQueen et al. 2024). The ex ante value of the healthcare technology obtained by healthy individuals are referred to as insurance value(Lakdawalla et al. 2017). Just as in Section 2.1, quality of life is substituted with the certainty equivalent quality of life in the GRACE framework due to the convexity of the utility function of risk-averse individuals. Figure 4 provides a conceptual example of how healthy, risk-averse members of the general population derive insurance value from a novel medical innovation using certainty equivalent quality of life. A more detailed example is available in Appendix Conceptual Example: Disease Risk Reduction.

Figure 4:

Conceptual representation of disease risk reduction.

To perceive an intervention’s insurance value, it’s important to frame what risk it averts. When a person with allergies pays for an epinephrine autoinjector, deriving comfort from carrying it around and experiencing anxiety whenever they forget to bring it with them to a restaurant, they are clearly appreciating the insurance value of that technology protecting them from anaphylaxis. People with food allergies have also reported a preference for a non-injectable form of epinephrine because the prospective of having to self-inject can be a source of anxiety and reason for hesitating to administer a dose (Shaker et al. 2023). So the epinephrine provides insurance value against anaphylaxis whereas the non-injectable formulation provides insurance value against needle phobia.

Indeed, studies eliciting willingness to pay for generous insurance coverage for an innovative treatment among samples of the healthy^[2] general population show that, across indications, a substantial percentage of treatment value is captured by the disease-free population. In a stated-preference survey of willingness to pay for coverage of treatments for lung cancer, 89.8 % of the treatment value was accrued by individuals without lung cancer, who value improved survival if they are diagnosed with lung cancer in the future (Shafrin et al. 2021). Another stated preference survey of the healthy general population found that members of the general population were willing to pay $1,004 per year (compared to conventional CEA projections of $45.30 per year) for gene therapies in Duchenne muscular dystrophy (Shafrin et al. 2023). In a stated preference survey of three interventions for multiple sclerosis (MS), one third of the total value of the treatments was accrued to those without an MS diagnosis (Liu, Shim, and Lakdawalla 2016).

To incorporate the value from disease risk reduction into the value assessment, GCEA recommends estimating GRA-QALY and RASA-WTP under the GRACE framework. Unlike stated preference surveys that have frequently been used to estimate this value component, the GRACE framework is easily generalizable to a variety of disease and treatment combinations because no matter what the disease of interest is, once patients risk preference over quality-of-life metric and the loss of quality of life by the disease are given, GRACE can always account for the value from disease risk reduction.

3.3 Value of Knowing

Diagnostic tests provide value to patients, practitioners, and caregivers as improved knowledge can lead to better decision making. These better treatment decisions result in changes to patient health outcomes (e.g. QALYs) and cost. Consider a lung cancer patient who uses an improved diagnostic test to identify whether they have anaplastic lymphoma kinase (ALK) gene-positive tumors. Based upon this knowledge, patients can access targeted therapies, such as alectinib, which has improved median overall survival for ALK-positive patients compared to traditional chemotherapy (Jiang et al. 2022). Costs may also change; costs may increase if perhaps alectinib is more expensive than traditional chemotherapy, but may also fall if use of more targeted treatment leads to reduced hospitalizations and medical costs. The ability of diagnostics tests to impact health outcomes and costs through better decision making is already captured by conventional CEA.

However, patients may also derive value from new diagnostics simply from having more knowledge about their health even if the outcome of the diagnostic test does not change decision-making. This “value of knowing” is not included in conventional CEA. Suppose that a patient is diagnosed with Huntington’s disease. Since there is currently no treatment, the diagnostic enables no direct health benefit within the conventional CEA framework. However, patients may be able to take additional vacations, spend more time with family, or save money to cover the more intensive care required as the disease progresses. These choices might improve patient QoL compared to not knowing they had the disease. In addition to the planning value enabled outside of a healthcare context, studies have demonstrated that there is value in “knowing for knowing’s sake” to patients who have undiagnosed symptoms (Asch et al. 1990). More broadly, stated preference surveys have determined that many patients derive planning value from diagnostics, whereby they can make utility-maximizing financial, work, reproduction, retirement, and end-of-life care plans (Lee et al. 2010; Neumann et al. 2001). Current research has examined the value of knowing as it is accrued to family members, and family members have reported deterioration of their own health and quality of life as a result of limited information in cancer diagnosis (Lavallée et al. 2019).

The “value of knowing” need not always be beneficial to patients. For example, consider the case where a patient is diagnosed with amyotrophic lateral sclerosis (ALS). ALS results in progressive, full body paralysis and early death with no cure available. Simply knowing that their future health may inevitably be poor can cause distress, fear, and disutility for patients, even while their current health is quite similar to that of a healthy member of the general population. This “negative” value of knowing is evidenced by the existence of genetic counselors and the prescription of behavioral therapy for many patients of severe disease (Dugan et al. 2003; Kim et al. 2021; González-Martín et al. 2023).

To estimate the value of knowing, stated preference methods can be utilized to determine the willingness to pay to gain the psychological benefit and increased certainty for making future plans. These preference surveys follow both conventional and behavioral economic theory that patients are willing to pay for a diagnostic(Kahneman 1979; Lee et al. 2010). Two forms of stated preference survey can be utilized to elicit the value of knowing from patients. Contingent valuation methods describe a hypothetical screening to patients and ask individuals how much they would be willing to pay to have access to that screening (Phillips et al. 2006). Conjoint analysis requires participants to compare hypothetical scenarios by ranking, rating, or selecting scenarios after evaluating them across a collection of attributes (Phillips et al. 2006). The value of knowing would be any difference between the willingness to pay for the diagnostic test and the ex ante direct health benefit to the patient enabled by the improved diagnostic. For example, if a patient is willing to pay $75,000 for a diagnostic that enables a 0.5 QALY gain from better treatment selection, then using standard WTP for a QALY the value of knowing is $75,000 – (0.5QALYs)*($100,000/QALY) = $25,000. One could then readily incorporate this value into GCEA.

3.4 Empirical Application #1: Using GRACE to Estimate Treatment Uncertainty & Disease Risk Reduction

GRACE incorporates the value of treatment uncertainty and disease risk reduction into a single unifying framework. GRACE accomplishes this through three key steps: addressing the change in generalized risk-adjusted QALYs (GRA-QALY), measuring the risk- and severity-adjusted willingness to pay (RASA-WTP), and deriving value metrics for GRACE such as an ICER or risk- and severity-adjusted net monetary benefit (RASA-NMB) (Box 1). A description of each phase follows in the section below. For a full derivation of how to use GRACE to calculate RASA-NMB and ICER under GRACE see Lakdawalla and Phelps (Journal of Benefit-Cost Analysis 2023) study.^[3]

4.1 Dynamic Net Health System Cost

While most conventional CEAs assume a drug’s market launch price is constant over time (Whittington et al. 2023), in practice prices never remain constant. Drug prices are dynamic and change over their life cycle; prices may increase prior to the drug’s loss-of-exclusivity (LOE) and later decrease due to competition, especially post-LOE from genericization or biosimilar entry (Bhattacharya and Vogt 2003; Conti and Berndt 2016; Goldman et al. 2018). For example, prices for enalapril, fluoxetine and ranitidine in the Netherlands fell by 61 %, 51 % and 69 %, respectively, after the introduction of the first generic drug (Boersma et al. 2005; Vondeling et al. 2018). Other analyses indicate that for drugs in the US that lost exclusivity and experienced generic entry between 2002 and 2014, average price decreased by 51 % in the first year and 57 % after the second year (Aitken 2016). In specific markets such as oral cancer therapies, prices fell by 60–90 % within 12 months of LOE (Padula et al. 2016). Moreover, more substantial price reductions have been reported with increased generic entry over time, such as the 96 % price decrease of imatinib relative to its brand-name counterpart (Campbell et al. 2019).

To better incorporate pricing dynamics, we recommend the incorporation of dynamic net health system costs into the value assessment framework, but the manner in which pricing is incorporated may vary by type of technology (e.g. small molecule vs. biologic vs. cell and gene therapy) (McQueen et al. 2023). According to the FDA, price decreases after generic entry range from 54 % with two generic competitors to more than 95 % with six or more generic entrants (Conrad and Lutter 2019). Yet, when considering biologic drugs specifically, biosimilar entry often leads to smaller and more variable price decreases that depend on the biosimilars’ eventual uptake; price decreases one-year after biosimilar entry can range from 2.4 % to 59.3 %, with the low end of the spectrum largely due to low biosimilar uptake (Hernandez et al. 2022). Furthermore, the timing of genericization depends on the type of health technology. Previous estimates indicate that generally drugs in the US do not experience genericization for nearly 13 years after initial launch and, for more innovative drugs, this exclusivity period rises to around 14 years (Kesselheim et al. 2017); in Europe, this exclusivity period is estimated to be at 11 years and may be subject to change following new legislation going forward (Miglierini 2022; “Reform of the EU pharmaceutical legislation” 2023). Moreover, certain drug types – such as gene therapy products or drugs that require a special device for administration – may never experience genericization due to their developmental complexity (Kolchinsky 2017). In the US, the recently passed Inflation Reduction Act (IRA) granted CMS the ability to lower prices of some drugs essentially by fiat, which provides yet another reason for dynamic modeling even if a drug is unlikely to face generic or biosimilar competitors (Arad and McClellan 2022; Whittington et al. 2023).

Implementing price reductions due to genericization can be done through two methodologies: (i) based on a proportional reduction from a branded drug’s manufacturer price (e.g. wholesale acquisition cost, or WAC, in the US) or (ii) based on a multiplier applied to a health technology’s marginal cost of production. While the former is more commonly used, previous percentage reduction estimates may fail to reflect important drug pricing trends. This is particularly relevant since the size of rebates in the US, or discounts elsewhere, has increased the gross to net pricing spread (Hernandez et al. 2020), applying different pricing reductions is necessary for different types of treatments, and (iii) prices may vary based on the disease for which the treatment is indicated.

Regarding the first point, researchers that wish to apply proportional reductions to simulate generic entry should do so based on net – rather than gross – prices, or should rely on more recent data on proportional price reductions. Consider the case where a drug was priced at $100 and after genericization fell to $5; this represents a 95 % discount. Nowadays, however, this same drug may have a wholesale acquisition cost of $200, but with a $100 rebate, for a net cost of $100. If one applied the 95 % discount to the gross price ($200), one would estimate the generic price to be $10 (i.e. $200 × (1–95 %)), which is double the actual generic price ($5). Thus, one should either apply the historical price discount (95 %) to the net price (i.e. $100), if available, or use more recent data which would indicate that prices fall by 97.5 % (i.e.  $ 200 − $ 5 $ 200 = 97.5 % ).

Second, analysts should apply different price reductions from WAC for small molecules, biologics, and cell/gene therapies given price decreases due to genericization may be lower for biologic products since biosimilars are more difficult to manufacture (Makurvet 2021), to be approved by regulatory agencies, and there are limited data available on how prices have changed for cell and gene therapy products. Finally, price decreases as a percentage of WAC are likely larger for rare disease and precision drugs given they are often priced at higher multiples of their production cost than widely used drugs to generate a comparable return despite having fewer patients to treat (Makurvet 2021).

While less commonly used, estimating future drug costs after LOE as a multiple of a drug’s marginal cost of production may be a more appropriate approach. This approach is more aligned to economic theory; in competitive markets – such as a generic drug market when a highly profitable drug goes off patent and when the barriers to entry are relatively low – prices fall towards these generic drugs’ marginal costs of production in equilibrium due to competition between multiple generic manufacturers. However, the information needed to utilize this approach may be more difficult for researchers to access. First, the marginal costs of production are not well known outside of the drug manufacturer. However, researchers could estimate cost of goods sold (COGS) for the product of interest – or analog products if necessary – based on financial reports filed with financial bodies, such as the Securities and Exchange Commission (SEC) in the US.

Second, there is a lack of consensus in the literature about the appropriate multiplier to use to infer the equilibrium price based on COGS. The best study to date estimates that gross margins are approximately 50 % in the US, which implies that the prices of generic drugs are two times the COGS for generic drugs (Sood et al. 2017).

Finally, while this approach does align more closely to economic theory, in practice not all drugs experience considerable competition when losing market exclusivity. Even then, given the tendency for policymakers to eventually target old drugs that remain expensive with policy like the IRA, a GCEA of a complex biologic could earn a dynamic pricing petal by exploring in a sensitivity analysis the impact of potential policy that lowered the drug’s price closer to its estimated COGS so that it was clear how much of a novel drug’s societal value hinged on it dropping in price in the future.

4.2 Dynamic Disease Prevalence

In addition to changes in drug prices over time, conventional CEA methods often exclude how the prevalence of the disease changes over time. This concept of dynamic disease prevalence – or “dynamic prevalence” for brevity – becomes non-trivial for some diseases when model horizons in CEA are sufficiently long. For example, consider the case of curative therapies such as recent advances for treating the hepatitis C virus. The disease’s prevalence is projected to decrease over time such that a novel hepatitis C virus drug decades into the future will treat a smaller patient population and thus the cost of treatments to the system may fall dramatically in the long run (Razavi et al. 2014). In contrast, new treatments for Alzheimer’s disease in the future may prove more costly to the health system given that the global prevalence of Alzheimer’s disease – and thus the population in need of treatment – is estimated to triple to nearly 153 million by 2050 (Nichols et al. 2022). As an even more salient example, as the burden of drug-resistant infections to society is expected to climb over time, the value of new antibiotics that come to market today may rise in the future even if physicians are likely to prefer treating current patients with older antibiotics to maintain the novel antibiotics’ potency in the future (Shafrin et al. 2022). Proposals to offer significant incentives for antibiotic development in the form of guaranteed subscriptions make sense if considering dynamic prevalence whereas conventional CEA would likely conclude that such high spending to treat the few patients who need these drugs today is not cost-effective (Bennet 2023; Glover et al. 2023).

Accordingly, to account for both dynamic net health system costs and dynamic prevalence over time, GCEA recommends the use of “stacked” cohorts that include impacts to current and future patients starting treatment. The stacked cohorts should include (i) the changes in prices due to competition and generic/biosimilar entry and (ii) the changes in prevalence over time. Patients who initiate therapies in the future may face lower total drug costs because the treatments are closer to genericization; also, there may be more or fewer patients who initiate treatment in the future due to dynamic prevalence. An example of how to apply the stacked cohort approach is described in Section 3.3.

4.3 Empirical Application #2: Stacked Cohort Approach for Incorporating Dynamic Net Health System Cost & Dynamic Prevalence

To demonstrate how to incorporate dynamic net health system costs and dynamic prevalence into GCEA, consider the following stylized example, where a hypothetical brand-name small molecule drug, Drug A, has just been approved by the FDA and has entered the market to treat a hypothetical chronic disease. Patients treated for the chronic disease need to take Drug A for 15 years and the annual WAC is $10,000 per person upon market entry; further, Drug A is predicted to always be available on the market. Our goal is to develop a CEA model with a lifetime model horizon of 70 years that incorporates the aforementioned pricing dynamics; for our example walkthrough, we utilize a 40-year model horizon for simplicity but report results for both the simplified model and full 70-year model in Table 6.

Implementing pricing dynamics requires three key steps: incorporating (i) genericization (dynamic net health system costs), (ii) stacked cohorts, and (iii) dynamic prevalence (Box 2). In the first step, researchers will need to first identify the branded drug’s exclusivity period and the timing of generic entry (Whittington et al. 2023). Following Drug A’s classification as a small-molecule product, we assumed that the generic entry would begin 13 years post-approval in line with the IRA’s proposed drug negotiation timeline and estimates provided by previous studies (Goldman et al. 2023; Kesselheim et al. 2017). Second, researchers should incorporate changes in annual drug prices pre- and post-genericization. We assumed that (i) real drug price remains constant prior to LOE and (ii) generic entry leads to a 51 % reduction in the first year after LOE and then falls by 95 % in all subsequent years (Aitken 2016; Conrad and Lutter 2019; Gupta et al. 2019). Finally, GCEA incorporates the annual uptake of the generic drug by patients (Whittington et al. 2023). For simplicity, our example assumes that there is a 100 % uptake of the generic molecule after LOE or else that the brand drug’s net price drops close to that of generics (as one would expect payors to rationally negotiate, though recent biosimilar uptake and biologic pricing dynamics post-LOE are more variable) (Hernandez et al. 2022). When incorporating all these inputs for genericization, Drug A’s annual cost (undiscounted) is $10,000 per year through year 13, decreases to $4,900 in year 14 and further falls to $500 for years 15 and beyond (Table 5). This approach incorporates genericization but in a static manner; future cohorts who use the treatment have not been taken into account.

To reflect the initiation of Drug A by future patients, new patient cohorts are introduced each year in our example until 10 new cohorts have been included and are “stacked” over the model horizon. While in practice we highly recommend using additional cohorts (e.g. 70 cohorts if the model extends 70 years since patients will forever benefit from the upgrade to outcomes that a novel drug offers, even if it’s later displaced by better treatments), we selected 10 cohorts to simplify this empirical example. As shown in Table 5, this is implemented such that only one patient cohort (Patient Cohort 1) will take Drug A in the first year of the model, two patient cohorts (Patient Cohort 1 and Patient Cohort 2) will take the drug during year 2, three patient cohorts will take the drug in year 3, and so on. These added cohorts will experience the same decrease in Drug A’s price upon generic entry, such that patient cohorts that would enter the model in year 14 or later will only experience the reduced post-genericization price throughout their entire treatment period (Figure 5 and Table 5).

Figure 5:

Percentage of treatment duration (15 years) that each patient cohort takes the generic drug in the 70-year model.

To determine how the prevalence of a disease will change over time – and thus the expected size and weights of future patient cohorts – researchers must determine the annual incidence of the disease for each year of the model. The model will start with a patient population size based on current prevalence, but future incident patient populations will be added over time. Moreover, prevalence should be estimated specifically for the population intending to use the treatment. Our example assumes that the disease’s annual incidence will steadily increase from 10 % during the first year to 12.0 % in 10 years (or 39.2 % in the 70-year model) (Table 5). The patient cohort weights are calculated as W c W , where W _c is the cohort size and W = ∑ c W c is the total size across all cohorts. Following this example, Patient Cohort 1 gets the least weight for having the smallest disease incidence (and thus cohort size) while Patient Cohort 10 (or Patient Cohort 70 in the longer model) receives the greatest weight in the analysis given it has the largest disease incidence.

With the incorporation of the above model inputs, we find that there is a significant change in the resulting, undiscounted average annual per patient treatment cost of Drug A; while the results were undiscounted here, researchers should discount drug costs over time. When using conventional CEA and ignoring any changes in price – such that Drug A’s cost was fixed at $10,000 – per patient cost of Drug A over the model horizon was $150,000 (calculated as the sum of Drug A’s costs over the 40-year model horizon for a single patient cohort) (Table 6). When only incorporating price decreases due to genericization, overall, per patient cost decreased to $135,400 (calculated similarly as the conventional CEA case). Adding new patient cohorts into the model further decreased the overall cost to $92,650, which was calculated as an average of the total costs across all patient cohorts, where each cohort was given an equal weight (Tables 5 and 6).

Table 6:

Comparison of overall per patient treatment costs between different model conditions.

Model condition	Factors considered			Average annual per patient costs
	Genericization	Stacked Cohorts	Dynamic Prevalence	Simplified (10 cohorts and 40 year model)	Comprehensive (70 cohorts and 70-year model)
Traditional CEA (TCEA)				$150,000	$150,000
TCEA considering generic entry				$135,400	$135,400
TCEA considering generic entry and future patients				$92,650	$19,980
Generalized CEA (GCEA)				$91,099	$12,958

Finally, when also considering the increasing prevalence of the disease over time, the overall cost decreased to $91,099, reflecting that most patients affected by the disease will start the treatment in the future when the drug has already gone generic (Table 6 and Figure 5). This was calculated as a weighted average of each cohort’s total costs with the weights being the normalized cohort size of each cohort (Table 5). When considering the full 70-year model horizon, the incorporation of genericization and future patient cohorts along with dynamic prevalence had an even more pronounced effect on the per patient average annual drug costs (Table 6). However, for cell and gene therapies, this effect on average annual drug costs may be more modest given these technologies may be less likely to go generic.

4.4 Option Value

Option value^[5] occurs when a treatment’s ability to either improve survival, health related quality of life, or disease progression allows patients to benefit from future innovations (Li et al. 2019, 2022). To better understand what option value is, consider a hypothetical patient who is diagnosed with non-small cell lung cancer. Assume that survival outcomes are poor among patients receiving the standard of care and a novel treatment – called OncoDrug – that extends patient survival for an additional year (12 months) has just come to market. When patients are considering taking this treatment, there are a number of additional treatments under development which could further improve but will come to market at a later date. In fact, consider the case where eight months after the release of OncoDrug, a new breakthrough treatment – OncoCure – is released with superior efficacy. Despite the first treatment (OncoDrug) only improving patient survival directly by one year, patients who receive OncoDrug survive long enough to receive OncoCure, and therefore the true value of OncoDrug is much higher than one year of additional survival. In short, the health improvements enabled by the second treatment are only attainable for patients if they receive the first treatment and therefore a portion of the health benefits incurred from the second treatment represent the option value of the first treatment.

Generally, option value can be assessed via two clinical mechanisms: prolonged survival and slowing disease progression. Examples of prolonged survival models include those in cancer, amyotrophic lateral sclerosis, and similar diseases where the option value is derived from survival gains that allow patients to live long enough to see the new treatment approved. Option value models that slow irreversible disease progression to assess option value are relevant for some cancer treatments as well as degenerative neurological conditions like Alzheimer’s disease, Parkinson’s disease, and multiple sclerosis. By slowing disease progression, the initial treatment allows more patients to be eligible for future innovations which could help them maintain a better quality of life. In this case, patients who fail to receive the first treatment may progress to such as poor health state that the benefit of subsequent may be minimal.

Empirically option value can be measured from two perspectives: ex ante and ex-post. Ex-post approaches are more accurate and can be useful for future price negotiations, such as those that will take place under IRA (Arad and McClellan 2022). The ex-post methodology utilizes real-world data (e.g. claims-based data) to quantify changes in healthcare utilization and patient survival (Thornton Snider et al. 2018). For instances where GCEA is used in this context, an example can be found in Section 9.1.2. Yet, in many cases, economic evaluation occurs to estimate the value of upcoming or newly approved treatments.

Therefore, the real-world data required for ex-post analyses are not available to researchers conducting value assessment. Nor do they know what innovations will be brought to market in the future. For this reason, GCEA recommends utilizing ex ante option value which relies on a series of assumptions about the efficacy, time to approval, and probability of approval for future innovative treatments that could benefit patients in the future (Becker, Murphy, and Philipson 2007; Lee et al. 2022; Li and Garrison 2020). While this approach is less accurate, it is more useful for decision-makers as ex ante option value can be estimated at the time of treatment launch. Moreover, necessary parameters can be readily estimated from data on the clinical trial pipeline, public statistics published by the FDA and others, and peer-reviewed publications. A key limitation of ex ante option value models is their reliance on stochastic parameters. Thus, a series of sensitivity analyses are required to address the underlying uncertainty of these parameter estimates. The differences in these approaches are summarized in Figure 6.

Figure 6:

Conceptual differences in real option value methods.

Option value provides non-trivial value in a variety of settings. In a study of the ex-post option value of ipilimumab in metastatic melanoma, patients diagnosed in 2013 experienced option value amounting to a 49 % increase in survival gains (Thornton Snider et al. 2018). Another study found that patients diagnosed with HIV/AIDS in 1995 and 1996 were willing to pay 400 % of the conventional value of azidothymidine therapy, because treatment with azidothymidine would allow them to access more effective highly active antiretroviral therapy later. In a separate, ex ante study of the option value of ipilimumab, incremental QALYs increased by 6.2 % while incremental costs only increased by 3.8 %, resulting in a 2.3 % decrease in the incremental cost-effectiveness ratio (Li et al. 2019). A separate ex ante study found that anaplastic lymphoma kinase gene-positive (ALK-positive) patients treated with first-line alectinib saw an increase in incremental QALYs of 12.9 % when accounting for the option value of the non-small cell lung cancer pipeline.

Implementing ex ante option value in practice requires four steps (Box 3). First, researchers should assess the time period when the focal drug would provide option value to patients. For example, if a clinical trial reports a 95 % confidence interval for a delay in disease progression, the mean value would be used to model the applicable time horizon for option value.

Second, researchers should forecast when the next innovative treatment targeting the same condition will become available to patients as well as its expected costs and benefits.^[6] For some diseases and treatments, such as Alzheimer’s disease, where innovation is slow and current interventions do not provide substantial health gains, treatments may not provide option value to patients. If a disease, such as lung cancer, has a rapid development pipeline and existing treatments have proven efficacious, researchers should use available data on drug approvals to forecast how far in the future the next innovative therapy is likely to enter the market.

The future health benefits, costs, and arrival of the predicted innovation can be forecasted by using either the trend or pipeline approaches (Figure 7). The trend approach forecasts survival and health-related quality-of-life trends based on their past trends on clinical outcomes and the demographical changes of the target population. Although this approach utilizes a variety of sources to produce robust projections of future efficacy, it is ultimately data intensive and relies upon the assumption that future trends can be predicted from past data. The pipeline approach requires early (or late if possible) clinical trial results and manufacturer pipeline data to forecast which treatments may become available to patients within a relevant time horizon and how effective the future treatments will be. This method assumes that the observed efficacy in early clinical trials is representative of the treatment’s true clinical benefit upon approval.

Figure 7:

Forecasting innovation for ex-ante option value.

The third step predicts the probability of treatment approval, the timing of approval, and real-world uptake (conditional on approval) for the future treatment. The probability and timing of approval can be obtained from literature-based estimates. For the uptake rate, one can use real-world data from disease analogs to estimate the likely speed at which new innovations are adopted. Additional precision can be gained by adjusting the expected time of arrival through the use of surrogate endpoints and special regulatory designations, such as orphan drug or breakthrough designations.

Finally, option value is calculated as net monetary benefit after allowing patients to begin utilizing the novel treatment in the future based on the expected value of the health benefit of the innovation. As ex ante option value largely relies on the stochastic parameters such as the likelihood of regulatory approval, the expectation of efficacy, the predicted time to market entry of new medical technologies, etc., it is imperative to confirm internal validity by conducting sensitivity or scenario analyses. A numerical example can be found in Section 9.1.

Incorporating option value into GCEA requires minimal changes to the overall model structure. At the time of expected arrival of the new treatment (determined in step 1 of Box 3) conditional on the specified probability of approval (determined in step 3 of Box 3), patients who are alive and eligible will begin to receive the new treatment. These patients, in both the treatment and comparator groups, will receive the health benefits (and incur subsequent costs) of the innovation (determined in step 2 of Box 3). In addition, option value can be combined with dynamic prevalence by running Monte Carlo simulations to model how patients who took the focal drug in subsequent year’s cohorts accrue option value.

4.5 Scientific Spillover

Treatment innovations provide value to society through knowledge spillovers or “scientific spillovers” that manufacturers can build upon to make future innovations. In fact, pharmaceutical firms routinely build on insights obtained from previous drug innovation, both successful and failed attempts. For example, the testing of compactin – the first cholesterol-reducing statin that was tested in animals in the 1970s – was discontinued due to the drug’s safety concerns. However, compactin’s failure did not mean the new statin class was without potential. After analyzing the issues, Merck determined the safety issues were specific to compactin but that the drug class still looked promising. Therefore, Merck resumed its own development of a chemical analog to compactin. That effort resulted in the approval of lovastatin, which then spurred further development of other chemically similar statins including simvastatin and atorvastatin (Frankel et al. 2023).

Moreover, these knowledge spillovers can also occur across different therapeutic classes; for instance, treatments for one autoimmune disease can incentivize treatment innovation for multiple different autoimmune diseases (Xie et al. 2022).

In other words, both failed and successful drug development efforts can generate new knowledge that catalyzes follow-on drug development that might prove even more valuable than the predecessor drug candidate. While a drug’s structural differentiation is not a direct source of value, novelty indirectly creates value as a source of exploration. Applying an old drug to a new disease area, new biological target, or new combination therapy, might also generate new scientific evidence that helps open up new avenues for research and/or new drug classes.

While the scientific spillover petal make sense intuitively, recent research has quantified the impact of scientific spillovers in the real world. Frankel et al. (2023) utilized the Tanimoto score – a measure that quantifies the share of chemical features a drug has with previously approved molecules – to link drug novelty to spillovers; using this score, the authors determined the number of successor drugs inspired by a focal drug and then calculated the sum of these successors’ annual revenues as the focal drug’s scientific spillover (Frankel et al. 2023). Moreover, the study demonstrated that more novel drug candidates generated larger spillovers than less novel drug candidates, even though novel drug candidates are themselves less likely to reach the market.

However, these methods have significant limitations for their use in CEA; the described method does not measure spillovers to health benefits directly, and the Tanimoto score is not applicable to biological drugs. While the former challenge remains underdeveloped in literature, the latter can be resolved using alternative methods that consider all drug types. Lanthier et al. (2013) developed an approach to group all new drugs into classes – based on FDA pharmacologic classes and the original indications of the drug – and to qualitatively measure the novelty of drugs within these classes as either “first-in-class” (first drug developed in a drug class), “advance-in-class” (drug provides major advancement in its class), or “addition-to-class” (drug is an incremental innovation) (Lanthier et al. 2013).

Alternatively, drug novelty can be measured on an order-of-entry basis for drugs, where drugs are grouped into classes with similar mechanisms of action and earlier drugs within these classes can be considered as more novel. Finally, one can extend the drug similarity method to biologics using historic pipeline data. For biologics with a novel target, the analyst could calculate the expected number of total/successful follow-ons based on the histories of novel biologics in similar disease areas.

We recommend that a new drug’s knowledge spillovers are incorporated into GCEA qualitatively by discussing the degree of drug novelty and likelihood of inspiring follow-on R&D (Box 4); scientific spillovers will not be directly incorporated into the estimate of treatment value, but should be analyzed by researchers (Frankel et al. 2023). Researchers can measure drug novelty using the methods described above or through expert clinical opinion; when possible, we recommend measuring novelty quantitatively using chemical similarity scores (e.g. Tanimoto) or based on the development entry order into a specific biological mechanism-of-action. Such measures provide an objective approach in measuring drug novelty compared to previous methods, which largely relied on the frequency of relevant patent citations (Frankel et al. 2023; Lanthier et al. 2013).

Qualitative assessments might further incorporate information about scientific excitement around a particular new drug or mechanism. How many scientific papers and distinct author groups publish papers during the development of the novel drug? If the scientific exploration around a novel drug target extends beyond the sponsoring pharmaceutical firm, the drug’s development is likely to generate knowledge and development spillovers beyond that focal drug – regardless of whether the focal drug succeeds in trials.

While more research is needed before implementation, researchers could alternatively consider adjusting the CE threshold used to value health gains in GCEA to increase by larger amounts for more novel drugs to reflect the drug’s larger spillover effects. Adjusting the CE threshold in this way reflects the fact that while more novel drugs can provide larger spillovers, these drugs are often riskier to successfully develop (Frankel et al. 2023); CEA should assign greater value to more novel drugs – beyond just their potential to bring larger patient health benefits – to reward and incentivize manufacturers for more novel innovations. This improves the efficiency of resource allocation by having the positive externality innate in a firm’s R&D investment decision-making process.

Moreover, as another alternative to the described approach, researchers could also estimate and incorporate a drug’s qualitative level of scientific spillovers into CEA based on its relative novelty using multi-criteria decision analysis (MCDA) (Gongora-Salazar et al. 2023; Phelps and Madhavan 2017; Thokala and Duenas 2012; Thokala et al. 2016). Using MCDA when groups are the decision-maker requires an additional expertise in social choice theory (Phelps and Madhavan 2021). While estimating the scientific spillovers value component is real, due to significant uncertainty in terms of approaches for forecasting this value and uncertainty around how a CE threshold could be adjusted, at this point we recommend including scientific spillover in a decision-maker’s set of contextual factors.

4.6 Societal Discount Rate

Discount rates are used in value assessment to indicate societal preferences for health benefits that occur sooner (rather than later) and to defer costs into the future. Observed human behavior provides an empirical and economic rationale for discounting both health costs and gains in health technology assessments, as surveyed people express a preference for benefits to accrue in the present and costs to occur in the future (Claxton et al. 2006; Gravelle and Smith 2001; Johannesson and Johansson 1997; Parouty et al. 2014). People who elect to invest in their future health face a clear opportunity cost by exchanging consumption of other goods and services for improved health (Claxton et al. 2006).

In recent years, conventional CEA practitioners most frequently employ a constant 3 % discount rate on both health benefits and costs. The practice of applying equal discount rates to health costs and benefits ensures that two interventions initiated at separate times, but with equal costs and benefits, will receive equal priority if the value of health is constant over time (Jit and Mibei 2015). The use of a rate of 3 % is often justified because past assessments utilize a constant rate of 3 %, and therefore future studies should do so as well to maintain comparability. In addition to academic precedent, many HTA bodies (e.g. NICE, ICER) recommend the use of a 3 % discount rate.

While important to reflect temporal preferences and promote comparability, the use of discount rates (and in particular the rate selected) has a significant impact on treatment value, particularly when costs and benefits are asynchronous. For example, pediatric indication and vaccine treatment benefits accrue over time, while treatment costs are a one-time event in the present period. In this case, standard discounting can significantly reduce the perceived health benefit compared to costs. The changes in the discounted QALYs and ICER in Table 7 are adapted from a CEA for human papillomavirus vaccination (Westra et al. 2012). The costs of vaccination take place in the present period for this study, while the benefits of immunity accrue well into the future. Under conventional, constant universal discounting of costs and benefits at a rate of 3 % (as used by most value assessments), the ICER increases by nearly five times the undiscounted value.

GCEA recommends using an empirically derived societal discount rate based on the positive approach or the normative approach, rather than simply following previous precedent (Box 5). The positive approach considers health as a tradable asset and therefore spending on healthcare can be interpreted as an investment in an asset whose opportunity cost should be compensated by future returns in health, represented by a risk-free interest rate. In a survey of experts on social discounting, the median reported value of the risk-free rate was 2 %, but suggested values ranged from between 0 % and 6 % (Drupp et al. 2018).

To employ the positive approach, the literature recommends utilizing the real rates of return on government bonds as a proxy to the risk-free rate. The rate selected should follow the time horizon specified within a value assessment. For example, if a lifetime horizon is used (commonly assumed to be 30 years), the 30-year rate on government bonds should be used as the discount rate. Although bond yields have in the past been consistent with a recommended annual discount rate of 3 % percent (e.g. when the Second Panel on Cost Effectiveness in Health and Medicine published their guidance in 2016), a trend toward lower real bond yields suggests an annual discount rate of 2 % for health technology assessment is now appropriate (Cohen 2024).

The normative approach utilizes the Ramsey equation to derive the societal discount rate that attains Pareto efficiency in an intertemporal optimization problem. The Ramsey equation implies that in the steady state of the Ramsey-Cass-Koopmans neoclassical model, the long-term social discounting factor (or the opportunity cost of the investment in the risk-free capital, i.e. risk-free rate) should be equivalent to the optimal rule of the intergenerational distribution of the marginal return on the risk-free capital. The determinant factors of the (extended) Ramsey equation include (i) societal pure time preference, (ii) wealth effect, which is a product of the consumption growth and risk aversion, and (iii) the precautionary effect which explains the uncertainty in consumption growth (Drupp et al. 2018).

The Ramsey equation is a useful alternative to the positive approach, despite some discussion of the limitations. First, uncertainty exists among several key determinants of the optimal discount rate such as the annual consumption growth rates, elasticity of the marginal utility with respect to consumption, and volatility of economic growth. Moreover, in a survey of economic experts, there existed significant differences between the rates calculated through the Ramsey equation (median = 3 %) and responses by the economic experts on their perceptions of the correct social discount rates (median = 2 %) (Drupp et al. 2018). This implies the Ramsey equation may not in fact be preferred in practice, but some additional research is needed.

Researchers may also use a combination of the positive and normative approaches to implement a societal discount rate within GCEA. In a survey of economists, 92 % determined that the societal discount rate fell between 1 % and 3 % with a median rate of 2 %(Drupp et al. 2018). This aligns with the 30-year rate suggested by the positive approach. Because model horizons may differ from 30 years and market conditions may evolve, if researchers use a 2 % rate at baseline it should be supplemented with sensitivity analyses utilizing rates between 1 % and 3 % (Drupp et al. 2018). Importantly, projections of slowing consumpton growth rates suggests that the all-important wealth effect term on the right side of the Ramsey Equation is decreasing, pointing to a lower discount rate based on the Ramsey framework (Cohen 2024).

Because the social discount rate depends on the societal preferences for trading off near-term and long-term health benefits and costs, there is no single theoretically supported consensus on a particular discount rate to use in health technology assessments. Therefore, researchers should justify why a particular discount rate and method are selected.

5.1 Patient-Centered Health Improvements

Making health value assessment more patient-centered has been a goal for many organizations, from patient advocates to researchers and beyond. Existing methods incorporated into conventional CEA often fail to capture a patient’s experience navigating treatment options. As the concept of patient-centeredness evolves, patients and other stakeholders must be engaged throughout both the treatment process as well as the establishment of best practice guidelines. Within GCEA, a central principle of the Patient-Centered Health Improvements petal is that the measurement of survival gains should not discriminate against those with disabilities.

Patient-centeredness faces a fundamental challenge: value assessment evaluates treatments at the population level, but treatment value may be unique for each person. For instance, some patients may prefer cancer treatments that extend survival regardless of the severity of adverse events incurred; for other patients, quality of life is paramount, even above and beyond survival gains.

While fully patient-centered health improvements are a multi-faceted challenge and will vary across diseases, GCEA argues that any quantitative measure of estimating the magnitude of health benefits should ideally capture at a minimum four key components. First, the measure should be built on microeconomic theoretical foundations. In other words, the specific health benefit measure should come from a conceptual framework based on individual patient preferences. Second, any metric used to measure health benefits should capture both quality-of-life and survival gains. Third, the health gains for patients with more severe diseases should be valued more than identical health gains for less severe diseases, as multiple studies have found that individuals are willing to pay more for identical QoL gains when these QoL gains occur from more severe disease states (Bleichrodt et al. 2003; Gyrd-Hansen 2005; Nielsen, Gyrd-Hansen, and Kjær 2021). Finally, survival gains should not discriminate against the disabled. Many health benefit metrics value life extensions less when these life extensions occur in a lower quality-of-life health state.

In the conventional CEA framework, the standard QALY approach has been used to measure health benefits. However, it possesses several limitations. First, although the use of the QALY is rooted in micro-theoretic foundations, it relies on an assumption of patient risk neutrality, which makes the QALY deviate from patient-centered sources of value (Lakdawalla and Phelps 2020). Since solving for the utility maximization problem under the risk-neutrality assumption, the conventional CEA framework equates the marginal gains in utility with those in health which are represented by QALY gains. However, it is typically assumed that patients are risk averse and thus, the conventional CEA based on the QALY approach does not give the true value of a treatment. Second, the standard QALY approach values all quality-of-life gains equally no matter what the disease is severe, or the baseline health of patients is poor (Bleichrodt et al. 2003; Gyrd-Hansen 2005; Nielsen, Gyrd-Hansen, and Kjær 2021). These may underestimate the value of a treatment of a severe disease or that for patients with low health endowment such as disabled or elderly.

The efforts to make value assessments patient-centered include life-years gained, decision modifiers, equal value life year (evLYG), and more recently the GRACE framework through GRA-QALY (Campbell, Whittington, and Pearson 2023b; Excellence 2022a; Luo et al. 2009). GCEA evaluates each approach by four criteria (Table 8).

Table 8:

Comparison of frameworks for addressing patient-centered health outcomes.^a

Measure of Health Benefit	Factors Considered
	Based in micro-theoretic foundations	Incorporates QoL and survival gains	QoL gains for severe disease valued more	Survival gains for disabled equivalent to able bodied
QALY
LY
Decision Modifiers
evLYG
HYT			-
GRACE

1. ^a *Yes, if patients are risk neutral. +Yes, if treatment has no QoL implications. −Yes, if QALY shortfall is met.

Analyses based on life-years gained simplify the health benefit of treatment by considering the survival benefit of treatment in isolation. Instead of weighting patient survival by their quality of life, life-year gained analyses sum patient survival gains to evaluate health benefit. By evaluating treatment benefit in terms of survival, life-year gained analyses value health benefits equally regardless of a patient’s baseline disability. By construction, life-year gained analyses exclude quality-of-life improvements that result from treatment. This implies that these analyses do not recognize the additional value of treating severe disease when treatments improve patient quality of life, despite being a source of value for patients. Life-years gained thus only aligns with microeconomic theory if and only if a treatment provides no quality-of-life benefit or patients do not care about quality-of-life gains (unlikely). Furthermore, life-year gains can be limited by factors unrelated to treatment efficacy, such as patient age and disease-specific mortality risks.

The equal value life year (evLYG) metric aims to address some of the limitations of the life-years gained metric and of the QALY metric by evaluating survival gains at a highly favorable, and uniform, assumed quality of life during periods of life extension. In the evLYG approach, additional life years are weighted by a QoL of an average individual within the general population (0.851) regardless of actual patient disability (Campbell, Whittington, and Pearson 2023b; Luo et al. 2009). The evLYG metric differs from life-year gained analyses because the evLYG considers the quality-of-life gains of treatment during periods of shared life, and only remains agnostic of disease severity or baseline disability during the additional survival gains period. By assuming the quality of life is the same during life extension for all conditions and individuals, the evLYG attempts to align treatment valuation with priorities for treating life extension equally (Nord et al. 1999; Representatives 2010; Rodgers 2023).

Although multiple CEA studies have included the evLYG, the decomposition and subsequent unequal weighting of health gains due to quality of life and survival is not based in microeconomic theory, and evLYG violates the weak and strong forms of the Pareto principle (Lin et al. 2023; Review 2023a, 2023b). As a result, evLYG biases value assessments against treatments that primarily improve patient quality of life in life extension. Additionally, the quality-of-life gains measured by evLY are assessed agnostic of disease severity or patient baseline health endowments. Finally, debate exists on the true willingness to pay for an evLYG or life-year gained and a range of estimates exist in the literature (Acaster et al. 2013; Campbell, Whittington, and Pearson 2023a, 2023b; Lakdawalla et al. 2018; Lakdawalla and Phelps 2020, 2021; O’Day and Mezzio 2021; Sawhney and Thakur 2023).

The Health Year in Total (HYT) metric attempts to address the limitations of the QALY by measuring patient health improvements on an additive scale (Basu et al. 2020). Similar to evLYG, HYT decomposes health gains into incremental life-year gains and QoL gains. However, rather than measure QoL gains over the period of survival in the comparator arm, HYT evaluates these gains for the total number of LY in the treatment arm. To do so, HYT makes two axiomatic assumptions. First, HYT assumes utility independence such that patient preferences over survival gains under uncertainty are not dependent on the quality of life enjoyed during survival, nor are preferences over QoL gains dependent on fixed longevity intervals. Second, HYT assumes a constant proportional tradeoff between QoL and life-years and as a result a patient’s marginal rate of substitution between additional survival and additional QoL is independent of that patient’s remaining life years. As a result of these assumptions and the additive framework used to construct HYT, the number of HYTs a patient can gain in a given year ranges from 0 to 2.

HYT – like evLYG – may be especially attractive for health technology assessment by CMS under the IRA since May 3, 2024 guidance documents states that QALYs cannot be used to inform maximum fair prices (Centers for Medicare and Medicaid Services 2024); however there are some potential logical inconsistencies that researchers who use HYT would need to be aware of. First, HYT can violate the independence of irrelevant alternatives axiom that is a fundamental pillar of rational choice theory, because survival gains in the comparator arm are dependent on the survival properties of an alternate, mutually exclusive treatment (the intervention) (Paulden et al. 2024). Second, the use of QoL estimates to compute patient utility requires that patients are alive and thus the counterfactual QoL in the comparator arm relied upon by HYT violates this definition. Third, HYT assumes that patients who experience additional survival experience the same QoL that they had prior to the period of additional survival. However, patients experiencing additional survival may have poorer health during survival compared to the health related QoL, among the alive, in the comparator arm. Fourth, empirical estimates suggest that HYT’s assumption of constant proportional tradeoffs is violated when patients are risk averse, as a patient’s marginal rate of substitution is correlated to their health allocations and risk preferences (Lakdawalla and Phelps 2022, 2020). Finally, additional research is required to establish cost-effectiveness thresholds for an additive scale, such as HYT, and at the time of publication the authors of this paper are unaware of any health technology assessments relying upon HYT to measure health gains (Basu et al. 2020).

Rather than replacing the standard QALY, NICE issued guidance on implementing decision modifiers in the assessment of rare and severe diseases. These modifiers increase willingness to pay for the treatment of severe disease if an indication meets a QALY shortfall threshold (Excellence 2022a). NICE defines absolute QALY shortfall as the difference in the level of QALYs to individuals with disease compared to the number of QALYs to comparable individuals (i.e. individuals with similar demographics) without the disease. If a disease meets this definition of severity, a scalar multiplier is applied to the modeled willingness to pay for a QALY to approximate the additional treatment benefit of treating severe disease. Because decision modifiers rely upon the QALY to initially calculate health benefits and QALY shortfall, modifiers incorporate both quality-of-life and survival gains in computing patient-centered health outcomes.

However, decision modifiers contain a number of limitations which make them poorly suited for implementation in patient-centered value assessment frameworks. First, they do not address all severe diseases, particularly those affecting older populations. For example, Alzheimer’s patients (who have an average age of onset of 80 years old for sporadic AD) are drawn from a population of patients with low expected QALYs and therefore do not meet the required thresholds for implementing severity modifiers. Second, the literature has also criticized severity multipliers for often being politically driven and containing logical inconsistencies (Paulden et al. 2014).

Rather than addressing the value of treating severe disease systematically, the decision to utilize a multiplier as well as the magnitude of the multiplier selected is often made arbitrarily. A consequence of this is an increased opportunity cost for other patient populations, whose treatments may already be under-valued due to bias against disability. For example, drugs approved under NICE’s end-of-life multiplier delivered an annual 12,401 QALYs to late-stage cancer patients but had an opportunity cost of 18,330 QALYs to patients with other diseases (Charlton 2023).

On the other hand, the GRACE framework uses a qualified patient-centered health benefit metric, GRA-QALY. The quality-of-life gains of GRA-QALY is a function of certainty equivalent obtained by relaxing the risk-neutrality assumption, which better fits consumer choice theory. Also, the GRA-QALY incorporates not only the quality-of-life gains but survival gains, which are calculated using the exchange rate between life years and QALYs. GRA-QALYs are monetized using RASA-WTP which models patients’ increased willingness to pay for health gains for severe disease. Furthermore, the monetary value of the GRA-QALY gains can address the issue of the conventional CEA regarding discrimination against individuals with poor health (e.g. disabled, elderly) because RASA-WTP additionally accounts for permanent disability at the baseline period.

Rather than use methods that inconsistently or incompletely address the multiple dimensions of patient-centered care, GCEA recommends using GRACE to incorporate additional value from patient-centered health outcomes. To address any potential limitation of GRACE, GCEA additionally recommends conducting life-year gained analyses as a sensitivity analysis. A discussion of implementing GRA-QALYs can be found under Section 3.4.

5.2 Equity

Cost effectiveness approaches – such as both conventional CEA and GRACE – capture treatment benefits to patients but often ignore how health benefits accrue differently across heterogenous patient populations. Aside from the additional value to patients of reducing health disparities across patient subgroups, there is a clear societal value in improving health equity. Numerous studies have demonstrated societal preferences for placing a higher value on health gains for disadvantaged individuals. Policy decisions (e.g. the creation of Medicaid and Federally Qualified Health Centers for low-income individuals) also suggest that society has strong willingness to pay for increased health equity (Lakdawalla et al. 2018). The passage of the Affordable Care Act expanded insurance access to an estimated 29 million US residents, many of whom represented minority groups. Because of the societal eagerness to promote equity considerations across institutions, health disparity reductions provide additional treatment value to be considered in value assessments.

Consider the case where there is a hypothetical disease called “billionaire-itis” which causes depression in the wealthiest of individuals and another disease called “low-income depression” which only occurs in individuals at the bottom of the wealth distribution. If treatments for both diseases came to market with identical safety and efficacy, conventional CEA would value each treatment equally. However, many members of society would say that the treatment for “low-income depression” patients should be valued more than the treatment for “billionaire-itis” because low-income individuals have relatively worse health outcomes at baseline. As such, the use of the new treatment for “low-income depression” would help reduce health disparities and add to societal value; conversely, the treatment for “billionaire-itis” would exacerbate health disparities, which would attenuate the treatment’s value to society.

Healthcare disparities inspire constant policy debate, with efforts such as the Affordable Care Act adopted to expand care access to under-represented groups. More recently, the COVID-19 pandemic renewed interests in health inequity, as many minority communities were disproportionately impacted by the pandemic. From a healthcare resource utilization perspective, Black and Hispanic individuals often report the highest rates of cost-related delays in care and lower access to high-quality medication. Furthermore, studies assessing health disparities in oncology have found that low socioeconomic status is correlated to poor access to high-quality care, lower screening rates, delays in treatment after diagnosis, and lower treatment adherence–all of which result in diminished patient health outcomes. As a result of these disparities additional benefit of medical innovation is derived from a treatment’s ability to reduce health disparities.

To incorporate the value of reduction in health inequality within a unified value framework, GCEA recommends using distributional cost-effectiveness (DCEA) (Box 6). DCEA enables researchers to estimate the health benefit of treatment while accounting not only for health disparities but financial disparities that exist between patient subgroups. DCEA relies upon societal preferences for reducing health disparities, and as a result awards more value – compared to conventional CEA – to treatments that disproportionally benefit underprivileged patient populations (Love-Koh et al. 2019). Societal preferences for reducing health disparities can increase the value of treatment by up to 3.38 times as many QALYs per patient (Love-Koh et al. 2019). In a study assessing mandates for lung cancer screening, approximately half of the treatment value was attributable to reductions in health disparities. Health equity improvements due to funding COVID-19 treatments resulted in a population-level health effect of more than 130,000 QALYs (Kowal et al. 2023).

Implementing DCEA to evaluate health equity in a US context requires a multi-step approach and the literature provides well-founded assumptions for where parameter data is lacking (Kowal et al. 2023). Box 6 provides one set of potential steps, but implementation can vary using the numerous methodologies available. Consider, for example, that we want to incorporate the societal value of a new treatment developed that cures lung cancer for patients with a history of smoking and tobacco use. Assume that among lung cancer patients there are two patient populations of smokers, wealthy smokers and poor smokers, and that the literature has documented that wealthy smokers have better outcomes at baseline than poor smokers.

The first step of DCEA is to identify and justify the selection of equity-relevant subgroups of patients based on demonstrated health inequities in the literature (Kowal et al. 2023). In the United States, evidence exists on baseline health disparity across race, ethnicity, and geography, but researchers should consider additional strata such as income quintile, level of educational achievement and other patient characteristics or combinations of traits relevant to the condition or population of interest. A number of indices exist for measuring health disparities, such as the social vulnerability index and more recently developed social vulnerability metric. ICER has developed the health improvement distribution index (HIDI) to examine differences in the composition of patient populations; but HIDI is an aggregate measure of representativeness, not a measure of disparities, and HIDI cannot be used in DCEA. Unfortunately, a variety of data needs exist to implement DCEA to inform healthcare decision-making in the US context. For example, mortality and baseline QALY data is unavailable for many races, ethnicities, as well as for Americans residing in rural areas. Furthermore, clinical trials used to assess treatment efficacy are often not representative of all types of patients while poor real-world evidence exists on patient outcomes across heterogenous patient populations.

Second, once sub-groups are selected, a conventional CEA model is used to calculate the group-level incremental QALYs and costs for both the intervention and comparator arms of the model.

Third, after calculating sub-group costs and benefits, the results are used to compute the subgroup-level net health benefit (NHB) of the intervention and comparator groups, which takes the form shown in (Equation (6)):

Where g denotes the subgroup, s represents the intervention or comparator group, and WTP refers to the standard willingness to pay for a QALY.^[7] In the example of a treatment that cures lung cancer in smokers, one would simply apply the treatment benefit in a conventional CEA framework but use subgroup-specific measures of quality of life, pharmaceutical costs, comparator clinical efficacy, medical costs and societal costs. For example, poor smokers may have worse overall quality of life due to untreated comorbidities or experience difficulty taking time off from work to receive regular infusions.

Fourth, researchers should select an appropriate inequality aversion parameter based on their model setting and leverage the sub-group specific data to estimate the opportunity costs for each patient subgroup. The most widely used metric for measuring health outcomes at baseline is quality-adjusted life expectancy (QALE). QALE represents a patient’s expected QoL accumulated over the duration of their survival. This differs from a QALY, which measures the direct health benefit accrued to a patient after getting a disease; instead, a QALE measures the quality-of-life-adjusted life expectancy for individuals from birth. To calculate subgroup-specific QALEs of the intervention and comparator groups, the baseline QALEs of each subgroup need to be calculated first by taking a weighted average of the subgroup, age-specific QoL. Age-specific QoL estimates can be obtained from publicly available data sets of healthcare survey data (e.g. Medical Expenditure Panel Survey), where the weights are the subgroup age-specific survival rates from the US Census data. Returning to the example of poor and wealthy smokers, wealthy smokers may have longer life expectancy as they can afford more preventative care over the course of their lives, and higher age-related quality of life as they enjoy more vacations than poor smokers. By isolating baseline health expectations and sub-group specific net health benefit, researchers can calculate the expected impact treatment has for both wealthy and poor smokers individually. For intervention and comparator groups, the subgroup-specific QALEs are derived by adding the subgroup-specific NHB which is equally distributed to the general population of the subgroup (Equation (7)):

where N _g represents the population size of the subgroup.

Fifth, for each of the intervention and comparator groups, calculate the equally distributed equivalent (EDE) QALYs. The EDE QALYs are analogous to the certainty equivalents discussed in Outcome Certainty; whereas certainty equivalent QALY gains incorporate risk preferences over uncertain outcomes, EDE QALYs incorporate inequality aversion preferences based on the distribution of health outcomes by group for both the intervention and comparator group. EDE QALYs are calculated using an Atkinson or Kolm inequality index, which estimates the social welfare resulted not only from the overall level of outcomes (QALYs in DCEA) of the general population but disparities (in terms of QALYs in DCEA) across subgroups or individuals.^[8] It is noteworthy that since NHB of each group is a function of opportunity costs of a treatment incurred to the group as well as health gains, the social welfare which is determined by the group-specific QALEs and the inequality index takes into account of the disparities by financial impact as well as by health impact. Once the preferred social welfare function is selected, researchers should select an accompanying inequality aversion parameter that is appropriate for their model setting and population. Consider, in the smoking example, that members of the general population strongly believe that poor smokers should enjoy the same access to cancer treatment as wealthy smokers. By applying a social welfare function to the QALEs of both wealthy and poor smokers, the additional societal benefit of curing poor smokers’ lung cancer is incorporated into the modeled health benefit.

Sixth, the difference between the intervention and comparator group EDE QALYs represents the total value of treatment when reductions in health disparities are accounted for. These incremental EDE QALYs can be calculated using standard calculations leveraged in CEA. A numerical example can be found in Section 9.2.

The steps detailed here are one feasible way to implement DCEA. A variety of methods (e.g. aggregate DCEA vs. full DCEA) and indices available that one may select based on objective, data availability, condition, etc.

In theory, incorporating health equity effects into GRACE would follow the same steps outlined in Box 6 for conventional CEA. Researchers would begin by defining subgroups within the patient population following the first step of implementing DCEA. Next, GRA-QALYs and RASA-WTP can be computed through the GRACE framework and used to construct the group and scenario-specific net health benefit using the formula above. Finally, the group, scenario-specific net health benefits can be utilized with the same Atkinson or Kolm social welfare functions to generate EDE GRA-QALYs, which capture the health equity effects of treatments. While US-specific estimates of inequality aversion parameters are still under development, GCEA recommends a value of 10.95 derived from UK estimates. However, future research is needed to fully integrate DCEA into GRACE.

5.3 Family and Caregiver Spillover

While diseases directly impact the health and costs experienced by patients, patients’ family members and friends are also affected by the financial and non-financial burdens of providing care. For instance, Alzheimer’s disease patients frequently have informal caregivers (usually a spouse or close family member) who have been shown to incur high costs and disutilities as a result of providing care (Haro et al. 2014; Neumann et al. 1999; Robinson et al. 2020). Moreover, cancer caregivers often perform a diverse range of tasks – such as advocating for their patients, aiding in treatment decisions, and performing many nursing-related tasks with little preparation – and have been found to be vulnerable to emotional strain and distress, negatively impacting both their quality of life and productivity (Chiu et al. 2022). However, despite the existence of these non-trivial impacts of providing informal care, conventional CEA models often exclude these caregiver and family spillover and underestimate the value a novel treatment can bring to society by reducing informal caregivers’ burdens (Grosse, Krueger, and Pike 2019a). ICER typically uses the payer or health system perspective as the base case and presents a modified societal perspective as a scenario analysis in most cases (Review 2023c).

Accordingly, GCEA recommends that caregiver and family burden elements are included into the base case in assessing the societal benefits of treatments, including the effect a treatment has in terms of caregiver QoL along with the treatment’s impact on their experienced cost and productivity. First, researchers should determine the treatment’s impact on caregivers’ QoL and the number of caregivers impacted. While studies tend to examine the impacts on primary caregivers only (Wittenberg, James, and Prosser 2019), caregiving for certain conditions can also impact other family members as well (Bobinac et al. 2010; Wittenberg et al. 2013). Researchers should include multiple caregivers in their analyses when relevant.

To determine caregiver QoL benefits, caregiver QALY gains between a comparator and intervention case in GCEA can be calculated. One study utilized survey and regression methods to compare the annual QALY decrement between close family members who were exposed to patients experiencing the after-effects of meningitis (i.e. cognitive problems, motor limitations, behavioral problems, etc.) and those who were not exposed to these after-effects (Al-Janabi et al. 2016). Moreover, another study provides a review of alternative methods – such as the use of different QoL measures or considerations of effects beyond the primary caregiver – that can also be used to estimate the QALY spillover effect of a treatment (Wittenberg, James, and Prosser 2019). In fact, literature recommends using EQ-5D inputs to minimize double counting between the disutility of unpaid caregivers and the monetary value of caregiving time (Park et al. 2022).

Furthermore, analysts should also consider instances when caregiver QoL ratings do not correlate well with reported patient QoL, notably when caregivers are treating patients with severe dementia (Landeiro et al. 2020b). As such, after the caregiver QALY gains are determined, these will be valued separately from patients’ risk-adjusted QoL and risk-adjusted longevity gains – captured as GRA-QALY gains within GCEA – and will be added together when determining the monetary value of the health benefits to society (Figure 8).

Figure 8:

Overview of implementing family and caregiver spillover into GCEA.

Second, researchers should assess the treatment’s impact on caregiver productivity, both on intensive (fewer hours worked) and extensive (exiting labor force) margins, which can be included into the cost side of value assessments. A majority of studies have estimated caregiver productivity costs as the time and wages forgone by informal caregiving. Yet, these methods only capture the partial impact caregiving can have on productivity, ignoring how employee absenteeism can have broader impacts to co-workers’ productivity as well (Berger et al. 2001). Instead, following the methods outlined in recent literature, GCEA recommends researchers estimate caregivers’ productivity losses using an augmented multiplier (Figure 8), which more comprehensively incorporates impacts on team member productivity and total compensation following caregivers’ reduction in work hours or labor force exit (Chiu et al. 2022). These losses represent a social economic loss which can be alleviated by treatment benefits.

Third, separate from productivity losses, researchers should also incorporate the treatment’s impact on non-healthcare sector costs – such as transportation, home renovation, and relocation costs – experienced by caregivers or family members to care for their loved ones (Mattingly et al. 2022). For instance, average monthly out-of-pocket costs for cancer caregivers ranged from $25 (breast cancer) to $1,223 (colorectal cancer), commonly covering expenses for care supplies, travel, and other accommodations (Coumoundouros et al. 2019). Fourth, researchers should also consider the healthcare sector costs caregivers experience due to challenges to their physical and emotional well-being caused by caregiver responsibilities (Park et al. 2022). In fact, caregivers often experience anxiety and depression during informal caregiving – especially for those caring for patients with dementia – and are associated with more frequent use of prescription medications, doctor visits, and use of outpatient procedures (“2021 Alzheimer’s disease facts and figures” 2021).

Finally, analysts may also consider broader caregiver spillover effects of the treatment (Figure 8). For instance, caregiving for a sick elderly parent may result in less attention or involvement towards the informal caregivers’ children, which may adversely impact their children’s future academic performance and social development (El Nokali, Bachman, and Votruba-Drzal 2010). A treatment that can reduce caregiver burdens can bring benefits to individuals beyond primary caregivers, and this should also be captured as societal benefits. Moreover, high caregiving burdens have been shown to lead to increased healthcare utilization by caregivers – including more frequent emergency department visits and provider visits (Goren et al. 2016) – which may lead to increases in healthcare costs experienced by society; treatments that can alleviate caregiving burdens have non-trivial value from their potential to reduce societal costs as well.

6.1 Community Spillover

Conventional CEAs for infectious disease treatments largely focus on the direct health impacts experienced by individuals who get sick. However, individuals who are not infected by the disease can also be indirectly impacted, one case being fears of contagion that often lead to spillover effects to communities at large due to the anxiety of future disease spread. As a salient and recent example, individuals who never got infected with COVID-19 were still highly impacted by the closing of schools, closing of businesses, and travel prohibitions due to the mandated “lockdowns.”(Ma et al. 2020).

As such, treatments that reduce infectious disease spread can provide substantial “community spillover” benefits to unaffected populations, and GCEA recommends that analysts consider these spillovers of treatments – when relevant – to capture their broader impacts on societal costs (Ma et al. 2020; McQueen et al. 2024). Following methods provided by recent literature, these spillover impacts resulting from the fear of contagion can be measured on a per-capita basis through changes in the behavioral impacts on businesses (e.g. school closures and decreased public transportation), economic impacts on businesses (e.g. cancelled major events and restricted travel) and impacts on the health sector (e.g. less health-seeking behavior due to risk of exposure and considering health outcomes of other diseases) (Figure 9) (Ma et al. 2020) Moreover, non-medical costs associated with protection items against disease spread (i.e. masks, hand sanitizers, etc.) can also be included into community spillover estimates.

Figure 9:

Overview of implementing community spillover into GCEA.

However, potential community spillovers are not limited to only infectious diseases. For example, treatments for hepatitis C that reduce the risk of patient’s need for a liver transplant can increase the number of donor organs available to patients whose diseases cannot be cured yet, as hepatitis C is a major cause of liver transplants (Zahid 2022). Moreover, as recent campaigns have reduced hepatitis C rates by remarkable amounts (Hassanin et al. 2021), full elimination of diseases by treatments can also provide spillover benefits to all future populations who will effectively no longer be at risk for this disease.

6.2 Productivity

Treatments that mitigate patient productive loss carry an additional source of value beyond direct health benefits. The productivity impacts of poor health include reduced productivity at work (presenteeism), reduced ability to go to work (absenteeism), unemployment due to disease, productivity loss due to premature death, and productivity gains of life extension. In accordance with ICER’s 2023 value assessment framework update, productivity impacts should be included in the societal perspective whenever possible. These are obtainable from literature or can be estimated when direct evidence for the condition does not exist (Review 2023c).

Disease-specific productivity impact estimates can easily be incorporated into relevant GCEA frameworks, but often differ in their methods. For example, one study measured the productivity loss of patients with multiple sclerosis through reduced working hours, absenteeism, and early retirement from the workplace (Kobelt et al. 2000; Robinson et al. 2020). A separate study of Alzheimer’s disease patients estimated the labor force participation rate of patients and the percentage of patients who claim to experience productivity losses due to their disease (Robinson et al. 2020). Recent efforts to generalize the productivity impacts of disease have extended this definition to include household production, non-household production, volunteering, and labor (Jiao and Basu 2023).

In addition to the standard productivity measures, there are some novel ways to capture productivity impacts that are less frequently used but valid, such as (i) non-labor income, (ii) spillover productivity impacts and (iii) macroeconomic impacts (Box 7).

For certain patients, non-market productivity (e.g. volunteering, looking after grandchildren) represents their primary source of economic output and therefore reductions in non-market productivity increase the burden of their disease. For example, one estimate in the literature found that patients aged 65–74 years generate 54.9 % of their economic output outside of labor markets (Grosse et al. 2019a). To capture this additional opportunity cost to patients, ICER’s 2023 value assessment framework advised researchers to include both market and non-market productivity and to value the time lost in each category at the same wage (Review 2023c). However, a limitation to including both forms of productivity loss is the lack of available studies that provide robust estimates of non-market productivity. When disease-specific measurements of patient non-market productivity loss are not available to researchers, GCEA recommends an algorithm proposed in the literature to estimate non-market productivity by age and QoL (Grosse et al. 2019b).

Within market productivity estimates, spillovers due to patient productivity losses may occur when people work in teams. In team-based settings, patients’ co-workers often also experience lower productivity due to their absence or lower productivity of their sick co-worker (or their co-worker who is a caregiver to a sick person, as noted above). For example, a factory employs workers on an assembly line where each individual specializes in the manufacture of a particular part of a larger good. When workers specialize, output is maximized across comparative advantages (Madiedo et al. 2020). If one worker has reduced productivity due to illness, they may lose their comparative advantage and the entire assembly line is reduced in its ability to produce goods. As a result, the illness of one patient has now affected the productivity of their entire team at work by more than the direct productivity loss related to their presenteeism and absenteeism.

When determining the value of productivity spillovers to teams, a teamwork multiplier can be constructed from existing literature estimates of substitution factors (measurement of how easily replaced a worker is by another) and team effect (measured as the percentage of time working on teams) (Nicholson et al. 2006; Pauly et al. 2002; Zhang et al. 2012). When team effects and fringe benefits are not included in productivity loss calculations, estimates of productivity loss were undervalued by 60–69 % of productivity loss for patients (Chiu et al. 2022).

Finally, some papers have estimated how illness impacts not only lower productivity at the individual and team levels but also the implications for the economy as a whole (Chen and Goldman 2018; Hafner et al. 2023). Namely, increased absenteeism and exits from labor force decrease the labor supply; decreased labor supply drives up wages in the broader economy and leads to reductions in gross domestic product (Hafner et al. 2023). Between 2000 and 2015 pharmaceutical innovation increased productivity in the US labor market by 5.5 million work days per year and $233 billion in wages per year (Chen and Goldman 2018). By comparison, in 2013 pharmaceutical net revenue for branded drug sales was $257 billion, implying that nearly all treatment costs were offset by societal productivity gains due to treatment (Hafner et al. 2023; Institute 2023).

6.3 Adherence

An additional element of value for new therapeutics is the degree to which they promote patient adherence, which allows more patients to benefit from treatment efficacy for longer durations. Ultimately, incorporating the adherence petal involves considering the possible divergence of an innovation’s performance in the clinical trial setting versus how it might work in the real world. A long-acting injectable atypical antipsychotic for the treatment of schizophrenia may be no more effective than an oral in a controlled clinical trial setting, but real world evidence has shown that adherence improves, discontinuation rates decline, and real-world hospitalization rates decline when using long-acting injectable atypical antipsychotics compared to daily oral medications (Kaplan et al. 2013; Marcus et al. 2015; Olivares et al. 2009).

Several factors have been identified to improve adherence, such as more convenient routes of administration, simpler dosing schedules, or combination treatments (Lakdawalla et al. 2018). Consider two treatments for a certain disease: the standard of care which requires inpatient infusions and a new innovation that allows treatment to be administered orally at home, once a day. Consider that for the population with this disease, driving to an infusion center for many patients is burdensome. The ease of administration for the new oral treatment may greatly improve adherence, as patients can administer the medication at home over a much shorter time period (swallowing a pill vs. a one-hour infusion) with limited transportation costs. This adherence-improving factor offers additional value to patients both through improvements in clinical outcomes and through additional utility due to convenience or tolerability.

For example, the enhanced effectiveness by the improved adherence not only increases the mean of future health outcomes of patients but reduces the uncertainty surrounding the outcomes. At the same time, for non-patients, it mitigates the impact of future disease risk (i.e. increases the value of disease risk reduction) as much as the effectiveness is enhanced by the improved adherence. Yet few CEA models have detailed how adherence should be incorporated into value assessment frameworks. When including adherence in value assessments, many studies rely on assumptions of perfect adherence or clinical trial adherence as a proxy which is not representative of real-world use patterns (van Onzenoort et al. 2011).

In practice, inclusion of treatment adherence in GCEA can be accomplished through changes in more traditional value dimensions including changes in effectiveness, costs, and utility. GCEA provides researchers with a series of key questions that help determine how adherence-improving factors can be included in value assessments (Box 8).

Once these questions are answered, a few methods exist to implement adherence into value assessments. If using a Markov model, adherence can easily be captured as patients transition during each cycle to an adherent or non-adherent state, with corresponding efficacy (improved health outcomes), costs (lower medical costs due to improved health), and utilities (less caregiver burden, higher GRA-QALYs from time spent in better health). One could also use pharmacokinetic/pharmacodynamic modelling to examine how suboptimal adherence is likely to impact real-world effectiveness (Shafrin et al. 2017). Ex-post value assessments can measure the true benefit to patients from adherence-improving factors, but rely on real-world data that are not readily available when value-based pricing decisions are made at drug launch.

6.4 Direct Non-medical Costs

As a consequence of their disease, many patients (and their families) incur direct non-medical costs to accommodate disability and diminished health-related quality of life. For example, parents of children with dyslexia often incur higher educational costs as tutors and other forms of remedial teaching become required to progress in school at the same rate as their peers. Other patients, such as those with amyotrophic lateral sclerosis, multiple sclerosis, Parkinson’s disease, Duchenne muscular dystrophy, and other diseases that impact ambulatory ability experience costs of retrofitting their homes with wheelchair ramps and arm railings.

Furthermore, while not included in conventional CEA pharmaceutical costs, patients incur additional travel costs in order to receive care such as medical evaluations, consultations, physical therapy, and infusions administered at an inpatient or outpatient facility. As demonstrated by Alzheimer’s patients, these direct non-medical costs provide a substantial burden and a benefit of medical innovation is reductions in these costs.

However, direct non-medical costs are highly specific to each disease and no single method can be readily generalizable across indications. As a result, GCEA recommends researchers consider the potential direct non-medical costs for a disease of interest and include those costs in their model.

7.1 Limitations

While GCEA presents a more comprehensive inventory for capturing novel and broader value of treatments, as with any work-in-progress framework, it does have several limitations. First of all, each element discussed in this paper provides its own, unique source of value to patients, but many cannot be valued in isolation and require the use of a broader unified framework to be incorporated into societal cost-effectiveness and value-based pricing decisions. For example, the values accounted for by the Outcome Certainty (the value of reducing uncertainty and the value of hope) and Disease Risk Reduction (the insurance value) petals are incorporated in a comprehensive manner using GRACE and cannot be valued in isolation for contextual purposes.

Furthermore, GCEA relies on a number of methodologies that themselves have limitations. For example:

1. Although being required for GRACE to generate the most patient-centered results, estimates of patient risk aversion over disease-specific health outcomes are limited. DCEA is highly sensitive to the inequality aversion, however, literature on this is also relatively sparse (Li et al. 2019; Li et al. 2022). Besides these key parameters, many disease-specific estimates of community spillover, family and caregiver spillover, and adherence may have limited data available for incorporation into GCEA in practice. Additional data collection is required to accurately model GCEA.

2. Quantitative approaches to estimate patient non-market productivity loss and translate drug novelty into treatment value itself due to scientific spillover have not yet been developed.

3. Predicting the trajectory of branded drug prices and the magnitude by which drug prices fall after loss of exclusivity is uncertain and must be based on analogous treatments, competitive landscapes, and evolving regulatory environment.

4. The GCEA equity petal, as currently constructed, focuses on reducing inequality in health outcomes. However, there are many other dimensions of inequality (e.g. income inequality) which also have value. Future work should consider if these other dimensions of inequality should be considered and, if so, how they can be integrated into the GCEA framework.

Conventional CEA, however, does not offer us a solution for all these limitations of GCEA. Rather, conventional CEA simply assumes a default value for every petal that it omits. By ignoring dynamics, for instance, conventional CEA assumes that a drug will never go generic. By ignoring productivity, it assumes that a patient and their caregiver will not be more productive – or this productivity is valued by society at $0 – which is also an extreme assumption. Thus, while GCEA limitations are important to note, GCEA does mark a significant improvement over conventional CEA’s simple but extreme assumptions surrounding the value of broader value elements.

Both conventional CEA and GCEA, while impressive to some for their quantitative precision, require humility in their interpretation precisely because of their limitations in capturing the revealed preferences of society and its many stakeholders. As the Second Panel on Cost Effectiveness and Health in Medicine states, “cost-effectiveness analysis is not by itself a sufficient decision-making standard and that it does not capture all relevant concerns.”(Sanders et al. 2016a)

7.2 A Path Forward

Despite these limitations and areas for future research, this current guidance on GCEA aims to provide researchers with a variety of improved methodological approaches over the conventional cost-effectiveness framework, which quantitatively omits all but, on occasion, a handful of the GCEA value flower’s petals. While the added value of many petals may be more challenging to calculate, this paper has extensively cited a growing body of research that has demonstrated that these values are real and should not be ignored. Indeed, the purpose of GCEA is to transition value assessment from the narrow payer perspective to incorporating real sources of value to our society that have been empirically measured in the literature.

In fact, GCEA’s feasibility is highlighted as value assessment literature is already moving in this direction and some studies can be considered to have taken on different ‘versions’ of GCEA. For example, a study evaluating the net social benefits for treating all hepatitis C-infected individuals included the petals of scientific spillover, family and caregiver spillover, patient-centered health improvements, productivity, adherence, dynamic net health system costs, disease risk reduction, and the value of hope (Moreno et al. 2017). A study on a recent treatment for early Alzheimer’s disease included the productivity and family and caregiver spillover petals (Whittington et al. 2022).

The limitations of GCEA are no reason for sticking with conventional CEA, which has far more limitations by the simple fact that it ignores – or assumes a value of 0 – for most of the petals of the flower. Even when a CEA doesn’t take various petals into account, possibly due to lack of data, its own limitations section should acknowledge that these values are missing and point out that their inclusion would likely alter the ICER and maybe even the conclusion about a product’s societal cost-effectiveness. In countries that rely on CEA to set coverage decisions, the recognition that petals were omitted due to lack of data should spark discussion of whether to err on the side of coverage. CEAs can always be refined and decisions changed once there are more data, but if innovation is discouraged because CEAs undervalue it, there won’t be anything to re-evaluate.

Countries that use CEAs to make decisions for their populations would ideally make sure that CEAs done for their narrow purposes are not written in a way that gives the impression that they are passing judgement on the value of a medicine for others. While it’s not clear why any country would ignore a demonstrably real value, doing so for reasons that are country-specific should be accompanied with an acknowledgement that this value could be present for others and therefore the assessed technology could be worth more to others.

Despite the clear need for more broadly capturing treatment value, additional research is needed to convert the GCEA inventory of value elements into a formal value assessment framework. For instance, the GCEA approach outlined in this paper does not explicitly recommend a specific willingness-to-pay per GRA-QALY threshold, nor how the additional of each GCEA petals would impact this threshold. Additionally, this current guidance does not describe how to address issues of potential double-counting. For instance, if caregiver and productivity impacts are incorporated in the value assessment, should the willingness to pay per QALY change and if so, how? Additionally, if one calculates the option value of an initial treatment, should that impact the value assessment conducted of the subsequent therapy (Towse 2022)?

Some suggest that if CEA calculates that a drug is not cost-effective and GCEA’s extra petals reduce the ICER below the cost-effectiveness threshold, then the threshold should be lowered. While this paper does not explicitly comment on the appropriate willingness to pay for health gains for different countries, GCEA should be seen as a series of methods to incorporate other forms of value that are well-recognized but currently unaccounted for in conventional CEA. Moreover, the more variables a GCEA includes, the wider the confidence interval of results. The range of possible answers should not be taken as a flaw with the approach but as a valuable warning not to assume that math can always substitute for the reveal preference of actual people. Whether one does conventional CEAs or GCEAs, the more one recognizes how much conventional CEAs have omitted what actually matters to people, the more humility one should have for what today’s GCEAs may get wrong about what we will someday discover we’ve long valued. Even if GCEAs only serve as reminders to keep an open mind about all that we have still failed to account for and to not to just ignore what we can’t compute, they will serve a useful purpose.