Access to Disease-Modifying Alzheimer’s Therapies: Addressing Possible Challenges Using Innovative Payment Models

Abstract

Context:

Aduhelm is the first approved disease-modifying treatment (DMT) for Alzheimer’s disease (AD). Under current payment models, however, AD DMTs – especially as they treat broader populations – will pose challenges to patient access since costs may accrue sooner than benefits do. New payment approaches may be needed to address this difference in timing.

Methods and Data:

We use the Future Elderly Model (FEM) which draws on nationally representative data sets like the Health and Retirement Study (HRS) to estimate the potential benefits due to hypothetical AD DMTs in four stylized treatment scenarios for patients with mild cognitive impairment (MCI) or mild AD, and develop a payment model to estimate the accrual of net costs and benefits to private and public payers.

Results:

The modeled AD DMTs result in clinical benefit of 0.30 to 0.55 QALYs gained per patient in the baseline treatment scenario and 0.13 to 0.24 QALYs gained per patient in the least optimistic scenario. Private payers may observe a net loss in patients 61–65 years old under the status quo (payment upon treatment). Constant and deferred installment payment models resolve this issue.

Conclusions:

Innovative payment solutions, such as installment payments, may be required to address misaligned incentives that AD DMTs may create among patients under the age of 65 years, and may help address concerns about the timing and magnitude of costs and benefits accrued to private payers.

Introduction

Disease-modifying therapies (DMTs) for Alzheimer’s disease (AD) are headed to the U.S. market. In 2021, the first possible DMT for the disease, aducanumab (Aduhelm), received accelerated approval by the FDA [1] targeting the prodromal and early dementia stages of AD, and two other drug candidates (Eli Lilly’s donanemab and Eisai’s lecanemab) have initiated rolling submissions [2, 3]. Under current payment models, these could pose significant challenges to payers and patients since costs may accrue sooner than benefits, especially if benefits prove durable. In 2020, the number of people estimated to suffer from clinical AD was 6.07 million which, in the absence of a therapy, is expected to grow to 13.85 million by 2060 [4]. Although it is yet to be determined how effective new DMTs will be at reducing prevalence and incidence of the disease over a longer term, even a modest improvement in clinical outcomes is likely to result in significant economic and societal benefits due to the enormous burden of the disease and the limitations of symptomatic treatments. Moreover, other AD DMTs remain in development and are nearing approval decisions of their own. These scientific developments have increased the urgency around the economic question of how to pay for the treatment of AD.

AD results in a significant economic and social burden (with annual health care costs of between $35,736 [5] and $52,755 [6] (in 2021 dollars) and informal caregiving costs of between $16,536 [5] and $36,649 [6] (also in 2021 dollars)). Given the aging of the U.S. population, it was estimated that the economic burden of both formal and informal care was $159–215 billion in 2010 (depending on the method used to value the cost of informal care), and expected to more than double by 2040 to $379–511 billion [5]. Zissimopoulos et al. [6] estimate that the delay of AD onset by 1 year could have a cumulative economic value of $183,227 per patient and a delay of AD onset by 3 years a cumulative value of $355,222 per patient (this was calculated as the difference between the value of additional AD-free years minus the change in formal and informal costs). The real-world, long-term impact of emerging therapies is not well understood as trials do not commonly track patient outcomes over a longer time horizon.

Aside from Aduhelm (aducanumab), which was approved by the U.S. Food and Drug Administration (FDA) in June 2021 using an accelerated approval pathway [1], other potentially disease-modifying therapies for AD are currently under development, with the majority of drugs in trials (82.5%) targeting the underlying biology with the potential of disease modification [7]. These potential DMTs under development target both preclinical and prodromal (also described as ‘mild cognitive impairment (MCI) due to AD’) and mild or moderate dementia patients. They have typically targeted patients as young as 50 years old in pivotal trials (with exceptions such as a Phase 2/3 trial of gantenerumab and solanezumab which may enroll patients as young as 18 years) [8]. These DMTs are expected to be administered in monthly infusions until treatment discontinuation (e.g. when a patient experiences severe adverse events or progresses in their disease). Different primary endpoints are used in trials, with Phase 3 DMT candidates active in January 2022 most commonly utilizing outcome measures in the cognitive and functional domains, such as the Clinical Dementia Rating-Sum of Boxes (CDR-SB) (NCT03486938, NCT03444870, NCT03443973, NCT03887455, NCT03605667), Alzheimer Disease Assessment Scale-Cognition (ADAS-Cog) (NCT03790709), ADAS-Cog 11 (NCT03823404, NCT04520412, NCT04187547, NCT03605667, NCT03446001), ADAS-Cog 12 (NCT04669028), ADAS-Cog 13 (NCT02051608), Alzheimer’s Disease Cooperative Study ADL Scale (ADCS-ADL) (NCT03823404, NCT03790709, NCT02051608, NCT03446001), Alzheimer’s Disease Cooperative Study - Clinical Global Impression of Change (ADCS-CGIC) (NCT04520412, NCT04669028), Preclinical Alzheimer Cognitive Composite 5 (PACC5) (NCT04468659), Free and Cued Selective Reminding Test (FCSRT) (NCT04098666), Alzheimer’s Disease Cooperative Study-Preclinical Alzheimer Cognitive Composite (ADCS-PACC) (NCT02008357, NCT02913664), and NIH Toolbox (NIH-TB) Cognition Battery (NCT02913664).

The emergence of such DMTs will create both new opportunities for the management of AD and MCI due to AD as well as result in challenges related to access and payment. In our work, we model the differential effect AD DMTs may have by therapeutic effect size and age and study the implications new therapies may have for private and public payers in the U.S. We explicitly model two alternative payment approaches which may reduce the risk of barriers to access in younger patients who are commonly covered by private insurance – constant and performance-based installment (deferred) payments. Deferred payment has previously been shown to result in incentives to treat patients sooner, which may have non-trivial clinical benefits [9].

Although only one therapy has so far been approved, this work may inform potential future considerations in the design of payment models for disease-modifying AD therapies. It is not our goal to focus entirely on aducanumab, in light of the uncertainty about its real-world effectiveness. Rather, we propose hypothetical effectiveness scenarios anchored in part on the best-case clinical trial results for aducanumab under the view that future therapies may eventually meet or exceed these.

Methods and Data

We use a microsimulation, the Future Elderly Model, which draws on nationally representative data from the Health and Retirement Study (HRS) and other sources, to predict clinical and economic outcomes of interest in a population of adults between 51 and 85 years under 4 different treatment effectiveness scenarios. We develop a payment model to estimate the net benefit/loss which may be attributable to AD DMTs by age, payer type and payment model, as discussed below. Our results provide a health care sector perspective of treatment benefits and costs, given the focus on direct costs and benefits that private and public payers will accrue [10, 11].

Future Elderly Model

The Future Elderly Model (FEM) was first applied to estimate the benefits of medical innovation in the U.S. elderly in 2005 [12]. The FEM is a simulation model that projects person-level health and economic outcomes using longitudinal data from the HRS and other surveys, including the National Health Interview Survey (NHIS), the Medical Expenditure Panel Survey (MEPS), and the Medicare Current Beneficiary Survey (MCBS). The FEM consists of multivariate models of cognitive and functional status, disease conditions, disability, mortality, nursing home entrance, caregiving, and others, based on patient-level characteristics, such as a history of smoking, age, marital status, and education. Earlier work has shown that functional decline follows cognitive decline, and is largely explained by it [13, 14] – the FEM itself follows this path dependency and allows for functional decline to be causally linked to cognitive decline in all adults.

The FEM predicts economic and clinical outcomes of interest at the patient level in 2-year waves and takes into account a range of individual-level characteristics. Cognitive function in the FEM is estimated as a first-order Markov model based on observed transitions in the Telephone Interview for Cognitive Status (TICS-27) score in the HRS sample. We weight our estimation sample by the probability of having a Clinical Dementia Rating score of 0.5 to mimic those included in clinical trials for AD DMTs. TICS-27 also impacts other transition models within the microsimulation: mortality, activities of daily living, instrumental activities of daily living, and nursing home entry. Consequently, an intervention on TICS-27 will impact these downstream outcomes. Functional limitations are modeled with an ordered probit for the number of functional limitations (0, 1, 2, 3 or more) and include walking, dressing, bathing, eating, getting in/out of bed, and using the toilet. This model is a function of previous functional limitations, demographics, chronic diseases, risk factors, and TICS-27.

Applying the transition model for TICS-27 yields an updated cognitive score based on the projected characteristics of the individual. In the intervention scenarios this cognitive score is adjusted to be consistent with the hypothetical intervention. For example, consider an individual who declined from a TICS-27 score of 10 to a TICS-27 score of 7 between two periods, a change of 3 points. In the 20 percent cognitive treatment scenario this 3-point decline would be reduced to a 2.4-point decline, yielding a score of 7.6.

Direct interventions on functional limitations are handled slightly differently. Functional limitations are modeled with an ordered probit model. We introduce an additional calibration term to the ordered probit model and then calibrate that parameter to yield a particular outcome, such as a two-year change in the proportion of patients with any limitations.

More information about the FEM is provided in the FEM’s Technical Documentation [15].

Treatment Scenarios

We model four treatment effect scenarios of AD DMTs in a population of Americans over the age of 50 years with mild cognitive impairment in FEM. These scenarios allow us to model the expected benefit of hypothetical DMTs at different levels of effectiveness in the cognitive and functional domains. We model cognitive benefit as a reduction in expected cognitive decline using TICS-27 and functional effect as a reduction in the expected development of difficulties with Activities of Daily Living (ADLs). The scenarios are:

• Scenario 1: 20% cognitive effect and 40% functional effect – approximately referencing the benefit observed in the EMERGE trial of Aduhelm (which reported a 22% reduction in decline on the cognitive and functional composite measure CDR-SB, an 18% reduction in decline on the cognitive measure MMSE, a 27% reduction in decline on the cognitive ADAS-Cog 13 measure, and a 40% reduction in the decline on functional ADCS-ADL-MCI measure¹);

• Scenario 2: 10% cognitive and 20% functional effect – calibrated at half of Scenario 1 (as a more pessimistic treatment effectiveness outcome);

• Scenario 3: 40% cognitive effect and 20% functional effect (as a sensitivity analysis testing the inverse relationship between cognitive and functional benefit relative to Scenario 1);

• Scenario 4: 20% cognitive effect and 10% functional effect (half of Scenario 3).

Our treatment scenarios assume the slower rates of decline are permanent. In sensitivity analyses, we investigate the consequences of shorter-lived effects. For each scenario, we compare the lifetime benefits and costs to matched controls – identical patients who are simulated to age naturally in the absence of an AD DMT – and report benefits and costs before and after the age of 65 years individually. Additional information about TICS-27 and ADLs, as well as the specification of the treatment effect under each scenario, is provided in Supplement 1.

Future Elderly Model Simulations

Simulated individuals enter the model based on their cognitive status. HRS respondents with a TICS-27 score of seven or higher, based on Crimmins et al. [16], who are likely to have a CDR score of 0.5, are selected. Individuals are excluded if they have a recent heart condition, recent stroke, or congestive heart failure. HRS does not allow for patient selection based on amyloid status, a hallmark of Alzheimer’s pathology which is among common inclusion criteria in AD clinical trials. In a large U.S. population study, 55.3% of MCI patients and 70.1% of dementia patients had positive amyloid Positron Emission Tomography (PET) results [17]. Select demographics characteristics of patients in our simulation are shown in Table 1.

Table 1:

Demographic characteristics of the simulated population

Cohort	Simulated observations at entry	Mean age at entry (years)	Median life expectancy (years)	Male % at entry	Mean years education at entry	Mean TICS at entry	Any ADL at entry	Any IADL at entry
51–55	2,288	53.8	27.0	26.4%	12.74	11.80	17.4%	11.0%
56–60	16,432	58.6	23.0	25.1%	11.90	11.60	26.5%	15.0%
61–65	20,488	62.9	19.0	27.5%	11.79	11.10	28.2%	16.5%
66–70	37,960	68.3	15.0	31.5%	11.68	11.62	20.6%	8.1%
71–75	63,752	73.0	13.0	34.4%	11.77	11.79	19.3%	8.2%
76–80	63,752	78.0	9.0	33.2%	12.17	11.73	20.1%	10.2%
81–85	53,040	83.0	7.0	41.2%	12.43	12.26	24.1%	14.0%
All	257,712	71.3	13.0	32.5%	11.96	12.06	22.2%	11.2%

Note: ADL = Activities of Daily Living, IADL = Instrumental Activities of Daily Living

Payment Modeling

We develop an economic model that translates gains due to treatment under each scenario and for each five-year age cohort. The model assigns a value of $150,000 to each quality-adjusted life year (QALY) gained [18], which is considered as the social welfare gain due to treatment. FEM estimates the effect of each of the modeled interventions on direct health care costs (which include costs associated with nursing home stays). The model estimates the net value accrued due to therapy under and over 65 years of age as the net balance of the social welfare gain due to treatment, the difference in health care costs, and the cost of therapy. The cost of therapy is assigned differently in each of the payment models studied, namely status quo (payment upon treatment), constant installment payment and pay-for-performance installments payments (see Figure 1 for a notional description of each of these payment models). We do not account for possible termination of effect and apply both installment approaches for the patients’ remaining lifetimes (this could be adjusted in the real world without the loss of generalizability). We study both constant and pay-for-performance installment options given the unique age profile of patients with cognitive impairment caused by Alzheimer’s disease, which is hypothesized to push benefits further into the future while costs are incurred early on (this assumption is specific to the most advanced disease-modifying drug candidates in the pipeline, many of which are biologic therapies).

Figure 1:

Notional Allocation of Treatment Costs by Payment Model

All future costs and benefits are discounted at 3% annually. We notionally assume all patients under 65 years are covered by a commercial plan (private payer) and patients over 65 years are covered by Medicare (public payer), although real-world insurance coverage under and over 65 years is more complex. Our modeling also abstracts from the challenges of a fragmented payment landscape in the U.S.

Payment at Time of Treatment

First, we model payment that occurs at the time of treatment (status quo). The cost of treatment is calculated as a fraction (20%) of social welfare gain (quantified based on the estimated QALY gain valued at $150,000 per QALY gained), drawing on historical estimates in multiple disease areas in which biomedical innovators appropriated less than 25% of the surplus produced by their innovations [19, 20].

Constant Installment Payments

Second, we model constant installment payments. We first estimate the mean, discounted life expectancy for each treated age group and distribute the cost of therapy, calculated as 20% of the value of the QALY gain, which is based on a $150,000 per QALY gained threshold. We allocate the cost of treatment to equal annual payments based on discounted years before (when applicable) and over 65 years of age. This payment model does not assume any down payment aside from the first annual installment payment.

Pay-for-Performance Installment Payments

Third, we model performance-based installment payments. We first estimate the discounted QALY gain under (when applicable) and over 65 years of age, and distribute the cost of therapy, calculated as 20% of the value of the QALY gain at a $150,000 per QALY gained threshold, based on the share of the QALY gain before and after the age of 65 years. This payment model is designed to match payment with the accrual of benefits over time.

Estimating Net Value by Payer

For each of these payment models, the net value accrued to the private payer (under 65 years of age) or public payer (over 65 years of age) is calculated as a difference between the discounted cost of therapy (accrued as described above based on the payment model), net difference in medical expenses (increase or decrease), and the welfare gain in QALY terms, valued at $150,000 per QALY gained. For instance, a therapy which results in a 0.10 QALY gain (of which a 0.05 QALY gain is observed before the age of 65 years) would have a value-based price of $3,000 under status quo based on a $150,000 per QALY gained threshold and the assumption that 20% share of societal gain is recovered by the manufacturer. In our model of status quo payment, it would be covered in full by private insurance for patients under 65 years of age. If such therapy reduces medical expenditures before 65 years of age by $5,000 and increases medical expenses over 65 by $5,000, the net benefit accrued to private payers would be calculated as follows: 0.05*$150,000 (social value) + $5,000 (medical expenditure savings) – $3,000 (cost of therapy) = $9,500. For public payers, the net benefit accrued under status quo (upfront payment) would be calculated as follows: 0.05*$150,000 (social value) – $5,000 (additional medical expenditures) = $2,500. In Supplement 3, we show how the net benefit results would change if the durability of treatment was limited to 2, 4, 6, 8, or 10 years, respectively.

Results

Using the Future Elderly Model, patients treated under each of the four treatment scenarios realize discounted QALY gains that follow different patterns based on the treatment effect assumed, as shown in Figure 2. Patients are expected to realize the highest discounted QALY gains in our baseline Scenario 1 (20% cognitive and 40% functional effect), with the highest clinical benefit in patients aged 56–60 years (0.55 QALYs gained), and the smallest clinical benefit in patients aged 81–85 years (0.30 QALYs gained), both assuming a persistent treatment effect. Reducing the treatment effect in half (Scenario 2) results in a reduction of QALY gains by between 47–50% depending on the age group, with the highest clinical benefit in patients aged 56–60 years (0.29 QALYs gained) and the lowest clinical benefit in the oldest patients aged 81–85 years (0.15 QALYs gained).

Figure 2:

Share of discounted QALY gain under and over 65 years of age, by treatment scenario and age cohort

In Scenarios 3 and 4, clinical benefit follows a similar trajectory; however, QALY gains relative to Scenarios 1 and 2, respectively, are somewhat smaller. For example, the highest reported QALY gain in Scenario 3 is in patients 56–60 years who expected to report 0.49 QALYs gained, or approximately 10% less than the same cohort in Scenario 1. The difference between Scenarios 2 and 4 is even a little bit bigger, with patients aged 56–60 years expected to report 0.25 QALYs gained in Scenario 4, or about 14% less than the same cohort in Scenario 2. The relatively smaller clinical benefit in Scenarios 3 and 4 relative to Scenarios 1 and 2 indicates that halting functional decline has a greater effect on improving quality of life than halting cognitive decline.

As also shown in Figure 2, patients younger than 65 years at time of treatment stand to accrue most of their QALY gains after the age of 65 years in all treatment scenarios. This confirms the hypothesis that under the status quo payment model, which pays for therapy at time of treatment, an incentive misalignment between payment and benefit accrual may occur. The underlying data for Figure 2 are shown in the Supplementary Table 1.

In Figure 3, we show the implications of each treatment scenario on medical expenditure savings and show the value-based cost of therapy (based on the average QALY gain and the value-based cost of treatment estimation described above). In all four scenarios, patients are expected to achieve medical expenditure savings (even when they live longer), reaching up to $7,020 for patients aged 66–70 years in Scenario 1 and $8,017 for patients of the same age in Scenario 3. These results indicate that cognitive benefit may have a relatively larger impact on medical expenditure savings than equivalent functional benefit. The relatively smallest medical expenditure savings are observed for patients treated at younger age (however, this may be affected by the timing of cost savings and the effects of discounting in our analysis). Across all treatment scenarios, patients aged 61–75 typically achieve the highest savings of medical expenditures, suggesting those patients may be most attractive to payers from a reimbursement perspective. In Figure 3, we also show the value-based estimate of the discounted, lifetime cost of therapy for each age group and treatment scenario. Given the QALY gains reported above, we observe relatively large differences between age groups. In Scenario 1, the value-based cost of therapy would be expected to range from $8,871 for patients aged 81–85 years to $16,465 for patients aged 56–60 years. In other scenarios, the value-based cost of therapy would also be highest for patients 56–60 years old ($3,673 in Scenario 2, $14,827 in Scenario 3, and $7,406 in Scenario 4). Patients treated at older age would be expected to have a lower value-based cost of therapy (accounting only for QALY gains and medical expenditure savings). Should the manufacturers be able to appropriate more than 20% of the surplus produced by their therapy, the value-based cost of therapy may increase. The underlying data for Figure 3 are shown in the Supplementary Tables 2 and 3.

Figure 3:

Medical expenditures estimates and value-based cost of therapy, by treatment scenario and age cohort

In Figure 4, we compare the net benefit or loss accrued to private and public payers by scenario, patient age cohort, and payment model. In some age cohorts, we find a net negative impact on private payers from providing access to an AD DMT, largely driven by the proximity to Medicare eligibility age, which prevents sufficient benefit accrual to outweigh the costs of treatment.

Figure 4:

Net costs or benefits accrued to private and public payer under treatment scenarios, by age cohort and payment model

Under all four scenarios in the status quo, payers are expected to accrue net negative benefits (i.e. losses) for patients in the 61–65 cohort, estimated to be −$9,708 in Scenario 1, −$280 in Scenario 2, −$10,348 in Scenario 3, and −$5,049 in Scenario 4.

Net losses are completely avoided under both constant and pay-for-performance payment models. Pay for performance installments would increase the financial incentive to private payers relative to constant installments. For example, payers’ net benefit for patients aged 61–65 years old in Scenario 1 would be $4,704 under pay-for-performance installments compared with $4,442 under constant installments.

As we show in sensitivity analyses (Supplement 3), there is limited need for innovative payment contracts if treatment durability is limited to 2 years only (with patients covered by private payers not expected to accrue a significant net loss due to therapy). However, for treatment durability of 4 years and longer, temporal misalignment of costs and benefit remains, and innovation in payment can help address it.

Discussion

Emerging DMTs in Alzheimer’s disease have the potential to improve people’s lives. Yet, ensuring access under existing payment arrangements may prove to be a challenge, particularly for patients under 65 years covered by commercial plans. Even if priced at a fraction of overall welfare gain, such therapies may result in net losses accrued to private payers, particularly for patients close to the Medicare eligibility age. With potential DMTs requiring payment at time of treatment (e.g. for amyloid-based biologic therapies used for a defined period of time), private payers may opt to delay the treatment of such patients. Alternatively, they may increase premiums or put such drugs on high co-pay formulary tiers, which could result in patients shouldering a larger share of the cost of treatment. Either of these approaches may result in suboptimal outcomes as fewer patients receive timely access to these emerging treatments.

We find that both constant and pay-for-performance installment payment models offer possible solutions. By spreading the cost of therapy over the remainder of a patient’s lifetime, private payers would no longer face the disincentive to provide care as their share of the treatment cost would no longer make the net benefit of treatment negative. Pay-for-performance installments may serve as the first-best solutions from an actuarial perspective; however, their implementation costs may outweigh their benefits. Constant installments result in net positive benefits to private payers for all age cohorts studied. Treatment durability of 2 years or less would obviate the need for innovative payment contracts, but durability of 4 years or longer creates misaligned incentives.

Should installment-based payment models be considered for implementation, assumptions about the long-term benefits of therapy may need to be made based on population models and simulations, such as the FEM, because clinical trials typically provide only a few years of relevant outcome data. The uncertainty about benefit accrual poses challenges to payment design [21]. In addition, public and private payers would need to cooperate and agree on a mechanism to transfer patients’ payment obligations between one another when they switch plans or become eligible for Medicare [22].

Other approaches not explicitly modeled here may be considered. For instance, a move to a single payer model or a transfer from public payers to commercial plans in order to incentivize the treatment of younger patients may offset net negative benefits they would otherwise incur. Such transfer may take different forms. Alternatively, patients not yet eligible for Medicare may receive a subsidy to pay for their AD treatment. Yet another potential solution is an Alzheimer’s risk pool which could be set up to provide access to both early screening, diagnosis and treatment of Alzheimer’s disease. Screening and early diagnosis may include cognitive and biomarker-based tests, requiring the reimbursement of primary care providers to administer them [23]. An Alzheimer’s-specific risk pool would provide both public and private payers with the option to make an actuarially fair contribution for patients over a certain age. In return, such an insurance facility could cover the cost of screening, diagnosis, and treatment for anyone with cognitive impairment due to Alzheimer’s disease, which would create the incentives to avoid delay in care and possibly also negotiate optimal drug prices with manufacturers (based on lifetime treatment value and relevant elements of value). A risk pool may also more optimally react to new insights about long-term efficacy of emerging AD DMTs, and have more flexible contracts with manufacturers (while reducing bureaucratic burden to many different payers across the nation).Although theoretically this solution may have many benefits, regulatory changes and buy-in from both public and private payers would be required for its success (which could later extend to other disease areas with unique treatment paradigms, such as gene and cell therapies). Our findings do indicate, however, that efficiency gains can be made when the temporal misalignment of costs and benefits is addressed. We leave the study of other payment models to future research.

While the implementation of innovation in payment may face many challenges, the economic impact of AD DMTs may result in substantive demand for new solutions [24]. Moreover, AD DMTs may encourage the conduct of health technology assessment (HTA) that would require the consensus of many stakeholders across the nation [25].

Limitations

Our work has several limitations. First, we use a stylized representation of the treatment effect under four distinct scenarios that is assumed to persist over the remainder of a patient’s lifetime. Given typical clinical trial duration, it will take many years before conclusions about lifetime benefits in QALY terms will be more precisely understood, resulting in uncertainty about the specification of treatment effect in our model and other similar research. The results in the main text should be viewed as upper bounds on the expected gains from innovative payment, because we assume that treatment permanently slows the rate of cognitive and/or functional decline. In sensitivity analyses, we show that gains are smaller when treatment effects are shorter-lived, but that gains persist so long as treatment effects last four years or more. Additionally, while we select patients who most closely resemble eligible populations in AD DMT clinical trials, the lack of biomarker testing in our selection restricts our model to patients with cognitive impairment due to any cause. It is possible that patients with amyloid pathology will observe greater or smaller benefits than expected, for instance if the treatment effect relative to a control group increases over time, or if patients with prodromal and mild AD experience a faster decline than our data indicate. Third, we quantify social welfare gain using a single threshold per QALY gained, which has its limitations. Some have argued, for example, that it does not account properly for differences between different patients’ response to a medication, or its externalities, and that it may result in discrimination against persons with disabilities and chronic illnesses [26]. Also, our estimates of treatment benefit (which translate to QALY gains) are based on TICS-27, CDR, and ADL outcome measures. Limited tools are available to ‘crosswalk’ between endpoints within cognitive and functional domains [27], making direct comparisons to clinical trial results less reliable. Additionally, our work estimates direct health care costs and benefits, such as the reduction in nursing home care expenditures, but we do not explicitly model the lower demand on informal care and possible improvements in the quality of life of caregivers, insurance value of a therapy, and its other elements of value [28]. Finally, mean benefits from AD DMTs may differ depending on the composition of the patient pool of individual insurers given that risk factors and comorbidities are not uniformly distributed across regions and demographic groups. Our analysis estimates mean benefits under select treatment scenarios based on nationally representative data.

Conclusions

We find that AD DMTs may provide patients with different levels of clinical benefit depending on age and treatment effects in the cognitive and functional domains. Our modeling indicates that payment at time of treatment for these therapies may result in challenges to access given the net negative benefit private payers are expected to incur in some patients under 65 years. We study two potential payment model solutions: constant and performance-based installments. Both avoid the net loss accrual to private payers. Depending on the long-term benefits and costs of AD treatments, such contracts may be negotiated, requiring the coordination of private and public payers in the coverage of AD DMTs. Other solutions, such as an AD-specific risk pool may be considered. Ultimately, the success of emerging DMTs for Alzheimer’s disease will depend on patients’ timely access to them. Our work suggests that new, innovative approaches may be necessary, particularly for patients covered by private payers.