2.1 Target Population and Settings

The target population of this study is individuals in Sweden diagnosed with early AD operationalized as MCI due to AD or mild AD dementia who are APOE ε4 non-carriers or heterozygotes with confirmed amyloid pathology and eligible for anti-amyloid treatments [21].

2.2 Intervention and Comparators

This study compared two treatment strategies for AD. The first strategy involves the use of lecanemab in addition to standard of care (SoC) in Sweden for MCI due to AD and mild AD dementia and SoC alone for moderate to severe AD dementia. SoC in Sweden primarily includes treatment with cholinesterase inhibitors for mild to moderate AD dementia and memantine for moderate to severe AD dementia [22]. The second strategy is SoC alone.

2.3 Model Overview

The health economic evaluation was based on a decision-analytic Markov cohort model with a 10-year time horizon and a 3-month cycle length from the perspective of Swedish formal care payers (including regions and municipalities). The cycle length was chosen to facilitate modelling changes in the frequencies of administering lecanemab over time as well as adverse events and the resulting costs and disutilities associated with lecanemab. The 10-year time horizon was chosen considering that 2007–2020 data from SveDem were used to model transitions from AD dementia to subsequent states, and the data after 10 years became sparse. The base-case model assumed that lecanemab would be administered to individuals with MCI due to AD or mild AD dementia for up to 3 years. Our model assumed a treatment scheme for lecanemab as intravenous infusions at a dose of 10 mg/kg biweekly during the initiation phase of 18 months, after which maintenance doses would be administered every 4 weeks. In the base-case model, the maximum duration of treatment with lecanemab was 3 years because only 3-year data on the efficacy of lecanemab (from the 18-month trial period and the non-randomized open-label extension phase) were publicly available at the time of the current study [23, 24]. Although extrapolation of treatment efficacy is common in economic models, we wanted our base-case analysis to rely on robust data and extrapolate the efficacy of lecanemab beyond the available data in scenario analyses.

The Markov cohort model included nine states: non-institutionalized MCI due to AD, non-institutionalized mild AD dementia, non-institutionalized moderate AD dementia, non-institutionalized severe AD dementia, institutionalized MCI due to AD, institutionalized mild AD dementia, institutionalized moderate AD dementia, institutionalized severe AD dementia, and death (Fig. 1). The starting population of the Markov cohort model was constructed to represent non-institutionalized patients with AD who are eligible for treatment with lecanemab in Sweden. The average age of the starting population was 70 years (range 50–90), 52% were female, 74% were in the MCI stage, and 41% were APOE ε4 non-carriers. The details of the study population are reported in the methods section of the electronic supplementary material (ESM) and Table 1 in the ESM.

Fig. 1

Model structure. AD, Alzheimer’s disease; LY, life-year; MCI, mild cognitive impairment; QALY, quality-adjusted life-year; SoC, standard of care

Transition probabilities for progression through AD dementia stages under SoC were estimated using SveDem linked to other Swedish registers (Fig. 2). Progression probabilities from MCI due to AD to subsequent stages under SoC were estimated using data from the National Alzheimer’s Coordinating Center (NACC) database or derived from previous studies [25]. For lecanemab combined with SoC, progression probabilities were adjusted with a hazard ratio (HR) of 0.69 reflecting the treatment efficacy of lecanemab [5]. The HR was only applied to transitions from MCI due to AD to mild AD dementia and from mild AD dementia to moderate AD dementia in community settings in the base-case model. This is because institutionalization in Sweden typically implies more advanced stages of AD, which contradicts the indications of lecanemab for early AD. Moreover, institutionalized older adults are generally frailer and more likely to have complex health problems, increasing the uncertainty around the efficacy and safety of lecanemab in this population [26]. We believe this population is unlikely to be a target group for lecanemab in Sweden. Therefore, treatment effects were not applied to this group in the base model but were considered in a scenario analysis.

Fig. 2

Swedish databases and their use in this study. Aβ, amyloid-beta; AD, Alzheimer’s disease; NPR, the Swedish National Patient Register; SOL, the Register of social care for elderly and persons with functional disability; SveDem, the Swedish Register of Cognitive/Dementia Disorders

The costs of care by AD stages and settings were estimated from the Swedish registers and applied to both treatment strategies. Additional costs associated with lecanemab administration, including costs related to infusion, monitoring, physician visits, and costs incurred by adverse events, were added to the lecanemab combined with SoC strategy. The drug cost for lecanemab was not included in the model because the study aim was to estimate the annual drug price of lecanemab.

The utility of each AD stage by settings was estimated through meta-analyses of previous studies (reported in the ESM). For lecanemab combined with SoC, disutilities due to serious adverse events were also considered. All costs were converted to 2023 SEK using the consumer price index from Statistics Sweden. In the base-case model, an annual discount rate of 3% for both costs and utilities was applied, as recommended by the Dental and Pharmaceutical Benefits Agency.

Life-years (LYs), quality-adjusted LYs (QALYs), and total costs for each treatment strategy were estimated for individual patient profiles, incorporating the treatment duration of lecanemab to account for lecanemab discontinuation. These estimates were then weighted by the proportion of each patient profile and duration of treatment with lecanemab to derive the average LYs, QALYs, and total costs for the starting population. The corresponding threshold annual drug price was calculated by dividing the net monetary benefit by the duration of treatment with lecanemab, assuming a willingness-to-pay (WTP) threshold of 1 million SEK per QALY [27].

The simulation model and calculation of LYs, QALYs, and costs were done using the R package ‘heemod’ (R version 4.3.3, R Foundation for Statistical Computing, Vienna, Austria). The economic evaluation was reported according to Consolidated Health Economic Evaluation Reporting Standards 2022.

2.4 Model Parameters2.4.1 Transition Probabilities Across AD-Institutionalization Stages

The transition probabilities for AD dementia for patients receiving SoC were estimated using cumulative hazards estimated from multi-state flexible parametric survival models based on individual-level data from SveDem linked to the Register of social care for elderly and persons with functional disability and the Swedish Cause of Death Register (Fig. 2) [16,17,18,19,20]. These registers contain information on longitudinal cognitive tests, institutionalization status, and dates of death in individuals with AD dementia. Individuals included in the survival analyses met the following criteria: (1) with a registered dementia diagnosis in SveDem between 1 May 2007 and 31 December 2020, (2) aged 50–90 years at the time of diagnosis, (3) with an abnormal cerebrospinal fluid (CSF) Aβ42/Aβ40 or Aβ42/p-tau181 ratio or receiving a diagnosis of AD in specialist settings and receiving a CSF test with its results not registered, and (4) without a history of stroke, bleeding disorders, or seizures. Details of the estimation of transition probabilities for the progression of AD dementia to next stages and evaluations of the fit of the multi-state flexible parametric survival models are reported in the Methods in the ESM and Fig. 1 in the ESM. The flexible parametric survival models were performed using the R package ‘flexsurv’.

For the MCI stage in SveDem, data were insufficient to estimate transition probabilities of progression from MCI due to AD to mild AD dementia, institutionalization, and death. Therefore, we used data from the NACC database to estimate the probabilities of transitioning from MCI to mild AD and from MCI to death with flexible parametric survival models. Individuals included in the analytical sample met the following criteria: (1) aged 50–90 years, (2) diagnosed with MCI, (3) exhibited abnormally elevated Aβ in CSF or on positron emission tomography, or were diagnosed with MCI due to AD in more than half of their evaluations, (4) were not APOE ε4 homozygotes, and (5) had no history of stroke. The progression from MCI to mild AD dementia was defined as having been diagnosed with dementia after the MCI diagnosis. The transition from MCI to death was defined as death occurring in the MCI stage without prior progression to mild AD dementia. We imputed the event time of conversion from MCI to AD dementia to address the issue of interval censoring in NACC data with R package “icenReg” [28].

For transitions from MCI due to AD to death and from AD dementia to next stages, the flexible parametric survival models included age and sex as the covariates. For transitions from MCI due to AD to mild AD dementia, the flexible parametric survival model additionally included APOE genotype (non-carriers vs. heterozygotes). The baseline characteristics of the SveDem sample and the NACC sample used for these survival models are reported in Table 2 in the ESM.

The probability of institutionalization in individuals with MCI was estimated using data from a previous Swedish study on institutionalization incidence in older adults, combined with HR for age for institutionalization in MCI from a prior study conducted in the Netherlands (Table 1) [29, 30].

Table 1 Input parameters for the base-case model2.4.2 Treatment Efficacy and Treatment Discontinuation of Lecanemab and Associated Adverse Events

The 18-month lecanemab trial reported an HR of 0.69 for progressing to the next stages, comparing the lecanemab group with the placebo group [5]. During the trial, 18.8% of the participants discontinued treatment [5]. The treatment efficacy and discontinuation probability reported from the trial were used as input parameters, and the treatment effect was applied to MCI due to AD and mild AD dementia in community settings in our model. We converted the 18-month discontinuation probability into a per-cycle probability, assuming a constant discontinuation rate over time, independent of demographic and genetic factors.

Evidence on the long-term efficacy of lecanemab is limited. Published data from the 2-year gap period of the trial, during which lecanemab treatment was paused, showed that Aβ reaccumulated and the rate of cognitive decline became similar between the lecanemab group and the placebo group [23]. On the other hand, data from the open-label extension phase showed that, when both the original treatment and placebo groups received lecanemab, their cognitive decline rate was slower than predicted by natural history models based on Alzheimer’s Disease Neuroimaging Initiative data for at least an additional 18 months [24]. Therefore, in the base-case model, we assumed that lecanemab was administered for up to 3 years and was effective when on treatment but no longer effective in slowing disease progression as soon as the treatment stopped. This was done by applying a waning factor of 1 for lecanemab after treatment discontinuation (i.e., 100% reduction of treatment effect resulting in a treatment effect HR for lecanemab of 1 instead of 0.69, no longer reducing the risk to progression to a more severe state), using the same equation as previously published [33].

Amyloid-related imaging abnormalities with oedema (ARIA-E) and with hemosiderin deposits (ARIA-H) are common adverse effects of lecanemab and can incur additional costs and affect quality of life [6]. We assumed that 70% of ARIA-E occurred in the first 3 months of treatment, 20% of ARIA-E occurred between 3 and 6 months after the treatment initiation, and 10% of ARIA-E occurred at a constant rate 6 months after the treatment. Isolated ARIA-H was assumed to occur at a constant rate [6]. The probability of mild to moderate ARIA as well as serious to severe ARIA by APOE ε4 genotypes were derived from the results from the core trial and open-label extension phases of lecanemab (Methods in the ESM) [6].

2.4.3 Health-Related Quality of Life

Health-related quality of life (HR-QoL) was expressed as health utilities for each AD stage. A systematic review summarized studies on HR-QoL reported by patients with dementia or their caregivers, published between January 1990 and April 2017 [34]. A researcher (YL) searched for additional literature in PubMed between April 2017 and September 2024 using the same search strategy as in the systematic review and extracted HR-QoL from these studies [34]. Another researcher (XX) reviewed the studies and the extracted data. Discrepancies in the extracted data were resolved through discussion between the two researchers. We excluded studies that (1) focused on dementia types other than AD, such as vascular dementia; (2) lacked information on the study setting; (3) did not report stage-specific HR-QoL; or (4) lacked information on standard deviations of HR-QoL. The identified studies, reasons for exclusion, and the final studies from which HR-QoL values were used in the present study are provided in the ESM. As the aim of the literature search was to extract input parameters for our economic evaluation rather than to conduct a comprehensive systematic review, no quality assessment or risk-of-bias assessment was performed.

We estimated HR-QoL for each AD stage through random-effects meta-analysis of the included studies using the R package ‘meta’. An exception was MCI due to AD in institutional settings, for which meta-analysis was not performed because only one study was available. HR-QoL for each AD stage was calculated separately for institutional and community settings. For MCI due to AD, self-reported HR-QoL was used. For all stages of AD dementia, caregiver-reported HR-QoL was used in the meta-analysis, considering that self-reported HR-QoL from individuals with dementia may not accurately reflect the HR-QoL of the dementia stages of AD. Heterogeneity among the studies was high across all AD stages, with I2 exceeding 85% for MCI due to AD in community settings and 95% for the other AD stages.

Disutilities due to serious adverse events were implemented for the treatment strategy of lecanemab combined with SoC. We used disutilities caused by the following conditions with similar symptoms: rash for serious infusion-related reactions and delirium for serious ARIA [31, 32]. The disutilities were applied additively to the state-specific utility in the model.

2.4.4 Costs and Resource Use

Annual costs of each AD stage by settings under SoC included costs of inpatient care, outpatient specialist care, prescribed drugs, and social care. The costs were estimated following a previously published approach using SveDem and Swedish healthcare registers (Fig. 2) [3]. Briefly, inpatient care and outpatient specialist care were identified through the Swedish national patient register, and the associated costs were calculated using year-specific diagnostic-related group weights from a Swedish database that contains information on the unit costs of inpatient and outpatient care (KPP). Costs of prescribed drugs were derived from the Prescribed Drug Register. The resource utilization of social care, including institutionalization, short-term residence, housing support, home service, and daytime activity, was derived from the Register of social care for elderly and persons with functional disability. The unit costs of each social care were derived from the municipality and region database (KOLADA). The unit costs of social care are reported in Table 3 in the ESM. Annual costs by AD stages were calculated by dividing the total costs incurred during that stage by the total person-years spent in the stage.

Costs incurred by treatment with lecanemab included costs of intravenous drug administration, clinical evaluations, monitoring, and management of adverse events. As data on the costs associated with lecanemab intravenous infusions were lacking, we used the average costs of administering intravenous medications for nervous system diseases as a proxy. These costs were estimated using data on reception fees, personnel fees, and material costs from KPP. Following recommendations for the appropriate use of lecanemab, we assumed that patients would undergo clinical evaluations by neurologists and magnetic resonance imaging (MRI) monitoring twice within the first 3 months, followed by three additional evaluations at 3-month intervals [21]. The unit costs for intravenous infusions, clinical evaluations by neurologists, and MRI monitoring are reported in Table 1 and in Table 3 in the ESM.

Costs associated with managing adverse events related to lecanemab were estimated following a similar approach in an appraisal of lecanemab by the UK National Institute for Health and Care Excellence and assumed that the costs would be incurred mainly by serious infusion-related reactions and moderate to serious ARIA (Table 3 in the ESM) [13]. Serious infusion-related reactions would be managed by pharmacological treatments. Management of mild to moderate ARIA would mainly involve a clinical assessment and monitoring with MRI, whereas serious ARIA would additionally involve inpatient care and pharmacological treatments.

2.5 Sensitivity Analyses

We conducted one-way sensitivity analyses to test the influence of uncertainties around the treatment efficacy of lecanemab and the utility of each AD stage (Table 4 in the ESM). Additionally, we examined the impact of varying the annual discount rate for costs and health effects on our results, setting it at 0% and 5%.

In contrast to probabilistic sensitivity analyses (PSA) in previous studies, which were typically performed by drawing random values from the distributions of the individual input parameters, we performed PSA combining non-parametric and parametric bootstrapping with 10,000 iterations. In each iteration, transition probabilities were estimated by randomly drawing from the multivariate normal distribution of the maximum likelihood estimates from the flexible parametric survival models, costs were estimated using non-parametric bootstrapping, and other input parameters, as outlined in Table 4 in the ESM, were resampled from their respective distributions. Subsequently, the Markov cohort model was run using these input parameters. The PSA was conducted for males and females aged 70 years, which represents the mean age of the starting population.

2.6 Scenario Analysis

We conducted a series of scenario analyses to evaluate the impact of varying assumptions on our results. These scenarios encompassed two primary categories: (1) assumptions regarding treatment efficacy and maximum treatment durations and (2) assumptions related to patient populations receiving lecanemab and discounting health benefit.

Scenario analyses for treatment efficacy and duration assumptions included (1) assuming the effect of lecanemab varied by subgroups (Methods in the ESM), (2) assuming a residual treatment effect of lecanemab after discontinuation with waning factors of 0.1, 0.25, and 0.5 (i.e., 10%, 25%, and 50% reduction of treatment effect each cycle after the treatment stops), (3) assuming that lecanemab was only administered for a maximum of 18 months, and (4) assuming that lecanemab was administered for maximum durations of 4, 5, 6, 7, 8, or 9 years or until progression to moderate AD dementia across the 10-year time horizon.

Scenario analyses concerning patient populations and discounting health benefit included (1) restricting lecanemab use to APOE ε4 non-carriers and applying their subgroup-specific effect estimate, (2) restricting lecanemab use to APOE ε4 heterozygotes and applying their subgroup-specific effect estimate, (3) assuming treatment with lecanemab for institutionalized patients, and (4) discounting costs at 3% and QALYs at 0%.