The Potential Economic Value of Lecanemab in Patients with Early Alzheimer’s Disease Using Simulation Modeling

AD Archimedes condition-event (ACE) is a patient-level simulator that captures complex interactions among key AD pathology components (e.g., underlying biomarkers such as Aβ measures and tau levels) and the clinical presentation of AD, based on various patient-level scales of cognition, behavior, function, and dependence (Fig. 1) [11, 12]. Comprehensive external validations using patient registries, clinical trials, and published literature have shown that AD ACE can correctly predict and provide reasonable transition estimates of AD dementia, institutionalization, and mortality over a long-term horizon [11, 12].

The model simulated the effects of disease modification and early intervention on disease progression with impacts on clinical and economic outcomes. For early AD, the relationships among changes were quantified using predictive mixed linear equations derived from long-term observational data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) [13]. As patients progress to more severe stages of AD that the ADNI study does not effectively represent, the AD ACE switches to Assessment of Health Economics in Alzheimer’s Disease II (AHEAD) equations for cognition and behavioral scales to make the model more representative and accurate across all stages of AD [14, 15]. In this study, AD ACE triggered a switch between the ADNI and AHEAD equations once a simulated patient progressed into moderate AD and the Mini Mental State Examination (MMSE) score fell below 15. Additionally, the model captured transitions to/between community and institutional care settings as patients progressed to more severe stages of AD.

The model captured the mean and incremental patient life years (LY), QALYs, delay to onset of AD dementia, and different treatment- and AD-related care costs. Direct and indirect care costs for patients and caregivers were reported separately in the community and institutional care settings. The overall QALYs were broken down to patient QALYs, caregiver QALYs lost, and QALYs lost to amyloid-related imaging abnormalities-edema/effusion (ARIA-E) adverse events (AE). An annual discount rate of 3% was applied to all health and economic outcomes in accordance with the recommendations for cost-effectiveness analyses in the USA [16].

An early AD population was selected for the base-case analysis; this aligned with the patient characteristics in both Study 201 and the ongoing phase III CLARITY AD trial (i.e., MCI due to AD and mild AD dementia and confirmed Aβ pathology). A total of 1735 ADNI participant profiles were available and 429 profiles were selected as inputs for the AD ACE simulator; this subgroup was matched on key Study 201 inclusion criteria: ages 50–90 years, Mini-Mental State Examination ≥ 22, and amyloid PET SUVr level ≥ 1.1 [7]. The mean baseline characteristics of the selected profiles aligned closely with the placebo and 10 mg/kg biweekly lecanemab groups of the trial (Table 1). A total of 2000 individual patient profiles were sampled with replacement from the selected subgroup of 429 ADNI patients and simulated separately on the lecanemab + SoC and SoC alone arms to capture AD disease trajectory and treatment effect.

The modeled base case explored the early AD population of MCI due to AD and mild AD dementia and confirmed Aβ pathology. Assessments were conducted for key patient subsets and alternative treatment stopping rules and dosing regimens. Sensitivity and scenario analyses were conducted to explore the impact of different settings and parameter uncertainty on the outcomes.

AD ACE was also recently used to study the long-term health outcomes of lecanemab in patients with early AD [17]. All the treatment settings and assumptions considered in the AD ACE model for the health outcomes study and this VBP assessment were aligned except the assumed rate of treatment discontinuation in the base-case scenario which resulted in slight difference in reported outcomes common between the two studies. No treatment discontinuation was assumed in the health outcomes study to help assess the true efficacy of treatment with lecanemab and its impact on clinical outcomes in study participants who strictly adhered to the protocol and planned treatment whereas in this study the observed rate of discontinuation in Study 201 was considered.

The BAN2401-G000-201 (Study 201) trial was conducted in accordance with the Declaration of Helsinki and the International Council for Harmonization and Good Clinical Practice guidelines and was approved by the institutional review board or independent ethics committee at each site. All patients provided written informed consent. An independent interim monitoring committee was responsible for oversight and conduct of the interim analyses and response-adaptive randomization design to evaluate the safety routinely and review the futility analysis results. This assessment is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors. Model parameters related to costs, mortality, disease progression equations, risk of transition to institutional care, and health-related quality of life for patients and caregivers were primarily informed by published literature. Other parameters in the model such as the baseline population, treatment effect, and treatment discontinuation were informed by the Study 201 outcomes. Additional details regarding each of these are provided below.

The natural history of AD progression for patients receiving SoC was modeled using AD ACE based on disease equations developed from longitudinal patient-level ADNI data for early AD [13] and published AHEAD equations for more severe stages of AD [14, 15]. CDR-SB thresholds were used to determine patients’ disease severity at baseline and over time in AD ACE (i.e., MCI due to AD < 4.5, mild AD ≥ 4.5 to < 9.5, moderate AD ≥ 9.5 to < 16, severe AD ≥ 16) [18]. Accordingly, the proportion of patients with MCI due to AD at the start of the simulation in AD ACE (i.e., CDR-SB score < 4.5 corresponding to CDR Global = 0.5) should be compared to the proportion of patients with CDR Global of 0.5 at baseline in Study 201. Disease severity levels assigned by ADNI on the basis of clinical evaluation of patients are also provided in Table 1 and should be compared to the reported proportion of patients with MCI due to AD diagnosis in Study 201. Model inputs such as patient utility, caregiver disutility, care cost, and hazards of mortality and institutional care changed with disease progression.

Mortality was modeled by applying hazard ratios (HR) to age-specific US general population survival [19] to naturally increase the probability of death across all AD severity stages with age. The HR estimates from Andersen et al. [20], a random, population-based cohort study, were used for the base-case settings (Table 1). These HR estimates were reported by AD severity defined by CDR-SB scores in Andersen et al. [20]. The results in Andersen et al. demonstrated that the presence and severity of AD dementia corresponded to reduced survival based on a 14-year study duration. The current modeling analysis assumed that MCI due to AD did not impact mortality risk. The HRs from Wimo et al. [21] were explored in a scenario analysis.

The incidence-based institutionalization estimates leveraging Consortium to Establish a Registry for AD (CERAD) data were used in the model to inform the probability of transitioning from the community to nursing home (i.e., institutionalized) care by AD severity level (Table 1) [22]. No risk of admission to institutional care was assumed for patients with MCI due to AD as a result of limited published evidence and minimal impact on the outcomes. Alternative prevalence-based institutionalization data from Davis et al. [23] that reflected the proportion of patients institutionalized at each AD severity level was explored in scenario analyses. AD severity reported in both selected studies were defined on the basis of CDR-Global score.

The treatment effect in this study was modeled on the basis of the key assumption that the effect of lecanemab on the clinical outcomes is correlated with amyloid PET level as a surrogate endpoint [24, 25]. In AD ACE the relationships between biomarkers of disease and clinical outcomes are based on correlations mainly observed in the ADNI data. Disease equations are evaluated repeatedly at subsequent time intervals every 6 months to estimate the AD disease trajectory of patients. An anti-amyloid DMT can be potentially modeled in AD ACE by imposing effects on estimated amyloid PET SUVr outcomes of a simulated patient. In this study, a calibration process to only adjust the predicted measures of amyloid PET SUVr at each time interval was used to model the treatment effects of lecanemab over time (Fig. 2). Amyloid PET is a predictor in all AD ACE disease equations; therefore, any calibrated reduction in amyloid PET SUVr at a given time interval impacts the prediction of amyloid PET SUVr and other modeled AD biomarkers and scales such as CDR-SB at later time intervals. Calibration is a process of adjusting the parameters of a model until a chosen sample reaches the known outcome within an acceptable range.

Reported data from the 10 mg/kg biweekly group of Study 201 on change in amyloid PET SUVr from baseline was used as the basis for calibrating mean amyloid reductions required at different time intervals to closely match the observed percentage change in CDR-SB from baseline in Study 201 during the first 18 months of treatment initiation. The data on change in amyloid PET from a model-based simulation study that explored the effect of continued treatment with 10 mg/kg intravenous biweekly dose of lecanemab [26] was then used to calibrate treatment effect beyond the Study 201 time horizon until the mean amyloid level reached the mean amyloid level observed in cognitively normal individuals in the ADNI data set [13]. Additional reduction in amyloid PET level was then calibrated to maintain the cognitively normal mean amyloid level for patients while they remain on lecanemab, an assumption validated by clinical experts. The calibration process only adjusted the estimated PET SUVr values over time and did not impact default AD ACE equations.

A lifetime simulation of 2000 sampled ADNI patients was used during the treatment effect calibration process where treatment discontinuation was not allowed. The resulting trajectories of amyloid PET and CDR-SB for both lecanemab + SoC and SoC arms (Fig. 2) indicated that the change relative to SoC achieved in amyloid PET SUVr and CDR-SB in the AD ACE model mirrored Study 201 results during the trial duration.

In the base-case scenario, patients on treatment were assumed to receive a 10 mg/kg intravenous biweekly dose of lecanemab, which resulted in a 26% change from baseline in CDR-SB. This change was estimated on the basis of the conventional mixed model for repeated measures (MMRM), which cannot generate a confidence interval around ratios (i.e., percentage reduction or percentage change from baseline). Consequently, a ± 15% variation was considered in the sensitivity analyses to explore the uncertainty surrounding lecanemab treatment effect. This was consistent with the reported range of percentage reductions estimated from six different statistical methods in a recent study [27]. The impact of alternative treatment dosing regimens and their potential impact on treatment effect during a maintenance phase were also explored in the scenario analysis. Outcomes reported from the model-based simulation study were used to inform alternative treatment effect scenarios in the model, where long-term maintenance dosing of lecanemab beyond the 18-month trial period was investigated [26]. The model-based simulation data suggested that less frequent dosing may be feasible to prevent re-accumulation of amyloid and maintain treatment effect beyond the trial duration [26]. For each dosing regimen scenario, the mean amyloid reduction was recalibrated in the maintenance time intervals by proportionally adjusting the mean amyloid reductions estimated in the base-case setting.

The discontinuation risk in Study 201 was 35.6% (217 subjects) in the lecanemab arm vs. 23.3% (57 subjects) in the placebo group over the 18-month trial period [9]. The risk was higher in the lecanemab 10 mg/kg monthly (92 [36.4%]) and biweekly groups (71 [44.1%]) for reasons considered unrelated to treatment (i.e., at the discretion of regulatory authorities). Approximately one-quarter (24.8%) of participants in the 10 mg/kg biweekly group discontinued treatment because of treatment-related AEs, subject choice, or inadequate therapeutic effect. This translated into a 17.3% annual risk of discontinuation, which was used in the base-case analysis. Alternative annual risk levels of 10% and 20% were explored in the sensitivity analyses. Progression to the moderate AD health state (i.e., CDR-SB ≥ 9.5) also triggered treatment discontinuation in the base case. Treatment stopping rules based on fixed treatment durations of 1.5, 3, and 5 years where patients discontinued treatment were also explored in scenario analyses. Once a patient discontinued treatment because of any of the modeled conditions in this study (i.e., risk, progression, or stopping rule), the mean calibrated amyloid reductions estimated over lifetime were not applied in time intervals beyond the discontinuation time, so the changes in amyloid level at each time interval were estimated solely on the basis of evaluation of the amyloid PET disease equation in AD ACE.

The incidence rates of serious AEs and treatment-emergent AEs other than ARIA-E (e.g., falls) and infusion reactions were similar in both treatment arms in Study 201 and were consistent with AEs observed in an early AD population. Infusion reactions were mainly mild to moderate (grade 1–2) and typically responded to prophylactic treatment. Sixteen cases of ARIA-E were observed in 161 patients in the 10 mg/kg biweekly group; three were symptomatic cases of headaches, visual disturbances, or confusion [9]. Hence, the model only accounted for the impact of ARIA-E in the first year with a 9.9% occurrence rate, and 18.8% of these cases were symptomatic. ARIA-E events only resulted in short treatment interruptions in Study 201; therefore, no treatment discontinuation was considered in the model as a result of ARIA-E events.

Patient utilities were informed by Landeiro et al. [28], a fixed-effect meta-analysis using studies from a systematic literature review. The same utility estimated on the basis of disease severity was used in the community and institutionalized settings (Table 1). The studies identified in the systematic review defined AD severity on the basis of MMSE, CDR-Global, Alzheimer’s Disease Assessment Scale-Cognitive Subscale, the Global Deterioration Scale, or a combination of two or more of these measures [28]. Values from Neumann et al. [29], which defined AD severity on the basis of the CDR-Global score, were considered in the scenario analysis. An estimated disutility of − 0.14 for headache [30] was also applied to patients experiencing ARIA-E for 12 weeks [9]. Caregiver disutilities were taken from Mesterton et al. [31]; patients were assumed to have one caregiver.

The study analyzed the direct and indirect community and residential care costs for patients and caregivers. US cost data across the full disease continuum are lacking, so inputs from multiple sources were combined as needed. The results from GERAS-US [32], a prospective, longitudinal cohort study adapted from GERAS I [33], were used to inform the community-based care costs for patients with MCI and mild AD. GERAS-US assessed the cross-sectional total societal costs for patients with MCI and mild AD, including direct medical and non-medical costs for healthcare resource use and indirect costs for caregiving. Both GERAS I and GERAS-US used the MMSE scale to define AD severity [32, 33]. GERAS I was conducted in multiple European countries and reported on community-based costs for patients with mild, moderate, and severe AD dementia; the study found that costs increased with disease severity. The mean monthly costs for patients with moderate and severe AD dementia were not reported in GERAS-US. Therefore, the mean relative ratio between estimated costs for mild AD vs. moderate and severe AD from GERAS I for three European countries was computed and applied to the community-based care costs for mild AD from GERAS-US to approximate mean US monthly costs. The computed mean relative ratios for mild-to-moderate AD (1.3) and mild-to-severe AD (1.9) were well aligned with the findings from Leon et al. [34] and Small et al. [35], on relative change in average cost of care by disease stage. Additionally, these relative ratios calculated on the basis of costs reported in GERAS I are in line with the average ratios (across all three age groups) used by the Alzheimer’s Association [36] for mild-to-moderate AD (1.3) and mild-to-severe AD (2.0).

The direct non-medical costs in the residential care setting were informed by Genworth’s Cost of Care Survey tool [37]. The 2021 monthly median cost for a private/semi-private room in a US residential care facility was $8477; the model did not adjust this cost by disease stage since it is usually a flat rate. Other residential care costs were assumed to be the same as community-based care costs for both patients and caregivers except for indirect non-medical costs for caregiving where only 44% of the informal care cost in the community-based setting was applied [38].

Patients receiving lecanemab accrued monitoring costs which were assumed to be the cost of five magnetic resonance imaging scans in the first year. The unit costs were collected from Centers for Medicare and Medicaid Services (CMS) Physician Fee Schedule [39]. Symptomatic treatment costs were accounted for in GERAS estimates and thus not included in the analysis. Medical resource use (MRU) associated with ARIA-E stratified by severity and type (symptomatic/asymptomatic) was informed by the Study 201 protocol. MRU costs were collected from the CMS database [40] and IBM® RED BOOK® [41]. The costs of diagnostics and screening (e.g., CSF and PET scans) were not considered in the base case because of uncertainties around requirements and potential variations in reimbursement policies; instead, these costs were explored in scenario analyses. All costs were inflated to 2021 US dollars using the US Bureau of Economic Analysis price index for personal consumer expenditures for healthcare [42]. The breakdown of all cost categories is provided in Table 2.