Population pharmacokinetics and exposure–response analyses of safety (ARIA‐E and isolated ARIA‐H) of lecanemab in subjects with early Alzheimer's disease

Abstract

Lecanemab (Leqembi®) was recently approved by health authorities in the United States, Japan, and China to treat early Alzheimer's disease (AD), including patients with mild cognitive impairment (MCI) or mild dementia due to Alzheimer's disease upon confirmation of amyloid beta pathology. Extensively and sparsely sampled PK profiles from 1619 AD subjects and 21,929 serum lecanemab observations from two phase I, one phase II, and one phase III studies were well characterized using a two‐compartment model with first‐order elimination. The final PK model quantified covariate effects of body weight and sex on clearance and central volume of distribution, ADA‐positive status, and albumin on clearance, and of Japanese ethnicity on central and peripheral volumes of distribution. Exposure to lecanemab was comparable between two lecanemab‐manufacturing processes. However, none of the identified covariates in the model had a clinically relevant impact on model‐predicted lecanemab C _max or AUC at steady state following 10 mg/kg bi‐weekly. Importantly, age, a well‐recognized risk factor for AD, was not found to significantly affect lecanemab PK. The incidence of ARIA‐E as a function of lecanemab exposure was modeled using a logit function with data pooled from 2641 subjects from the phase II and phase III studies, in which a total of 177 incidences of ARIA‐E were observed. The probability of ARIA‐E was significantly correlated with model‐predicted C _max and predicted to be higher in subjects homozygous for APOE4. The incidence of isolated ARIA‐H was not associated with lecanemab exposure and was similar between placebo and lecanemab‐treated subjects.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

It is well known that there is an association of occurrence of ARIA with the treatment of anti‐amyloid antibodies in AD.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This article describes serum lecanemab population pharmacokinetics and exposure–response analysis for pre‐specified adverse events of special interest ARIA‐E and isolated ARIA‐H.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The PK of lecanemab was well characterized by linear, two‐compartment model, and identified covariates were consistent with other monoclonal antibodies. The exposure–response analysis for safety showed that ARIA‐E was significantly correlated with steady‐state lecanemab serum concentration and predicted to be higher in APOE4 homozygous subjects, whereas the incidence rate of isolated ARIA‐H was independent of lecanemab exposure and was similar between placebo and lecanemab‐treated subjects.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

The developed PK and PK/PD models provide insights into the effect of lecanemab dosing on the incidence of ARIA‐E and isolated ARIA‐H.

INTRODUCTION

Alzheimer's disease (AD) is a progressive, neurodegenerative disorder of unknown etiology and the most common form of dementia among older people. Risk factors for AD are increasing age, genetics, and family history. ¹ While several genes are associated with an increased risk of AD, the ε4 allele of the apolipoprotein E (APOE4) is the strongest known genetic risk factor. ² , ³ AD is defined biologically by the presence of two abnormal protein deposits: extracellular deposits of brain amyloid plaques (comprising β‐amyloid (Aβ) peptides) and neurofibrillary tangles (comprising abnormal tau protein). Biomarker, clinicopathological, and cohort studies indicate that the disease process commences 10–20 years before the clinical onset of symptoms. ⁴ , ⁵ , ⁶ The “amyloid cascade” hypothesis postulates that neurodegenerative processes in AD are driven by accumulation of aggregated Aβ species, which result from an imbalance between Aβ production and Aβ clearance in the brain. ⁷

Lecanemab is a humanized, high‐affinity monoclonal antibody binding to soluble amyloid‐beta (Aβ) protofibrils, which have been shown to be more toxic to neurons than monomers or insoluble fibrils. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ Lecanemab was safe and well tolerated in two phase I studies, with dose‐proportional exposure. ¹⁴ In a dose‐finding phase II study (Study 201/ClinicalTrials.gov Identifier: NCT01767311), lecanemab treatment led to a dose‐dependent reduction in brain amyloid, a slowing of clinical decline across a number of outcome measures, and directionally consistent biomarker changes at 18 months in subjects with early AD (mild cognitive impairment (MCI) due to AD or mild AD dementia). ¹⁵ In a phase III confirmatory study (Study 301/Clarity AD/ClinicalTrials.gov Identifier: NCT03887455), lecanemab 10 mg/kg bi‐weekly significantly reduced markers of amyloid in early AD and resulted in reduced decline in measures of cognition and function compared to placebo at 18 months. ¹⁶ In both studies, lecanemab was generally well tolerated. The most frequently observed treatment‐emergent adverse events were amyloid‐related imaging abnormalities–edema/effusion (thereafter ARIA‐E), defined as signal hyperintensities on fluid‐attenuated inversion recovery MRI sequences due to parenchymal fluid accumulation or sulcal fluid effusion, ¹⁷ and infusion reactions. ¹⁵ , ¹⁶ Most ARIA‐E events occurred within the first 3 months of treatment, were generally mild to moderate in severity, and occurred more frequently in APOE4 homozygous carriers. ¹⁵ , ¹⁶ In the phase III study core phase, treatment‐emergent amyloid‐related imaging abnormalities–hemorrhage (thereafter ARIA‐H) adverse events, a type of ARIA with hemosiderin deposits, were observed. ARIA‐H can occur concurrently with ARIA‐E or as isolated ARIA‐H without ARIA‐E. Isolated ARIA‐H events occurred throughout the course of treatment, while ARIA‐H concurrent with ARIA‐E tended to occur early in treatment. The rate of isolated ARIA‐H was low (<9%) and similar in placebo and lecanemab‐treated subjects.

Here, we report updated results from population PK and exposure–response (thereafter ER) analyses of lecanemab correlating lecanemab exposure at steady state with safety end‐point ARIA‐E in patients with early AD on pooled data from the phase II and the phase III studies, and graphical ER for safety end‐point isolated ARIA‐H on data from the phase III study.

METHODS

Study design and treatments

Study designs and treatment regimens across all studies are detailed in Table S1.

Phase II Study 201 Core phase was a double‐blind, parallel‐group, placebo‐controlled, multicenter study utilizing a dose‐finding adaptive randomization design to evaluate the safety, tolerability, and efficacy of lecanemab in subjects with MCI due to AD or with mild AD dementia. Study 201 Core randomized 856 subjects across 6 treatment groups: placebo, 2.5 mg/kg bi‐weekly, 5.0 mg/kg monthly, 5.0 mg/kg bi‐weekly, 10 mg/kg monthly, or 10 mg/kg bi‐weekly for 18 months. The open‐label extension (OLE) phase of Study 201 was initiated to allow subjects to receive open‐label lecanemab 10 mg/kg bi‐weekly for up to 60 months (5 years). ¹⁵ All subjects randomized to lecanemab in Study 201 Core and OLE phase received Process A drug product. Process B drug product was introduced into the Study 201 OLE Phase in the United States, Canada, and South Korea from May 2020, and in Japan from July 2020.

Phase III Study 301 Core phase was an 18‐month, multicenter, double‐blind, placebo‐controlled, parallel‐group trial involving patients with early Alzheimer's disease. Eligible participants were randomly assigned in a 1:1 ratio to receive intravenous lecanemab (10 mg/kg) or placebo bi‐weekly. The randomization was stratified according to clinical subgroup: MCI due to AD or mild AD dementia, the presence or absence of concomitant medication for symptoms of AD at baseline (e.g., acetylcholinesterase inhibitors, memantine, or both), apolipoprotein APOE4 carriers or non‐carriers, and geographic region. During the trial, participants underwent serial blood testing for serum lecanemab and plasma biomarkers. OLE subjects received open‐label lecanemab 10 mg/kg bi‐weekly. ¹⁶ All subjects randomized to lecanemab in Study 301 Core received Process B drug product.

All studies were approved by relevant institutional review boards/ethics committees and conducted in accordance with the International Conference on Harmonization (ICH) of Technical Requirements for Registration of Pharmaceuticals for Human Use and all applicable local Good Clinical Practice (GCP) guidelines, including the Declaration of Helsinki.

Bioanalytical assays and safety assessments

Serum lecanemab concentrations were measured by validated immunoprecipitation–liquid chromatography–tandem mass spectrometry (IP/LC–MS/MS) methods using anti‐human IgG antibody to precipitate lecanemab from a serum sample. The assay was validated with a calibration range of 0.500 (lower limit of quantitation (LLOQ)) to 150 μg/mL. The intra‐run and inter‐run QC in‐precision is generally less than 9.2%, except 18.8% at LLOQ level. For the first phase I study (Study 101), a validated enzyme‐linked immunosorbent assay (ELISA) was used for the measurement of serum concentrations of lecanemab. Lecanemab antidrug antibody (ADA) and neutralizing ADA (NAb) were measured in serum using validated immunoassays.

As part of safety magnetic resonance imaging (MRI), both ARIA‐E and ARIA‐H were monitored and recorded throughout the core and OLE of phase II and III studies. In both studies, safety MRI was conducted at weeks 9 and 13, and 6, 12, and 18 months of treatment. Additional MRI assessment could have been conducted if warranted.

Modeling software

Population PK parameters were estimated using the first‐order conditional estimation with interaction and logistic regression analysis for ARIA‐E was conducted using the conditional estimation method with Laplacian, implemented in NONMEM software (version 7.4.3 or higher; ICON Development Solutions, Ellicott City, MD, USA) aided by PDx‐Pop (version 5.2; ICON Development Solutions). R software (version 4.1.3) was used for the pre‐ and postprocessing of NONMEM, output creation of diagnostic plots, and graphical visualization. Covariate plots were created using R package coveffectsplot (version 1.0.2) and VPCs using PsN (version 5.3.0) and R package xpose4 (version 4.7.2). Simulations were performed using the R package mrgsolve (version 1.0.6). The final NONMEM model codes and abbreviated datasets are provided in Texts S1–S4.

Population PK analysis

The population PK analysis was performed using pooled data from two phase I studies (101 ¹⁴ and 104), and Studies 201 and 301 Core and OLE. The population PK model for lecanemab utilized in the current analysis was developed as previously described. ¹⁸ Lecanemab PK was best described using a linear two‐compartment model with first‐order elimination following intravenous administration parameterized for clearance, central and peripheral volumes of distribution, and inter‐compartmental clearance. Model development steps consisted of structural model selection, covariate search, and final model evaluation.

The impact of covariates in the final PK model on the lecanemab exposures at steady state in terms of area under the curve (AUC_ss) and maximum serum concentration (C _ss,max) was evaluated using forest plots generated using the coveffectsplot R package. Lecanemab exposures at steady state (5th and 95th percentiles) across the represented cut points of the represented range of each covariate were simulated and compared with a reference exposure range based on population typical (median) covariate values. For each covariate of the final model, with the other model covariates at their respective typical values, the median and 90% CIs of each covariate effects were generated based on 1000 simulations using final estimates of typical subject values and the variance–covariance matrix from final PK model. Individual empirical Bayes estimates of the PK parameters from the final model were used to derive C _ss,max, steady‐state minimum serum concentration (C _ss,min) and average steady‐state concentration (C _ss,av), which were then used in the graphical and logistic regression analysis for ARIA‐E and graphical analysis for isolated ARIA‐H.

ER analysis for safety (ARIA‐E and isolated ARIA‐H)

The ER analysis for ARIA‐E was performed with pooled data from Study 201 Core and Study 301 Core and for isolated ARIA‐H with data from Study 301 Core only, since no data were available from Study 201 at time of the analysis.

The incidence of ARIA‐E during treatment was modeled using a logistic regression model. The logit model was of the form:

$$\text{Logit}\left[P\left(Y_i=1\right)\right]=B+\text{Slope}\times {\text{Exposure}}_i+E_{\mathrm{cov}}$$

where:

$P\left(Y_i=1\right)$ denotes the probability that the occurrence of ARIA‐E during the study, Y _i for subject i at some timepoint during the study is equal to 1, B is a baseline logit, slope is the effect of lecanemab exposure modeled as a linear function, exposure is the lecanemab serum exposure for subject i, and E _cov is the effect of covariates as predictors. Lecanemab exposures explored were model‐predicted C _ss,max, C _ss,min, or C _ss,av during treatment.

A set of graphical analyses were performed to assess whether the incidence of treatment‐emergent isolated ARIA‐H was related to model‐predicted lecanemab exposure (C _ss,max, C _ss,av, and C _ss,min).

Covariate analysis

In each analysis, covariates were selected based on clinical relevance and biological plausibility and tested using a stepwise approach. Covariate selection criteria included convergence of the estimation and covariance routines, reasonableness and adequate precision of typical subject and variance estimates (relative standard error (RSE) <50%), and statistically significant difference in objective function (likelihood ratio test) value (OFV) criterion: ≥6.635‐point decrease in OFV (p < 0.01) for forward inclusion, and ≥10.828‐point increase in OFV (p < 0.001) for backwards elimination to support the retention of covariates in the final model.

In the PK analysis, effects of body weight and age at study entry, albumin, dose, sex, race/ethnicity, ADA status at sample level (positive or negative), and ADA titer at the time of the PK sample on clearance and of body weight, age, sex, and race/ethnicity on central and peripheral volumes of distribution were evaluated as covariates. Additionally, the compatibility between two manufacturing processes, Processes A and B, was also evaluated. In the categorical covariate analysis for ADA, both ADA negative conclusive and ADA negative inconclusive were assigned as ADA negative because no systematic exposure difference was observed. ¹⁹ All PK observations with missing ADA status were assumed to be ADA negative.

In the ER analysis for ARIA‐E, effects of body weight, age, race/ethnicity, sex, clinical subgroup (MCI vs mild AD), APOE4 carrier status (carrier vs. non‐carrier), APOE4 genotype (heterozygous and homozygous vs. non‐carrier), ADA status at subject level (positive vs. negative) and NAb status (positive vs. negative) at subject level, concomitant AD symptomatic medications, and baseline Mini‐Mental State Examination (MMSE) were evaluated as predictors of ARIA‐E incidence.

Model evaluation

The final PK model was evaluated using goodness‐of‐fit plots, prediction‐corrected visual predictive checks (pcVPC), ²⁰ and non‐parametric bootstrapping. ²¹ , ²² For ARIA‐E, the final logistic regression model was evaluated using non‐parametric bootstrapping and by overlaying model‐predicted proportion of subjects experiencing ARIA‐E with the observed proportion at each C _ss,max quartile. For estimates of precision, asymptotic RSEs and non‐parametric bootstrap 95% confidence intervals (CI) were obtained for each of the model parameters.

RESULTS

Population PK analysis

The population PK dataset included 21,929 serum lecanemab observations from 1619 subjects, of which 653 (3.0%) were from Study 101, 395 (1.8%) from Study 104, 7991 (36.4%) from Study 201 Core and OLE, and 12,890 serum lecanemab observations (58.8%) from Study 301 Core and OLE. A total of 614 PK samples were excluded from the PK analysis of which 337 were with serum concentrations below the limit of quantification (BLQ) or BLQ with TAD over 2000 h, 61 with missing PK sampling time, 107 with CWRES >5, 46 with serum concentrations above 600 mg/mL, considered as outliers as these observations are inconsistent within the same subject's profiles and are much higher than those for other subjects, and 30 with TAD over 2000 h, which is more than 10‐fold the average elimination half‐life of 7 days. Summaries of subject baseline demographics and covariates for the population PK dataset are presented in Table S2. Correlation plots for continuous and categorical covariates in the PK dataset are presented in Figure S1, and there was no apparent correlation between covariates except for body weight which was lower in females compared to males and lower in Asians including Chinese, Japanese, and Koreans compared to Whites and Black/African Americans.

A previously described, ¹⁸ two‐compartment PK model was used in the current analysis as a base model to develop the population PK analysis on the pooled dataset from the updated dataset with 1619 subjects. Lecanemab PK was described by a two‐compartment linear model parameterized for the clearance (CL), the volumes of distribution of the central (V1) and peripheral (V2) compartments, and intercompartmental clearance (Q). The results from base PK model development, covariate univariate analysis, and multivariate analysis with backward deletion are presented in Tables S3–S5, respectively. In the base PK model development several models were evaluated with either combined or proportional error models for residual variability, with and without IIV on Q, including and excluding observations (DV) > 600 μg/mL, including and excluding DV with time after dose (TAD) > 2000 h, or excluding observations with conditional weighted residuals (CWRES) > 5. Following base PK model development, a model assessing exposure comparability to lecanemab for manufacturing Process B relative to Process A, using F with inter‐individual variability (IIV), with covariance between CL and V ₁, without IIV for Q, with combined error model for residual variability, excluding DV > 600 μg/mL and TAD >2000 h, and all observations with CWRES >5 converged successfully and resulted in the lowest objective function value (Table S3) and therefore this model was used in all subsequent analysis for covariate effects (Tables S4 and S5).

The final population PK parameters for the final covariate model are presented in Table  1 . All key model parameters were estimated with good precision (%RSE ≤4.23% for core model parameters and ≤11.8% for covariate effects) and the bootstrap confidence intervals and the median values of the distribution of bootstrapped parameter values are consistent with the original parameter estimates and CIs based on asymptotic parameter SE from the final PK model. Consistent with the previously published PK model, ¹⁸ the final model contained statistically significant covariate effects of time‐variant ADA status, body weight, albumin, and sex on clearance, of sex and body weight on central volume of distribution, of Japanese race/ethnicity on peripheral volume of distribution (V ₂), and slightly lower exposure (9.6%) to Process B compared to Process A. Additionally, in the final PK model from the current analysis, Japanese race/ethnicity was significant in central volume of distribution (V ₁). For comparison purposes, the previously published PK model, with all abovementioned effects without the effect of Japanese race/ethnicity on V1, was fitted to the new pooled dataset from the updated dataset with 1619 subjects by setting maximum evaluation (MAXEVAL) to zero. This resulted in higher objective function value of 173949.983 compared to that of 169260.786 from the new analysis, supporting the validity of the newly developed final PK model.

TABLE 1.

Population PK parameters and bootstrap CIs for the final lecanemab covariate model.

Parameter (Units)	NONMEM		Bootstrap
	Estimate	%RSE	Median	95% CI
PK parameters
CL (L/h)	0.0154	1.60	0.0154	0.0147–0.0160
V 1 (L)	3.24	0.799	3.24	3.18–3.30
V 2 (L)	2.00	4.09	2.02	1.83–2.21
Q (L/h)	0.00718	4.23	0.00701	0.00155–0.0125
F (comparability) for process B	0.904	0.750	0.904	0.890–0.918
Covariate effects
Weight ~ CL (exponent)	0.353	10.5	0.344	0.250–0.460
Albumin ~ CL (exponent)	−0.374	9.71	−0.372	−0.482 to −0.261
Females ~ CL (ratio to males)	0.791	2.17	0.791	0.758–0.824
ADA positive ~ CL (ratio to ADA negative)	1.13	0.860	1.13	1.09–1.17
Weight ~ V 1 (exponent)	0.513	5.01	0.514	0.469–0.558
Females ~ V 1 (ratio to males)	0.868	1.04	0.868	0.853–0.884
Japanese ethnicity ~ V 1 (ratio to non‐Japanese)	0.920	1.58	0.920	0.896–0.945
Japanese ethnicity ~ V 2 (ratio to non‐Japanese)	0.671	11.8	0.665	0.475–0.835
Interindividual variability (%CV)
CL	34.9	4.36	34.8	33.3–36.3
Correlation CL ~ V1 (R)	0.144	27.9	0.145	0.0682–0.205
V 1	12.2	6.53	12.2	11.2–13.1
V 2	94.6	7.28	95.5	83.7–106
F	8.51	21.5	8.54	5.81–10.6
Residual variability
Proportional (%CV)	21.0	1.23	21.0	20.4–21.5
Additive (SD; μg/mL)	1.12	16.8	1.12	<0–1.62

Abbreviations: %RSE, percent relative standard error of the estimate = SE/parameter estimate × 100; ADA, antidrug antibody; CI, confidence interval; CL, clearance; CV%, square root of variance × 100; F, comparability; PK, pharmacokinetic; Q, intercompartment clearance; V ₁, central volume of distribution; V ₂, peripheral volume of distribution.

$$\mathrm{CL}=0.0154*{\left(\frac{\mathrm{BW}}{72}\right)}^{0.353}*{\left(\frac{\mathrm{ALB}}{43}\right)}^{-0.374}*{0.791}^{\mathrm{SEX}}*{1.13}^{\mathrm{ADA}}$$

$$\mathrm{V}1=3.24*{\left(\frac{\mathrm{BW}}{72}\right)}^{0.513}*{0.868}^{\mathrm{SEX}}*{0.920}^{\mathrm{JPN}}$$

$$\mathrm{V}2=2.00*{0.671}^{\mathrm{JPN}}$$

$$F=1*{0.904}^{\text{FORM}.}$$

ADA, 0 (negative) or 1 (positive); ALB, albumin; BW, body weight; SEX, 0 (male) or 1 (female); JPN, 0 (non‐Japanese) or 1 (Japanese); FORM, 0 (Process A) or 1 (Process B).

Eta shrinkage (%): ETA_CL, 6.02%; ETA_V₁, 29.3%; ETA_V₂, 25.8%; and ETA_F1, 60.2%.

The impact of the significant covariates in the final PK model (Table 1) on predicted steady‐state PK exposures (AUC_ss and C _ss,max) after 10 mg/kg bi‐weekly is illustrated using forest plot in Figure 1, which displays a relative change in steady‐state AUC and C _max with 90% CI for each covariate to the reference subject values, defined as a 72 kg male, non‐Japanese subject with albumin of 43 g/L, administered Process A drug with all PK samples ADA negative. As depicted, none of the covariates indicated a clinically meaningful covariate effect of lecanemab exposure; CIs of all covariate effects were within or overlapped the reference 0.8–1.25 interval.

FIGURE 1.

Effect of covariates on lecanemab AUC and C _max at steady state after 10 mg/kg bi‐weekly. Covariate effects are expressed as lecanemab exposures at steady state relative to a reference subject (72 kg male, non‐Japanese subject with albumin of 43 g/L who was administered Process A drug product with all PK samples ADA negative). Body weight and albumin test categories (49 and 99 kg for body weight, 39 and 48 g/L for albumin) represent the 5th and 95th percentiles of PK analysis set, respectively. Gray area indicates the acceptance interval (0.80–1.25). AUC, area under the curve; CI, confidence interval; C _max, maximum concentration; PK, pharmacokinetic.

For the final PK model, goodness‐of‐fit plots did not indicate systematic model deficiencies (Figure S2), except for a slight trend in CWRES in Study 101 for which the reason is unknown. However, there may be a non‐linear PK at lower doses (<1 mg/kg) used in Study 101, and possibly this could also be due to the difference in PK assay, where an ELISA assay was used for Study 101 compared to the IP/LC–MS/MS used for all other studies. Results of pcVPC for the final PK model, stratified by study, are presented in Figure 2. Overall, the pcVPC plots showed good agreement between simulated and observed data, indicating that lecanemab time course has been reasonably well defined by the final PK model with good predictive performance.

FIGURE 2.

Prediction‐corrected visual predictive check plots of observed and predicted lecanemab concentrations for final PK model stratified by study.

For the final PK model, the bootstrap medians were concordant with the population‐predicted values, indicating the final model was valid and stable and produced well‐estimated parameters (Table 1).

Logistic regression analysis for ARIA‐E and graphical analysis for isolated ARIA‐H

The analysis dataset for the incidences of treatment‐emergent ARIA‐E included data from 1789 subjects enrolled in Study 301 Core and 852 subjects enrolled in Study 201 Core. In total, there were 2641 subjects in the ARIA‐E dataset, of whom 1499 subjects received lecanemab and 1142 subjects received placebo. A total of 177 ARIA‐E‐positive subjects were observed in Studies 201 Core and 301 Core, of which 160 subjects received lecanemab (129 following 10 mg/kg bi‐weekly) and 17 subjects received placebo. The analysis dataset for the probability of isolated ARIA‐H included data from 1789 subjects enrolled in Study 301 Core, of whom 150 subjects experienced treatment emergent isolated ARIA‐H events of which 80 subjects received lecanemab 10 mg/kg bi‐weekly and 70 subjects received placebo. Summaries of subject baseline demographics and covariates for the population PK/PD dataset are presented in Table S2.

Initially, a graphical analysis of observed incidence of ARIA‐E categorized by quartiles of average model‐predicted lecanemab C _ss,max, C _ss,min, and C _ss,av suggested that ARIA‐E rates increase with exposure. Logistic regression models explored this relationship, and steps taken in the development of the base logistic regression model for ARIA‐E are summarized in Table S6. There was some apparent correlation among C _ss,max, C _ss,min, and C _ss,av either for pooled data from phase II and 3 studies or just from phase II study. While all exposure metrics were statistically significant and correlated with increasing incidence of ARIA‐E, C _ss,max was the statistically best predictor of ARIA‐E in this analysis. Accordingly, a model with a linear drug effect, with C _ss,max as the exposure parameter predicting ARIA‐E incidence, was selected as the base model for the subsequent univariate addition of covariates as additional predictors of ARIA‐E.

From the univariate addition of covariates, only APOE4 genotype was identified as a statistically significant predictor of ARIA‐E (Table S7). The population parameter estimates for the final logistic regression model for ARIA‐E are presented in Table 2. All key model parameters were estimated with good precision (%RSE <9.9% for core parameters and <36% for covariate effects), the bootstrap confidence intervals are generally narrow, and the median values of the distribution of bootstrapped parameter values are consistent with the parameter estimates and asymptotic standard errors from the final logistic regression model. For each APOE4 genotype, model‐predicted incidence rate and trend are consistent with the observed incidence of ARIA‐E from Studies 201 and 301 (Figure 3).

TABLE 2.

Population parameters and bootstrap CIs for the final logistic regression model for incidence of ARIA‐E.

Parameter	NONMEM		Bootstrap
	Estimate	%RSE	Median	95% CI
Logit = INT + SLP*C ss,max + Cov APOE4Hetero + Cov APOE4Homo
Intercept (INT)	−4.89	5.38	−4.91	−5.55 to −4.39
Slope of lecanemab exposure effect (SLP; per C ss,max unit in μg/mL)	0.00666	9.82	0.00670	0.00540–0.00790
Odds ratio of slope of lecanemab exposure effect (SLP; per C ss,max unit in μg/deciliter)	1.95	9.82	1.95	1.72–2.20
Cov APOE4Hetero: effect of APOE4 heterozygous	0.640	35.8	0.630	0.236–1.15
Odds ratio of effect of APOE4 heterozygous	1.90	35.8	1.88	1.27–3.16
Cov APOE4Homo: effect of APOE4 homozygous	1.91	12.7	1.92	1.52–2.51
Odds ratio of effect of APOE4 homozygous	6.75	12.7	6.82	4.57–12.30

Abbreviations: %RSE, percent relative standard error of the estimate = SE/parameter estimate * 100; APOE4, apolipoprotein E4; CI, confidence interval; C _ss,max, maximum lecanemab serum concentration at steady state; INT, intercept; SLP, slope of lecanemab exposure effect.

FIGURE 3.

Model‐predicted and observed % ARIA‐E incidence rate versus model‐predicted lecanemab maximum serum concentration at steady state (C _ss,max) by APOE4 genotype. In the top panel, filled circles represent observed incidence of ARIA‐E for each lecanemab C _ss,max quartile (1Q–4Q) and placebo, plotted at the median C _ss,max of each APOE4 genotype. Whiskers represent 95% confidence interval of the observed ARIA‐E incidence. Solid simulated lines represent the model‐predicted % incidence of ARIA‐E in each APOE4 genotype. The shaded areas represent the 95% confidence interval of the predicted incidence. In the bottom panel, the range of model‐predicted C _ss,max values is displayed by quartile.

The incidence of isolated ARIA‐H events was plotted by quartiles of C _ss,max, C _ss,av, and C _ss,min, and is presented in Figure 4. A linear trend line was added to help visualize any potential trend in isolated ARIA‐H incidence with increasing exposure. This graphical analysis showed little apparent correlation between lecanemab treatment and exposure levels with isolated ARIA‐H incidence across APOE4 carrier status and genotypes, indicating that the incidence rate of isolated ARIA‐H appears to be independent of lecanemab treatment and exposure. The incidence appears to be higher in APOE4 carriers, particularly in subjects with homozygous APOE4 genotype, suggesting a potential contribution of APOE4 homozygous status to the incidence rate of isolated ARIA‐H. Based on these results, exposure relationship for isolated ARIA‐H was not further explored with modeling approaches.

FIGURE 4.

Isolated ARIA‐H incidence by APOE4 genotype as a function of model‐predicted lecanemab C _ss,max, C _ss,av, and C _ss,min by APOE4 genotype. C _ss,max, Steady‐state maximum serum concentration; C _ss,av, steady‐state average steady state concentration; C _ss,min, steady‐state minimum serum concentration. In the top panel, filled circles represent the observed proportion of subjects with isolated ARIA‐H for exposure quartiles plotted at the median exposure of each group. Whiskers represent 95% confidence intervals for observed proportion of subjects with isolated ARIA‐H. The blue, green, and red lines represent linear smooth of data points in each genotype group; the gray area is the associated 95% confidence interval of the linear smooth. In the bottom panel, the range of model‐predicted exposure values in the entire Study 301 Core dataset is displayed by quartile.

For the final logistic regression model for ARIA‐E, the bootstrap medians were concordant with the population‐predicted values, indicating that the final model was valid and stable and produced well‐estimated parameters (Table 2).

DISCUSSION

We present results of a comprehensive population PK and exposure–response (ER) analysis for the incidence of ARIA‐E, following the addition of new data from the completed confirmatory phase III study to the existing dataset from the lecanemab phase II proof‐of‐concept study. Additionally, we present the results from a graphical ER analysis for isolated ARIA‐H on data from phase III study.

Consistent with the previous population PK analysis based on pooled data from two phase I studies and a phase II study, ¹⁸ lecanemab PK in the current updated analysis after pooling additional data from the phase III study was well described by a two‐compartment model with linear elimination from the central compartment. Similarly, the final PK model from the updated analysis included the same significant covariate effects of body weight, albumin, sex, and time‐varying ADA on CL, body weight and sex on V1, and Japanese ethnicity/race on V2, with similar extent of effects, but with an additional minor effect of Japanese ethnicity/race on V1. The exposure to lecanemab, assessed as F in the PK model, was comparable between manufacturing Processes A and B with a population estimate of 0.904 (95% CI: 0.890–0.918), which is slightly lower than the estimate of 0.998 from the previous analysis ¹⁸ However, in the previous PK analysis, only 5.1% of the observations in the PK dataset were following the administration of Process B from the open‐label extension of the phase II study, whereas in the pooled dataset used in the updated PK analysis, 60.8% of the observations where following process B, mainly from phase III study 301. As such the parameter estimate of F for PK comparability of 0.904, which was well estimated (%RSE = 0.75%), from the updated PK analysis is considered more reliable. The terminal half‐life of lecanemab for the typical patient estimated from PK model is approximately 14.5 days, slightly longer than that of ~9.5 days from a previous population PK analysis. ¹⁸ Lecanemab CL and V ₁ increased with increasing body weight with exponents of 0.353 and 0.513, respectively, and both parameters were slightly lower in females compared to males, 20.9% for CL and 13.2% for V1. These findings are consistent with results from our previous PK analysis, ¹⁸ and with those for donanemab ²³ and other IgG mAbs. ²⁴ The rationale for the small estimated difference in PK of mAbs between males and females is unclear, however, there may be differences in the lymph flow rate and expression of the Fc receptor. ²⁵ Lecanemab clearance was found to decline with increasing albumin levels with an exponent of 0.374. The neonatal Fc receptor (FcRn) facilitates IgG and albumin homeostasis by recycling them across cell membranes back to the central circulatory system. ²⁶ Thus, a higher albumin concentration could be an indicator of an increased number of FcRn and related reduced lecanemab elimination. However, none of the significant covariates identified in the final PK model had a clinically relevant effect on steady‐state exposure to lecanemab in terms of both C _max and AUC (Figure 1), and as such no dose adjustment is required for the recommended dosing regimen of 10 mg/kg bi‐weekly to account for any of the significant covariates identified in the final PK model. Age, which is a risk factor for AD, was not found to significantly affect lecanemab PK.

Given the association of the occurrence of ARIA with the treatment of anti‐amyloid antibodies, ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ both ARIA‐E and ARIA‐H were pre‐specified as adverse events of special interest and closely monitored in the lecanemab phase II proof‐of‐concept study and confirmatory phase III study. In this analysis, a logistic regression model has been developed to evaluate ARIA‐E based on pooled data from the phase II and III studies. Various model‐predicted lecanemab exposure parameters at steady state and several other covariates such as APOE4 genotype were evaluated as predictors of the incidence of ARIA‐E. The model‐predicted incidence rates and trends are consistent with the observed incidence of ARIA‐E in the pooled dataset. The incidence of ARIA‐E is statistically better predicted by C _ss,max than C _ss,av or C _ss,min, and is correlated with APOE4 genotype status (Figure 3), although all exposure parameters correlated with the ARIA‐E incidence well and were statistically significant. The odds ratio for the incidence of ARIA‐E in the presence of lecanemab exposure at the study 301 model predicted range of C _ss,max between 58 and 544 μg/mL was estimated to be 1.47 and 37.5, respectively. At the PK model‐predicted mean lecanemab C _ss,max of 305 μg/mL in subjects receiving 10 mg/kg bi‐weekly in the phase III study, ARIA‐E incidence rate is predicted to be higher at 28.0% (95% CI: 22.6%–34.1%) in APOE4 homozygous carriers compared to 9.85% (95% CI: 7.96%–12.1%) in APOE4 heterozygous carriers and 5.45% (95% CI: 3.75%–7.84%) in APOE4 non‐carriers. These results are consistent with observed ARIA‐E incidence rates in the phase III study of 32.6%, 10.9%, and 5.4% in APOE4 homozygous carriers, APOE4 heterozygous carriers, and APOE4 non‐carriers, respectively, ¹⁶ and comparable to those from a previous ER analysis on data from the phase II finding study. ³¹

For lecanemab, similar to bapineuzumab, donanemab, and aducanumab, the occurrence of ARIA‐E appears to be dose dependent, and increased incidence is associated with the ε4 allele of APOE4. Comparison between trials would suggest that lecanemab has lower ARIA‐E than some of the other published Aβ immunotherapies, including aducanumab (~35%), ³² donanemab (~24%), ³² and gantenerumab (~35%). ³³ There are a number of possible reasons why there might be a lower incidence of ARIA with lecanemab at 10 mg/kg bi‐weekly dose including: (a) physicochemical and/or pharmacological profile; (b) degree of brain penetration; (c) specificity of the antibody for vascular amyloid (i.e., the different Aβ target binding site); (d) immune response to the antibody/amyloid complex; and (e) binding profile to Aβ, where lecanemab has 10–15 times stronger binding to protofibrils than to fibrils (Magnusson et al. ¹¹ ). ³¹

For isolated ARIA‐H, graphical analysis of data from phase III study showed that the incidence rate had no apparent correlation with treatment or lecanemab exposure at 10 mg/kg bi‐weekly dose across APOE4 genotypes. The incidence rate of isolated ARIA‐H appeared to be lower in heterozygous APOE4 carriers or non‐carriers than in homozygous APOE4 carriers.

In conclusion, the PK of lecanemab was well characterized in patients with AD by a linear, two‐compartment model, and identified covariates were consistent with data from other studies with amyloid beta‐targeting monoclonal antibodies and beyond. ARIA‐E was correlated with maximum lecanemab serum concentration and incidence was higher in APOE4 homozygous carriers than APOE4 heterozygous carriers and APOE4 non‐carriers. Isolated ARIA‐H incidence rate had no apparent correlation with treatment or lecanemab exposure at 10 mg/kg bi‐weekly dose across APOE4 genotypes. These results alongside the positive results from the statistical analysis for the primary efficacy end‐point Clinical Dementia Rating–Sum of Boxes (CDR‐SB) and secondary efficacy end‐points Alzheimer's Disease Assessment Scale‐Cognitive Subscale (ADAS‐Cog14), Alzheimer's Disease Composite Score (ADCOMS), and Alzheimer's Disease Cooperative Study–Activities of Daily Living Scale for Mild Cognitive Impairment (ADCS‐MCI‐ADL) ¹⁶ support the currently recommended therapeutic dosing of lecanemab at 10 mg/kg bi‐weekly intravenously. However, enhanced clinical vigilance for ARIA is recommended during the first 14 weeks of treatment with lecanemab. Risk of ARIA, including symptomatic ARIA, increased in apolipoprotein E ε4 homozygotes compared to heterozygotes and non‐carriers. Hence, if a patient experiences symptoms suggestive of ARIA, clinical evaluation should be performed, including magnetic resonance imaging (MRI) scanning if indicated.

AUTHOR CONTRIBUTIONS

Z.H. wrote the manuscript. All authors designed the research. O.M., Z.H., and Y.C. performed the research and analyzed the data.

FUNDING INFORMATION

The studies were funded by Eisai Co., Ltd.

CONFLICT OF INTEREST STATEMENT

O.M. and Z.H. are employees of Eisai Ltd., B.A.W., Y.C., B.L., S.R., N.P., and L.R. are employees of Eisai Inc., and S.H., O.T., and S.Y. are employees of Eisai Co., Ltd.

Supporting information

Figure S1.

Table S1.

Text S1.

Majid O, Cao Y, Willis BA, et al. Population pharmacokinetics and exposure–response analyses of safety (ARIA‐E and isolated ARIA‐H) of lecanemab in subjects with early Alzheimer's disease. CPT Pharmacometrics Syst Pharmacol. 2024;13:2111‐2123. doi: 10.1002/psp4.13224