Population pharmacokinetics of adebrelimab – Support of alternative flat dose regimen in extensive‐stage small‐cell lung cancer

Abstract

Adebrelimab, a novel anti‐PD‐L1 antibody, has been approved by the National Medical Products Administration of China as an intravenous infusion for use in combination with carboplatin and etoposide as first‐line treatment for extensive‐stage small‐cell lung cancer in 2023. A two‐compartment model with empirical time‐varying CL for adebrelimab was established based on data from 263 patients receiving body weight‐based doses from two clinical studies. Significant covariate effects of baseline body weight, albumin levels, tumor size, neutrophil counts, and presence of anti‐drug antibodies were identified on CL of debrelimab, none of which were clinically significant or warranted dose adjustment. The degree of decrease in CL was higher in patients who responded to treatment with adebrelimab than in non‐responders. Adebrelimab exposures (AUC, C_trough, or C _max) were not identified as a statistically significant factor related to efficacy or safety endpoint in the exposure–response analysis. Distribution of simulated exposure metrics from the flat dose regimen (1200 mg q3w) was similar to the marketed weight‐based dosing regimen (20 mg/kg q3w), supporting the alternative flat dose regimen in the clinic.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Adebrelimab, a novel anti‐PD‐L1 antibody, has been approved by the National Medical Products Administration (NMPA) of China as an intravenous infusion for use in combination with carboplatin and etoposide as first‐line treatment for extensive‐stage small‐cell lung cancer (ES‐SCLC). Adebrelimab is administered currently at a body weight‐based dose of 20 mg/kg q3w.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This is the first published analysis for population pharmacokinetics of adebrelimab and exposure–response relationship. PopPK‐derived exposures were used to compare the flat dosing with the current body weight‐based dosing regimen and ultimately to potentially recommend a flat dose in Asian patients with ES‐SCLC.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study provides a population pharmacokinetics model with empirical time‐varying CL for adebrelimab. None of the statistically significant exposure metrics was identified in the exposure–response analysis. A flat dose of 1200 mg for adebrelimab was predicted to provide comparable exposures with the currently marketed body weight‐based dosing regimen of 20 mg/kg in Asian patients with ES‐SCLC.

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

These analyses support the NMPA approval of an alternative flat dosing regimen of 1200 mg every 3 weeks in combination with carboplatin and etoposide as first‐line treatment for patients with ES‐SCLC, with the potential benefit of practical advantages providing greater convenience to healthcare providers.

INTRODUCTION

Adebrelimab, a humanized IgG4 monoclonal antibody against programmed cell death ligand‐1 (PD‐L1), expressed on tumor cells and tumor‐infiltrating immune cells, allows tumor cell killing through blocking the binding of PD‐L1 to PD‐1 receptors expressed on T cells and via activating signaling pathways of T cells. Other antibodies with a similar PD‐1/PD‐L1 mechanism, also known as immune checkpoint inhibitor (ICI), have been developed for the treatment of various cancer types in recent years. ¹ , ² , ³ , ⁴ , ⁵ , ⁶

Adebrelimab at dose levels of 3 mg/kg q3w (n = 3), 10 mg/kg q3w (n = 13), 20 mg/kg q3w (n = 13), and 10 mg/kg q2w (n = 12) were studied in a dose‐escalation trial that included patients with advanced solid tumors (NCT03474289). ⁷ The maximal tolerated dose was not reached at the highest dose level while exposures were approximately dose proportional. The ex vivo PD‐L1 target was occupied rapidly by adebrelimab within 10 min following the first dose administration. In a phase III trial (NCT03711305, CAPSTONE‐1) of ES‐SCLC patients, adebrelimab combined with carboplatin and etoposide significantly improved overall survival (OS) compared to placebo plus carboplatin and etoposide, with an acceptable safety profile. ⁸ Median OS was 15.3 months (95% CI: 13.2–17.5) for the adebrelimab group, versus 12.8 months (95% CI: 11.3–13.7) for the placebo group, and the hazard ratio was 0.72 (95% CI: 0.58–0.90). Recent studies also suggested that combination treatments with adebrelimab are a promising therapeutic strategy for other cancers. ⁹ , ¹⁰

Adebrelimab has been approved for the first‐line treatment of ES‐SCLC at a dose of 20 mg/kg administered intravenously every 3 weeks (20 mg/kg q3w) in China. The objectives of this analysis were to (i) build and qualify a population pharmacokinetic (PopPK) model of adebrelimab, (ii) to assess and describe potential relationships between adebrelimab exposures and safety and efficacy endpoints, and (iii) to support a flat dose recommendation by comparing adebrelimab exposure between weight‐based and flat dosing regimens, considering both (i) and (ii).

METHOD

Studies

Data from two clinical studies were used in the PopPK and exposure–response (ER) analyses. The study of adebrelimab (NCT03474289, SHR‐1316‐I‐101) was a phase I, open‐label, non‐randomized, dose escalation/expansion study, where 41 patients with advanced tumors were included in the popPK dataset. The pivotal study (NCT03711305, SHR‐1316‐III‐301) was a phase III, multicenter, randomized, double‐blind, placebo‐controlled study to evaluate the efficacy and safety of adebrelimab versus placebo in combination with carboplatin and etoposide as a first‐line treatment for ES‐SCLC. A total of 222 patients with at least one PK concentrations after administration in the adebrelimab group of the pivotal study were included in the popPK dataset. A summary of the demographic characteristics and covariates is provided in Table 1. The study protocols were approved by institutional review boards and independent ethics committees at each site. All patients provided written informed consent.

TABLE 1.

Demographic and disease‐related covariates in the studied population.

Covariate a	SHR‐1316‐I‐101	SHR‐1316‐III‐301	Total population
Number of patients	41	222	263
Nominal dose, n (%)
3 mg/kg q3w	3 (7.3)		3 (1.1)
10 mg/kg q2w	12 (29.3)		12 (4.6)
10 mg/kg q3w	13 (31.7)		13 (4.9)
20 mg/kg q3w	13 (31.7)	222 (100)	235 (89.4)
Age, years	53 (18–68) [0]	62 (37–73) [0]	61 (18–73) [0]
Baseline weight, kg	61.0 (42.5–87.5) [0]	64.0 (38.1–97.0) [0]	64.0 (38.1–97.0) [0]
Baseline height, cm	168 (150–180) [0]	167 (143–185) [1]	167 (143–185) [1]
Sex, n (%)
Male	20 (48.8)	177 (79.7)	197 (74.9)
Female	21 (51.2)	45 (20.3)	66 (25.1)
Race, n (%)
Asian (Chinese)	41 (100)	222 (100)	263 (100)
Neutrophil count, 109/L	4.0 (1.7–10.3) [0]	4.2 (1.7–11.4) [0]	4.2 (1.7–11.4) [0]
Platelet, 109/L	220 (107–396) [0]	235.5 (103–533) [0]	235 (103–533) [0]
ALT, IU/L	14.0 (5.7–55.1) [0]	18.7 (4.0–151.2) [0]	18 (4.0–151.2) [0]
AST, IU/L	19.5 (10.4–196.4) [0]	22.0 (9.0–109.2) [0]	21.5 (9.0–196.4) [0]
Total protein, g/L	74.9 (50.8–84.5) [0]	70.05 (52.3–88) [0]	71.1 (50.8–88) [0]
Albumin, g/L	43.1 (29.6–51.1) [0]	41.0 (26.6–53.2) [0]	41.4 (26.6–53.2) [0]
Creatinine, μmol/L	66.0 (39.0–103.0) [0]	68.0 (37.0–146.0) [0]	67.7 (37.0–146.0) [0]
Creatinine clearance, mL/min	104.6 (48.0–164.6) [0]	87.8 (33.9–187.4) [0]	90.5 (33.9–187.4) [0]
Hepatic impairment, n (%)
Normal hepatic function	34 (82.9)	187 (84.2)	221 (84.0)
Mild hepatic impairment	5 (12.2)	35 (15.8)	40 (15.2)
Moderate hepatic impairment	2 (4.9)	0 (0)	2 (0.8)
Renal impairment, n (%)
Normal renal function	30 (73.2)	104 (46.8)	134 (51.0)
Mild renal impairment	5 (12.2)	99 (44.6)	104 (39.5)
Moderate renal impairment	6 (14.6)	19 (8.6)	25 (9.5)
Cancer type, n (%)
Lung cancer	8 (19.5)	222 (100)	230 (87.5)
Others	33 (80.5)	0 (0)	33 (12.5)
Metastatic disease, n (%)
Metastasis	41 (100)	222 (100)	263 (100)
Clinical stages, n (%)
III stage	0 (0)	8 (3.6)	8 (3.0)
IV stage	39 (95.1)	214 (96.4)	253 (96.2)
Missing	2 (4.9)	0 (0)	2 (0.8)
ECOG performance status, n (%)
Fully active (0)	0 (0)	33 (14.9)	33 (12.5)
Restricted in physically strenuous activity (1)	41 (100)	189 (85.1)	230 (87.5)
PD‐L1 expression, n (%)
Negative	7 (17.1)	189 (85.1)	196 (74.5)
Positive	6 (14.6)	24 (10.8)	30 (11.4)
Missing	28 (68.3)	9 (4.1)	37 (14.1)
Anti‐drug antibody, n (%)
Negative	13 (31.7)	177 (79.7)	190 (72.2)
Positive	27 (65.9)	44 (19.8)	71 (27.0)
Missing	1 (2.4)	1 (0.5)	2 (0.8)

Abbreviations: ALT, alanine transaminase; AST, aspartate transaminase; ECOG, Eastern Cooperative Oncology Group; PD‐L1, programmed death ligand 1.

^a Continuous covariates are reported as median (range) [number of missing].

Pharmacokinetic sampling and bioanalytical methods

The intensive PK sampling schedule following adebrelimab administration was designed to describe its distribution and elimination in the phase I study (NCT03474289). The sparse PK sampling schedule (within 0.5 h before administration of cycle 1, 2, 4, 6, 9, 13, 17) was designed to assess PK after single and repeated dosing in the pivotal study (NCT03711305). Adebrelimab was quantified by enzyme‐linked immunosorbent assay (ELISA). The lower limit of quantification (LLOQ) of the adebrelimab assay in human serum was 0.4 μg/mL.

Population PK model development

Data exploration, such as logscale concentration–time curves, suggested that PK of adebrelimab followed a two‐compartment distribution model. Two‐compartment models with linear, non‐linear, or parallel linear and non‐linear CL were assessed. Given that multiple PD‐1/PD‐L1 inhibitors have shown time‐dependent CL, the sigmoid maximal inhibitory clearance model (Equation 1) was used to describe the CL change of adebrelimab, the same as for other immune checkpoint inhibitors such as nivolumab, atezolizumab, durvalumab, and avelumab. ¹¹ , ¹² , ¹³ , ¹⁴ In addition, the simpler exponential decay model (Equation 2), alone or in combination with time‐independent or non‐linear CL was also explored. ¹⁵

$${\mathrm{CL}}_{i.t}=\text{TVCL}*\exp \left(\frac{I_{\max ,i}*{t_i}^{\text{HILL}}}{{{\mathrm{TC}}_{50}}^{\text{HILL}}+{t_i}^{\text{HILL}}}\right)$$ (1)

$${\mathrm{CL}}_{i.t}=\text{TVCL}*\exp \left(-K_{\mathrm{des}}*t_i\right)$$ (2)

Here, CL _i,t is CL of the ith individual at time t. TVCL is the typical value of CL at Time = 0. I _max,I represents the maximal change in CL relative to baseline CL for the ith individual, and TC₅₀ is the time at which the change is half of its maximum in CL, while HILL is an exponential shape parameter. K _des is the decay coefficient of time–dependent clearance.

The PK parameters were assumed to have a log‐normal distribution. Residual variability was described using either additive, proportional or combined error model. Model selection was informed using a log‐likelihood ratio test (i.e. significant objective function value [OFV] reduction), together with an assessment of goodness‐of‐fit plots, acceptable precision in parameters' estimation, and scientific and physiological plausibility.

Covariates' screening on the PopPK model

Covariates were evaluated for their effect on adebrelimab PK parameters using a stepwise covariate modeling (SCM) strategy. Based on the predefined criteria and graphical evaluation from the pharmaco‐statistical model (PSM), the relationships between following baseline covariates and CL were considered: body weight (BW), age, sex, cancer type, ECOG performance status, creatinine clearance (CLCR), albumin (ALB), alanine (ALT), and aspartate (AST) aminotransferase, sum of the longest diameters for all target lesions at baseline (SLD), absolute neutrophil count (NEUT), platelet, renal, and hepatic impairment stage. The assessed post‐baseline covariates on CL included subject‐level anti‐drug antibodies (ADA) presence and combination with chemotherapy. Baseline body weight, sex, and albumin were assessed on V ₁ and V ₂.

The relationship between continuous covariates and PK parameters was described using power models (Equation 3) while categorical covariates were tested using a linear function (Equation 4).

$${\mathrm{Par}}_i={\mathrm{Par}}_{\mathrm{ref}}*{\left(\frac{{\mathrm{Cov}}_i}{{\mathrm{Cov}}_{\mathrm{med}}}\right)}^{{\theta }_{\mathrm{Par},\mathrm{Cov}}}$$ (3)

$${\mathrm{Par}}_i={\mathrm{Par}}_{\mathrm{ref}}*\left(1+{\theta }_{\mathrm{Par},\mathrm{Cov}}\right)$$ (4)

Par _i is the parameter value for the ith individual, and Par_ref is the parameter value when the covariate was the standard reference value or median value. Cov _i is the covariate value of the ith individual, and Cov_med is the covariate median value in the population. θ _(Par,Cov) is a parameter describing the covariate effect on the parameter.

The forward selection method with an OFV decrease of at least 6.63 (associated with a p‐value < 0.01) was used to add a covariate–parameter relationship in sequence. At the end of the forward inclusion, a backward elimination step with an OFV decrease of at least 10.83 (p < 0.001) was performed to reach the most parsimonious final PopPK model. Goodness‐of‐fit plots and visual predictive checks (VPCs) were performed at key model‐development steps (i.e. PSM and final models) to assess the model predictive performance. The NONMEM control stream for the final model is provided in Data S1.

Illustration of the impact of covariates using forest plots

Forest plots that illustrate the effect of covariates on exposure metrics at steady state and CL at baseline of adebrelimab was generated based on simulations with uncertainty in fixed‐effect parameters. The reference condition was the approximate median value and the test conditions were the 5th and 95th percentiles for continuous covariates. For categorical covariates, the conditions were the same mode of the final model. A total of 1000 sets of fixed‐effect parameters were sampled from the estimated value and sandwich matrix of the final model, and then the ratio of CL and exposure metrics at the test and the reference condition were calculated for each set of fixed‐effect parameters. The 5th to 95th percentiles of the CL ratio are taken as the 90% confidence interval.

Exposure–response analyses

Data from ES‐SCLC patients receiving adebrelimab combined with carboplatin and etoposide arm was included in the exposure–efficacy analysis. In consideration of a suspected time‐varying clearance of adebrelimab, analysis using steady‐state exposure metric might lead to biased results of the exposure–efficacy relationship. ¹¹ The relationship between exposure metrics (AUC_0–21day, C _trough) after the first dose and overall response rate (ORR) assessed by blinded‐independent central review (BICR) were evaluated using logistic regression modeling. Pooled data from patients treated with adebrelimab from CAPSTONE‐1 study and the first‐in‐human phase I study was included in the exposure–safety analysis. Logistic regression models were developed to explore the relationship between adebrelimab exposure metrics (C _avg, C _max) after first administration or at steady state, and grade ≥3 adverse event (AE), serious AE (SAE) or immune‐mediated adverse events (IMAE).

Simulations of exposure metrics for comparing of flat dosing versus weight based dosing

The uncertainty in parameters, between‐subject variability and residual error was accounted for in simulations. Firstly, 200 sets of parameters were sampled from the multivariate normal distribution of parameters in the final model. Secondly, 500 sets of η value were sampled from the interindividual variability (IIV) variance–covariance matrix, resulting in 100,000 sets of individual parameters. Thirdly, 100,000 sets of covariates variables at the patient level were generated by resampling with the replacement of the individuals in the population (n = 263), then assigning individual parameters. Finally, 1000 virtual patients were sampled without replacement from the pool of 100,000 individuals. The R code for the sampling process is provided in Data S2. The exposure metrics, C _trough, C _max, and AUC after a single dose and at steady‐state, were simulated for weight‐based and flat‐dosing regimens. These exposure metrics were compared by quartiles of weight between different regimens.

RESULTS

PopPK model with time‐varying clearance

The best structure of the adebrelimab popPK model was a two‐compartmental, with time‐varying linear CL described by the empirical time‐dependent sigmoid‐maximal model. The PK parameters were assumed to have a log‐normal distribution and the interindividual variability (IIV) was included in CL, V ₁, and V ₂. The list of model development is provided in Table S1. The covariance between the random effects for CL, V ₁, and V ₂ was estimated by full covariance matrices. The maximal change in CL relative to baseline (I _max) was assumed to have a normal distribution. The residual error was described by a combined additive and proportional error model (Table 2).

TABLE 2.

Parameter estimates for the final PopPK model.

Parameter	Pharmaco‐statistical model			Final model
	Estimate	RSE	Shrinkage	Estimate	RSE	Shrinkage
OFV	13116.293	/	/	12943.442	/	/
Condition number	38.31	/	/	39.17	/	/
Fixed effect parameter a
CL (L/day)	0.25	2.9%	/	0.238	2.7%	/
V 1 (L)	3.21	2.6%	/	3.23	2.6%	/
Q (L/day)	0.705	10.8%	/	0.703	10.1%	/
V 2 (L)	1.74	8.8%	/	1.69	7.5%	/
I max	−0.365	12.5%	/	−0.349	13.7%	/
TC50 (day)	73.9	15.6%	/	74.4	13.1%	/
HILL	1.88	19.7%	/	1.97	20.3%	/
ADA on CL	/	/	/	0.185	24.1%	/
Albumin on CL	/	/	/	−0.861	15.2%	/
Body weight on CL	/	/	/	0.710	13.6%	/
Neutrophil count on CL	/	/	/	0.159	28%	/
SLD on CL	/	/	/	0.103	27.2%	/
Body weight on V 1	/	/	/	0.577	24.8%	/
Body weight on V 2	/	/	/	1.8	18.2%	/
Interindividual variability b
IIV on CL (ω 2 CL)	0.0796	15.2%	8.7%	0.0416	13.1%	13.8%
Cov (CL, V 1)	0.0235	27%	/	0.0104	49.1%	/
IIV on V 1 (ω 2 V1)	0.0378	30.2%	42.6%	0.0296	28.1%	50.1%
Cov (CL, V 2)	0.0727	35.9%	/	0.0371	38.3%	/
Cov (V 1, V 2)	0.0587	23.3%	/	0.0308	38%	/
IIV on V 2 (ω 2 V2)	0.225	28.6%	37%	0.128	36.2%	44.3%
IIV on I max (ω2 Imax)	0.0461	29.1%	33.7%	0.0534	26.8%	28.2%
Residual variability
Proportional residual error (σ 2 Prop)	0.0222	18.8%	13.9%	0.0224	17.9%	13.3%
Additive residual error, μg/mL (σ 2 add)	45.3	51.9%	13.9%	32.6	65%	13.3%

Abbreviations: ADA, antidrug antibody; CL, clearance; Cov, covariance; HILL, exponential shape parameter; IIV, interindividual variability; I _max, maximal change in CL relative to baseline; Q, inter‐compartmental clearance; RSE, relative standard error; SLD, sum of the longest diameters for all target lesions of baseline; TC₅₀, the time at which the change is half of its maximum in CL; V ₁, central volume of distribution; V ₂, peripheral volume of distribution; ω ², variance of omega, σ ², variance of sigma.

^a The baseline values for ALB, BW, NEUT, and SLD were included in the final model. The impact of ADA was evaluated as a subject‐level covariate (i.e. dichotomous 1 for a patient during the full treatment if the patient experienced treatment induced or boosted ADA positivity at any point). In the forward selection procedure, the relationships of eight covariate‐parameter pairs (ALB–CL, BW–CL, BW–V2, ADA–CL, NEUT–CL, SLD–CL, BW–V1, and SEX–CL) were sequentially included. The SEX–CL relationships were removed from the full model in the backward elimination procedure. The linear covariate relations for ADA were added on CL, and the power relation for other covariates was used to add on parameters.

^b The exponential model for interindividual variability of CL, V ₁, and V ₂ was used. The additive model for interindividual variability of I _max was used.

The final PopPK model included the covariate effects of body weight, SLD, albumin, and neutrophil count at baseline, and treatment‐emergent ADA at the subject level on CL and body weight on V ₁ and V ₂. ADA‐positive patients had higher CL than ADA‐negative patients. Higher baseline albumin, lower body weight, lower neutrophil count and lower SLD levels were associated with reduced CL. The IIV for CL decreased from 28.8% before to 20.6% after the inclusion of the covariates in the final model. The estimated exponent of the body weight on CL was 0.71, which was close to the theory‐based allometric exponent of 0.75. Higher baseline body weight was associated with increased V ₁ and V ₂. The final model parameter estimates are provided in Table 2. Goodness‐of‐fit plots and distributions of random effects are shown in Figures S1 and S2. The correlations between the covariates included in the final model are shown in Figure S3. Prediction corrected VPC (PcVPC) showed good predictive performance in Figure S4.

The predicted variation in CL over time from the final model during the treatment period is shown in Figure S5. Based on a simulation using post hoc individual PK parameters of ES‐SCLC patients, the geometric mean of CL at baseline and at steady state (at 20th administration cycles) was 0.25 L/day (geoCV: 22.8%) and 0.177 L/day (geoCV: 27.6%), respectively, and the volume of distribution at steady state was 4.91 L (CV: 18.5%). The geometric mean of the terminal half‐life of adebrelimab at steady state was estimated at 19.9 days (CV:23.9%), which is consistent with most of the mAbs.

Influence of covariates on exposure

Forest plots of the impacts of covariates on exposure at steady state in the final model are shown in Figure 1. ADA‐positive patients had a reduction of 16% in AUC_tau,ss compared to ADA‐negative patients. Higher albumin levels with the 95th percentile of 48.3 g/L led to a 14% increase in AUC_tau,ss compared to a typical patient. Conversely, a reduction of 18% in AUC_tau,ss was predicted in patients of lower albumin levels with the 5th percentile of 33.1 g/L. Body weight with the 95th and 5th percentiles, 84.3 kg and 47.1 kg, also had a slight impact on CL of 8% increase and 9% reduction under 20 mg/kg Q3W regimen, respectively, compared to a typical patient. Baseline neutrophil counts and tumor size had minor impacts on AUC_tau,ss (<10%). As with C _max, the impacts of all covariates were within the no‐effect range of 80%–125% of the point estimate of the covariates. However, patients with positive ADA or a lower ALB level had a reduction 22% and 25% in C _trough, respectively. The impacts of all the covariates on baseline CL were in the no effect range of 80%–125% (Figure S6).

FIGURE 1.

Forest plots of the impact of covariates on adebrelimab exposure according to the final PopPK model. Left labels indicate the covariate being evaluated with the 5th and 95th percentiles of the continuous covariate distributions.

Influence of responder status on clearance decay over time

A total of 222 ES‐SCLC patients treated with adebrelimab combined with carboplatin and etoposide were included in the exposure–efficacy analyses. ⁸ The individual time‐varying CL values relative to baseline CL stratified by response to treatment are shown in Figure 2a,b for non‐responders and responders, respectively. Median observed decrease in CL at the end of treatment from baseline was 26.2% in responders, whereas 15.3% in non‐responders. Similar distribution of baseline CL between responders and non‐responders was observed (Figure 2c), while responders had lower CL compared to non‐responders at steady state (Figure 2d). Given the difference in treatment duration between responders and non‐responders (i.e. shorter treatment duration in non‐responders), the simulation of 20 dosing cycles was carried out for the sake of comparability, and the similar result was observed (Figure S7). As expected and previously reported for other immune checkpoint inhibitors such as nivolumab, atezolizumab, durvalumab, and avelumab, ¹¹ , ¹² , ¹³ , ¹⁴ it seems the responder status alters the magnitude of reduction in CL from baseline to steady state. Therefore, using exposure metrics derived from the first administration (AUC_0–21day, C _trough) was a more rational selection for the exposure–efficacy analysis.

FIGURE 2.

Change in estimated clearance CL over time from baseline stratified by response.

Exposure–response analysis

The geometric mean of predicted AUC_0–21day after the first dose for ES‐SCLC patients was 3320 μg*day/mL (range: 2160–5680; geoCV: 13.2%), and the geometric mean of AUC_tau,ss was 6260 μg*day/mL (range: 2160–12,400; GeoCV: 33.6%). The geomean of C _max and C_trough after first dose was 393 μg/mL (range: 309–540; geoCV: 8.80%) and 83.3 μg/mL (range: 27.0–156; geoCV: 26.1%), respectively, and the geomean of C _max,ss and C _trough,ss was 579 μg/mL (range: 351–874; geoCV: 20.7%) and 185 μg/mL (range: 27.0–448; geoCV: 46.8%), respectively. The distribution of the predicted exposure metrics of patients in the dose‐escalation phase 1 study and CAPSTONE‐1 study is shown in Figures S8 and S9, respectively.

The ORR assessed by BICR was 73.0% (162/222) in ES‐SCLC patients of the adebrelimab group. The results of the univariate regression analysis showed neither AUC_0–21day nor C_trough at cycle 1 was significantly related to the probability of response (Figure 3). Exposure–safety analyses were performed on the pool of 41 patients with solid advanced tumors in the phase 1 trial and 222 ES‐SCLC patients in the CAPSTONE‐1 study. In the analyses, the C _avg,ss was used instead of AUC, in addition to C_max,ss, because of the different dosing schedules. AEs of grade ≥3 occurred in 96 (36.0%) of 263 patients in the pooled data, and SAE occurred in 53 of them (20.2%). The risk of grade ≥3 AE and SAE was not statistically related to C _avg,ss (p = 0.685; p = 0.195, respectively) or (p = 0.506; p = 0.180, respectively) in the univariate logistic regression modeling (Figure 4). A relatively lower incidence (29.7%, 78/263) of immune‐mediated adverse events (IMAE) was found in the pooled data, and no significant association between the exposure at steady state and IMAE risk was observed in the analysis (p = 0.078 for C _avg,ss; p = 0.218 for C _max,ss; Figure S10). In addition, when looking at the first administration, there was no significant relationship between adebrelimab exposures after the first administration (C _avg,ss and C _max) and any of the safety endpoints (grade ≥3 AE, SAE, and IMAE, Figures S11 and S12).

FIGURE 3.

Observed ORR by quartiles and predicted probability versus adebrelimab AUC_0–21day (a) and C _trough (b) at first dose in univariate analysis. Blue solid curve and gray shaded areas represent the predicted probability and 95% confidence interval by a logistic regression model. Black square and red error bars represent actual ORR and 95% CI by exposure quartiles. Small vertical black lines at the top and bottom represent response events (0: non‐responder; 1: responder) and the predicted value of the exposure metric. The bottom box plot with whiskers shows the distribution of the exposure metric.

FIGURE 4.

Incidence by quartile and predicted probability of grade ≥3 AE and SAE versus adebrelimab C _avg,ss and C _max,ss at steady state in the exposure–safety analyses. (a) grade ≥3 AE versus C _avg,ss; (b) grade ≥3 AE versus C _max,ss; (c) SAE versus C _avg,ss; (d) SAE versus C _max,ss. Blue solid curve and gray shaded areas represent the predicted probability and 95% confidence interval by the logistic regression model. Black square and red error bars represent incidence of grade ≥3 AE and 95% CI by exposure quartiles. Small vertical black lines at the top and bottom represent AE events (0: no 1: yes) and the predicted value of exposure. The bottom box plot with whiskers shows the distribution of the exposure metric for groups of different dose and schedule.

Comparison of weight‐based versus flat‐dosing regimens

The median baseline body weight of patients included in the PopPK model was 64.0 kg (range: 38.1–97.0 kg, Figure S13), as expected from the Asian patient population. The exposure of flat dosing (1200 mg Q3W) and weight‐based dosing (20 mg/kg Q3W) regimen was simulated by the PopPK final model. As shown in Figure 5 and Table S2, the predicted exposures from the flat dosing regimen were similar to those from the body weight‐based dosing. The geometric mean of AUC_0–21day simulated by flat dosing and weight‐based dosing regimen were 3110 and 3280 day*μg/mL, respectively, with AUC_tau,ss of 6710 and 7080 day*μg/mL, respectively. The coefficients of variations for both AUC_0–21day and AUC_tau,ss with the flat‐dosing were close to those of the weight‐based dosing independently of the body weight quartile groups, illustrating the overall similar variability of 23.2% versus 20.0% for AUC_0–21day, respectively, and 37.7% versus 37.1% for AUC_tau,ss, respectively. The distribution of the simulated AUC_0‐21day and AUC_tau,ss in all weight groups nearly overlapped, even though slight differences in AUC between the lowest and highest body weight quartiles were observed as expected (Figure 5b,d and Table S2). Similarly, predicted C _trough and C _max after first administration and at steady state (C _trough,ss and C _max,ss) are consistent with those reported for AUC, as shown in Figures S14, S15 and Tables S3, S4. However, there is no specific concern about loss of efficacy for the highest body weight quartile or more adverse events for the lowest body weight quartile, when considering the flat exposure–efficacy and exposure–safety relationships, as the predicted exposure range for the flat dosing regimen remains within the exposure range following the body weight‐based regimen.

FIGURE 5.

Distribution of simulated exposure metrics for flat (1200 mg q3w) and weight‐based (20 mg/kg q3w) dosing using the final population pharmacokinetic model of adebrelimab. Box plots overlaid with dot of AUC_0–21day (a) and AUC_tau,ss (c) for the entire simulation population. Box plots of AUC_0–21day (b) and AUC_tau,ss (d) for the subsets grouped by quartiles of weight. AUC_0–21day, area under the curve during the first dosing interval. AUC_tau,ss, area under the curve during dosing interval at steady state.

DISCUSSION

In this study, the pharmacokinetics of adebrelimab was described by a two‐compartment model with time‐varying linear CL. The empirical sigmoid‐E _max model has been applied to describe the time‐varying clearance of the ICI. ¹⁶ The baseline values of body weight, albumin, neutrophil count, SLD, and post‐baseline ADA at the subject level were retained as significant covariates in the final PopPK model. Apart from impacts of the ADA positivity and the lower ALB level, which resulted in a reduction 22% and 25% in C _trough, respectively, the impact of all covariates on CL, C _max,ss or baseline CL for adebrelimab were within the range of 80%–125%. Nevertheless, this implies that none of the covariates has a clinically significant impact when considering the flat exposure–response relationship. This observation of time‐varying clearance has recently been reported in multiple ICIs, such as nivolumab, atezolizumab, durvalumab, and avelumab, and has largely been attributed to disease status. ⁵ , ¹¹ , ¹² , ¹³ Cancer cachexia is known to cause rapid degradation of proteins, including therapeutic antibodies. A treatment‐induced improvement of the disease status reduces cachexia. ¹⁷ In addition, tumor shrinkage with effective treatment might decrease the antigen burden, which could lead to the gradual decrease of the clearance (i.e. via saturation of target‐mediated drug disposition pathway at lower drug concentrations). The mechanism implies the reduction of clearance in the responders, which is consistent with the observation of adebrelimab in the present analysis.

The recommended dose of 20 mg/kg q3w of adebrelimab was selected based on the adequate safety and tolerability assessment obtained in the phase 1 dose‐escalation study. No significant exposure–efficacy relationship was observed in ES‐SCLC patients treated with adebrelimab at 20 mg/kg q3w in the pivotal study (CAPSTONE‐1). We noticed that there was a non‐significant decrease of over 10% in the lower ORR for the 1st quartile (representing the values ≤25th percentile) of AUC_0–21day and C _trough. In addition, it was found that patients in the 1st exposure quartile had a lower ALB and BW, and higher SLD at baseline, as well as a higher baseline CL in this study (Figures S16 and S17). A series of studies indicated that a higher clearance of therapeutic monoclonal antibody was associated with a greater disease severity, characterized by lower levels of ALB and BW, as well as a larger SLD at baseline. ¹¹ , ¹⁸ , ¹⁹ This implied that the slight association between lower response and first quartile of AUC_0–21day and C _trough was correlational rather than causal, both being caused by baseline disease burden. Moreover, the present exposure–safety analysis based on pooled data from the phase I study and the pivotal study identified no significant relationships between adebrelimab exposures and risk of grade ≥3 AE, SAE, and IMAE.

The publication from Wang et al. suggested that with the allometric scaling exponent on CL close to or higher than the value of 0.68, more benefit would be expected with body weight‐based dosing as compared to a flat dose regimen. ²⁰ In the present adebrelimab PopPK analysis, the estimated exponent on clearance was 0.71, which was close to the theory‐based allometric scaling value. In addition, the estimated allometric scaling exponents on volumes of central and peripheral compartments were not close to the theory‐based value of 1 (i.e. 0.577 and 1.80, respectively), resulting in an influence of body weight on C _max and C _torugh values. Therefore, it seemed more appropriate to compare the predicted exposures between weight‐based and flat dosing regimens based on the final PopPK model. The predicted exposures and associated variability for the flat‐dose regimen were similar to the weight‐based dosing regimen, strongly suggesting that the flat dose is an alternative dosing regimen. The simulated AUC_0–21day distribution of the lowest body weight quartile for the flat dose was in the same range as the distribution in the highest body weight quartile for the weight‐based dosing regimen, and vice versa. Based on the median body weight of the patients included in the data set of 64.0 kg, a theoretical flat dose of 1280 mg would have resulted in closer geometric mean predicted exposures to 20 mg/kg body weight‐based dose, thus explaining the slightly lower geometric mean predicted exposures obtained with the 1200 mg flat dose. However, when considering the quartile analysis of body weight categories from the data set we have in the present analysis, the 1200 mg flat dose seems to be the dosing regimen associated with the closest predicted exposures to the 20 mg/kg body weight‐adjusted dose. On the other hand, there was no significant exposure–efficacy relationship, implying comparable benefit–risk profile between 1200 mg flat dose and 20 mg/kg weight‐based dose, further justifying the support of an alternative flat dosing regimen.

A model‐based approach has been applied to assess the comparability between weight‐based and flat dosing for PD‐1 or PD‐L1 blocking antibodies, such as nivolumab, atezolizumab, and avelumab. ²¹ , ²² , ²³ Overall, flat dosing is expected to provide several practical advantages, with the potential benefit of simplified administration providing greater convenience to healthcare providers including decreased risk of dosing errors during drug preparation and minimized amount of drug wasting. From a practical point of view, administering the 1200 mg dose of adebrelimab, available in 600 mg vials, will require 2 vials with no resulting waste. A guidance was recently published by the FDA on the PK‐based criteria for supporting alternative dosing regimens for PD‐1/PD‐L1 inhibitors. ²⁴ This guidance generally recommends alternative dosing regimens with model‐predicted geometric means of exposure metrics to be within 20% of those of the reference dosing regimen used to establish efficacy in clinical trials, further strengthening our recommendation of the flat 1200 mg dose of adebrelimab as an alternative regimen.

In conclusion, an empirical time‐varying linear CL model described adebrelimab population PK in 222 ES‐SCLC patients and 41 advanced solid tumors patients. The significant covariates retained on CL, V₁ or V₂ from the final model had an impact ranging within the 80%–125% boundaries for most exposure metrics, without the need for any dose adjustment. No significant exposure efficacy or exposure safety was observed. A flat dose of adebrelimab 1200 mg Q3W would most likely provide overall a similar pharmacokinetic profile to those of 20 mg/kg Q3W experimented in previous clinical trials. The flat dose regimen has been since then already applied to new clinical trials of adebrelimab based on the present findings.

AUTHOR CONTRIBUTIONS

P.C., Y.Z., and N.D. wrote the manuscript. Y.Z., N.D., and K.S. designed the research. P.C. and Y.W. performed the research. Y.Z., K.M., and W.S. analyzed the data.

FUNDING INFORMATION

This study was sponsored by Jiangsu Hengrui Pharmaceuticals.

CONFLICT OF INTEREST STATEMENT

P.C., Y.Z., Y.W., K.M., W.S., N.D., and K.S. are employees of Jiangsu Hengrui Pharmaceuticals.

Supporting information

Data S1.

Figure S1.

Table S1.

Chen P, Zhang Y, Wang Y, et al. Population pharmacokinetics of adebrelimab – Support of alternative flat dose regimen in extensive‐stage small‐cell lung cancer. CPT Pharmacometrics Syst Pharmacol. 2024;13:1238‐1251. doi: 10.1002/psp4.13155