Clinical Pharmacokinetics and Pharmacodynamics of Marstacimab, an Anti-tissue Factor Pathway Inhibitor Monoclonal Antibody, in Adolescent and Adult Participants with Hemophilia

Abstract

Background and Objectives

Marstacimab, a monoclonal antibody that targets tissue factor pathway inhibitor (TFPI), was developed for prophylactic treatment of hemophilia, with or without inhibitors. A nonlinear mixed-effects modeling approach was used to characterize plasma marstacimab and TFPI concentrations and identify covariates impacting marstacimab concentration.

Methods

Population modeling using nonlinear mixed-effects modeling (NONMEM) 7.5.0 software was performed with marstacimab and total TFPI concentration data pooled from 213 participants across 6 clinical trials including healthy volunteers (n = 63) and participants with hemophilia (n = 150). Participants received subcutaneous marstacimab at doses ranging from 30 mg to 450 mg. Plasma samples to determine marstacimab and total TFPI were analyzed using validated assays. An E_max pharmacokinetic (PK)/pharmacodynamic (PD) model was developed to link model-predicted free TFPI concentrations to peak thrombin.

Results

Marstacimab and total TFPI concentrations were adequately described with a target mediated drug disposition (TMDD) model with nonlinear clearance. Body weight was the key structural covariate. After adjusting for body weight, no clinically relevant effect of age (adolescent vs adult), race (Asian vs non-Asian), participant status (healthy vs hemophilia) or mild hepatic impairment was seen. There was good agreement between observed and model-predicted peak thrombin levels in adults and adolescents, with no clinically relevant differences between the populations.

Conclusions

A TMDD model with first-order absorption and quasi–steady-state approximation adequately characterized marstacimab PK and total TFPI concentrations. An E_max model adequately described the relationship between model-predicted free TFPI and peak thrombin. The PK/PD simulation indicated that no changes in dosing regimen based on age or body weight were warranted.

Clinical Trials Registration

ClinicalTrials.gov: NCT02531815, NCT02974855, NCT03363321, NCT03938792, NCT04832139, NCT04878731.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40262-026-01695-5.

Key Points

A population pharmacokinetic model of marstacimab was developed using data from 6 clinical studies in healthy volunteers and participants with hemophilia.

The model identified body weight as the key structural covariate. After adjusting for body weight, no clinically relevant effect of age, race, health status or mild hepatic impairment on marstacimab pharmacokinetics was seen.

No clinically relevant differences in peak thrombin, a key biomarker for blood coagulation, were observed between adults and adolescents suggesting that no changes in dosing regimen for age were warranted.

Introduction

Blood clot formation is a highly regulated process controlled by a series of enzymatic activations of coagulation factor proteins in the extrinsic and intrinsic coagulation pathways that converge in the common pathway leading to the conversion of fibrinogen to fibrin to form a stable clot. Upon cessation of bleeding, negative feedback loops ensure coagulation is terminated to prevent further clotting and thrombosis [1].

Hemophilia A and hemophilia B are rare X-linked genetic bleeding disorders arising from a deficiency in coagulation factors VIII (FVIII) and IX (FIX), respectively. Deficiency of FVIII or FIX activity results in the inadequate activation of the intrinsic pathway and insufficient formation of blood clots to regulate bleeding [2]. While people with hemophilia retain a limited ability to stop bleeds via the intact extrinsic pathway, this is not sufficient to control major bleeds or to prevent spontaneous bleeds. People with severe hemophilia (FVIII or FIX activity <1% of normal) experience frequent spontaneous and traumatic bleeding events in the joints, muscle and soft tissues [3]. Treatment of hemophilia A and B involves the use of intravenous (IV) factor replacement therapy delivered prophylactically (standard of care) or on-demand to control bleeding events. However, development of neutralizing antibodies may limit efficacy of factor replacement and the treatment burden of frequent IV infusions is associated with poor adherence. Thus, novel treatment alternatives to factor replacement therapies and bypassing agents have been developed to address this unmet need in people with hemophilia [2].

Tissue factor pathway inhibitor (TFPI) is a multiple Kunitz domain protease inhibitor that inactivates both activated factor X (FXa) and activated factor VII (FVIIa) to negatively regulate blood clotting via the extrinsic pathway [4]. Marstacimab is a human immunoglobulin G isotype subclass 1 (IgG1) antibody targeted to the functional Kunitz domain 2 (K2) of TFPI to relieve inhibition of the extrinsic pathway and promote blood clot formation [5, 6]. Marstacimab is currently approved for the treatment of hemophilia A and B with and without inhibitors [7, 8] and its safety, efficacy, pharmacokinetic (PK) and pharmacodynamic (PD) profiles have been established from previous clinical studies [6, 9–14].

The Phase I, single ascending dose study established safety of subcutaneous (SC) marstacimab in healthy volunteers up to 450 mg [9]. Results from the study showed that exposure (area under the curve [AUC] and maximum concentration [C_max]) increased in a greater than dose-dependent manner with associated increases in plasma TFPI (indicative of target binding), peak thrombin generation and D-dimer (indicative of clot formation). These data provided the rationale for a once-weekly (QW) dosing regimen with potential to overcome the treatment burden of frequent IV infusions. The Phase Ib/II multiple ascending dose study of adult males with severe hemophilia A and B with or without inhibitors showed that marstacimab was well tolerated and significantly reduced mean annualized bleeding rates (ABR) compared with on-demand and pre-treatment factor replacement therapy over 3 months [12]. Exposure generally increased in a dose-related manner with steady-state concentration reached by Day 57 alongside treatment-related increases in TFPI and clotting biomarkers [12]. Efficacy and safety were maintained for a further 12 months in the Phase II extension study [13, 14], which informed the dosing strategy for the pivotal Phase III study in adolescents and adults with severe hemophilia A or moderately severe to severe hemophilia B with or without inhibitors [13, 14].

The aim of this current analysis was to develop a population pharmacokinetic (popPK) model, based on PK data from healthy volunteers and patients with hemophilia, and to characterize the factors that contribute to variability in marstacimab concentration levels. Additionally, an E_max PK/PD model was developed to link popPK model-predicted free target (TFPI) concentrations to a key clotting biomarker, peak thrombin.

Methods

Study Population

Clinical Studies and Data Collection

Data used for the analysis were obtained from 6 clinical studies (Table 1) [9–14]. The study design, participants, and timing of blood sample collection varied among these studies: three Phase I studies in healthy volunteers and participants with hemophilia (N = 69) (NCT02531815, NCT04832139, NCT04878731) [9–11]; one Phase Ib/II study of participants with hemophilia (NCT02974855) [12] and subsequent Phase II study of participants with hemophilia (NCT03363321) [13] (N = 28); and one Phase III study of participants with hemophilia (non-inhibitor cohort, N = 116) (NCT03938792) [14]. All 6 studies were approved by institutional review boards or ethics committees according to the requirements of the specific countries of origin. All participants provided written informed consent prior to initiation of study medications. Dosing regimens, and PK and PD sampling times are summarized in Table 1. Plasma samples to determine marstacimab and total TFPI were analyzed using validated assays (Supplementary Methods). The lower limit of quantification (LLOQ) of the assay used for samples included in this popPK model was 100 ng/mL for marstacimab; concentrations below LLOQ were reported as below limit of quantification (BLQ). The LLOQ for total TFPI was 10ng/mL; concentrations below LLOQ were reported as BLQ. The upper limit of quantification for total TFPI was 1000ng/mL.

Table 1.

Clinical studies included in popPK and PD analysis

NCT [references]	Phase	Study design	Population	N	Dosing regimen	PK/PD sampling
NCT02531815 [9]	I	Double-blind, placebo-controlled, single ascending dose escalation study	Healthy volunteers	41	Single SC dose cohorts of 30 mg, 100 mg, 300 mg Single IV, dose cohorts of 150 mg and 440 mg An additional single SC dose cohort of 300 mg SC of Japanese participants	Predose on Day 1, 1 h, 2 h (IV cohort only), 4 h, 8 h (IV cohort only), 12 h, 24 h (Day 2), 48 h (Day 3), 72 h (Day 4), Day 7, Day 14, Day 21, Day 28, Day 35, Day 42, Day 56, Day 70 and Day 84
NCT02974855 [12]	Ib/II	4-arm, open-label, multiple ascending dose study designed to evaluate the safety, tolerability, PK, PD, and efficacy of SC marstacimab	Participants with severe hemophilia	26	Multiple SC dose cohorts Participants without inhibitors: 300 mg SC QW, 150 mg SC QW with a 300 mg SC loading dose, 450 mg SC QW Participants with inhibitors: 300 mg SC QW	Predose on Day 1, Day 2, Day 4, Day 8, Day 15, Day 22, Day 29, Day 30, Day 33, Day 57, Day 85 and Day 113
NCT03363321 [13]	II	Multicenter, open-label study to evaluate the long-term safety, tolerability and efficacy of SC marstacimab	Participants with severe hemophilia	22 (participants continued from Study 2, n=20; participants were enrolled de novo, n=2)	Participants from marstacimab 300 mg SC dose of study NCT02974855 continued in study NCT03363321	Predose on Day 1, Day 29, Day 85 and Day 169
NCT03938792 [14]	III	Open-label study comparing standard treatment to marstacimab prophylaxis (non-inhibitor cohort)a	Adolescent and adult severe (coagulation factor activity <1%) hemophilia A participants with or without inhibitors or moderately severe to severe hemophilia B participants (coagulation factor activity ≤ 2%) with or without inhibitors	116	Participants without inhibitors: 150 mg SC QW with a 300 mg SC loading dose Individual participants who met protocol specified dose escalation criteria based upon break-through bleed were allowed to have the dose increased to 300 mg SC QW	Predose (Day 0), Day 7, Day 28, Day 60, Day 120, Day 240, Day 300 and Day 360 (only adult participants)
NCT04832139 [10]	I	Open-label, randomized, 4-period, 2-sequence, crossover study to evaluate the bioequivalence of marstacimab prefilled syringe device and prefilled pen device following SC administration	Healthy adult male participants	22	A single dose of marstacimab 300 mg SC was administered using a prefilled syringe or prefilled pen device	Predose (0 h), 1 h, 2 h, 4 h, 8 h, 12 h, 24 h, 48 h, 72 h, Day 7, Day 14 and Day 21 after a single administration of marstacimab
NCT04878731 [11]	I	Single-arm, open-label, non-randomized, non-controlled multicenter study to evaluate the PK, PD, safety, and tolerability of a single SC dose of marstacimab	Chinese adult participants with severe hemophilia	6	A single dose of marstacimab 300 mg SC	Predose (−2 h to −5 min), 1 h, 2 h, 4 h, 8 h, 12 h, 24 h, 48 h, 72 h, Day 7, Day 14, Day 21 and Day 28 after a single administration of marstacimab

h hour, IV intravenous, NCT national clinical trials, PD pharmacodynamic, PK pharmacokinetic, popPK population pharmacokinetics, QW once weekly, SC subcutaneous

^aOnly participants from the non-inhibitor cohort were included in this analysis. The inhibitor cohort was ongoing at time of this analysis

Data for Analysis

All participants treated with marstacimab and for whom marstacimab plasma concentrations (from at least 1 post-dose visit) were available were included in the analysis. The analysis dataset for popPK and PK/PD models included participant identification, dosing information, time of sample collection, marstacimab and total TFPI concentrations, demographic information, and biomarker values including peak thrombin for PK/PD modeling.

Data Exclusions

For the Phase III study (NCT03938792) [14], PK, PD (total TFPI, peak thrombin) and immunogenicity data at Day 360 were not included in the analysis. Marstacimab concentration was expected to reach steady state by 3 months [12] and therefore Day 360 values were not expected to significantly affect estimation of model parameters. Accordingly, an early snapshot of PK, antidrug, antibody (ADA)/neutralizing antibody and relevant PD biomarkers (excluding Day 360 data) were utilized for creating the dataset used for this modeling exercise. Only data from the non-inhibitor cohort were included (the inhibitor cohort of this study was ongoing at the time of data analysis). One participant was excluded from the analysis as they had only received a single loading dose of marstacimab and the single plasma marstacimab concentration value available post-dose was BLQ.

Missing Data and Imputations

For the popPK analysis, all concentrations reported as below the LLOQ for marstacimab were set to missing (missing dependent variable indicator was set to 1). Any concentration data with missing date/time of sample collection or data/time of dose were excluded.

Table S1 shows the number of BLQ PK observations and imputed total TFPI and peak thrombin observations in the different studies, populations and overall. Most of the BLQ values in PK occurred in the healthy volunteer studies because low doses (30 mg SC QW and 100 mg SC QW) were studied in the Phase I study (NCT02531815) and samples were taken for an extended period of time after a single dose in the Phase I prefilled pen device study (NCT04832139).

Even though the percentage of marstacimab concentrations below LLOQ was slightly greater than the accepted threshold of 10%, since the majority of BLQ concentrations came from the studies in healthy volunteers and this percentage was lower for participants with hemophilia, an M3 approach was not considered for modeling. There were no systematic trends observed in BLQ concentrations, and excluding these samples was not expected to have a significant impact on structural model development. For the total TFPI and peak thrombin observations, few (0.2% and 1.9%, respectively, for participants with hemophilia, 0% for healthy volunteers) baseline observations were imputed to the median baseline levels of the respective studies (Table S1).

For the PK/PD modeling, peak thrombin records were not included in the model if samples were taken within 4 days of a participant taking replacement factors in response to a bleed. The effect of a standard or extended half-life replacement factor therapy remains for approximately 96 hours after bleed treatment, so the clinical biomarker levels measured shortly after the bleed may be affected by the treatment of the bleed.

Modeling: Strategy and Software

Translational PK/PD modeling showed that a target mediated drug disposition (TMDD) model with an additional nonlinear elimination adequately described the in vitro marstacimab PK [15]. Using this model as a starting point, a preliminary popPK analysis of marstacimab was conducted with pooled data from the Phase I study of healthy adult participants (NCT02531815) and partial data from the Phase Ib/II study of participants with hemophilia (NCT02974855). Subsequently, the model was updated with all participants from the Phase Ib/II study. The results of this updated popPK analysis using complete Phase I and Phase II data demonstrated that a TMDD model with quasi–steady-state approximation (QSSA) accurately described the concentration–time course of marstacimab and total TFPI and was determined to be an appropriate initial structural model for the current analysis with Phase I, Phase II, and Phase III studies.

Data were analyzed using nonlinear mixed-effects modeling (NONMEM) methodology as implemented by the software program NONMEM version 7.5.0 within an internally validated Pfizer analysis platform (Improve version 4.0.4). Nonlinear mixed-effects modeling was used to estimate the population parameters, mean and interindividual variance (IIV), and to identify potential covariates that explain IIV in the parameters. Stochastic Approximation Expectation-Maximization with NONMEM serial version was used for all model runs. Visual predictive checks (VPCs) and/or bootstrap simulations (N = 1000) were conducted using the software program Perl-speaks-NONMEM (PsN) version 5.3.0. Pre- and post-processing of NONMEM software output to generate goodness-of-fit (GOF) plots were performed using R software (R 4.1.3 and 4.2.1). The R package mrgsolve (version 1.0.4) was used for generating simulated PK and PK/PD results. Data were graphically examined for any potential outliers and the analysis used the best practices consistent with current regulatory guidance for popPK modeling [16]. The analysis followed the steps of building a base model and a final model after inclusion of covariates of scientific/clinical relevance [15].

Model Development

PopPK Model

Based on prior modeling experience [15] and the non-linearity observed in AUC_last and maximum concentration (C_max) of the Phase I study over the range of doses studied, marstacimab PK was initially described using a TMDD model defined in terms of CL (linear clearance), V_c (central volume), Q (inter-compartmental flow rate), V_p (peripheral volume), k_a (first-order absorption rate constant), and A_lag (lag time).

Bioavailability (F) after SC administration was estimated using the IV data from healthy volunteers in the Phase I study (NCT02531815). Marstacimab concentration data and total TFPI data were used to estimate model parameters. The IIV in the PK parameters was modeled assuming a log-normal parameter distribution (see Supplementary Appendix for details), and the percent coefficient of variation (% CV) for IIV was calculated for base and final models. Separate residual error models were included in the model for PK and total TFPI. Different approximations of the TMDD model formulation, such as Michaelis–Menten approximation and QSSA were tested to arrive at the model which best describes marstacimab concentration and total TFPI concentration. Other models, e.g., a 2-compartment model with or without nonlinear elimination for PK, were also explored.

Body weight (with estimation of allometric scaling constants) was included as structural covariate on CL, V_c, Q and V_p. The structural model was further explored in case of model non-convergence or poor GOF diagnostics.

Close attention was given to evaluating the stability of the models throughout the model development process. To avoid ill conditioning, inspection of the covariance ($COV) step output at every stage of model development was performed to verify that no extreme pairwise correlations of the parameters were encountered.

E_max PK/PD Model

An E_max PK/PD model was developed to describe peak thrombin, a key clotting biomarker, and its relationship to free TFPI predicted from the popPK model. Since there is an inverse relationship between free TFPI and peak thrombin, the reciprocal of model-predicted free TFPI was used as the derived independent variable in the modeling analysis using the Phase I (NCT02531815) and Phase Ib/II (NCT02974855) studies (see Supplementary Appendix for details). The PK/PD modeling was only performed for participants with hemophilia in this analysis as prediction of peak thrombin in the hemophilia population is of key interest.

Interindividual variance was explored on all the parameters in the E_max model equation. Similar to the strategy applied for the popPK model, close attention was paid to evaluate stability and suitability of the models via objective function value (OFV), GOF plots, inspection of the covariance step and prediction-corrected VPCs (pcVPCs).

Inclusion of Covariates and Final Model Development

Covariate-parameter relationships were evaluated based on mechanistic plausibility, exploratory analysis and clinical interest. Covariates were initially plotted against individual η values to identify any relationships. It was assumed that not all the parameters in the TMDD model were dependent on the different covariates.

The Phase III study (NCT03938792) could include participants aged up to 75 years as well as those with moderate renal impairment (eGFR 30 to < 60 mL/min/1.73 m²). However, exploratory analyses to evaluate the effects of older age group (65+ years of age) and moderate renal impairment on marstacimab pharmacokinetics were not explored as only one participant was aged >65 years and there were no participants with moderate renal impairment in the dataset.

A full modeling approach was used for covariate modeling and at every stage of model development to assess the effect of various (continuous or categorical) covariates on marstacimab PK parameters. Correlations between covariates were checked, and if strong correlations existed, the covariate that was more clinically relevant or with greater significance was included. For covariate modeling, plots of the η values and/or post hoc estimates of the parameters versus covariates was used to guide covariate parameterizations where trends were observed.

Covariates screened and added to the base model included body weight, race, participant status (healthy vs hemophilia), hemophilia type and ADA status on CL and body weight on V_c, Q and V_p (Table 2). The parameterization of the continuous and categorical covariates is described in the Supplementary Appendix.

Table 2.

Covariates evaluated in the popPK model

PK parameter	Covariates
CL	Weight, race, participant type (healthy vs hemophilia), hemophilia type, ADA status
Vc	Weight
Q	Weight
Vp	Weight
F	Injection site (abdomen, arm, thigh, not available)a

ADA antidrug antibody, CL linear clearance, F bioavailability fraction, popPK population pharmacokinetics, Q inter-compartmental clearance, V_c volume of distribution, V_p peripheral volume of distribution

^aPercentage of injection sites for marstacimab 150-mg dose group for the Phase III study (NCT03938792): abdomen: 67.0%; arm: 21.4%; thigh: 11.3%; not available: 0.3%

Injection Site Analysis for Marstacimab PK

To understand the effect of site of injection on marstacimab PK parameters, data from the 150-mg dose cohort of the Phase III study (NCT02531815) were selected. This cohort had the largest number of injections across all studies and consists of one injection per dose (except the loading dose) compared with other studies where multiple injections per dose were administered at multiple injection sites (Table S2). The majority (67%) of injections were administered in the abdomen (Table 2). There was no available information regarding the site of injection for 0.4% of injections; in these instances, the injection site was fixed to the abdomen for bioavailability purposes. Injection site was added as a covariate on the absorption parameters, keeping the other parameters of the model fixed to the values obtained in the comprehensive model developed previously (see Supplementary Appendix for details). Due to sparse sampling in this cohort, injection site was only added as a covariate for bioavailability and not on other absorption parameters such as k_a and A_lag.

Assessment of Model Adequacy (Goodness of Fit)

At all stages of model development, assessment of model adequacy was conducted via GOF criteria: change in OFV, visual inspection of diagnostic plots, precision of the parameter estimates, and decreases in between-participant and residual variability. Goodness of fit plots were examined to assess model adequacy, possible lack of fit, or violation of assumptions. Standard diagnostic plots included: (a) population predictions (PRED) and individual predictions (IPRED) with dependent variable, (b) conditional weighted residuals (CWRES), individual weighted residuals (IWRES) with PRED, IPRED and independent variable (time), and (c) η values (etas) with continuous and categorical covariates.

Quantile-quantile (QQ) plots were used to visualize the distribution of η_s to ensure approximately normal distributions. Shrinkage towards the mean was also assessed to confirm model assumptions. In addition, plots of η values in the final model with each covariate were compared to similar plots for the base model to check that the final model accounted for trends observed with the base model. For base and final models, 95% confidence interval (CI) around the parameter estimates was generated based on standard errors generated from the NONMEM covariance step or by non-parametric bootstrapping.

Assessment of Model Predictive Performance (Validation)

The performance of the final model was evaluated by simulating data using final parameter estimates (fixed and random effects) and conducting a posterior predictive check. The concordance between individual observations and simulated values as well as the distribution of observed and simulated data was also assessed.

For both the popPK and PK/PD models, pcVPCs were performed to assess whether the final model described the central tendency and variability in the observed data by comparing the observed median and quantiles to the median and quantiles of the simulated dataset; i.e., data that would arise using the observed data structure if the hypothesized models were true.

Assessment of Marstacimab Dose in Adult and Adolescent Participants

A dosing regimen of 300-mg SC loading dose followed by 150 mg SC QW was selected for both adolescents and adults in the Phase III study based on predictions from a previously developed popPK and PK/PD model from Phase I and Phase II studies [15]. Data from adult and adolescent participants in the Phase III study were compared with data from a previously developed popPK and PK/PD model to assess marstacimab dose in adolescents and confirm the dose strategy for adults and adolescents. Simulations were performed for a population of males aged 12 to <18 years with weight distribution taken from the Centers for Disease Control and Prevention growth charts [17]. Approximately 500 adolescent males in each age group were simulated, providing a total of 3000 simulated male subjects. Weights were kept between 3rd and 97th percentiles of the age range for adolescent males. Observed weights were compared with simulated weights. Observed marstacimab, total TFPI and peak thrombin concentrations for adults and adolescents were compared with model predictions for the adolescent population.

Results

Baseline Demographic/Covariates for Analysis

A total of 213 unique male participants (healthy volunteers, n = 63; participants with hemophilia, n = 150) from 6 marstacimab studies were included in the analysis (Table 1). Baseline continuous and categorical variables for the pooled study population are summarized in Tables 3 and 4, respectively. Median (range) age was 33 (13–66) years and median (range) body weight was 71.1 (35–120) kg. Mean (SD) concentration of total TFPI was 132.5 (28.5) ng/mL and peak thrombin was 20.7 (18.8) nM. Most participants were White (33.3%) or Asian (32.4%), ADA negative (76.4%), had hemophilia A (56.8%) and had normal renal function (78.4%).

Table 3.

Participant demographic table: continuous covariates

Variable	Statistic	Phase I study; NCT02531815 [9]	Phase Ib/II; Phase II NCT02974855 [12]/NCT03363321 [13]	Phase III (Adol) NCT03938792 [14]	Phase III (adults) NCT03938792 [14]	Phase I NCT04832139 [10]	Phase I NCT04878731 [11]	All studies combined	Healthy volunteer studies	Hemophilia participant studies
Baseline age (years)	N	41	28	19	97	22	6	213	63	150
	Mean	36.1	35.5	14.7	35.9	41.6	30.8	34.4	38.0	32.9
	SD	10.6	12.4	1.5	12.4	10.9	2.5	12.8	10.9	13.3
	Median	36.0	34.5	14.0	35.0	43.0	31.5	33.0	39.0	31.0
	Min	19.0	19.0	13.0	18.0	22.0	27.0	13.0	19.0	13.0
	Max	55.0	63.0	17.0	66.0	56.0	34.0	66.0	56.0	66.0
Baseline weight (kg)	N	41	28	19	97	22	6	213	63	150
	Mean	75.3	73.0	54.7	71.6	79.5	59.2	71.5	76.8	69.2
	SD	11.0	13.5	12.8	15.6	12.5	12.5	15.1	11.6	15.9
	Median	75.3	74.9	53.0	71.0	80.6	56.5	71.1	77.6	69.6
	Min	52.3	50.7	35.0	43.2	58.0	46.0	35.0	52.3	35.0
	Max	98.8	96.0	74.1	120.0	100.2	79.4	120.0	100.2	120.0
Baseline eGFR (mL/min/1.73 m2)	N	41	28	19	97	22	6	213	63	150
	Mean	94.9	125.2	181.8	111.5	94.3	133.8	115.2	94.7	123.9
	SD	16.2	32.4	33.3	23.0	18.2	25.7	33.4	16.8	34.9
	Median	93.2	123.4	184.8	106.7	89.6	131.6	106.7	92.5	116
	Min	71.3	80.1	114.7	76.1	68.6	103.1	68.6	68.6	76.1
	Max	150.2	214.2	233.1	188.4	129.5	176.3	233.1	150.2	233.1
Baseline total TFPI (ng/mL)	N	41	27	18	96	–	6	188	41	147
	Mean	134.9	148.3	122.0	129.2	–	130.3	132.5	134.9	131.9
	SD	26.4	37.7	20.5	27.0	–	11.3	28.5	26.4	29.1
	Median	136.0	144.0	122.5	126.5	–	131.5	130.0	136.0	128.0
	Min	92.7	100.0	90.0	81.0	–	110.0	81.0	92.7	81.0
	Max	193.0	271.0	155.0	210.0	–	143.0	271.0	193.0	271.0
Baseline peak thrombin (nM)	N	41	28	16	79	–	6	129a	41	129
	Mean	95.8	23.0	14.7	21.5	–	14.9	20.7a	95.8	20.7
	SD	39.2	14.0	7.8	21.8	–	15.3	18.8a	39.2	18.8
	Median	93.2	21.0	12.6	16	–	9.2	15.9a	93.2	15.9
	Min	42.0	6.6	6.0	3.6	–	4.9	3.6a	42.0	3.6
	Max	183.6	60.0	33.0	141.6	–	45.3	141.6a	183.6	141.6

Adol adolescents, eGFR estimated glomerular filtration rate, max maximum value, min minimum value, N number of participants, SD standard deviation, TFPI tissue factor pathway inhibitor

^aThe combined column for baseline peak thrombin shows summary statistics for all hemophilia studies only, i.e., excluding the Phase I study in healthy volunteers [9]

Table 4.

Participant demographic table: categorical covariates

Variable	Statistic	Phase I study; NCT02531815 [9]	Phase Ib/II; Phase II NCT02974855 [12]/NCT03363321 [13]	Phase III (Adol) NCT03938792 [14]	Phase III (adults) NCT03938792 [14]	Phase I NCT04832139 (PFS vs PFP study) [10]	Phase I NCT04878731 (China study) [11]	All studies combined	Healthy volunteer studies	Hemophilia participant studies
Race	Total	41	28	19	97	22	6	213 (100%)	63 (100%)	150 (100%)
	White	33	16	0	1	21	0	71 (33.3%)	54 (85.7%)	17 (11.3%)
	Black or African American	3	12	0	0	1	0	16 (7.5%)	4 (6.3%)	12 (8%)
	Asian	5	0	7	51	0	6	69 (32.4%)	5 (7.9%)	64 (42.7%)
	American Indian	0	0	0	0	0	0	0 (0%)	0 (0%)	0 (0%)
	Native Hawaiian	0	0	12	44	0	0	56 (26.3%)	0 (0%)	56 (37.3%)
	Other	0	0	0	1	0	0	1 (0.5%)	0 (0%)	1 (0.7%)
Administration	Total	41	28	19	97	22	6	213 (100%)	63 (100%)	150 (100%)
	IV	12	0	0	0	0	0	12 (5.6%)	12 (19%)	0 (0%)
	SC	29	28	19	97	22	6	201 (94.4%)	51 (81%)	150 (100%)
ADA status	Total	41	28	19	92	22	6	208 (100%)	63 (100%)	145 (100%)
	ADA positive	15	3	2	21	6	2	49 (23.6%)	21 (33.3%)	28 (19.3%)
	ADA negative	26	25	17	71	16	4	159 (76.4%)	42 (66.7%)	117 (80.7%)
Participant status	Total	41	28	19	97	22	6	213 (100%)	63 (100%)	150 (100%)
	Healthy	41	0	0	0	22	0	63 (29.6%)	63 (100%)	0 (0%)
	Hemophilia A	0	25	15	76	0	5	121 (56.8%)	0 (0%)	121 (80.7%)
	Hemophilia B	0	3	4	21	0	1	29 (13.6%)	0 (0%)	29 (19.3%)
Baseline renal impairment	Total	41	28	19	97	22	6	213 (100%)	63 (100%)	150 (100%)
	Normal	27	24	19	80	11	6	167 (78.4%)	38 (60.3%)	129 (86%)
	Mild	14	4	0	17	11	0	46 (21.6%)	25 (39.7%)	21 (14%)
	Moderate	0	0	0	0	0	0	0 (0%)	0 (0%)	0 (0%)
	Severe	0	0	0	0	0	0	0 (0%)	0 (0%)	0 (0%)

ADA antidrug antibody, Adol adolescents, IV intravenous, SC subcutaneous

Base Model Results

Marstacimab PK was described using a TMDD model with QSSA with additional nonlinear clearance for drug (marstacimab) added to the model (Fig. 1). Full parameter estimates of the base model are provided in (Table S3).

Fig. 1.

Schematic of the final popPK model. TMDD model with quasi-steady-state approximation and additional nonlinear degradation of the drug. Weight added as a structural covariate on CL, Q, V_c and V_p. C_free concentration of free drug, CL clearance, IV intravenous, k_COM complex elimination constant, k_syn synthesis rate, k_deg degradation rate, K_m Michealis–Menten constant, popPK population pharmacokinetics, Q intercompartmental clearance, SC subcutaneous, sTFPI soluble tissue factor pathway inhibitor, TMDD target mediated drug disposition, V_c, central volume of distribution, V_p peripheral volume of distribution, V_max maximum elimination rate

Population and individual predictions of marstacimab concentrations showed good agreement with observed total marstacimab concentration, with some deviation from the model observed at higher concentrations above the clinically relevant dosing regimen of 300 mg (Figure S1A). Residual-based diagnostic plots showed no systematic bias in the model with population or individual predictions and time from the first dose (Figure S1B). Prediction-corrected VPC showed the base model adequately described marstacimab drug concentrations across all studies (Figure S1C). Supplementary Figure S2 shows the pcVPC of marstacimab for the Phase III study. For total TFPI concentrations, GOF was demonstrated with population and individual predictions (Figure S3A), no systematic bias was observed (Figure S3B), and pcVPC results showed the observed data were in good agreement with the model (Figure S3C). The effect of weight on IIV parameters is shown in Supplementary Figure S4. Overall, diagnostic plots for the base model showed that it was able to characterize marstacimab PK and total TFPI concentrations well and could be used to study the effect of covariates in the model.

Parameter Estimates

All clinically important covariates were considered and their effect on key model parameters was quantified to arrive at the final model. The relative standard error (RSE) values of the parameter estimates show that most PK parameters were estimated precisely with RSE <30% (Table 5). The population estimate for marstacimab CL was 0.0188 L/h (11.3% RSE). Parameter estimates for V_c and V_p were 3.61 L (9.2% RSE) and 4.99 L (10.5% RSE), respectively, giving a total volume of distribution of 8.60 L. Bioavailability was estimated to be 70.5% (4.4% RSE) with a first-order K_a of 0.00742 (1/h) (7.3% RSE) and A_lag of 2 h (2.3% RSE). Weight exponent on V_c and CL was estimated to be 1.86 (18.5% RSE) and 1.14 (32.0% RSE), respectively.

Table 5.

Parameter estimates of the final popPK model

Parameter	Population estimate	SE	RSE/CV (%)	Bootstrap estimate	Bootstrap 95% CI	ETA shrinkage SD (%)
Vc (L)	3.61	0.33	9.15	3.12	2.29, 4.15	–
CL (L/h)	0.0188	0.00212	11.3	0.0181	0.0118, 0.0226	–
Q (L/h)	0.00489	0.000501	10.2	0.00554	0.00389, 0.0189	–
Vp (L)	4.99	0.527	10.5	4.13	2.39, 34.3	–
Km for nonlinear drug clearance, Km, (nM)	4.31	0.429	9.94	3.56	1.78, 4.6	–
Vmax for nonlinear drug clearance, Vmax (nM/h)	0.53	0.0422	7.96	0.605	0.377, 0.81	–
F	0.705	0.0314	4.44	0.69	0.554, 0.744	–
Ka (1/h)	0.00742	0.000541	7.29	0.00695	0.00583, 0.00938	–
Baseline total TFPI (nM)	3.75	0.0571	1.52	3.76	3.64, 3.87	–
kdeg (1/h)	0.0161	0.000921	5.72	0.0159	0.014, 0.0186	–
kcom (1/h)	0.000754	9.31 × 105	12.3	0.000694	0.000551, 0.000928	–
KSS	72.2	3.64	5.04	73.8	65.9, 82.3	–
Lag time (h)	2	0.0449	2.25	1.98	1.23, 2.59	–
Additive residual error on total TFPI (log scale)	0.158	0.00154	0.971	0.157	0.143, 0.174	–
Additive residual error on marstacimab (log scale)	0.392	0.000991	0.252	0.391	0.341, 0.444	–
Weight exponent on Vc	1.86	0.345	18.5	1.77	1.19, 2.49	–
Weight exponent on CL	1.14	0.365	32	1.15	0.568, 1.71	–
Additional effect on CL for healthy population	0.199	0.211	106	0.163	−0.124, 0.842	–
Additional effect on CL for Asian participants	0.346	0.189	54.8	0.34	0.109, 0.689	–
Additional effect on CL for ADA +ve participants	0.229	0.191	83.2	0.196	−0.0375, 0.477	–
Additional effect on CL for mild renal impairment	−0.168	0.118	70.4	−0.152	−0.34, 0.121	–
Additional effect on CL for HB participants	0.118	0.162	138	0.0881	−0.15, 0.402	–
IIV on Vc	0.232	0.0819	48.1	0.162	0.0414, 0.486	39.07
Correlation coefficient between IIVs for CL and Vc	−0.527	0.172	41.4	−0.113	−0.265, 0.00799
IIV on CL	0.262	0.0592	51.2	0.246	0.126, 0.407	33.284
IIV on Q (fixed)	0.0225	0	15	0.0225	0.0225, 0.0225	82.841
IIV on Vp (fixed)	0.0225	0	15	0.0225	0.0225, 0.0225	59.542
IIV on Km (fixed)	0.0225	0	15	0.0225	0.0225, 0.0225	81.72
IIV on Vmax	0.101	0.043	31.8	0.105	0.02, 0.29	42.135
IIV on F	0.0332	0.0585	18.2	0.0575	0.00106, 0.208	79.918
IIV on Ka	0.278	0.0504	52.7	0.201	0.116, 0.405	31.537
IIV on baseline total TFPI	0.0249	0.00324	15.8	0.0249	0.0178, 0.0325	17.713
IIV on Kdeg	0.023	0.0183	15.2	0.0219	4.99 × 105, 0.215	75.072
IIV on Kcom (fixed)	0.0225	0	15	0.0225	0.0225, 0.0225	85.671
IIV on KSS constant	0.105	0.0301	32.3	0.105	0.0493, 0.166	37.039
IIV on lag time (fixed)	0.0225	0	15	0.0225	0.0225, 0.0225	36.942

For this analysis, the hemophilia A subgroup (patient type), non-Asian (race), ADA −ve, (ADA status type) and no renal impairment were treated as the default (most represented) population type for the covariate analysis. Bootstrap estimates were obtained from a bootstrap run with N = 1000 samples using the software program Perl-speaks-NONMEM version 5.3.0 bootstrap routine. %RSE is provided for population (theta) parameters, %CV is provided for IIV parameters. Some of the IIVs were fixed to 15%CV. Residual variability was calculated on PK and total TFPI via an additive error model on the log scale with a theta-rized (fixed) sigma parameter

ADA antidrug antibody, CI confidence interval, CL linear clearance, CV coefficient of variation, ETA represents the addition of a random variable drawn from a normal distribution on a parameter resulting in inter-individual variability on this parameter, F bioavailability fraction, HB hemophilia B, K_a transfer rate from SC to central compartment, IIV interindividual variance, K_com degradation rate of marstacimab-TFPI complex, K_deg degradation rate of free soluble TFPI, K_m Michealis-Menten constant, K_SS quasi-steady-state parameter, Q inter-compartmental flow, RSE relative standard error, SE standard error, SC subcutaneous, SD standard deviation, TFPI tissue factor pathway inhibitor, Vc central volume, V_max maximum elimination rate, Vp peripheral volume

Model parameters associated with estimating the covariate effects were less precisely estimated due to fewer participants/observations for the different sub-groups based on the covariate (Table 5). The additional effects of parameters on CL estimates were a 22.9% increase for ADA positive status, a 16.8% decrease for mild renal impairment (estimated glomerular filtration rate [eGFR] < 90 mL/min/1.73 m²), 11.8% increase for hemophilia B, 19.9% increase for healthy volunteers, and 34.6% increase for Asian race. However, these changes in CL were not considered significant as estimate precision was low (RSEs >50%) and bootstrap 95% CI contained zero.

Although a 28.8% decrease in CL values was seen in post hoc estimates for adolescents versus adults, this effect reduced to 3.1% after accounting for the weight differences (Table 6). Similarly, the weight-adjusted decrease in V_c for adolescents was approximately 25.9%. After accounting for weight differences, the effect of participant status (healthy vs hemophilia) was 12.1%, Asian race (vs non-Asian) was 31.9% and hepatic impairment was −11.1%. The plot for different races shows a clear overlap between the individual estimates of CL, demonstrating no clinically relevant difference in PK between Asian and non-Asian participants (Figure S5).

Table 6.

Absolute and weight-normalized linear clearance and central volume

Variable	Type 1	Type 2
Age type	Adolescents	Adults
n	19	131
Median CL (L/h)	0.0143	0.0201
Difference in CL from adults (%)	−28.8	–
Median weight-adjusted CL (CL/W) (L/h/kg)	0.000279	0.000288
Difference in CL/W from adults (%)	−3.13	–
Median Vc (L)	2.54	4.48
Difference in Vc from adults (%)	−43.2	–
Median weight-adjusted Vc (Vc/W) (L/kg)	0.0452	0.0609
Difference in Vc/W from adults (%)	−25.857	–
Participant type	Healthy	Hemophilia
n	63	150
Median weight-adjusted CL (CL/W) (L/h/kg)	0.000322	0.000287
Difference in CL/W from hemophilia participants (%)	12.135	–
Race type	Asian	Non-Asian
n	64	86
Median weight-adjusted CL (CL/W) (L/h/kg)	0.000326	0.000247
Difference in CL/W from non-Asians (%)	31.934	–
Hepatic impairment type	Mild impairment	None
n	15	135
Median weight-adjusted CL (CL/W) (L/h/kg)	0.000256	0.000288
Difference in CL/W from no hepatic impairment participants (%)	−11.1	–

For calculating hepatic impairment type, participants who had bilirubin ≤ 40 µmol/L and AST ≤ 20 µmol/L were considered as having no hepatic impairment

AST aspartate aminotransferase, CL linear clearance, CL/W weight-adjusted linear clearance, V_c central volume, V_c/W weight-adjusted central volume

Final PopPK Model Results

Final Model Evaluation: GOF and pcVPC of the Final Model

Diagnostic plots and pcVPCs of marstacimab (Fig. 2) illustrate final model appropriateness. The diagnostic plots of observed marstacimab concentration with population and individual predictions (Fig. 2A and Figure S6) indicate good agreement between observed values and the population/individual predictions from the model for marstacimab concentrations. Residual-based diagnostic plots of CWRES and IWRES versus PRED, IPRED and time after first dose (in days) showed no systematic trend, indicating that the final model described the data reasonably well (Fig. 2B). A comparison of the final diagnostic plots of marstacimab to similar base model plots of marstacimab (Fig. 2 and Figure S1) demonstrated no obvious covariate trends. In the pcVPCs, the median values and corresponding 90% CIs (5th and 95th percentile points) for simulated and observed marstacimab plasma concentrations were similar (Fig. 2C); the observed median was generally contained within the 90% CI of the simulated median data, which indicated that the model adequately described the central tendency of the marstacimab concentration–time profile. Additional pcVPCs for the Phase III study stratified by age and treatment group also demonstrated adequacy of fit (Figure S7). Figure S7C and S7D also show the QQ plots and distribution of etas, respectively.

Fig. 2.

Assessment of final popPK model adequacy for predicting for marstacimab concentrations by A goodness-of-fit plots with population and individual predictions for marstacimab concentrations (linear scale), B residuals versus population and individual predictions for marstacimab concentration and C visual predictive checks of marstacimab concentrations by dosing regimen. A Line of Unity is shown in gray. Loess fit of data is shown in blue. B Loess fit of residuals is shown in blue line. C Prediction-corrected VPC of marstacimab concentration for all patients divided into 150-mg QW and 300-mg QW dosing regimens. The solid red line shows median of the observed data, the dashed black line shows the same for the simulated data. The pink shaded region shows 90% prediction interval around the simulated median. The dotted black lines show 5th and 95th quantiles of the simulated data and the dashed red lines show the same for the observed data. The shaded blue regions show 90% prediction interval around 5th and 95th percentile of the simulated data. The blue circles show observed data. CWRES conditional weighted residuals, IWRES individual weighted residuals, popPK population pharmacokinetics, QW once-weekly, VPC visual predictive check

Diagnostic plots and pcVPCs of total TFPI (Fig. 3) illustrate appropriateness of the final model. The GOF plots demonstrate good agreement between the observed total TFPI concentrations and population/individual predictions (Fig. 3A). Residual-based diagnostic plots of CWRES and IWRES versus PRED, IPRED and time after first dose (in days) for total TFPI concentrations (Fig. 3B) showed no systematic bias in the model with population or individual predictions and time after the first dose. A comparison of the final diagnostic plot and base model plot of total TFPI (Fig. 3 and Figure S3) demonstrated no obvious covariate trends. The pcVPC of total TFPI concentrations for participants stratified by dosing regimen type showed that the model was able to adequately explain the observed total TFPI data (Fig. 3C). Additional pcVPCs of total TFPI concentrations for the Phase III study stratified by age and treatment group also demonstrated that the observed data are described well by the model (Figure S8).

Fig. 3.

Assessment of final popPK model adequacy for predicting for total TFPI concentrations by A goodness-of-fit plots with population and individual predictions for total TFPI concentrations, B residuals versus population and individual predictions for total TFPI concentration and C visual predictive checks of total TFPI concentrations by dosing regimen. A Line of Unity is shown in gray. Loess fit of data is shown in blue. B Loess fit of residuals is shown in blue line. C Prediction-corrected VPC of total TFPI for all patients divided into 150-mg QW and 300-mg QW dosing regimens. The solid red line shows median of the observed data, the dashed black line shows the same for the simulated data. The pink shaded region shows 90% prediction interval around the simulated median. The dotted black lines show 5th and 95th quantiles of the simulated data and the dashed red lines show the same for the observed data. The shaded blue regions show 90% prediction interval around 5th and 95th percentile of the simulated data. The blue circles show observed data. CWRES conditional weighted residuals, IWRES individual weighted residuals, popPK population pharmacokinetics, QW once-weekly, TFPI, tissue factor pathway inhibitor; VPC, visual predictive check

Secondary PK Parameters

The final popPK model was used to calculate secondary PK parameters (Table 7) for all participants with hemophilia using the empirical Bayesian estimates of the model parameters from the final popPK model and simulating a dosing regimen of either 150 mg or 300 mg SC QW in adult and adolescent groups (Table S4). The mean minimum concentration at steady state (C_min,ss), maximum concentration at steady state (C_max,ss) and average concentration at steady state (C_avg,ss) were close to each other, indicating a low peak-to-trough ratio in marstacimab concentrations at steady state for the 150 mg and 300 mg (adults) sub-groups (Table 7). Comparing the marstacimab concentrations between 150 mg and 300 mg (adults) showed a greater than linear increase in the secondary PK parameters. The accumulation ratio was calculated as the ratio of AUC for the dosing interval of 168 hours at steady state to the AUC after first dose (see Supplementary Appendix for details). Median accumulation ratio was estimated to be approximately 4. The mean effective half-life (t_1/2) of marstacimab, calculated from the accumulation ratio, was estimated to be 16–18 days and median elimination half-life was estimated to be 7–10 days, with 90% of marstacimab expected to be eliminated by the end of approximately 1 month after the last dose.

Table 7.

Summary of secondary PK parameters by dosing regimen

Parameter	150 mg (adults)	300 mg (adults)	150 mg (adolescents)	300 mg (adolescents)
Participants (N)	99	32	17	2
Cmin,ss (ng/mL)
Mean (%CV)	13,700 (90.4)	46,500 (48.3)	27,300 (53.2)	60,700 (99.9)
Geometric mean	8320	41,200	23,400	42,900
Median (2.5–97.5 percentile)	9400 (500–46000)	47,300 (14,000–94,300)	28,000 (8150–53,000)	60,700 (20,000–101,000)
Cmax,ss (ng/mL)
Mean (%CV)	17,900 (77.5)	58,200 (45.1)	34,700 (48.5)	71,700 (91.4)
Geometric mean	12,800	52,200	30,500	54,700
Median (2.5–97.5 percentile)	13,600 (1710–53,000)	57,300 (19,600–109,000)	32,400 (11,400–62,200)	71,700 (27,700–116,000)
Cavg,ss (ng/mL)
Mean (%CV)	16,500 (81.2)	54,200 (45.7)	32,100 (49.5)	68,100 (94.1)
Geometric mean	11,400	48,500	28,100	50,900
Median (2.5–97.5 percentile)	12,600 (1280–50,800)	54,700 (17,700–104,000)	31,300 (10,300–59,200)	68,100 (25,100–111,000)
AUCss (ng·h/mL)
Mean (%CV)	2,770,000 (81.2)	9,100,000 (45.7)	5,390,000 (49.5)	11,400,000 (94.1)
Geometric mean	1,910,000	8,150,000	4,720,000	8,550,000
Median (2.5–97.5 percentile)	2,110,000 (214,000–8,530,000)	9,200,000 (2,970,000–17,500,000)	5,260,000 (1,730,000–9,950,000)	11,400,000 (4,210,000–18,700,000)
ACCR
Mean (%CV)	4.86 (58.1)	4.72 (61.2)	4.88 (50.3)	3.94 (24.1)
Geometric mean	4.19	4.14	4.32	3.89
Median (2.5–97.5 percentile)	4.37 (1.42–12.5)	3.9 (1.8–11.3)	4.46 (1.7–10.1)	3.94 (3.31–4.58)
Effective t1/2 (days)
Mean (%CV)	21 (65.6)	20.3 (69.2)	21.1 (56.7)	16.6 (28)
Geometric mean (days)	17.1	17.1	17.9	16.3
Median (2.5–97.5 percentile)	18.7 (3.95–58)	16.4 (5.98–52.6)	19.1 (5.48–46.4)	16.6 (13.5–19.7)

ACCR accumulation ratio, AUC area under the curve, AUC_ss area under the curve at steady state, C_avg,ss average concentration at steady state, C_min,ss minimum concentration at steady state, CV coefficient of variation, t_1/2 half-life

Dose-proportionality was assessed using the predicted individual AUC_ss across different dose groups (150- to 450-mg dose range) and is shown in Supplementary Figure S9. A linear model between log(AUC_ss) and log(DOSE) was used to estimate the value of the slope (β) and intercept (µ) (see Supplementary Appendix for details), which were estimated to be 1.8 and 5.5, respectively (Table S5). However, the 95% CI on the slope was wide (−2.15, 5.77), indicating that the value of β did not significantly differ from 1 and therefore dose non-linearity could not be concluded despite the greater than proportional increases observed in AUC_ss.

The effect of injection site on bioavailability was also analyzed by estimating the difference in bioavailability from the abdomen as a multiplicative factor (see Supplementary Appendix for details). The 3 main injection sites from the 6 marstacimab studies were the abdomen, arm and thigh, although higher doses were administered in multiple injections of 150 mg each across multiple injection sites (Table S2). Marstacimab bioavailability was slightly lower in the arms (0.951; F_arm = 67.0%) and in the thigh (0.959; F_thigh = 67.6%) compared with the abdomen (Table S6). However, there was considerable overlap in the parameter estimates for the different injection sites implying minimal effect of site of injection on bioavailability (Figure S10).

E_max PK/PD Model of Free TFPI and Peak Thrombin

The final popPK model was applied to calculate the model-predicted free TFPI concentration (i.e., TFPI unbound from marstacimab) for each participant (i.e., empirical Bayesian estimates) and was linked to observed peak thrombin concentration using the E_max PK/PD model. An inhibitory relationship was observed between model-predicted free TFPI concentration and peak thrombin concentration, with peak thrombin concentration plateauing at low concentrations of model-predicted free TFPI (Fig. 4).

Fig. 4.

Model-predicted free TFPI concentrations and observed peak thrombin concentrations. The observed peak thrombin data is faceted by different clinical studies in participants with hemophilia [11–14]. The blue line is obtained using equation 5 in the Supplementary Information and population estimates obtained from the PK/PD model and shows the overall relationship between model-predicted free TFPI and observed peak thrombin values. PD pharmacodynamics, PK pharmacokinetics, TFPI tissue factor pathway inhibitor

Parameter estimates for the PK/PD model along with the bootstrap median and 95% CI obtained from 1000 bootstrapped samples are shown in Table 8. The model parameters were estimated precisely with %RSE values within 20% and maximum %CV ~35%. The observed peak thrombin values with population and individual predictions showed good agreement with the model (Fig. 5A). However, there was a significant residual additive error estimated, which pointed towards the high degree of variability often seen in peak thrombin observations. Residual-based diagnostic plots (Fig. 5B) showed no systematic trends or bias in the model fits, although a higher spread of residuals was observed when compared to the corresponding plots for marstacimab concentrations (Fig. 2B), which also supports the higher variability observed in peak thrombin concentrations. Prediction-corrected VPC of the entire patient population group showed there are no clinically relevant differences in peak thrombin levels for the 300-mg SC once-weekly dose as compared to the levels seen with 150-mg SC once-weekly dose (Fig. 5C), suggesting a lack of excessive pharmacology associated with the higher dose. As such, an optional escalated dosing regimen of 300 mg SC QW is supported by the PK/PD analysis. The PK/PD model characterized the peak thrombin observations for all patients well, including the Phase III study (Fig. 6).

Table 8.

Parameter estimates of the final PK/PD model

Parameter	Population estimate	SE	RSE/CV	Bootstrap estimate	Bootstrap 95% CI
Baseline peak TGA, PBASE (nM)	37.8	4.45	11.8	37.5	11.8–50.2
EC50 (nM)	2.92	0.359	12.3	2.92	1.77–4.19
Maximum peak TGA, PMAX (nM)	91.4	12.2	13.3	91.6	74.8–153
Hill coefficient (fixed)	5	0	0	5	5–5
Additive error peak TGA (nM)	26.3	0.51	1.94	26.2	24.2–28.5
IIV on PBASE	0.104	0.0863	32.2	0.0987	0.0115–0.223
IIV on EC50	0.0283	0.0351	16.8	0.02	0.00165–0.132
IIV on PMAX	0.111	0.0473	33.3	0.103	0.00341–0.183

Bootstrap estimates were obtained from a bootstrap run with N = 1000 samples using the software program Perl-speaks-NONMEM version 5.3.0 bootstrap routine. Residual variability was calculated on peak thrombin via an additive error model on normal scale with a theta-rized (fixed) sigma parameter. % RSE is calculated for the population (theta) parameters, % CV is calculated for IIV parameters

CV coefficient of variation, EC₅₀ half maximal effective concentration, IIV interindividual variance, P_BASE baseline peak thrombin, PD pharmacodynamics, PK pharmacokinetics, P_MAX maximum peak thrombin, RSE relative standard error, TGA thrombin generation assay

Fig. 5.

Assessment of final popPK model adequacy for predicting for peak thrombin concentrations by A goodness-of-fit plots with population and individual predictions for peak thrombin concentrations, B residuals versus population and individual predictions for peak thrombin concentration and C visual predictive checks of peak thrombin concentrations by dosing regimen. A Line of Unity is shown in gray. Loess fit of data is shown in blue. B Loess fit of residuals is shown in blue line. C Prediction-corrected VPC of peak thrombin for all patients divided into 150-mg QW and 300-mg QW dosing regimens. The solid red line shows median of the observed data, the dashed black line shows the same for the simulated data. The pink shaded region shows 90% prediction interval around the simulated median. The dotted black lines show 5th and 95th quantiles of the simulated data and the dashed red lines show the same for the observed data. The shaded blue regions show 90% prediction interval around 5th and 95th percentile of the simulated data. The blue circles show observed data. CWRES conditional weighted residuals, IWRES individual weighted residuals, popPK population pharmacokinetics, QW once-weekly, VPC visual predictive check

Fig. 6.

Visual predictive checks of peak thrombin A for the Phase III study divided by treatment group and B model-predicted free TFPI as the independent variable. Prediction-corrected VPC of peak thrombin for all patients divided into 150-mg QW and 300-mg QW dosing regimens. The solid red line shows median of the observed data, the dashed black line shows the same for the simulated data. The pink shaded region shows 90% prediction interval around the simulated median. The dotted black lines show 5th and 95th quantiles of the simulated data and the dashed red lines show the same for the observed data. The shaded blue regions show 90% prediction interval around 5th and 95th percentile of the simulated data. The blue circles show observed data. QW once-weekly, TFPI tissue factor pathway inhibitor, VPC visual predictive check

Marstacimab Dose in Adult and Adolescent Participants

A comparison of observed weights of adolescent population in the Phase III study with weights of approximately 3000 simulated male subjects is shown in Fig. 7A. Comparison of data from adolescents and adult participants in the Phase III study to model predictions showed a good agreement between observed and predicted marstacimab concentration (Fig. 7B), total TFPI (Fig. 7C) and peak thrombin (Fig. 7D) for adults and adolescents. Although, median peak thrombin levels were slightly lower in adolescents (40–50 nM) compared with adults (60–70 nM), given the good overlap in individual values and the variability seen in peak thrombin, differences were not clinically relevant.

Fig. 7.

Observed and model-predicted PK, TFPI and peak thrombin in adult and adolescent participants with hemophilia following SC QW dosing of marstacimab (150 mg SC QW with 300-mg loading dose or 300-mg SC QW). A Overlay of observed weights of adolescent population in the Phase III study[14] on top of weights of approximately 3000 simulated male subjects. B–D Overlay of observed marstacimab concentration, total TFPI and peak thrombin, respectively for adults (red circles) and adolescents (blue circles) on top of predictions for adolescent population made from the model. The gray bands show 95% prediction intervals in these plots from simulated adolescent population. The dashed blue and red lines show the 5th and 95th percentile of the observed data for marstacimab concentrations, total TFPI and peak thrombin for adolescents and adults, respectively. PK pharmacokinetics, QW once-weekly, SC subcutaneous, TFPI tissue factor pathway inhibitor

Discussion

The present analysis used a semi-mechanistic TMDD model with additional nonlinear clearance of marstacimab and QSSA to characterize the concentration–time profiles of marstacimab and total TFPI. Using pooled data of 213 participants with hemophilia and healthy volunteers from 6 studies, the TMDD model accurately described the marstacimab PK and total TFPI results from the populations examined. Population and individual predictions showed good agreement with marstacimab and total TFPI concentrations, although deviations were observed at higher non-clinically relevant marstacimab concentrations (i.e., concentrations above 50,000 ng/mL). The PK/PD E_max-type model accurately described the relationship between model-predicted free TFPI and peak thrombin.

In the full popPK model, parameter estimates for V_c and V_p together with the volume of distribution suggested limited distribution of marstacimab into peripheral tissues. Estimations of bioavailability, absorption rate, lag time and marstacimab CL were characteristic of SC administered monoclonal antibodies that display limited movement out of the vascular space due to their large protein structures [18, 19].

Clinically relevant covariates of interest were evaluated in the full model, and the differences in medians of the post hoc estimates of CL and V_c between the adolescent patient population and adults were estimated to be −28.8 and −43.2%, respectively. However, the weight-adjusted difference in CL and Vc between adolescents and adults was −3.1 and −25.9%, respectively, showing that weight accounted for most of the observed difference between these age groups. Although weight is an important covariate for PK with trends observed for decreasing concentrations with increasing weight, these trends are not seen with peak thrombin, the PD endpoint. A popPK model together with an E_max PK/PD model demonstrated the feasibility of a fixed (i.e., body weight-independent) QW SC dosing regimen for marstacimab with comparable PD and ABR across body weight ranges with no new safety or efficacy signals identified in the models at either low or high body weights [20]. The absence of a narrow therapeutic window supports the potential for a fixed-dose regimen [20]. Furthermore, the efficacy and safety of marstacimab 150 mg SC QW in adolescents and adults (aged 13–66 years) with hemophilia A or B across a range of body compositions (weight range: 35–120 kg) was demonstrated in the Phase III study [14].

Weight-adjusted differences in CL due to race (Asians vs non-Asians) was estimated to be 31.9% and not considered to be clinically relevant. In the Phase I marstacimab study of Chinese participants with severe hemophilia without inhibitors, marstacimab exposure was similar to those seen in Japanese and White participants from the Phase I and Ib/II studies [9, 11, 12]. Treatment-related trends in PD parameters were also similar to those observed in non-Asian participants with severe hemophilia [11, 12]. Furthermore, the Phase III study comprised 47.7% Asian participants and ABR reductions were consistent with those observed in White participants [14]. Similarly, differences in CL due to hepatic impairment, participant health status, ADA status, renal impairment, and hemophilia type were not found to be significant enough to warrant any dose adjustments (as the 95% CI bootstrap estimates for these covariates includes 0).

Monoclonal antibodies characterized by TMDD typically display a nonlinear PK profile as a consequence of high affinity binding kinetics with target molecules [21]. Marstacimab binds to TFPI with high affinity (K_D = 3.7 nM) [22] and, while lower than affinity of other antibodies in its class [23, 24], is expected to influence serum marstacimab concentrations. Secondary PK parameters calculated from the final model suggest a greater than linear increase in C_avg,ss between the SC QW 150-mg and 300-mg doses, although statistically significant deviation from dose-proportionality was not established in the 150- to 450-mg dose range. The other SC administered anti-TFPI monoclonal antibodies, concizumab (approved for treatment of hemophilia A and B with inhibitors) and befovacimab (discontinued), are also characterized by nonlinear, dose-dependent PKs consistent with TMDD [25]. Although these monoclonal antibodies all target TFPI, their unique binding epitopes result in differences in their binding affinities (concizumab K_D = 0.025 nM; befovacimab K_D < 0.01 nM), which may also contribute to differences in their safety profiles [15, 23, 24].

The median accumulation ratio of marstacimab was estimated to be approximately 4. The peak-to-trough ratio (C_max,ss to C_min,ss) was found to be small at steady state, and the mean effective half-life was approximately 16–18 days, matching with the QW dosing regimen of marstacimab [7, 8, 16]. The limitations of standard of care clotting factor concentrates include their short half-life requiring frequent IV infusions, which often leads to inadequate adherence [2]. The relatively long half-life of marstacimab reduces the need for frequent dosing and may improve treatment adherence. Furthermore, SC administration can improve adherence by reducing treatment burden associated with IV infusions and bypassing venous access [2]. A choice of injection site may also improve tolerability, particularly if some injection site locations are perceived to be less painful than others. While some studies have suggested site selection can influence monoclonal antibody bioavailability [19, 26], the results presented here show minimal effect of site of injection on bioavailability. These results support the current method of administration that allows for injection at various sites [7, 8].

The relationship between model-predicted free TFPI and peak thrombin was adequately described by the E_max model, although the residual error was estimated to be high due to the large degree of variability observed in the peak thrombin observations. Prediction-corrected VPC for PK, total TFPI and peak thrombin show that the popPK and PK/PD models can describe the marstacimab concentrations, total TFPI and peak thrombin for both adults and adolescents. A study is ongoing in pediatric populations and dosing will be further explored.

A comparison between the predictions for adolescent population from the previous model developed using Phase I and Phase II studies [15] and the observed data from the adolescent population in the Phase III study showed good agreement between the model predictions and observed data. The comparison confirmed predictions that even though marstacimab concentrations were higher in adolescents, they did not result in a clinically relevant difference in the key clotting biomarker peak thrombin as there was a high degree of overlap with adults in observed peak thrombin levels [14]. Both the observed data and model predictions showed that the recommended dosing strategy of a 300-mg SC loading dose followed by 150 mg SC QW produced peak thrombin values considered to be clinically higher upon starting marstacimab prophylaxis [7]. The PK/PD simulation results indicated that no changes in dosing regimen between adolescents and adults were warranted and support the same dose selection for adolescents and adults used in the Phase III study [14].

There are few assumptions and limitations of this modeling analysis that should be mentioned. Most prominently, there were no observed data available for free TFPI to compare model estimates of free target with observed values. In addition, a nonlinear binding of marstacimab to the endothelial bound part of TFPI is assumed, which is a simplification of drug binding to TFPI localized into endothelium, plasma and platelets. This assumption was made due to lack of observed data for the different forms of TFPI in the clinical studies. For the PK/PD model, a Hill coefficient was fixed for the E_max model based on model stability and best fit to the peak thrombin data. The approach to exclude BLQ concentrations for estimation was not expected to have a significant impact on the model since most of the BLQ values come from sub-therapeutic doses and from sampling time points taken beyond the dosing interval of 7 days. Furthermore, the inclusion of covariates was driven by choice of clinical significance of the covariate, rather than statistical significance. Time after first dose was used instead of time after last dose to report residuals for the popPK model; this showed there was no significant change in simulated and observed drug concentration as the time progressed.

Conclusions

A TMDD model with additional nonlinear clearance, first-order absorption and QSSA described marstacimab and total TFPI concentrations well. Based on the covariate analyses, participant status, age group, ADA status, race, hemophilia type, renal or hepatic impairment did not have a clinically significant effect on PK of marstacimab. No significant effect of marstacimab injection site selection was observed on its PK. An E_max PK/PD model adequately described the relationship between model-predicted free TFPI and peak thrombin. Pharmacokinetic/pharmacodynamic simulation results indicated that no changes in dosing regimen between adolescents and adults were warranted.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This study was sponsored by Pfizer.

Declarations

Conflict of interest

Satyaprakash Nayak, Akiyuki Suzuki, Patanjali Ravva, and Sangeeta Raje are employees of Pfizer and may own stock or stock options.

Ethics approval

All 6 studies were approved by institutional review boards or ethics committees according to the requirements of the specific countries of origin.

Consent to participate

All participants provided written informed consent prior to initiation of study medications.

Consent for publication

Not applicable.

Availability of data and material

Upon request, and subject to review, Pfizer will provide the data that support the findings of this study. Subject to certain criteria, conditions, and exceptions, Pfizer may also provide access to the related individual de-identified participant data. See https://www.pfizer.com/science/clinical-trials/trial-data-and-results for more information.

Code availability

Not applicable.

Author contributions

All authors participated in data interpretation and critical review and revision of this manuscript and provided approval of the manuscript for submission.