Relationship between Dose, Factor IX Activity Levels and Bleeding Probability for rIX‐FP Prophylaxis in Hemophilia B: A Repeated Time‐to‐Event Analysis

Abstract

In hemophilia B, pharmacokinetic (PK)‐guided dosing of extended half‐life factor IX (EHL‐FIX) concentrates can secure targeted FIX exposure. Target FIX activity levels in plasma should be individually set primarily taking bleeding tendency into account, alongside the presence of target joints, physical activity, and preferred dosing schedules. In other words, both PK and pharmacodynamics (PD) are relevant when individualizing therapy. Our objective was to examine the relationship between dose, FIX activity levels, and bleeding specifically for EHL‐FIX concentrate recombinant fusion protein linking coagulation factor IX with albumin (rIX‐FP). Data from hemophilia B patients with endogenous FIX activity level ≤ 2 IU/dL from five clinical trials were combined. Bleeding probability was described with a parametric repeated time‐to‐event (RTTE) model. Data included 2,493 FIX activity levels and 514 bleeds from 114 unique patients with a median age of 26 years (range: 1–61) followed for a median of 416 days (range: 6–1,233). Joints were the most frequent bleeding site (46%), and more than half of the bleeds were trauma‐related (52%). Overall, 60% and 40% were categorized as damage‐causing or nuisance bleeds, respectively. A baseline hazard of 7.3 bleeds per year was calculated when FIX activity levels were set at zero. The probability of all bleeding decreased by 50% when the FIX activity level was 12 IU/dL. Variability in bleeding hazard between individuals with similar FIX activity levels was substantial (182%). Simulations showed that targeting trough FIX activity levels to 20 IU/dL resulted in a median annual bleeding rate (ABR) of zero (range: 0–3).

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Currently, no publications exist on the relationship between prophylactic FIX exposure and the probability of bleeding for extended half‐life factor IX concentrates in hemophilia B.

• WHAT QUESTION DID THIS STUDY ADDRESS?

In this study, we are the first to quantify the relationship between dose, factor IX (FIX) activity levels, and bleeding for the extended half‐life FIX concentrate rIX‐FP (Idelvion®) in a large population with hemophilia B using a repeated time‐to‐event model.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

A FIX activity level in the central compartment of 12 IU/dL decreased the bleeding hazard by 50% compared with a situation without FIX in the central compartment. Annual bleeding rates (ABR) were simulated when trough FIX activity levels of 1, 3, 5, 10, and 20 IU/dL are maintained. ABR decreased from three and two (1 IU/dL) to zero (20 IU/dL) for damage‐causing and nuisance bleeds.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

This model can be used to individualize treatment based on severe hemophilia B patients' pharmacokinetic and bleeding profiles.

Hemophilia B is an inherited bleeding disorder caused by coagulation factor IX (FIX) deficiency. People with severe and some with moderate hemophilia B are treated prophylactically with FIX concentrates to prevent spontaneous bleeding, and additionally on demand to treat acute bleeding or to prevent bleeding in, for example, perioperative settings. If not adequately treated, repetitive bleeding typically leads to joint and muscle damage. Several FIX concentrates with varying pharmacokinetic (PK) properties exist. ¹ Albutrepenonacog alfa (Idelvion®), a recombinant fusion protein linking recombinant coagulation factor IX with recombinant albumin (rIX‐FP), is one of three extended half‐life (EHL)‐FIX concentrates with an improved FIX terminal half‐life compared with the standard half‐life (SHL) recombinant FIX concentrates. ² , ³ , ⁴ , ⁵ , ⁶ All three are able to significantly lower the frequency of intravenous infusions for patients.

Variability in the PK of FIX concentrates between patients is well‐known, even when they are dosed according to bodyweight. ¹ To overcome this variability in exposure, PK‐guided dosing is recommended. ⁷ , ⁸ , ⁹ , ¹⁰ Constructed population PK models are used to derive individual PK parameters allowing the calculation of an individualized dosing regimen. These dosing regimens typically target factor activity levels above a specified plasma trough factor activity level. Increasingly, plasma factor activity levels ≥ 3 IU/dL are prescribed instead of historical factor activity levels ≥ 1 IU/dL. ¹¹ The relationship between plasma FIX activity levels and their therapeutic effect is, however, complex. Unlike factor VIII (FVIII) in hemophilia A, where plasma FVIII activity levels generally correlate with the hemostatic effect, FIX is able to bind type IV collagen in the extravascular space where it contributes to hemostatic protection but is not measured in the plasma. ¹² , ¹³ , ¹⁴ To integrate these pharmacodynamic (PD) effects, bleeding events as well as other risk factors, such as the presence of target joints and physical activity, should be addressed in order to specify which FIX exposure or peak and trough factor activity levels need to be targeted. ¹⁵ , ¹⁶ , ¹⁷

Although PK‐PD relations are underlined to be important when establishing the treatment plan in hemophilia B patients, little has been published on the exposure‐effect relations of available FIX concentrates. ¹⁸ Interestingly, the exposure‐effect relationship of FVIII concentrates in hemophilia A has recently been investigated, with publications describing FVIII concentrates octocog alfa (rFVIII; Kovaltry®) ¹⁹ and lonoctocog alfa (rFVIII‐SingleChain; Afstyla®), ²⁰ applying a repeated time‐to‐event (RTTE) analysis. In an RTTE analysis, the likelihood of an event occurring over time is described using a hazard function. This specifies the risk of an event, such as a bleeding episode, at any given timepoint. The use of RTTE analysis, however, has to our knowledge not been applied for FIX concentrates. Not only is FIX PK distinctly different from FVIII PK, it is also less well characterized and described as more complex. ²¹ Therefore, our aim was to achieve a deeper understanding of these relations by constructing a PK‐RTTE model for rIX‐FP to evaluate the relationship between prophylactic dose, FIX activity levels, and observed bleeding events.

METHODS

Participants and data

Data from five clinical trials evaluating the safety, efficacy, and PK of rIX‐FP (Idelvion®; CSL Behring GmbH) were used to examine the relationship between rIX‐FP concentrate dose, FIX activity levels, and bleeding in persons with severe hemophilia B (endogenous FIX activity level ≤ 2 IU/dL). FIX activity levels were measured with the one‐stage assay (OSA) as previously described by Santagostino et al. ³ This study excluded all FIX activity levels measured after subcutaneous administration, during on‐demand treatment, bleeding episodes, or medical procedures, as well as all bleeding episodes that occurred around medical procedures or during on‐demand treatment.

The first clinical trial from which data were used was a first‐in‐human phase I study (CSL654_2001), including patients ≥ 12 years of age with severe hemophilia B. ³ Participants received a single intravenous dose of either 25, 50, or 75 IU/kg rIX‐FP. Seven participants received a second dose of rIX‐FP at a median of 2 months (range: 1.8–7.3) after their previous rIX‐FP dose. The second clinical trial was a phase I/II study (CSL654_2004), including patients ≥ 12 years of age with severe hemophilia B who were previously prophylactically treated with FIX replacement therapy. ⁴ Participants received a single dose of 25 IU/kg rIX‐FP for PK analysis. Subsequently, participants were treated with rIX‐FP for approximately 5 months with doses based on their respective PK profiles to evaluate safety and efficacy. The third clinical trial was a phase II/III study (CSL654_3001), including patients ≥ 12 years of age with severe hemophilia B who received previous FIX replacement therapy for at least 150 days. ⁵ In the trial, participants received routine weekly prophylaxis with doses of 35–50 IU/kg for the first 26 weeks, after which participants were allowed to switch to 10 or 14‐day prophylaxis with a dose of 75 IU/kg. The fourth clinical trial was a phase III study (CSL654_3002), including patients < 12 years of age with severe hemophilia B who have received previous factor replacement therapy for at least 50 days (< 6 years of age) or 150 days (6 to < 12 years of age). ²² Data were collected from these participants after receiving a single dose of 50 IU/kg rIX‐FP for PK analysis, after which doses of 35–50 IU/kg rIX‐FP were assigned based on the investigator's opinion. The fifth clinical trial was a phase III extension study (CSL654_3003), including patients who previously completed a CSL654 study or patients with previously untreated severe hemophilia B ≤ 18 years of age. ²³

All trials were open‐label and registered on clinicaltrials.gov under NCT01233440, NCT01361126, NCT01496274, NCT01662531, and NCT02053792, respectively. Bleeding episodes were recorded by participants or their caregiver in an electronic diary. All bleeding episodes were treated with rIX‐FP, except in CSL654_2001, in which the participants' previously used FIX concentrate was administered to treat acute bleeds. Ethics approval, approval by the relevant national authorities, and informed consent for every participant were obtained prior to trial enrolment. Persons (n = 28) who participated in multiple trials had no dosing or bleeding data recorded between subsequent trials. As a result, these occasions were analyzed as separate individuals in the population PK‐RTTE analysis performed in this study.

Population PK model

A population PK model for prophylactic rIX‐FP was constructed in a stepwise manner. During construction, the number of PK compartments was evaluated. PK parameters were expressed by clearance (CL), intercompartmental clearance (Q) and volume (V). Interindividual variability (IIV) of these parameters was estimated. Residual error was described with a combined additive and proportional model. During the covariate analysis, stepwise forward inclusion and backward elimination was applied. ²⁴ Reductions in the objective function value (OFV) of 3.84 (p < 0.05, χ ² ‐distribution, 1 df) and > 6.64 (p < 0.01, χ ² ‐distribution, 1 df), respectively, were required for covariate inclusion. Age, height, bodyweight (WT), lean bodyweight (LBW), fat‐free mass (FFM), body mass index (BMI), country, location of treatment center, and study were available and explored as covariates. We evaluated candidate models by examination of PK parameter estimates, their respective residual standard errors (RSE), OFV, goodness‐of‐fit (GOF) plots, and prediction corrected visual predictive checks (pcVPC).

As several FIX activity levels before the first rIX‐FP dose were greater than 2 IU/dL, a distinction between endogenous FIX activity and exogenous residual FIX activity levels was made. Exogenous residual FIX activity levels were subjected to a residual decay correction procedure as described in earlier work. ²⁵ Endogenous FIX activity was estimated by means of a modified sigmoid function (Eq. 1 ). This modification was done to ensure that the individual baseline FIX activity level was within the range of 0–2 IU/dL, in accordance with the inclusion trial criteria.

$${\mathit{\text{BASE}}}_i=2*\left(\frac{1}{1+\exp \left(-{\eta }_i*1.55\right)}\right)$$ (1)

in which BASE _i represents the estimate of the endogenous baseline FIX activity level of individual i and η _i  ~ N(0,1) represents the individual's estimation of the baseline. η _i is transformed using the sigmoid function, which is used to obtain a uniform prior for BASE _i between 0 and 2 IU/dL. Here, η is scaled by a factor of 1.55 to optimally approach a uniform distribution (further information can be found in Figure S1 ). As no trainable parameters are included in BASE _i , no bootstrap analysis over this estimate could be performed.

Repeated time‐to‐event model

The likelihood of a bleed was evaluated with an RTTE model which describes the occurrence of repeated events, such as bleeds over time simultaneously with possible predictors, such as FIX activity levels, using parametric survival analysis. ¹⁹ , ²⁶ For this analysis, all patient‐reported bleeds in the described clinical studies, except bleeds associated with medical procedures, were used. Subsequently, bleeds were categorized as either damage‐causing or nuisance bleeds. Damage‐causing bleeds were defined as all bleeds not categorized as nuisance bleeds. Nuisance bleeds were defined as bruising, mucosal bleeds, and bleeds of an unknown cause. We concluded that if the location of a bleed is unknown, it is likely not severe enough to be considered damage‐causing.

An RTTE model uses a hazard function to describe the spontaneous rate at which a bleed can occur. We evaluated exponential, Gompertz, Weibull, and Log‐Logistic functions ²⁷ to describe the time to bleeding distribution. IIV on the overall bleeding hazard was explored. At the end of the individual observation period, right censoring was applied. The baseline hazard is presented by the scale (λ; year⁻¹), which describes the bleeding hazard over time with no FIX activity level present. The relationship between FIX activity levels in the central compartment and the bleeding hazard was evaluated with linear, exponential, maximum inhibitory (I _max) and sigmoidal I _max PD models. The relationship between FIX activity levels and bleeding hazard in the peripheral compartment was evaluated as well.

The potency of the FIX activity levels in the central compartment to lower the bleeding hazard is described by the IC₅₀ value (IU/dL). Furthermore, it was assessed if successive bleeds are dependent on time since the previous bleed using a Markov dependence hazard function. In this function, time was described as the time since the previous bleed. ¹⁹ Covariates that were tested for a significant impact on the bleeding hazard are age, body weight, pre‐study ABR, and location of treatment center.

Model development and evaluation

Models were developed using NONMEM software (v7.4.1, Icon Development Solutions, Gaithersburg, MD, USA) ²⁸ and evaluated based on estimated parameter precision, scientific plausibility, and the OFV. First‐order conditional estimation with interaction method (FOCE‐I) was used to estimate PK parameters. For nested models, a change in OFV (ΔOFV) of −6.64 (p < 0.01, χ ² ‐distribution, 1 df) was considered significant. The Akaike information criterion (AIC) was used to compare and select between non‐nested models. Robustness of the PK parameter estimates was assessed by bootstrap analysis. The adequacy of the RTTE model to describe the data was examined by performing Kernel‐based visual hazard comparison and by observed and model‐predicted comparison of Kaplan–Meier curves. ²⁹ Data visualization and evaluation were performed in R (version 4.3.1), Pirana (version 2.9.9) and PsN (version 4.8.1). ³⁰ , ³¹

Simulations

Simulations were performed with weekly dosing while targeting trough FIX activity levels of 1, 3, 5, 10, and 20 IU/dL to evaluate the relationship between steady‐state trough FIX activity level and annual bleeding hazard. A virtual set of 1,000 patients with a body weight of 70 kg and typical PK parameters resulting from the developed population PK model was used. As doses of clotting factors are generally adjusted to obtain a specific trough activity level, only IIV in bleeding hazard was simulated. The analysis was performed separately to evaluate damage‐causing and nuisance bleeds specifically.

RESULTS

Participants and data

Data from the described five clinical studies were combined. These studies included 25, 17, 63, 27, and 18 people with severe hemophilia B who were treated with intravenous rIX‐FP (Tables 1 and S1 ). This resulted in 150 study participants, of whom 114 were unique. As seven patients received a second dose of rIX‐FP in CSL_2001, the PK analysis was performed with 157 individuals. At study inclusion, median age was 26 years (range: 1–61) and median body weight was 64 kg (range: 11–130). For the development of the population PK model, 2,493 FIX activity levels were available. Three samples (0.1%) had FIX activity levels below the limit of quantification and were excluded in the analysis. A total of 514 bleeds were modeled during a median follow‐up period of 416 days (range: 6–1,233). Joint bleeds were the most frequently reported (238/514; 46%) and trauma‐related bleeding was the most frequently reported cause (268/514; 52%). Of the bleeds, 60% and 40% (306/514 and 208/514) were categorized as damage‐causing and nuisance bleeds, respectively.

Table 1.

Baseline patient characteristics

	Number (n, median)	% or range
Patient characteristics
Participants	114
Follow‐up time (days)	416	(6–1,233)
Age (years)	26	(1–61)
Weight (kg)	64.0	(11–130)
Height (cm)	172.0	(78–190)
BMI (kg/m2)	21.6	(16–35)
Location of treatment center
Africa	2	(2)
Asia	10	(9)
Europe	72	(63)
North America	7	(6)
Middle East	18	(16)
Oceania	5	(4)
Pharmacokinetic data
rIX‐FP dose per week (IU/kg)	38.0	(6.5–71.4)
FIX activity levels per patient	15	(2–60)
Bleeding data
Total bleeds	514
Patients without bleeding	31	(27)
No. of bleeds per patient	3	(0–33)
Pre‐study ABR	3	(0–50)
During‐study ABR	3	(0–52)
Location
Joint	238	(46)
Muscle	26	(5)
Other	156	(31)
Unknown	94	(18)
Cause
Spontaneous	198	(39)
Trauma	268	(52)
Unknown	48	(9)
Severity
Damage‐causing	306	(60)
Nuisance	208	(40)

ABR, annual bleeding rate; cm, centimeter; FIX, factor IX; IU, international units; kg, kilogram; rIX‐FP, recombinant fusion protein linking coagulation factor IX with albumin.

Population PK model

A two‐compartment model with a central and a peripheral compartment adequately described our data (Table 1 ). The addition of a second peripheral compartment improved the fit of the model to the data; however, the resulting parameter estimates of both peripheral compartments were imprecise. Therefore, a second peripheral compartment was not included in the final model. Age, height, WT, LBW, FFM, BMI, country, location of treatment center, and study were explored as covariates. In the univariate analysis, the separately weight‐related covariates WT, FFM, and LBW were all significantly related to CL, V1, and V2. We chose, however, to scale these parameters to WT as it is used in the clinical setting. We had a dataset with a relatively large range of body weight, which allowed for direct estimation of allometric exponents. Q was, however, not normalized for WT, as it was not found to improve the model (ΔOFV = 0.010). Age, country, location of treatment center, and study were not found to be of significance. The location of the treatment center and study were tested on the residual errors as well; however, no significant impact was found. Furthermore, interindividual variability (IIV) was significant on CL, V1, and V2 (p < 0.01), along with correlations between IIV on CL and V1 and between IIV on V1 and V2. The final PK parameters of the population PK models are presented in Table 2 . The pcVPC shows adequate model performance for indicated body weight categories of < 60 kg, 60–80 kg, and > 80 kg, respectively, although a slight divergence was seen between the observations and the 90% confidence interval of the 5th percentile (Figure S2 ). The developed population PK model was used to estimate the individual FIX activity levels over time. Empirical Bayesian estimates of the PK parameters were used to calculate the individual FIX activity level at the time of bleeding. The median FIX activity level at which any damage‐causing or nuisance bleeding occurred was 21.4 IU/dL (IQR: 13.6–34.7), 23.2 IU/dL (IQR: 13.7–37.5) and 20.4 IU/dL (IQR: 13.5–30.6), respectively.

Table 2.

Final estimates of the altered population pharmacokinetic model for rIX‐FP

Parameters	Estimate (RSE %) (Shr.) (RSE %) (Shr.)	Bootstrap estimate (95% CI*)
CL (dL/h)	0.509 (2)	0.510 (0.49–0.53)
WT exponent on CL	0.588 (5)	0.589 (0.52–0.67)
V1 (dL)	60.0 (3)	59.7 (57–63)
WT exponent on V1 (and V2)	0.780 (4)	0.778 (0.71–0.85)
Q (dL/h)	0.354 (19)	0.350 (0.26–0.52)
V2 (dL)	33.3 (12)	32.3 (25–43)
WT exponent on V2 (and V1)	0.780 (4)	0.778 (0.71–0.85)
IIVa on CL (%)	18.6 (9) [12]	18.7 (15–22)
IIV on V1 (%)	19.6 (14) [14]	19.1 (14–26)
Correlationb IIV CL and V1 (%)	61 (28)	65 (35–85)
IIV on V2 (%)	130.4 (13) [20]	123.8 (71–185)
Correlation IIV V1 and V2 (%)	51 (26)	47 (21–65)
Proportional error (%)	20.6 (13)	20.1 (15–26)
Additive error (IU/dL)	2.00 (34)	1.93 (1–3)

CL, V1, and V2 were normalized for a typical patient with a body weight of 70 kg. CL, clearance; V1, central volume of distribution; Q, intercompartmental clearance; V2, peripheral volume of distribution; WT, body weight; BL, baseline FIX activity level; IIV, interindividual variability; RSE, residual standard error; Shr., shrinkage; CI, confidence interval. The error structure for factor IX activity levels is described by Y = IPRED + (IPRED*EPS(Proportional) + EPS(Additional)).

^a IIV coefficient of variation calculated as: $\sqrt{e^{{\omega }^2}-1}*100\%$.

^b Correlation calculated as: $\frac{\text{covariance}}{\sqrt{{\text{variance}}_1}*\sqrt{{\text{variance}}_2}}*100\%$.

^*95% CI, nonparametric 95% confidence interval from bootstrap results (n = 1,000).

$\mathrm{CL}={\theta }_{\mathrm{CL}}*{\left(\frac{\mathrm{WT}}{70}\right)}^{0.588}*e^{{\eta }_{\mathrm{CL}}}$.

$\mathrm{V}1={\theta }_{\mathrm{V}1}*{\left(\frac{\mathrm{WT}}{70}\right)}^{0.780}*e^{{\eta }_{\mathrm{V}1}}$.

$Q={\theta }_Q$.

$\mathrm{V}2={\theta }_{\mathrm{V}2}*{\left(\frac{\mathrm{WT}}{70}\right)}^{0.780}*e^{{\eta }_{\mathrm{V}2}}$.

Repeated time‐to‐event model

An exponential hazard function described the time to bleeding data best, which indicated a constant bleeding hazard over time. The Weibull and log‐logistic functions were not selected, as both faced minimization issues indicating the data did not contain sufficient information to support reliable estimation of all parameters. A Gompertz hazard function had a slightly lower OFV than the exponential model (ΔOFV = −5.23). The shape parameter, which described the declining hazard over time, however, was estimated with low precision (RSE = 44%) and was close to zero. Given that the Gompertz hazard function is essentially an exponential hazard function with an added shape parameter, ³² the exponential hazard function was selected as the final distribution.

The effect of the FIX activity level in the central compartment on the bleeding hazard was statistically significant (ΔOFV = −99.6; p < 0.01) and was described with an Imax model, which shows a decreased bleeding hazard at higher FIX activity levels. Relating hazard to FIX activity levels in the peripheral compartment did not improve the fit (ΔOFV = 57.345). Successive bleeds were not found to be dependent, as the addition of a Markov element in the bleeding hazard did not improve the model. The individual bleeding hazard was best described by Eq. 2 :

$$h_i\left(t\right)=\lambda *\left(1-\frac{\mathrm{FIX}{\text{activity level}}_i\left(t\right)}{\mathrm{FIX}{\text{activity level}}_i\left(t\right)+{\mathrm{IC}}_{50}}\right)*e^{{\eta }_i}$$ (2)

in which h _i (t) describes the individual bleeding hazard at time t, λ the bleeding hazard in absence of a FIX activity level, FIX activity level the individual FIX activity level in the central compartment at time t, IC₅₀ the FIX activity level at which 50% of the maximum inhibition of the bleeding hazard occurs, and η the random effect describing the IIV in the bleeding hazard.

The parameters of the RTTE model are summarized in Table 3 . The scale (λ) was 7.34 years⁻¹ and the IC₅₀ value was 12 IU/dL, which is indicative of a reduced bleeding hazard of 50% compared with a situation with a FIX activity level of zero. A visual interpretation of the RTTE model estimates can be found in Figure 1 . In case a virtual constant FIX activity level of 1 or 3 IU/dL would be present over a year, 6.5 or 5.7 bleeds are predicted, respectively. In turn, this would mean that an annual bleeding rate (ABR) of 6.5 (95% C.I. 3.9–9.1) and 5.7 (95% C.I. 3.6–7.7) is to be expected for these respective FIX activity levels. Figure 2 presents the time profiles of FIX activity levels, bleeding hazard, and the probability of not having a bleed for two illustrative study participants. The IIV in bleeding hazard was high, with a coefficient of variation (CV) of 182%. The parameters of the RTTE models for damage‐causing and nuisance bleeds are presented in Table 3 .

Table 3.

Final estimates of the repeated time‐to‐event (RTTE) model

Parameters	Estimate (RSE %) (Shr.) (RSE %) (Shr.)
All bleeds
Scale, λ (year−1)	7.34 (24)
IC50 (IU/dL)	12.0 (30)
IIV Hazard (CV%)	182 (10) [20]
Damage‐causing bleeds
Scale, λ (year−1)	3.33 (28)
IC50 (IU/dL)	12.2 (36)
IIV Hazard (CV%)	275 (11) [23]
Nuisance bleeds
Scale, λ (year−1)	2.36 (31)
IC50 (IU/dL)	12.3 (41)
IIV Hazard (CV%)	175 (13) [31]

CV, coefficient of variation; IU, international units; RSE, residual standard error; Shr., shrinkage.

Figure 1.

Bleeding hazard versus FIX activity level on the basis of the final repeated time‐to‐event (RTTE) model. The solid red line shows the median relationship between bleeding hazard (year⁻¹) and constant FIX activity levels (IU/dL) based on the final model parameters. The shaded area shows the 95% confidence interval calculated on the basis of the relative standard errors of the parameter estimates. The relationship describes a virtual scenario in which a hemophilia B patient would have a constant FIX activity level over the course of a year.

Figure 2.

Time profiles of factor IX (FIX) activity levels, bleeding hazard, and the survival probability for two illustrative patients. Patient A (66 kg, 53 years old, treated with 15 IU/kg weekly after initial PK analysis) experienced 5.1 bleeds per year during the study. Patient B (72 kg, 20 years old, treated with 17 IU/kg weekly after initial PK analysis) experienced 1.2 bleeds during the study. The top panels describe the individually predicted FIX activity levels based on the PK observations (red dots). The panels in the center show the predicted bleeding hazard according to the model, taking bleeding events into account (red crosses). These panels demonstrate the inverse correlation between bleeding hazard and FIX activity levels. The bottom panels show the probability of not experiencing a bleed (survival probability), which decreases consecutively when the bleeding hazard is higher and returns to 100% after a bleed occurs.

The fit of the RTTE model to the data is visualized in Figures 3 and S3 , utilizing Kaplan–Meier curves and the Kernel‐based visual hazard comparison, ²⁹ respectively. The observed and simulated Kaplan–Meier curves overlap for the six bleeding events visualized, indicating an adequate fit of the RTTE model to describe bleeding probability. Covariate effects on the bleeding hazard were tested, but no significant covariate was found.

Figure 3.

Observed and simulated Kaplan–Meier plots for the first six bleeds during study duration. The solid thick line in each panel demonstrates the observed percentage of individuals not experiencing a bleed, n, while the shaded area demonstrates the simulated percentage of subjects that do not experience a bleed, n, using the presented repeated time‐to‐event (RTTE) model. Ninety patients had at least one bleed; 75 patients had at least two bleeds; 60 patients had at least three bleeds; 50 patients had at least 4 bleeds; 44 patients had at least five bleeds; and 37 patients had at least six bleeds. Vertical lines in the solid line indicate censoring. The observed and simulated Kaplan–Meier curves overlap, indicating that the model describes the bleeding data adequately.

Simulations

In the virtual population, administration of weekly prophylactic 70, 560, 1,120, 2,590, 5,530 IU rIX‐FP resulted in steady‐state trough FIX activity levels of 1, 3, 5, 10, and 20 IU/dL, respectively. In Figure 4 , the simulated ABR when maintaining these trough FIX activity levels is displayed, with differentiation into damage‐causing and nuisance bleeds. This figure illustrates that a median person encounters a median of three (range: 0–18) damage‐causing and two (range: 0–10) nuisance bleeds per year at 1 IU/dL, two damage‐causing (range: 0–14) and one (range: 0–7) nuisance bleed(s) at 3 IU/dL, one (range: 0–10 and range: 0–7) damage‐causing and one (range: 0–6 and range: 0–3) nuisance bleed at both 5 and 10 IU/dL, respectively, and no (range: 0–4 and range: 0–2, respectively) damage‐causing or nuisance bleed at 20 IU/dL. Noteworthy, nuisance bleeds appear to occur less frequently than damage‐causing bleeds. This is probably caused by the higher scale factor for damage‐causing bleeds (Table 3 ). Interestingly, the median simulated ABR drops to zero with trough FIX activity levels between 10 and 20 IU/dL, indicating that few to no bleeds occur when these trough FIX activity levels are targeted.

Figure 4.

Simulation of annualized bleeding rates (ABR) when varying trough factor IX (FIX) activity levels are maintained during a prophylactic regime. Virtual patients (n = 1,000) were assumed to have typical PK parameters. Administration of weekly prophylactic doses of 70, 560, 1,120, 2,590, and 5,530 IU resulting in trough FIX activity levels of 1, 3, 5, 10, and 20 IU/dL, respectively, are visualized. Whiskers of the boxplot range represent the 80% prediction interval. For both damage‐causing (red) and nuisance bleeds (blue), the percentage of persons with zero bleeds (top row) and median ABR (bottom rows) are displayed. The percentage of persons with zero bleeds increases in the presence of higher targeted trough FIX activity levels. Furthermore, lower FIX activity levels suffice for nuisance bleeds as opposed to damage‐causing bleeds. Importantly, these simulations evaluate the effect of fluctuating FIX activity levels during prophylaxis with this intravenously administered specific FIX concentrate, which differs from Figure 1 , which presents the results when FIX activity levels are constant.

DISCUSSION

In this study, an RTTE model was developed associating rIX‐FP (Idelvion®) dose, FIX activity levels, and bleeding in patients with hemophilia B receiving rIX‐FP prophylaxis. By doing so, an exposure–effect and therefore PK–PD relationship for hemophilia B patients receiving rIX‐FP prophylaxis was described. To our knowledge, no RTTE model is available yet for clinical use in hemophilia B. This is especially relevant in hemophilia B as FIX is known to bind to type IV collagen in the extravascular space and still contribute to hemostasis and therefore therapeutic effect outside of the plasma compartment. An exponential hazard function described the data best, meaning that a constant baseline bleeding hazard was present. A constant FIX activity level of 12 IU/dL decreased the bleeding hazard by 50%. With increasing trough FIX activity levels, ABR decreased from 3 and 2 (trough FIX activity level 1 IU/dL) to zero (trough FIX activity level 20 IU/dL) for both damage‐causing and nuisance bleeds, respectively. Our simulations suggest that damage‐causing bleeds are harder to prevent. An ABR of zero is obtained with trough FIX activity levels between 10 and 20 IU/dL, which are higher compared with the advocated trough FIX activity levels ≥ 3 IU/dL. ¹¹

Population PK model

To our knowledge, one population PK model has been published describing the PK of rIX‐FP. ³³ Interestingly, in this model, the volume of the central compartment augments with increasing doses. This would mean that as rIX‐FP doses increase, the rise in FIX concentration in the central compartment becomes progressively smaller, implying a nonlinear trend. A hypothesis for this phenomenon may be the lungs. When intravenously administered, a drug will first pass through the lungs before reaching the systemic circulation. In a recent study, an EHL‐FIX concentrate coupled to IgG (rFIX‐Fc; Alprolix®) was demonstrated to pool in the lungs. ³⁴ Both rFIX‐Fc and rIX‐FP are internalized by the neonatal Fc receptor (FcRn) and FIX is able to bind collagen type IV, ¹² which is expressed in the lungs. ³⁵ , ³⁶ This may also (in part) explain the lower increase of FIX activity as measured in the systemic circulation or central compartment after increasing intravenously administered doses. Dose dependency on the volume of the central compartment of rFIX‐Fc, on the contrary, has never been demonstrated. Nevertheless, there is no sound pharmacological explanation for this phenomenon. In our view, this relation between the volume of the central compartment and bodyweight‐adjusted dose may also be due to treatment selection bias. In trial CSL654_3001, ⁵ the treating physician was allowed to select a pre‐specified rIX‐FP starting dose. This lack of randomization in the starting dose allocation makes the model susceptible to treatment selection bias. ³⁷ , ³⁸ In essence, patients with a high bleeding risk due to relatively lower FIX exposure as a result of a higher volume of distribution may be dosed with a higher pre‐specified rIX‐FP starting dose. Therefore, in our study, a newly developed population PK model was constructed which adequately described these clinical trial data without the inclusion of a dose‐dependent parameter. Differences between the published model and the newly developed model are primarily observed in the parameter estimates of the peripheral compartment, with 15.8 (95% CI: 12.1–19.5) versus 33.3 (95% CI: 25.8–40.8) dL for the published and newly developed population PK model, respectively.

Repeated time‐to‐event model

In our analysis, no decreasing baseline hazard over time in the RTTE model was found. This suggests that the bleeding hazard at the start of the study and after 1 year of follow‐up is equivalent. Moreover, the IIV in bleeding hazard was high with a coefficient of variation (CV) of 182%. This demonstrates that patients with similar FIX activity levels may present with a variable number of bleeds. Specifically, it is important to recognize that this discrepancy may be due to less reliable reporting of the bleeds with which the model is built. These bleeds are noted as self‐reported bleeds; therefore, reporting bias may have influenced the bleeding observations used in the RTTE analysis. Such a reporting bias may also explain the lack of a time‐dependent correlation between successive bleeds. More specifically, we realize that multiple smaller bleeds may have been reported as one prolonged bleed. Still, no time‐dependency was observed in RTTE analyses in hemophilia A as well, ¹⁹ , ²⁰ which coincides with our results.

In hemophilia B, there is increasing evidence that the extravascular compartment is of importance for the hemostatic efficacy. ¹³ , ³⁹ , ⁴⁰ Although the FIX activity level in the central compartment is a crucial biomarker, it may not be entirely indicative of the hemostatic effect. ⁴¹ In turn, FIX activity levels in the central compartment are not readily comparable between different EHL‐FIX therapies. Since the FIX activity level is measured in the central compartment and the rate at which drug transfer occurs between the central and extravascular compartments is to a large extent unknown, this could indicate that the FIX activity level in the central compartment may not be the sole predictor of the hemostatic efficacy. Furthermore, our model predicted a higher hazard in patients on an on‐demand treatment regimen compared with patients on prophylaxis. Notable differences were found when patients solely on an on‐demand regimen were modeled using the RTTE model, in which the scale (λ) was 22.6 years⁻¹ and the IC₅₀ value was 4 IU/dL (data not shown). We speculate that prophylaxis, which leads to a more continuous and prolonged exposure to the drug, is more likely to load the extravascular compartment and in turn could decrease the bleeding hazard for patients. Interestingly, this concept of local FIX sequestering has been suggested in earlier reports. ¹³ , ⁴² Together, this suggests that the extravascular compartment should be accounted for to predict the bleeding risk more precisely in hemophilia B in future studies. In the clinical setting, however, it is not possible to measure FIX activity outside of the plasma compartment. To obtain a better understanding of the relationships between plasma FIX activity and develop a full picture of the hemostatic response in hemophilia B, physiology‐based PK modeling may be used. ³⁴ PBPK models allow estimating FIX concentrations in other compartments and, subsequently, investigate the relationship with the observed response. Continued efforts are made by our group to realize this model.

Comparison to other studies

To our knowledge this is the first study in which an RTTE analysis is applied in hemophilia B. Earlier, this approach was applied in hemophilia A. Abrantes et al. and Bukkems et al. have examined the relation between FVIII activity levels and bleeding for FVIII concentrates BAY 81–8,973 and rFVIII‐SingleChain, respectively. ¹⁹ , ²⁰ These studies reported bleeding hazards of 3.0 and 6.4 years⁻¹ with IC₅₀ values of 8.15 and 8.90 IU/dL, respectively. The observed bleeding hazard of 7.3 years⁻¹ and the IC₅₀ value of 12 IU/dL in the present study can be considered more or less comparable. Comparison of PD parameter values between FVIII and FIX as mentioned above, however, is complicated by the effect of extravascular concentrations of the latter.

Strengths and limitations

We are the first to apply RTTE modeling for an EHL‐FIX concentrate (rIX‐FP) administered in hemophilia B patients, which is a powerful method to describe events that vary in time, such as bleeding events. Further strengths are the rich and reliable clinical trial data, supported by the large patient cohorts, prolonged follow‐up times, and substantial amount of PK data.

Limitations of this study include the self‐reporting of bleeds by study participants. This may have led to over‐ or underreporting of the actual number of bleeds.

CONCLUSION

In conclusion, the developed population PK‐RTTE model describes the relationship between a prophylactic dosing regimen with rIX‐FP concentrate, FIX activity levels, and the probability of bleeding in patients with hemophilia B. The simulations provide insights into which trough FIX activity levels suffice to prevent bleeding in clinical practice. Still, characterization of bleeding remains a challenge, with varying bleeding rates at similar baseline activity levels in individual patients. ⁴³ This underlines clinical observations that individuals show significantly different bleeding (tendencies) when similar FIX activity levels are targeted. Nevertheless, tailoring of the dose on both the individual PK and individual bleeding risk is of additional value in hemophilia A, ⁴⁴ and hence highly likely to be of similar additional value in hemophilia B. In future studies, other EHL‐FIX concentrates with different extravascular distribution characteristics should be evaluated, as these may have different exposure‐effect relations. To conclude, the developed RTTE model for prophylactic treatment with rIX‐FP could be used to aid in dose tailoring for an individual to reach sufficient FIX activity levels. External validation of the RTTE model with independent data, however, is recommended first.

FUNDING

This study was funded by the SYMPHONY NWO‐NWA consortium.

CONFLICTS OF INTEREST

M.H.C. has received investigator‐initiated research grants over the years from the Netherlands Organization for Scientific Research (NWO), the Netherlands Organization for Health Research and Development (ZonMw), the Dutch “Innovatiefonds Zorgverzekeraars,” Baxter/Baxalta/Shire, Pfizer, Bayer Schering Pharma, CSL Behring, Sobi, Biogen, Novo Nordisk, Novartis, and Nordic Pharma, and has served as a steering board member for Roche and Bayer. All grants, awards, and fees go to the Erasmus MC as institution. R.A.A.M. has received grants from governmental and societal research institutes, such as NWO, ZonMW, Dutch Kidney Foundation, and Innovation Fund, and unrestricted investigator research grants from Baxter/Baxalta/Shire/Takeda, Bayer, CSL Behring, Octapharma, and Sobi. He has served as an advisor for Bayer, CSL Behring, Merck Sharp & Dohme, and Baxter/Baxalta/Shire/Takeda. All grants and fees are paid to the Amsterdam UMC as institution. All other authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

S.F.K. wrote the manuscript. S.F.K. analyzed the data. All authors, S.F.K., M.H.C., and R.A.A.M. designed and performed the research.

Supporting information

Data S1.