Bridging Pharmacokinetics and Pharmacodynamics: A PBPK/PD Model‐Based Approach for Deferasirox Dosing in Transfusion‐Dependent Thalassemia

ABSTRACT

Patients with transfusion‐dependent thalassemia (TDT) require lifelong blood transfusions, resulting in excessive iron accumulation and necessitating effective chelation therapy. Deferasirox (DFX) is the primary oral iron chelator for managing iron overload; however, the response to this treatment varies substantially within different individuals, potentially because of differences in its pharmacokinetics (PK) and pharmacodynamics (PD). This study aimed to develop a physiologically based pharmacokinetic–pharmacodynamic (PBPK/PD) DFX model, integrating hepatic‐ and transfusion‐derived iron burdens to assess their impact on DFX PK and optimize dosing. The model was developed using clinical PK data from Caucasian and Thai populations, comprising healthy individuals and patients with TDT. TDT‐specific physiological parameters were incorporated into the TDT model. The verified model was applied to predict the targeted DFX dose required to achieve a 25% reduction in the liver iron concentration (LIC) from baseline after 6 months of treatment based on the baseline LIC and blood transfusion regimen. The model demonstrated high predictive accuracy across populations, identifying the effects of iron levels on DFX clearance. Simulations revealed that patients with higher baseline LIC were more likely to achieve the targeted reduction, whereas those with lower LIC required higher doses because of slower iron mobilization. A reduced blood transfusion regimen was associated with improved therapeutic outcomes at the same DFX dose. The PBPK/PD model proposed targeted DFX doses to achieve a 25% reduction based on baseline LIC levels and transfusion regimen, emphasizing the requirement for individualized dosing strategies based on iron burden and blood transfusion patterns to maximize clinical outcomes.

Study Highlights.

• What is the current knowledge on the topic?

  • ○
    Deferasirox (DFX), the primary oral iron chelator for transfusion‐dependent thalassemia (TDT), showed variability in patient responses, likely because of differences in pharmacokinetics and pharmacodynamics. The effects of the iron burden, particularly the liver iron concentration (LIC) and blood transfusion‐derived iron, on DFX clearance and efficacy remain unclear.
• What question did this study address?

  • ○
    This study investigated how hepatic and transfusion‐derived iron burden, particularly baseline LIC and blood transfusion regimens, affects DFX pharmacokinetics and therapeutic response using a PBPK/PD modeling approach to support optimized dosing strategies.
• What does this study add to our knowledge?

  • ○
    This study is the first DFX PBPK/PD model that incorporates hepatic and transfusion‐derived iron burden in patients with TDT. The model revealed that body iron levels can increase DFX clearance and enable individualized dose predictions based on baseline LIC and transfusion regimen. Furthermore, a higher baseline LIC enhances the treatment response at standard doses, whereas a lower LIC or higher transfusion burden may require dose adjustments.
• How might this change clinical pharmacology or translational science?

  • ○
    The findings support a personalized approach to DFX dosing, emphasizing the necessity of tailoring treatment according to the baseline LIC and transfusion regimen. This approach could improve treatment efficacy, reduce iron overload complications, and minimize unnecessary dose escalations, contributing to personalized medicine.

1. Introduction

Thalassemia, a group of inherited autosomal recessive hemoglobin disorders, is highly prevalent from sub‐Saharan Africa through Southeast Asia and is increasingly observed in Western regions due to rising migration [1, 2]. In Thailand, approximately 1% of the population is affected [3]. Thalassemia syndromes that are more severe and require regular blood transfusions are known as transfusion‐dependent thalassemia (TDT) [1]. TDT patients are at heightened risk of complications due to secondary iron overload from chronic blood transfusions, leading to severe morbidity and mortality if not effectively managed [1].

Deferasirox (DFX) is an oral iron chelator categorized as a Biopharmaceutics Classification System Class II drug, with high permeability and low solubility [4]. DFX exhibits high oral bioavailability (~70%), low volume of distribution (0.18 L/kg), and binds extensively to albumin [5, 6]. The primary metabolism of DFX is glucuronidation, mainly through uridine diphosphate glucuronosyltransferase (UGT) 1A1 and 1A3; however, its precise metabolized fractions are unclear [7, 8]. A minor proportion (~8%) is oxidized via cytochrome P450 (CYP) enzymes, primarily CYP1A and CYP2D6 [8]. The primary route of DFX excretion is feces, with an unclear extent of unchanged drug via biliary excretion, and minimal renal clearance [8]. Notably, DFX exhibits a secondary peak following the maximum concentration (C _max) owing to enterohepatic recirculation (EHR), which is well‐documented in animal models; however, its quantification remains uncertain in humans [5, 8, 9].

Despite DFX's extensive use, treatment responses have varied, with treatment failure of 22% to 70% [10, 11, 12]. In Thailand, 44% of TDT patients exhibit inadequate response [13], increasing the risk of iron overload complications. One key factor influencing DFX effectiveness is its pharmacokinetics (PK), particularly the trough concentration (C _trough) and area under the concentration–time curve from 0 to 24 h (AUC_0–24h) [12, 14]. Although direct evidence is limited, baseline liver iron concentration (LIC), a key biomarker of body iron, may influence DFX PK [11]. Patients with higher baseline LIC exhibited lower DFX AUC and C _trough but greater LIC reductions after 12 to 24 months of treatment [15, 16, 17]. Thus, baseline LIC may affect DFX clearance, as indicated by variation in PK parameters, resulting in varied responses and necessitating dose adjustment.

Physiologically based pharmacokinetic (PBPK) modeling provides a robust alternative for evaluating the influence of body iron burden on DFX PK. By integrating physiological and drug‐specific parameters along with intrinsic and extrinsic factors, PBPK models enable hypothesis testing, dose optimization, and enhance personalized treatment strategies [18].

Thus, this study employed a PBPK/PD modeling approach to explore the relationship between body iron burden and DFX PK, particularly focusing on the iron‐mediated modulation of drug clearance. The pharmacodynamic (PD) model incorporated body iron based on DFX's mechanism. By leveraging PBPK/PD simulations, we aimed to elucidate the underlying iron‐dependent alterations in DFX disposition and to predict targeted doses tailored for different iron burdens.

2. Materials and Methods

2.1. Modeling and Simulation Software

The population‐based PBPK simulator (Simcyp Version 22; Certara UK Ltd., Sheffield, UK) was used to model and simulate DFX PBPK/PD. COPASI software (version 4.44) was used to estimate the PD parameters [19]. The plasma concentration–time profiles were digitized using WebPlotDigitizer (https://automeris.io). PK profiles and PD plots were generated using R (version 4.4.2; R Foundation for Statistical Computing, Vienna, Austria).

2.2. Clinical PK Data for Modeling and Simulation

Clinical observational data regarding DFX—comprising PK profiles, PK parameters, and PD data—were collected from published studies involving healthy and TDT Caucasian populations. Model training and verification included eight datasets from six studies on healthy Caucasians and nine datasets from four studies in TDT Caucasian patients. One PK dataset of healthy Thai volunteers was used to verify the healthy Thai population model (Government Pharmaceutical Organization, Thailand; unpublished data). TDT Thai population data included one PK dataset, C _trough at steady state, and two PD datasets (Siriraj Thalassemia Center (STC), Mahidol University; unpublished data). All data usage was approved by the Siriraj Institutional Review Board (SIRB protocol no. 237/2565) and was conducted in accordance with ethical standards. Table S1 presents a summary of the studies that were used in the PBPK/PD model.

2.3. TDT Population Models

The TDT Caucasian population model was adapted from the “Sim‐Healthy Volunteer” population in the Simcyp library, with adjusted physiological, demographic, and biochemical parameters. Age, weight, and height were modified based on correlations derived from reported means and standard deviations using bootstrapping (n = 800). Because the demographic data in the literature were not stratified by sex, the TDT Caucasian population model assumed a uniform distribution for both sexes. Biochemical parameters, including serum creatinine and hematocrit, were adjusted using the STC data (Table S2). Parameters without available data remained the same as those in the “Sim‐Healthy Volunteer” population.

The TDT Thai population model was adapted from the Healthy Thai population model from a previous study [20]. Demographic data were modified using patient information from the STC, whereas physiological and biochemical parameters were followed by the TDT Caucasian population model (Table S2).

2.4. Model Development, Verification, and Application

A stepwise middle–out approach was employed to establish the DFX PBPK/PD model. The simulation for each scenario involved 10 trials that closely matched the observed clinical characteristics, including the number of subjects, age range, proportion of females, and scheduled meals. Figure 1 presents the overall workflow for developing, verifying, and applying the PBPK/PD model. The model was initially developed in healthy Caucasian populations and subsequently adapted to TDT Caucasian populations. The model was then adopted for the Thai healthy and TDT populations.

FIGURE 1.

Workflow for the development and application of the DFX PBPK/PD model and the TDT population model. ADME, absorption distribution metabolism and excretion; C _trough, trough concentration; DFX, deferasirox; IV, intravenous administration; LIC, liver iron concentration; MD, multiple dosing; PBPK, physiologically based pharmacokinetic; PD, pharmacodynamic; PO, oral administration; SD, single dosing; TDT, transfusion‐dependent thalassemia.

2.5. DFX PBPK Model

A DFX PBPK model was developed using physicochemical and ADME data. Initial model construction was based on clinical data from a single 130‐mg intravenous (IV) dose to estimate the clearance and steady‐state volume of distribution (V _d,ss) using the whole‐body PBPK method 2 [5, 21]. Because of the limited human ADME data, the UGT pathway and unchanged biliary excretion were assumed as a single biliary route, accounting for 70% of the IV clearance (CL_iv) based on human mass balance data [8]. The intrinsic clearances of biliary, CYP1A2, and CYP2D6 were calculated using the retrograde model within the reverse translation tool in the simulator with 70%, 6%, and 2% of CL_iv, respectively, and the residual clearance was assigned as additional clearance via human liver microsomes [5, 8]. Renal clearance was set at 0.5% of the CL_iv based on urinary excretion data [8].

The model was then extended to oral administration using clinical data from 375 mg and 20 mg/kg oral doses [5, 22]. Absorption was modeled using the Advanced Dissolution, Absorption, and Metabolism (ADAM) model with a suspension formulation and pH‐dependent solubility [21, 23, 24, 25]. The effective permeability coefficient in humans (P _eff,man) was initially predicted from Caco‐2 cell data and was optimized to fit the observed PK. The EHR was optimized with 65% of biliary‐excreted DFX reabsorbed to reflect the observed oral administration PK. Table 1 summarizes the input parameters of the DFX model. Table S1 presents the datasets used for development and verification.

TABLE 1.

Final input parameters for DFX PBPK/PD model.

Parameters (unit)	Value		References
Physicochemical properties
Molecular weight (g/mol)	373.4		PubChem
Log Po:w	6.3		PubChem
Compound type	Monoprotic acid		Heinz et al. [26]
pKa	3.7		Steninhauser et al. [27]
Blood/Plasma ratio	0.61		Waldmeier et al. [8]
Fraction of unbound in plasma	0.004		Weiss et al. [6]
Absorption
Model type	ADAM		Ezuruike et al. [21]
PappCaco‐2pH7.4:7.4 (cm/s)	5.5 × 10−6		Optimized against the observed PK profiles in Séchaud et al. [5, 22]
Formulation	Suspension		Deferasirox prescribing information [23]
Fraction of API Dissolved (%)	82		Durdunji et al. [24]
Solubility pH 1.2 (mg/mL)	0.004
Solubility pH 6.8 in 0.5% Tween 20 in USP phosphate buffer (mg/mL)	0.62
Solubility pH 8.77 (mg/mL)	0.9		Ediz et al. [25]
Solubility pH 8.91 (mg/mL)	1.07
Distribution
Model type	Full PBPK
Volume of distribution at steady state (L/kg)	0.181		Séchaud et al. [5]
K p scalar	0.84		Optimized against V d,ss in Séchaud et al. [5]
Elimination
Model type	Enzyme kinetics
CYP1A2 CLint (μL/min/pmol)	0.3445		Retrograde from CLIV of 3.53 L/h (Séchaud et al. [5]), 6% of the dose (Waldmeier et al. [8])
CYP2D6 CLint (μL/min/pmol)	1.0029		Retrograde from CLIV of 3.53 L/h, 2% of the dose (Waldmeier et al., 2010 [8])
Bile CLint (μL/min/106 cells)	58.186		Retrograde from CLIV of 3.53 L/h, 70% of CLIV (Assumption)
Renal clearance (L/h)	0.018		Waldmeier et al. [8]
Additional clearance, Liver, HLM (μL/min/mg)	52.391		Retrograde from CLIV of 3.53 L/h, residual clearance
Percentage available reabsorption (%)	65		Optimized against the PK profiles in Séchaud et al. [5, 22]
PD parameters
Population	Healthy	TDT
Storage iron (sFe) compartment volume (L)	1.69		Peters et al. [28]
Chelatable iron (cFe) compartment volume (L)	5.40
Initial storage iron concentration (μM)	2828.48	Various	Obrzut et al. [29]
Initial chelatable iron concentration (μM)	Depending on the initial storage iron concentration		Barry [30]
k in (μM/h)	0.28	0.07	Ganz [31]; Porter [32]
k out (μM/h)	0.28	0.28	Ganz [31]
k blood (μM/h)	0	Various
V max,efflux (μM/h)	43		Estimation
Kmefflux (μM)	1 × 106		Estimation
a (g/mg)	1.66		Aoki et al. [33]
b (μM−1)	0.036		Gehrke et al. [34]
c (unit less)	0.135		Chung et al. [35]
V max,uptake (μM/h)	0.306687		Estimation based on Parkes et al. [36]
Kmuptake (μM)	15.2		Parkes et al. [36]
k a (h−1)	1 × 10−6		Estimation based on Parkes et al. [36]
k on (μM−2)	200		Manual optimized
k off (h−1)	0.96		Suzuki and Iki [37]
k eli (h−1)	0.4		Manual optimized

Abbreviations: a, effect of liver iron concentration on hepcidin constant; API, active pharmaceutical ingredient; b, effect of chelatable iron on hepcidin constant; c, effect of hepcidin on efflux iron transporter constant; CL_int, intrinsic clearance; efflux, efflux transporter; HLM, human liver microsome; k _a , effect of chelatable iron on uptake iron transporter constant; k _blood, rate constant of iron input from blood transfusion; k _eli, elimination rate constant of DFX and iron complex; k _in, rate constant of iron absorption; Km, Michaelis–Menten constant; k _off, dissociation rate constant of DFX and iron complex; k _on, formation rate constant of DFX and iron complex; k _out, rate constant of iron loss; K _p , partition coefficient; Log P_o:w, logarithm of n‐octanol/water partition coefficient; P_appCaco‐2_pH7.4:7.4, apparent permeability in Caco‐2 cell system at pH 7.4 both side; pKa, negative base‐10 logarithm of the acid dissociation constant; uptake, uptake transporter; V _d,ss, volume of distribution at steady‐state; V _max, maximum velocity of iron transport.

2.6. PD Model

Figure 2 presents the structure of the PBPK/PD model. The PD model was developed based on the mechanism of action of DFX and human iron homeostasis using a simplified two‐compartment structure: storage iron and chelatable iron [7, 31]. The storage iron compartment simulated the LIC—the primary iron storage site—where iron exists mainly as ferritin and hemosiderin, which are not readily chelatable [31]. Its volume was set to the physiological liver volume. The chelatable iron compartment represented all iron forms available for the chelation of non‐transferrin‐bound iron (NTBI) and labile iron pools, excluding transferrin‐bound iron, which was assumed to remain constant across physiological states. Its volume was set to the total blood volume. The iron absorption (k _in) and excretion (k _out) rate constant parameters were calibrated to reflect daily iron turnover in healthy individuals. Under TDT conditions, the model accounted for adequate regular blood transfusions with iron overload. k _in was reduced, and k _out remained unchanged compared with those in healthy populations, and additional iron input was added from regular blood transfusions (k _blood). The iron transport dynamics between the two compartments were governed by the uptake and efflux transporters, representing NTBI and ferroportin transporters, respectively [35, 38]. Chelatable iron directly increased the expression of uptake transporters (k _a ), whereas storage and chelatable irons influenced efflux transporter activity via hepcidin (parameters a, b, and c in Figure 2) [33, 34, 35].

FIGURE 2.

Structure of the DFX PBPK/PD model. Major pathways are depicted with bold arrows, whereas minor pathways are represented with lighter arrows. Dashed gray arrows indicate the activation effects, whereas blunt‐ended lines denote the inhibitory effects.

The PD model parameters were derived from physiological data, in vitro experiments, LIC progression rates, and daily transfusional iron in healthy individuals and TDT patients. COPASI software was used for simultaneous parameter estimation [30, 39]. However, the lack of dynamic LIC and chelatable iron data limited PD model verification. Table 1 summarizes the final PD parameter input values. The ordinary differential equations (ODEs) used to describe the compartmental dynamics were defined as follows:

$$\begin{array}{c}\frac{\mathrm{d}\left[\mathrm{cFe}\right]}{\mathrm{dt}}=k_{\text{in}}+k_{\text{blood}}-k_{\text{out}}-\left(\frac{V_{\max ,\text{uptake}}\times \left[\mathrm{cFe}\right]}{{\mathrm{Km}}_{\text{uptake}}+\left[\mathrm{cFe}\right]}+\left({\mathrm{k}}_a\times \left[\mathrm{cFe}\right]\right)\right)\hfill \\ +\frac{\left(\frac{V_{\max ,\text{efflux}}\times \left[\mathrm{sFe}\right]}{{\mathrm{Km}}_{\text{efflux}}+\left[\mathrm{sFe}\right]}\times {\left(a\times \frac{\left[\mathrm{sFe}\right]}{2886.2}\times {\mathrm{e}}^{-b\times \left[\mathrm{cFe}\right]}\right)}^c\right)\times {\mathrm{Vol}}_{\mathrm{sFe}}}{{\mathrm{Vol}}_{\mathrm{cFe}}}\hfill \end{array}$$ (1)

(2)

where, [cFe] represents the chelatable iron concentration, [sFe] denotes the storage iron concentration, and V _max, Km, uptake, efflux, and Vol correspond to the maximum transport velocity, Michaelis–Menten constant, uptake transporters, efflux transporters, and compartment volumes, respectively.

2.7. Combination of the DFX PBPK and PD Models

The PD model and parameters were transferred to the Custom Lua Model in Simcyp. The PD model was combined with the DFX PBPK model using ODE to describe systemic DFX binding to chelatable iron at a 2:1 (DFX:iron) ratio, assuming rapid binding and dissociation of two DFX molecules [4]. The ODEs of DFX and iron were as follows:

$$\begin{array}{c}\frac{\mathrm{d}\left[\mathrm{cFe}\right]}{\mathrm{dt}}=k_{\text{in}}+k_{\text{blood}}-k_{\text{out}}-\left(\frac{V_{\max ,\text{uptake}}\times \left[\mathrm{cFe}\right]}{{\mathrm{Km}}_{\text{uptake}}+\left[\mathrm{cFe}\right]}+\left(k_a\times \left[\mathrm{cFe}\right]\right)\right)+\frac{\left(\frac{V_{\max ,\text{efflux}}\times \left[\mathrm{sFe}\right]}{{\mathrm{Km}}_{\text{efflux}}+\left[\mathrm{sFe}\right]}\times {\left(a\times \frac{\left[\mathrm{sFe}\right]}{2886.2}\times {\mathrm{e}}^{-b\times \left[\mathrm{cFe}\right]}\right)}^c\right)\times {\mathrm{Vol}}_{\mathrm{sFe}}}{{\mathrm{Vol}}_{\mathrm{cFe}}}\\ -\left(\left(k_{\mathrm{on}}\times \left[\mathrm{cFe}\right]\times {\left(\left[{\mathrm{DFX}}_{\mathrm{sys}}\right]\times f_{\mathrm{u}}\right)}^2\right)-k_{\mathrm{off}}\times \left[\mathrm{DFX}―\text{Iron complex}\right]\right)\end{array}$$ (3)

$$\begin{array}{c}\frac{\mathrm{d}\left[\mathrm{sFe}\right]}{\mathrm{dt}}=\left(\frac{\left(\frac{V_{\max ,\text{uptake}}\times \left[\mathrm{cFe}\right]}{{\mathrm{Km}}_{\text{uptake}}+\left[\mathrm{cFe}\right]}+\left(k_a\times \left[\mathrm{cFe}\right]\right)\right)\times {\mathrm{Vol}}_{\mathrm{cFe}}}{{\mathrm{Vol}}_{\mathrm{sFe}}}\right)\hfill \\ -\left(\frac{V_{\max ,\text{efflux}}\times \left[\mathrm{sFe}\right]}{{\mathrm{Km}}_{\text{efflux}}+\left[\mathrm{sFe}\right]}\times {\left(a\times \frac{\left[\mathrm{sFe}\right]}{2886.2}\times {\mathrm{e}}^{-b\times \left[\mathrm{cFe}\right]}\right)}^c\right)\hfill \end{array}$$ (4)

$$\begin{array}{c}\frac{\mathrm{d}\left[{\mathrm{DFX}}_{\mathrm{sys}}\right]}{\mathrm{dt}}=\frac{\mathrm{d}\left[{\mathrm{DFX}}_{\mathrm{sys},\text{PBPK}}\right]}{\mathrm{dt}}-\left(2\times \left(k_{\mathrm{on}}\times \left[\mathrm{cFe}\right]\times {\left(\left[{\mathrm{DFX}}_{\mathrm{sys}}\right]\times f_{\mathrm{u}}\right)}^2\right)\right.\hfill \\ \left.-k_{\mathrm{off}}\times \left[\mathrm{DFX}―\text{Iron complex}\right]\right)\hfill \end{array}$$ (5)

(6)

where, [DFX_sys] represents the systemic DFX concentration after the iron interaction, [DFX_sys,PBPK] denotes the systemic DFX concentration predicted by the PBPK model, and [DFX–Iron complex] refers to the concentration of the DFX–iron complex. The parameters f _u, k _on, k _off, and k _eli represent the fraction of unbound DFX in the plasma, complex formation rate constant, complex dissociation rate constant, and complex elimination rate constant, respectively. f _u and k _off were obtained from literature [6, 37], and k _on and k _eli were manually optimized with a training dataset until the simulated PK profile closely matched the observed clinical profile in TDT populations.

2.8. Verification of the DFX PBPK/PD Model

The PBPK/PD model was verified using data from healthy Caucasian and TDT populations to evaluate its predictive performance. For healthy Caucasians, the same datasets used in the PBPK model verification were used to maintain consistency. In TDT patients, independent datasets covering various DFX doses and LIC levels were used to verify both PK and PD. PD verification was based on changes in LIC from baseline to 6 or 12 months post‐DFX treatment. Given that DFX reaches PK steady‐state after 3 days of daily dosing, and to ensure PD steady‐state was achieved after DFX treatment, the verification simulations were run over a 30‐day duration and then extrapolated to estimate 6‐ and 12‐month LIC values using linear regression.

The verified PBPK/PD model was subsequently adapted and verified for Thai healthy individuals to ensure its predictive performance. The population model was then modified to represent Thai TDT patients and evaluated by simulating against Thai clinical data to assess its accuracy in capturing DFX PK and PD.

2.9. Model Application

The verified PBPK/PD and Thai TDT population models were used to predict the targeted DFX dose required to achieve a 25% reduction of LIC after 6 months of treatment based on LIC levels and blood transfusion regimen, with a particular focus on patients with severe iron overload. Simulations were performed using 10 virtual subjects, aged 18 to 60 years, with equal sex distribution, across 10 independent trials, employing the same running duration and extrapolation as those used in the verification step.

2.10. Model Evaluation

The model was deemed to have successfully satisfied the PK simulation at all stages if it fulfilled the following acceptance criteria: (1) Visual Predictive Check: > 90% of observed DFX plasma concentration–time points must fall within the 5th–95th percentile of simulations. (2) R‐value (Equation 7): PK parameters (C _max and AUC_∞ or AUC _t ) were within the two‐fold error range (0.50–2.00), a commonly accepted criterion [40]. For the PD evaluation, the simulated LIC values were considered valid if they were within the two‐fold error range and the observed mean ± standard deviation range. The model accuracy and precision were assessed using the average fold error (AFE) and absolute average fold error (AAFE), Equations (8 and 9); values between 0.80 and 1.25 indicating good predictive performance without systematic bias [41].

$$\begin{array}{c}R-\text{value}=\frac{\text{Simulated}}{\text{Observed}}\end{array}$$ (7)

$$\begin{array}{c}\mathrm{AFE}={10}^{\frac{\sum \log \left(\frac{\text{Simulated}}{\text{Observed}}\right)}{\text{Number of observation}}}\end{array}$$ (8)

$$\begin{array}{c}\text{AAFE}={10}^{\frac{\sum \left|\log \left(\frac{\text{Simulated}}{\text{Observed}}\right)\right|}{\text{Number of observation}}}\end{array}$$ (9)

3. Results

3.1. PBPK/PD Modeling and Evaluation

The whole‐body PBPK/PD model of DFX was successfully developed using a stepwise middle‐out approach. Initially, a PBPK model was constructed for healthy Caucasian populations (PBPK model performance in Figure S1 and Table S3), followed by the development and integration of a PD model. The resulting PBPK/PD model was subsequently adapted for the Thai healthy population and, ultimately, the Thai TDT population.

3.2. Healthy Populations

The whole‐body PBPK/PD model of DFX successfully captured the plasma concentration–time profiles in both healthy Caucasian and Thai populations (Figure 3). The model adequately described DFX PK across IV and oral administration. The iron storage compartment was parameterized using the LIC values reported in the general population [29]. However, the PD response could not be evaluated in healthy populations because of limited data. All R‐values were within the two‐fold error range (Table S4), with AFE and AAFE values of 1.05 and 1.19, respectively, confirming good model accuracy and precision and the absence of systematic bias.

FIGURE 3.

Simulated and observed plasma concentration–time profiles of the DFX PBPK/PD model in healthy Caucasian (blue; a, b, c) and Thai (cyan; d) populations following a single DFX dose of 250 mg to 30 mg/kg. The colored lines represent the simulated mean, whereas the shaded areas indicate the 5th to 95th percentiles. Observed clinical data are depicted as triangle markers, with those including upright lines representing means ± standard deviations (SDs) and those without indicating the mean values. The inserted figure presents the data on a semilogarithmic scale.

3.3. TDT Populations

The TDT population models for both the Caucasian and Thai groups were adapted from their respective healthy population models, with ethnicity‐specific demographic, physiological, and biochemical parameters modified (Table S2). The whole‐body PBPK/PD model combined with the TDT population models successfully simulated plasma concentration–time profiles of DFX across various oral dosing regimens, body iron levels, and transfusion‐derived iron input (Figure 4). All PK R‐values fell within the acceptable two‐fold error range (Table S4). Furthermore, the AFE and AAFE values were 0.78 and 1.36, respectively, supporting the model's reasonable accuracy and moderate precision in simulating DFX PK in TDT populations.

FIGURE 4.

Simulated and observed plasma concentration–time profiles of the DFX PBPK/PD model in Caucasian (red; a–e) and Thai (orange; f–h) populations with TDT. Profiles include single‐dose (SD) oral administration (PO; a, d, e, f) ranging from 20 to 35 mg/kg, multiple‐dose regimens (MD; b, c) of 10 and 20 mg/kg/day, and trough concentrations at steady state (C _trough,ss) of 30 and 40 mg/kg/day (g, h) under varying liver iron concentrations and blood transfusion regimens. Colored lines indicate the simulated means, shaded areas represent the 5th to 95th percentiles, and the observed clinical data are shown as dot markers. Markers with upright lines denote means ± standard deviations (SDs), whereas those without indicate the mean values.

The model's PD performance was assessed by simulating LIC levels after 6 and 12 months post‐DFX treatment across varying doses, baseline LIC levels, and transfusion‐derived iron input (Figure 5 and Table S5). The simulated post‐treatment LIC values generally fell within the two‐fold error range, except for datasets 14–16. Nevertheless, they were consistent with the observed changes from the baseline values (Figure S2), supporting the model's reliable predictive performance in TDT populations.

FIGURE 5.

Predicted versus observed mean values for C _max (a), AUC_∞ (b), and liver iron concentration (LIC) (c) at 6–12 months after DFX treatment. The solid line represents the line of unity, whereas the dashed lines indicate the twofold error range.

3.4. Targeted DFX Dose Prediction

The verified PBPK/PD and Thai TDT population models were applied to predict the DFX dose required to reduce LIC by 25% from baseline after 6 months of treatment. The 25% reduction was considered a key threshold for reducing iron overload complications and associated risks. The predictions were performed using baseline LIC levels of 10, 15, 22.5, and 30 mg/g, with transfusion regimens of 1 to 3 units every 3 to 4 weeks. The results indicated that at the standard DFX dose of 20 to 40 mg/kg/day, the predicted LIC reduction varied according to baseline LIC levels and transfusion burden (Figure 6). Patients with higher baseline LIC values (e.g., 22.5 and 30 mg/g) generally exhibited greater absolute LIC reduction, whereas those with lower LIC values (e.g., 10 and 15 mg/g) required higher DFX doses to achieve the desired reduction (Table S6).

FIGURE 6.

Application of the DFX PBPK/PD model to predict liver iron concentration (LIC) at 6 months after the initiation of DFX treatment at doses of 20, 30, and 40 mg/kg/day, considering baseline LIC and blood transfusion regimen. Colored dots represent the mean LIC, with error bars indicating the 5th to 95th percentiles. The shaded green area denotes the reduction target, corresponding to a desired ≥ 25% reduction in LIC at 6 months.

4. Discussion

DFX is widely used to manage chronic iron overload [23]. However, substantial interindividual variability in DFX response has been reported among Caucasian, Taiwanese, and Thai patients [10, 11, 12, 13]. Assessing treatment efficacy and adjusting doses may take up to 6 months, increasing the risk of complications [23, 32]. Although the current guideline recommends increasing initial DFX doses for patients with high transfusion rates, this strategy remains imprecise, aiming only for negative iron balance rather than specific LIC reduction [32]. Moreover, DFX demonstrated higher drug exposure in healthy individuals than in TDT patients, possibly because of differences in body iron burden, though evidence is limited [22, 42]. To address these gaps, the PBPK/PD model of DFX was developed and used to investigate and optimize dosing strategies. Using a top‐down approach, glucuronidation and biliary excretion were assumed into a single clearance pathway, accounting for 70% of CL_iv based on the DFX and metabolites recovered in feces [8]. Furthermore, EHR was suspected from the secondary peak observed in several DFX PK studies; however, the exact fraction in humans remains unclear [5, 22, 42, 43]. In our model, EHR accounted for 65% of biliary excretion, adequately capturing the observed PK profiles and correlating with animal data [9]. Based on Yang et al.'s equation [44], the biliary‐excreted unchanged DFX is < 10% of the dose. Assuming it is reabsorbed similarly to the absorption fraction (~80%), up to 37% of the dose may reflect glucuronide‐mediated EHR. A previous DFX PBPK model was developed for drug interaction purposes; however, it overpredicted the observed data [45], and direct comparison with our model was limited because of differences in objectives and model assumptions. Due to limited data on the fractional contribution of DFX metabolic pathways, the PBPK/PD model could not mechanistically delineate the characterization of UGT‐mediated EHR. Consequently, the model is not suitable for evaluating drug–drug interactions involving UGT enzymes. Further data on metabolic fractions and enzyme kinetics are required to refine the model for such mechanistic applications.

The PD model was derived from human iron metabolism and simplified into a physiologically relevant two‐compartment structure that adequately captured DFX's PD. The PD model was trained particularly using LIC data from both healthy and TDT populations to fit changes over time [30, 39]. In contrast, the chelatable iron was not directly trained and remained as a model‐calculated value. However, this study focused solely on LIC as the primary marker of the effects of PD. Although more comprehensive models incorporating multiple compartments to describe human iron homeostasis are available [46, 47], our two‐compartment model prioritized simplicity and integration with the PBPK model. Despite its reduced complexity, the model demonstrated adequate accuracy and precision in predicting LIC changes following DFX treatment, supporting its use for evaluating the PD effects.

Overall, the DFX PBPK/PD model exhibited good predictive performance, with AFE and AAFE values of 0.96 and 1.24, respectively, for PK across all populations. Furthermore, > 90% of the observed data points across most datasets fell within the 5th–95th percentile of the simulations, except for datasets 12 and 19, with the small sample size likely underrepresenting population variability, limiting the model's predictive generalizability. However, the simulated PK profiles closely reflected the observed trends, supporting the model's reliability (Figure 4d,f). Consistent overprediction observed in healthy Thai volunteers (Dataset 18) reflects ethnic‐specific physiology that is not fully represented in the current model, indicating the need for refinement using ethnicity‐specific parameters. For PD, datasets 14–16 exceeded the two‐fold error range, likely due to using averaged transfusional iron inputs across the four groups (datasets 14–17) (Table S1), rather than group‐specific values, which may obscure variability. Moreover, while the clinical study adjusted DFX dose, our model fixed the dose, which may explain why Dataset 11, with no dose changes over 6 months, showed close agreement between predictions and observations, whereas Dataset 16, despite the same total dose and population, had dose adjustments over 1 year, resulting in greater prediction discrepancies.

To investigate the effects of body iron levels on DFX PK, we compared simulations of the same 20 mg/kg single dose at 0.98‐mg/g and 8.54‐mg/g LIC using datasets 6 and 9 (Figures 3b and 4a). In addition to the differences in population models, the body iron burden was a key variable in these simulations. The PBPK/PD model captured this variation, aligning with the observed data and showing that elevated LIC levels were associated with approximately 40% higher DFX clearance compared to healthy populations [42]. We further compared datasets 12 and 13 (3.2‐mg/g and 11.20‐mg/g LIC, respectively; Figure 4d,e), involving the same population and dosage. Although the model underpredicted dataset 12, it reproduced the trend: higher LIC correlated with approximately 30% increase in DFX clearance. This finding is consistent with reports showing that clearance between the high‐ and low‐LIC groups tends to reflect differences. However, other factors beyond LIC likely influenced variability because dataset 12 had a small sample size and the study focused on treatment response [11]. Together, these results support a possible 30%–40% increase in DFX clearance with higher iron levels. However, because of the similar LIC differences between these two studies [11, 42] and limited data for further validation, no definitive LIC–clearance relationship could be established. Additional studies administering consistent DFX doses across varying LIC levels are needed.

The TDT population models were adapted from their respective healthy models, incorporating ethnicity‐specific demographics and physiological and biochemical parameters, while retaining other default values. Although altered CYP and UGT activity has been reported in TDT patients, the extent remains unclear [48, 49]. Severe iron overload may impair liver function and promote fibrosis, potentially affecting drug‐metabolizing enzymes and transporters (DMET) [50], complicating PK and treatment outcomes. Despite these simplifications, the modified TDT population models adequately reproduced the observed PK and PD. Further refinement of the disease‐associated DMET function could enhance predictive accuracy.

Figure 6 presents the predicted LIC levels at 6 months post‐DFX treatment in Thai TDT patients, considering baseline LIC, transfusion regimens, and DFX dosing. Considering the 5th–95th percentile, simulations suggested that 20 mg/kg/day was likely sufficient to achieve a 25% LIC reduction for patients with high baseline LIC (> 15 mg/g) across most transfusion frequencies, whereas those with lower baseline LIC required 30–40 mg/kg/day to reach the same likelihood. This finding is consistent with clinical observations showing greater absolute LIC reductions in patients with higher iron overload who undergo the same chelation therapy [15]. Among patients with similar baseline LIC, the transfusion burden strongly influenced treatment outcomes. Lower transfusion regimens (e.g., 1 unit every 3–4 weeks) were associated with a higher likelihood of achieving the target LIC reduction than higher transfusion regimens (e.g., 2–3 units every 3 weeks). In patients with low baseline LIC and high transfusion rates, even the highest recommended dose may be inadequate. The model explained that at lower LIC levels, the conversion of storage iron to chelatable iron is slower. This conversion is regulated by hepcidin and iron levels, thus resulting in less chelatable iron available for DFX to remove. Consequently, higher DFX doses are required to achieve the same therapeutic effect. In contrast, patients with high baseline LIC required lower DFX doses. One possible explanation is that severe iron overload may impair DMET activity, increasing the DFX proportion that can form iron chelation complexes. However, this mechanism is a hypothesis with limited clinical evidence. Further data are required to validate this potential influence. Overall, these findings suggest that the PD response is shaped by the interplay between baseline LIC, transfusion intensity, and iron metabolism, reinforcing the necessity for individualized DFX dosing.

In conclusion, the DFX PBPK/PD model was successfully developed and verified in both healthy and TDT populations, including Caucasian and Thai cohorts. The TDT models incorporated physiological modifications to reflect disease‐specific changes, and clinical data confirmed the model's predictive accuracy for DFX PK and PD. The results highlight the significant impact of body iron burden on DFX clearance, with baseline LIC and transfusion regimen being key determinants of treatment response. Higher baseline LIC facilitated greater reductions, while lower LIC required higher doses due to slower iron mobilization. Lower transfusion frequency improved therapeutic outcomes. This model supports personalized chelation strategies to enhance efficacy and reduce risks of under‐ or overtreatment. Future studies with larger datasets are needed to refine predictions and extend applicability across diverse patient populations.

Author Contributions

W.S., X.P., S.E., S.S., V.K., and V.Y. wrote the manuscript; W.S., U.U., C.L., T.P., and V.Y. designed the research; W.S., X.P., K.Y., S.E., and V.V. performed the research; W.S., S.E., V.V., T.P., and V.Y. analyzed the data; P.T., S.K., V.K., I.T., and S.S. contributed new reagents/analytical tools.

Conflicts of Interest

X.P. is employee of Certara UK Limited and may hold shares in Certara. P.T., S.K., V.K., and I.T. are employees of the Government Pharmaceutical Organization, Thailand, and declared no competing interests. The other authors declare no conflicts of interest.

Supporting information

Data S1: cts70355‐sup‐0001‐Supinfo.docx.

Sakares W., Udomnilobol U., Pan X., et al., “Bridging Pharmacokinetics and Pharmacodynamics: A PBPK/PD Model‐Based Approach for Deferasirox Dosing in Transfusion‐Dependent Thalassemia,” Clinical and Translational Science 18, no. 9 (2025): e70355, 10.1111/cts.70355.