Optimal dosing for vascular anomalies paediatric patients with population pharmacokinetic model of sirolimus

Abstract

Aim

Sirolimus is currently used off‐label for paediatric patients with vascular anomalies. However, the optimal dosage regimen for paediatric patients remains controversial. This study aimed to determine the optimal dosing regimen of sirolimus in these patients using a population pharmacokinetic (PK) model.

Methods

Prospective study data (seven subjects) were used as the structural model without covariates, while the retrospective study data (21 subjects) were used as the structural model with size and maturation functions for evaluation. The model was assessed using general model‐building criteria, and a bootstrap stepwise covariate modelling method was used to identify relevant covariates. Simulations were performed for various scenarios to determine the optimal dosing regimen for maintaining sirolimus trough concentrations within a specific range. NONMEM software (Version 7.5) with the first‐order conditional estimation method with interaction was used.

Results

The final sirolimus model was developed using a two‐compartment model, which was implemented by a maturation function using postmenstrual age (PMA), allometrically scaled body weight to account for size differences, and haemoglobin as a covariate on clearance. Simulations were conducted to propose an optimal dosing regimen. A once‐daily dose of 0.01–0.1 mg/kg over 1 month maintained the trough concentration within the range of 5–15 ng/mL, taking into account patient‐specific weight and the influence of haemoglobin constrained on PMA. The optimal dose varied by group, within the range of 0.06–0.09 mg/kg.

Conclusions

We developed a robust and reliable PK model of sirolimus for paediatric patients with vascular anomalies. Paediatric optimal dosing regimens should be based on individual growth patients.

What is already known about this subject

• Due to its therapeutic narrow, sirolimus requires mandatory monitoring to ensure safety and efficacy.

• Sirolimus has been empirically used off‐label to treat paediatric vascular anomalies in various countries, despite the lack of paediatric‐specific exposure information.

What this study adds

• A population pharmacokinetic model of sirolimus encompassing neonates to adolescents with vascular anomalies was successfully developed.

• Optimal initial dosage regimens were suggested, taking into account paediatric growth and haemoglobin levels.

1. INTRODUCTION

Sirolimus was approved by the U.S. Food and Drug Administration (FDA) in 1999 for the prevention of organ rejection following renal transplantation. Recently, sirolimus has been prescribed off‐label for the treatment of paediatric patients with vascular anomalies. Sirolimus inhibits the mechanistic target of rapamycin complex (mTORC) pathway. The activation of mTORC1 pathway involves vascular endothelial growth factor (VEGF) generation and angiogenesis. In a systematic review, sirolimus was reported to improve the prognosis of vascular anomalies. ¹ The Pharmaceuticals and Medical Devices Agency (PMDA) has approved sirolimus for the treatment of refractory vascular tumours and refractory vascular malformations, recommending therapeutic drug monitoring (TDM) to maintain a trough concentration within 15 ng/mL.

Sirolimus is primarily metabolized by the cytochrome P450 enzyme CYP3A4, with additional contributions from CYP3A5 and CYP2C8. Its oral bioavailability is affected by intestinal CYP3A content as well as by efflux and uptake transporters, including P‐glycoprotein (P‐gp) and organic anion‐transporting polypeptides OATP1B1 and OATP1B3, which influence intestinal absorption and hepatic uptake. ² , ³ , ⁴ Sirolimus is characterized by highly variable absorption and elimination, which lead to substantial blood concentration differences among patients receiving the same doses. ⁵ TDM has demonstrated the utility of dose individualization to ensure that sirolimus target concentrations are within defined efficacy and safety limits. ⁶

Drug development for paediatric patients has been particularly challenging due to ethical barriers, limited sampling and insufficient subject enrollment, which complicate the evaluation of paediatric pharmacokinetics (PKs) and hinder the determination of appropriate dosing regimens. Paediatric patients exhibit significant differences in PK profiles compared to adults, owing to variations in maturation, organ function and enzyme activity, necessitating careful dose adjustments to ensure both efficacy and safety. To address these challenges, the introduction of the population approach in paediatric studies has provided more flexibility, while retaining sufficient information for parameter estimation. ⁷ The nonlinear mixed‐effects modelling technique allows for the estimation of typical PK parameters, along with the assessment of within‐ and between‐subject variability. ⁸ Additionally, it helps identify patient characteristics that influence PK variability, such as age, sex, body weight, race and ethnicity, liver and renal function and genotype. ⁹

The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) E11 guideline, modelling and simulation (M&S), is one such tool that can help avoid unnecessary paediatric studies and ensure that appropriate data are generated from the smallest number of paediatric patients. ¹⁰ The usefulness of M&S in paediatric drug development includes, but is not limited to, clinical trial simulation, dose selection, choice and optimization of study design, endpoint selection and paediatric extrapolation. By utilizing M&S, researchers can minimize the number of paediatric patients required for clinical studies, while still generating robust data for safe and effective treatment. Additionally, according to FDA guidelines, M&S can help reduce uncertainty about drug dosing in paediatric populations. Model‐informed drug development has been used for regulatory applications in paediatric drug development.

In recent studies, the population PKs of sirolimus have predominantly been investigated using TDM data. The dosing regimen approved by the PMDA was extrapolated from adult PK data without observing peak levels in paediatric patients, which presents a limitation in evaluating PKs in paediatric patients. Additionally, some studies have reported that its half‐life is shorter in paediatric patients compared to adults. ¹¹ , ¹² , ¹³

The PMDA approved a dosing regimen based on body surface area (BSA) for patients with refractory vascular tumours and refractory vascular malformations. According to the label, the starting dose of sirolimus is 2 mg once daily for patients with a BSA of 1.0 m² or more, and 1 mg once daily for those with a BSA of less than 1.0 m². Thereafter, the dosage is adjusted according to the blood trough concentration and the patient's condition but should not exceed 4 mg once a day. However, when the dosing regimen is based on BSA, paediatric patients up to 8 years of age are required to take 1 mg, which may result in high drug exposure in younger patients. ¹⁴

This study aimed to evaluate the overall PKs of sirolimus in Korean paediatric patients with vascular malformations by integrating TDM and clinical trial data. Using a population PK model, this approach provides safe usage information and improves our understanding of the PK profile of sirolimus.

2. METHODS

2.1. Characteristics and clinical study of patients with vascular anomalies

The patients were enrolled both retrospectively and prospectively. The patients were aged between 1 month and 18 years old. Patients with vascular anomalies were included if they met at least one of the following criteria and had participated, were currently participating, or were scheduled to participate in a sirolimus clinical trial: (1) Patients diagnosed with a vascular anomaly (vascular tumours or malformations) with at least one associated complication (coagulopathy, chronic pain, recurrent cellulitis, ulceration, visceral and/or bone involvement, organ dysfunction or severe deformity). (2) Vascular tumours that failed to respond to or relapsed after treatment with one or more of the following: steroids, interferon‐α, chemotherapeutic agents, propranolol or surgical/localized treatments. (3) Vascular malformations that were difficult to treat surgically or locally, failed treatment or relapsed after such interventions.

The sampling scheme was designed to adequately characterize both absorption and distribution phases in the prospective cohort, while reflecting routine therapeutic drug monitoring practice in the retrospective cohort. Early post‐dose sampling points were selected to capture peak concentrations and distribution kinetics, whereas trough samples were obtained to evaluate steady‐state exposure. This combined approach was intended to provide sufficient structural information for model development while incorporating real‐world clinical data.

In the retrospective study, samples for PK analysis were collected at trough levels on Days 3, 7 and nearly 1 month after sirolimus administration. In the prospective study, samples for PK analysis were collected 1.5 h post‐administration of sirolimus in patients who had already received the drug. For patients who had not previously been administered sirolimus, in the case of the syrup formulation, samples were collected at 1, 4 and 12 h post‐dosing on Day 1, as well as at 1, 4 and 12 h on Day 2, and at pre‐dosing and 1 h post‐dosing on the day of the first outpatient visit. For the tablet formulation, samples were collected at 2, 4 and 12 h post‐dosing on Day 1, as well as at 2, 4 and 12 h on Day 2, and at pre‐dosing and 2 h post‐dosing on the day of the first outpatient visit.

Additional information was collected from participants in both studies, including subject identification number, age, date of birth, diagnosis, start date of treatment, end date of treatment, age at diagnosis, height and weight on the day of dosing, pharmacogenetic information, administered dose amount and time, sampling times, TDM results, concomitant drugs, laboratory tests (e.g., Complete blood count, chemistry and urinalysis), adverse events and discontinued history.

Whole blood samples for PK analysis were collected using EDTA tubes to determine blood sirolimus concentration. The samples were analysed using high‐performance liquid chromatography–tandem mass spectrometry in accordance with bioanalytical method validation guidelines. The calibration curve ranged from 1.0 to 26.3 ng/mL and demonstrated acceptable linearity (r ² = 0.99). The lower limit of quantification (LLOQ) was 1.0 ng/mL, defined based on acceptable accuracy and precision criteria, ensuring reliable quantification of sirolimus concentrations within the analytical range.

Pharmacogenomic analysis was conducted on whole blood samples collected for PK analysis. Genotyping focused on identifying polymorphisms in drug‐metabolizing enzymes (CYP2C8, CYP3A4 and CYP3A5) and transporters, specifically the ABCB1 gene variants rs1045642 (C3435T), rs1128503 (C1236T) and rs2032582 (G2677T/A). DNA was extracted from whole blood using a DNA extraction kit. The presence of these genes was determined through polymerase chain reaction (PCR) amplification. The results for each genotype and variant were confirmed using direct sequencing and MassArray. All clinical trials and sample analyses were conducted at Samsung Medical Center in Seoul. (IRB No. 2023‐09‐075).

2.2. Population PK model development and evaluation

The blood sirolimus concentration was transformed using the natural logarithm. A population PK model for sirolimus was initially developed using prospective study data. The structural model with size and maturation functions was validated using retrospective data. When the structural model with size and maturation functions adequately explained the retrospective study data, the prospective and retrospective study data were integrated to explore covariates because certain covariate data were insufficient in the prospective dataset. This integration aimed to improve the model's accuracy and ensure the inclusion of all relevant covariates. Subsequently, the structural model with size and maturation functions was re‐estimated and updated during the covariate search process.

The structural model incorporating body size and maturation functions was re‐established using the integrated dataset to allow for covariate exploration. Due to anticipated limitations in structural parameter identifiability arising from the integration of prospective and retrospective data, prior information from previously published population PK models in a comparable patient population was considered. Literature‐based parameter estimates were incorporated using the $PRIOR NWPRI subroutine in NONMEM. Both THETA and OMEGA parameters were implemented with prior means and variances specified according to the parameter estimates and their corresponding variances reported in the reference study. Sensitivity analyses were conducted by varying the prior mean and variance by ±50% to evaluate the robustness of the model to prior specification. Following the establishment of the prior‐informed base model, covariate effects were systematically explored.

For covariate exploration, a stepwise covariate search (SCM) was initially conducted using the updated model, incorporating both forward addition (p < .05) and backward elimination (p < .01). Additionally, we used bootstrap SCM (Boot SCM) to ensure reliable covariate selection and avoid overfitting the parameters with the same statistical criteria as the SCM on each. Covariates were selected based on their inclusion frequencies across the bootstrap iterations. Covariates with an inclusion frequency of 80% or higher were considered highly stable and were included in the final model. Those with an inclusion frequency between 20% and 80% were more carefully evaluated, considering their theoretical relevance and interpretability before deciding on inclusion. Covariates with inclusion frequencies below 20% were excluded from the model due to their low selection stability. This criterion was applied to ensure a balance between model robustness and theoretical justification.

Model evaluation was performed through numerical criteria and graphical methods. The selection criteria for the best PK model structure were primarily based on statistically significant reductions in OFV, using the likelihood ratio test (ΔOFV > 3.84, corresponding to p < .05 for 1 degree of freedom), in combination with goodness‐of‐fit (GOF) plots, and prediction‐ and variability‐corrected visual predictive checks (pvcVPC). ¹⁵ Bootstrap was used to evaluate the stability and performance of the final model through 1000 resampling replications.

The population PKs of sirolimus were analysed using NONMEM software (Version 7.5, ICON, Ellicott City, MD, USA) with the first‐order conditional estimation method with interaction (FOCE INTER).

2.3. Simulation

Simulations were conducted to determine the appropriate dosing regimen for sirolimus in paediatric patients with vascular anomalies, based on the established final population PK model. The target range was defined as a trough concentration of 5–15 ng/mL at steady state. Simulations were performed to identify the dosing regimens that maximized the proportion of patients achieving the target trough concentration.

2.4. Nomenclature of targets and ligands

Key protein targets and ligands in this article, including mechanistic target of rapamycin (mTOR) and its ligand sirolimus (rapamycin), are hyperlinked to corresponding entries in the IUPHAR/BPS Guide to PHARMACOLOGY (http://www.guidetopharmacology.org) and are permanently archived in the Concise Guide to PHARMACOLOGY 2023/24. ¹⁶

3. RESULT

3.1. Characteristics and clinical study of patients with vascular anomalies

A total of 28 paediatric patients with vascular disease were enrolled in this prospective and retrospective study, providing 107 samples for model development. Of these, 74 samples were from the prospective study, which involved seven patients. According to ICH guidelines, four infants, 15 children and nine adolescents participated in the study. Patient demographics and characteristics are summarized in Table 1.

TABLE 1.

Demographic characteristics.

Demographic information		N or mean ± SD (min–max)
N		28
Sex (M/F)		11/17
Study (retrospective/prospective)		21/7
Age (year)		8.7 ± 5.6 (0.3–18.3)
Infant (n = 4)		1.1 ± 0.6 (0.4–1.9)
Children (n = 15)		7.4 ± 3.2 (2.1–11.5)
Adolescents (n = 9)		15.2 ± 2.3 (12.3–18.3)
Weight (kg)		37 ± 24.7 (6.3–83.8)
Height (cm)		126.4 ± 35.5 (62.2–170)
Body surface area (m2)		1.1 ± 0.5 (0.3–1.9)
Treatment duration (day)		593.9 ± 771.6 (0–3303)
Concomitant drug administration (Y/N)		1/27
Clinical laboratory test
Alanine transaminase (ALT)		23.1 ± 32.7 (6–173)
Bilirubin		0.4 ± 0.3 (0.2–1.4)
Haemoglobin		12.6 ± 1.2 (10.2–14.6)
Genotype
CYP2C8	*1/*1	28
CYP3A4	*1/*1	25
	*1/*18	3
CYP3A5	*1/*1	1
	*1/*3	11
	*3/*3	16

3.2. Population PK model development and evaluation

The structural model was best described as a two‐compartmental model using prospective study data. The structural model with size and maturation functions incorporated allometric scaling of individual volume of distribution (V _d) and clearance (CL) to a standard 70‐kg subject and a maturation function based on PMA to account for age‐dependent changes in clearance. The maturation function was fixed to literature‐based values reported in Japanese paediatric populations, as PMA was not directly available in our dataset. PMA was derived from chronological age by assuming a gestational age (GA) of 40 weeks.

The structural model with size and maturation functions was evaluated using the results from a retrospective study. The model demonstrated strong performance in prediction‐based diagnostics, with most of the retrospective data being well predicted and explained by the population PK model, as described in Table S1. Therefore, the prospective and retrospective study data were integrated. The structural model with body size and maturation functions was re‐established using the integrated dataset to explore the covariates. However, after model re‐establishment, elevated relative standard errors were observed across multiple parameters, indicating parameter instability. To address this issue, literature‐based prior information derived from the two‐compartment population PK model reported by Miyazaki et al. (2024) was incorporated into the structural parameters of the two‐compartment model, with prior distributions implemented for both the fixed‐effect (THETA) and corresponding inter‐individual variability (OMEGA) parameters. Implementation of this prior reduced parameter uncertainty and improved overall model stability. Sensitivity analyses demonstrated that varying the prior mean and variance by ±50% resulted in changes of less than 15% in the final parameter estimates, confirming that the model results were not overly sensitive to prior specification. Following the establishment of this prior‐informed base model, covariate effects were systematically explored.

During covariate exploration, CL was tested with haemoglobin (HGB), bilirubin (BIL), CYP3A4, CYP3A5 and CAT3A5 (categorized as CYP3A5 alleles 1 and 3). For V _d,c , the covariates tested included duration of treatment (DTT), HGB, BIL and formulation (FORM).

As a result of SCM, haemoglobin was included as a covariate for CL during forward selection but was excluded in backward elimination. In the subsequent Boot SCM, out of 500 iterations, haemoglobin levels were selected as a covariate for CL in 138 iterations, and this was included in the final model. No significant effects were observed for covariates such as BIL, CYP genotype, FORM and DTT.

The final model was described by following equation, and the model scheme was shown in Figure S1.

$$\begin{gathered} V_{\mathrm{d},\mathrm{c}\left(\text{individual}\right)}=V_{\mathrm{d},\mathrm{c}\left(\text{population}\right)}\times \left(\frac{\text{Weight}}{70}\right) \\ {\mathrm{CL}}_{\text{individual}}={\mathrm{CL}}_{\text{population}}\times {\left(\frac{\text{Weight}}{70}\right)}^{0.75}\times \mathrm{MF} \\ \times \left(1-0.155\times \left(\mathrm{Hgb}-12.6\right)\right)\times {\mathrm{e}}^{{\eta }_2} \\ \text{Maturation function}\left(\mathrm{MF}\right)=\frac{{\mathrm{PMA}}^{\text{Hill}}}{{{\mathrm{TM}}_{50}}^{\text{Hill}}+{\mathrm{PMA}}^{\text{Hill}}} \\ {Q}_{\text{individual}}=Q_{\text{population}}\times {\left(\frac{\text{Weight}}{70}\right)}^{0.75} \end{gathered}$$

V _{d,c(individual)} and V _{d,p(individual)} represent the individual central and peripheral volumes of distribution, respectively, while V _{d,c(population)} and V _{d,p(population)} represent their corresponding population‐level typical values. CL_individual and Q _individual refer to the individual clearance and intercompartmental clearance, and their corresponding typical population values are CL_population and Q _population, respectively. The model incorporated covariates such as HGB and PMA to explain part of the variability in clearance. The maturation effect on clearance was modelled using a Hill function, where TM₅₀ represents the postmenstrual age (PMA) at which 50% of adult clearance is achieved, and Hill is the Hill coefficient. η is the variability between subjects.

The final model estimates and bootstrap validations are summarized in Table 2. The median parameter estimates from the bootstrap validation closely matched those from the final population model, confirming the accuracy and reliability of the PK parameter estimates. The GOF plot for the final model showed that the relationship between individual predictions (IPRED) and observed concentrations was close to linear, with conditional weighted residuals (CWRES) falling within the range of −3 to 3, as shown in Figure S2. Additionally, the VPC plot for the final model demonstrated that most of the observed values were within the 95% prediction interval of the predicted values, as shown in Figure 1A.

TABLE 2.

Parameter estimates of final model and bootstrap validation.

		RSE a (%)	Bootstrap
Parameter	Estimate		Median	95%CI b (2.5% CI to 97.5 CI)
K a (h−1)	0.175	56.6	0.151	0.077–0.287
V c /F (L/70 kg)	102	92	82.5	9.6–265.7
V p /F (L/70 kg)	725	28.8	781.0	435.3–1109.9
CL/F (L/h/70 kg)	14.7	9	14.7	12.0–18.4
Q/F (L/h/70 kg)	24.7	37.1	23.3	9.0–45.2
TM50 (week)	62.9 FIX	‐	62.9 FIX	‐
Hill	2.94 FIX	‐	2.94 FIX	‐
COVHGB	−0.155	28.1	−0.155	−0.225 to −0.068
OMEGA
IIV_CL/F	0.124	15.5	0.122	0.105–0.148
IIV_V c /F	1.9	9.7	1.9	1.85–1.96
SIGMA
RV (proportional)	0.13	19.5	0.12	0.082–0.236

Abbreviations: CL/F, apparent clearance; COVHGB, typical population value of haemoglobin; Hill, Hill coefficient; IIV, inter‐individual variability; K _a, absorption rate constant; Q/F, apparent intercompartment clearance, TM₅₀, postmenstrual age at which 50% of adult clearance is reached; RV, residual variability; V _c/F, apparent central volume of distribution, V _p/F, apparent peripheral volume of distribution.

^a RSE, relative standard error.

^b CI, confidence interval.

FIGURE 1.

Visual predictive checks plots of the final model. Black dot: observation; blue area: 95% confidence intervals of 5th and 95th percentiles; red area: 95% confidence intervals of median; red solid line: median of simulated data; blue solid line: 5th and 95th percentiles of simulated data. Red dashed line: median of observed data; blue dashed line: 5th and 95th percentiles of observed data.

3.3. Simulation

The simulations accounted for weight, age, and haemoglobin levels, with dosages ranging from 0.01 to 0.1 mg/kg, administered once daily for 1 month.

To account for paediatric growth, GA was assumed to be 40 weeks and PMA was calculated accordingly. Weight and age data were generated using random numbers based on weight‐for‐age charts provided in the 2017 Pediatric Growth Chart Handbook by the Korean Disease Control and Prevention Agency (KDCA). ¹⁷

The covariates were categorized as follows: weight was divided into nine groups (5–10, 10–20, 20–30, 30–40, 40–50, 50–60, 60–70, 70–80 and 80–90 kg); PMA was divided into five intervals (66–92, 92–118, 118–144 and >144 weeks); and haemoglobin levels were grouped into five ranges (10–11, 11–12, 12–13, 13–14 and 14–15 g/dL).

Initially, the simulation was divided into 12 groups based on weight and age, with 100 subjects per group, and the appropriate dose was determined for each group. The effect of haemoglobin levels was assessed after completing the simulations based on weight and age, followed by additional simulations for each group to determine the optimal dose. The appropriate dose for each group was defined as the dose that resulted in a trough concentration (C _trough) within the target range of 5–15 ng/mL of steady‐state sirolimus.

The appropriate doses for each group, based on weight and PMA, are listed in Table 3. At steady state, the percentiles of subtherapeutic range below 5 ng/mL, target range between 5 and 15 ng/mL, and toxicological range above 15 ng/mL were compared across different doses, as illustrated in Figure S3. The effect of haemoglobin on sirolimus PKs was simulated for each group based on the dosing regimens listed in Table 3. When haemoglobin levels exceeded 13 g/dL, the median C _trough typically remained within the target range, although some predictions slightly surpassed the upper limit, as shown in Figure 2. Consequently, C _trough may exceed the target range, indicating the need for continuous monitoring.

TABLE 3.

Recommended dosing regimen by group at normal haemoglobin levels based on simulation.

Group	Dose (mg/kg)	Group	Dose (mg/kg)
WT 5–10 kg, PMA 66–92 weeks	0.08	WT 5–10 kg, PMA 92–118 weeks	0.09
WT 10–20 kg, PMA 118–144 weeks	0.09	WT 10–20 kg, PMA > 144 weeks	0.09
WT 20–30 kg, PMA > 144 weeks	0.08	WT 30–40 kg, PMA > 144 weeks	0.06
WT 40–50 kg, PMA > 144 weeks	0.06	WT 50–60 kg, PMA > 144 weeks	0.06
WT 60–70 kg, PMA > 144 weeks	0.08	WT 70–80 kg, PMA > 144 weeks	0.07
WT 80–90 kg, PMA > 144 weeks	0.06

Abbreviatios: PMA, postmenstrual age; WT, weight.

FIGURE 2.

Target attainment of C _trough at steady state by haemoglobin level for each group.

4. DISCUSSION

This clinical study, which included both prospective and retrospective data, evaluated infant children, and adolescent patients with vascular anomalies. The included patients had various diagnoses, including Kaposiform hemangioendothelioma, lymphangioma and venous malformations. The single dose of the medication was administered based on BSA, ranging from 0.27 to 1.04 mg/m², or based on weight, ranging from 0.01 to 0.04 mg/kg, with repeated dosing throughout treatment.

In general, paediatric patients exhibit physiological differences from adults, which can lead to variations in PKs. Regarding drug metabolism, enzymatic maturation occurs at different rates for each metabolic enzyme during growth, with most enzymes reaching maturity by 1 year of age. ¹⁸ , ¹⁹ To account for this, the maturation function was incorporated into the model using PMA in a sigmoidal E _max model, allowing CL to converge to adult values after a certain age. Since patient GA data were unavailable, it was assumed to be 40 weeks for all patients, and PMA was calculated accordingly. Although we attempted to estimate the maturation effect from the data, the age range of the study population was insufficient to reliably capture this effect, and the parameters could not be robustly estimated. TM₅₀ is a physiologically based parameter rather than a drug‐derived parameter, reflecting glomerular filtration rate–related maturation. Therefore, the TM₅₀ and Hill coefficient parameters of the E _max model were fixed to reflect age‐dependent changes in CL based on experimental data. ²⁰ Furthermore, changes in organ size relative to body size and the resulting alterations in CL and V _d were accounted for using allometric scaling.

In this study, PK differences based on genotypes were explored; however, marked differences were not found. According to Chen et al., the CYP3A5 genotype is a covariate for CL, and Djebli et al. also observed the influence of the CYP3A5 genotype in a cohort of kidney transplant recipients. ²¹ , ²² Genotypic variations were not included as covariates in this study because most patients with the *3/*3 allele also had a normal CYP3A4 genotype, leading to minimal metabolic differences. Other studies have identified several covariates, including age, ALT, sex, treatment duration, drug formulation, co‐administration of CYP3A4 inducers and use of frozen samples.

In this study, population PK model‐based simulations were performed to evaluate mg/kg‐based dosing regimens, incorporating not only therapeutic drug monitoring (TDM) data but also intensive PK sampling from clinical studies. This approach enabled a more comprehensive characterization of sirolimus PKs in paediatric patients, accounting for the known high variability in both absorption and elimination.

Population model‐based simulations predicted that the proposed dosing regimen (mg/kg) would be higher than the overall clinical trial results. This dose increase appeared to be due to the change to once‐daily dosing. If increasing the single dose leads to an elevated risk of adverse effects, it may be necessary to reduce the dose and switch to a twice‐daily regimen. Our dosing regimen was suggested as a mg/kg‐based dose. The reason for proposing the dose in mg/kg rather than BSA is that because the disease is a congenital disorder observed after birth, using BSA‐based dosing may lead to a higher risk of overdose. ²³ , ²⁴ Furthermore, referencing the studies by Mizuno et al., the reported regimens were administered twice daily, and when converted to mg/kg or BSA terms, the median daily doses were comparable to or lower than the mg/kg‐based doses predicted by our population PK model. These findings support that our proposed mg/kg‐based regimen provides equivalent systemic exposure while preserving patient safety across the population.

While the proposed dosing regimen provides guidance for the majority of paediatric patients included in this study, certain age groups were not represented in the dataset. Simulations outside the observed age range, including newborns and infants younger than 3 months, were not performed to avoid extrapolation beyond the available data. Consequently, the model cannot directly inform dosing in these very young patients, which represents a limitation of the present study. Further studies including these age groups are warranted to extend the applicability of the dosing recommendations.

Our PK model was developed using data from a single hospital, which may limit its generalizability. Therefore, to further assess the robustness and predictive performance of the model, external validation was conducted using data from diverse cohorts and the established literature. This will allow for a more comprehensive evaluation of the accuracy and applicability of the model across different populations and clinical settings.

This study established a robust and reliable population PK model of sirolimus in paediatric patients with vascular anomalies by integrating prospective and retrospective data. Our findings suggested that optimal sirolimus dosing regimens in paediatric patients with vascular anomalies should be based on growth represented by body weight, maturation reflected by PMA and haemoglobin levels, as these factors are important for achieving target drug concentrations.

AUTHOR CONTRIBUTIONS

Conceptualization: all authors; methodology: J. K., K. H. Y., H. Y. J., S. J., H‐y. Y., J‐w. C., S. L.; formal analysis: J. K., K. H. Y., H. Y. J., S. J., W. J., J. J., H‐y. Y., J‐w. C., S. L.; data curation: J. K., K. H. Y., H. Y. J., S. J., W. J., J. J., H. J., K‐B. R., J‐Y. H.; original draft: S. J., H‐y. Y., J‐w. C., S. L.; review and editing: all authors; supervision: H‐y. Y., J‐w. C., S. L., K. H. Y.; project administration: H‐y. Y., J‐w. C., S. L., K. H. Y., J. K., K. S. K.; funding acquisition: S. Y.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

CLINICAL TRIAL REGISTRATION

This clinical trial was conducted as part of a government‐funded, investigator‐initiated research program. Although the trial was not registered in a public clinical trial registry, it was approved by the Institutional Review Board of the study site and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines.

Supporting information

Figure S1. Model scheme of Population pharmacokinetic model for sirolimus. Ka; absorption rate constant, Vc/F: apparent central volume of distribution, Vp/F: apparent peripheral volume of distribution, CL/F: apparent clearance, Q/F: apparent intercompartment clearance, TM50: postmenstrual age at which 50% of adult clearance is reached, Hill: Hill coefficient, Hgb: Hegmoglobin level, ηi: difference of an individual parameter from the population average.

Figure S2. Goodness of fit plot for the final model: (A) DV vs. PRED, (B) DV vs. IPRED, (C) IWRES vs. IPRED, (D) CWRES vs. TIME. DV: dependent variable (observed concentration), IPRED: individual prediction, PRED: population prediction, IWRES: individual weighted residual, CWRES: conditional weighted residual.

Figure S3. Percentiles of target range attainment across different doses: (A) Therapeutic level, (B) Subtherapeutic level, (C) Toxicologic level.

Table S1. Summary of numeric predictive check.

Jeong S, Kim J, Jung W, et al. Optimal dosing for vascular anomalies paediatric patients with population pharmacokinetic model of sirolimus. Br J Clin Pharmacol. 2026;92(9):3115‐3123. doi: 10.1002/bcp.70595

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.