Development and Use of a Population Pharmacokinetic Model for Characterizing the Pharmacokinetics of Vonoprazan in Pediatric Patients

ABSTRACT

Gastroesophageal reflux disease affects about 20% of children. While proton pump inhibitors are a standard treatment, vonoprazan offers more rapid, potent, and prolonged gastric acid suppression in adults. Despite its promising pharmacological properties, vonoprazan is not yet approved for pediatric use. We aimed to develop and iteratively refine a population pharmacokinetic model to guide pediatric dosing for vonoprazan, characterize its pharmacokinetics in children and adolescents, and evaluate drug exposure across age groups. We developed a two‐compartment linear pharmacokinetic model using data from eight adult (18–54 years) studies, one adolescent (≥ 12 to 17 years) study, and one study in children (≥ 6 to < 12 years). The model incorporated delayed absorption via three transit compartments parameterized by mean transit time, along with estimates for absorption rate constant and relative bioavailability. To capture variability, we quantified between‐subject differences in clearance and central volume, accounting for correlation, and applied a combined additive and proportional error model to residual variability. Investigating relevant covariate effects allowed us to refine the model further. The model structure remained unchanged as pediatric data were added. The most significant parameter shifts occurred in the age effect on absorption rate and the weight effect on central volume. Predictions aligned closely with observed vonoprazan concentrations across age groups. Model‐based simulations demonstrated consistent steady‐state exposure among adults, adolescents, and children, with simulated pediatric doses aligning with the reference adult dose. In conclusion, the final two‐compartment model successfully predicted vonoprazan pharmacokinetics and confirmed dose selection in children and adolescents aged 6 to 17 years.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    Proton pump inhibitors are the standard treatment for pediatric gastroesophageal reflux disease, but their acid‐suppressing effects vary. Vonoprazan, a potassium‐competitive acid blocker, offers more potent and sustained acid inhibition and is approved for adults. Regulatory guidance supports the use of PopPK modeling and adult‐to‐pediatric extrapolation to inform pediatric dosing
• What question did this study address?

  • ○
    Can an iterative, popPK model‐informed approach using adult, adolescent, and pediatric data accurately characterize vonoprazan pharmacokinetics and support pediatric dose selection across age groups?
• What does this study add to our knowledge?

  • ○
    A two‐compartment PopPK model described vonoprazan pharmacokinetics in individuals aged 6 to 54 years, with no age‐ or weight‐related differences in clearance. Simulations predicted that 10‐ and 20‐mg doses in children aged ≥ 1 to < 6 years achieve adult‐comparable exposures, while lower doses may be needed for infants aged 1 month to < 1 year.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    This modeling framework supports efficient pediatric dose selection, reduces reliance on pediatric clinical trials, and informs regulatory decision‐making for vonoprazan.

Sequential population pharmacokinetic (PopPK) model development for vonoprazan in pediatric subjects. Adult data were used to establish the initial model, which was sequentially updated with data from adolescents (Study VPED‐102) and children (Study VPED‐103) to characterize vonoprazan pharmacokinetics and support pediatric dose selection.

1. Introduction

Gastroesophageal reflux disease (GERD) occurs when the reflux of acidic gastric contents causes troublesome symptoms and/or complications [1]. While symptoms can vary by age, heartburn and regurgitation are the most characteristic symptoms of GERD [1]. Infants and young children may have less specific presentations, including vomiting, feeding refusal, abdominal pain, or excessive crying. GERD affects around 20% of children and adolescents, highlighting its clinical relevance [1, 2, 3].

Proton pump inhibitors (PPIs) are the current mainstay of pharmacologic treatment for GERD in children and adolescents [4], demonstrating greater efficacy than antacids and H₂‐receptor antagonists [5]. According to the latest guidelines from the European Society for Pediatric Gastroenterology, Hepatology and Nutrition (ESPGHAN) and the North American Society for Pediatric Gastroenterology, Hepatology and Nutrition (NASPGHAN), PPIs should be the first‐line treatment for reflux esophagitis and typical GERD symptoms in infants and children [6].

In adults, vonoprazan, a potassium‐competitive acid blocker (P‐CAB), has provided more rapid, potent, and durable suppression of gastric acid secretion than conventional doses of PPIs [7, 8]. As of June 2026, vonoprazan is the only P‐CAB approved for use in the United States. For adults, authorized indications include the healing of erosive esophagitis (EE) and relief of heartburn associated with EE; maintenance of healing of all grades of EE and relief of heartburn associated with EE; relief of heartburn associated with non‐erosive GERD; and as part of Helicobacter pylori treatment regimens. Vonoprazan undergoes extensive metabolism, primarily via hepatic pathways, with only a small portion of the unchanged drug excreted in urine [9]. Vonoprazan is a potential alternative to PPIs for treating GERD in adolescents and children. However, as of June 2026, it is not approved for pediatric use in any country.

The approval and subsequent use of vonoprazan in infants, children, and adolescents involves addressing the well‐understood challenges common to many pediatric clinical trials. These include ethical and practical constraints, small sample sizes (which may limit the generalizability of study results), and developmental changes across age groups (such as potential differences in drug absorption, metabolism, or clearance) [10]. Extrapolation of efficacy data from adults to pediatric populations is generally considered acceptable in drug development, particularly when the disease pathophysiology and pharmacologic response, such as with antisecretory agents for GERD, are sufficiently similar across age groups. In this context, pharmacokinetic bridging studies between adult and pediatric populations are crucial for informing and supporting the development of appropriate pediatric dosing. In GERD, the similarity in disease characteristics and treatment response supports the use of adult efficacy data to guide pediatric drug development.

Alongside efficacy extrapolation, regulatory agencies emphasize the importance of population‐pharmacokinetic (popPK) modeling in pediatric drug development and pharmacokinetic assessments [10, 11]. PopPK modeling is a statistical method that describes key pharmacokinetic parameters as typical population values while also accounting for individual differences. It also includes important covariates, including age, weight, and sex [12, 13]. This approach enables the prediction of drug concentrations, optimization of dosing regimens, and analysis of exposure dynamics. PopPK modeling is especially valuable in high‐variability populations and specialized groups, including pediatric patients.

The objective of this analysis was to develop and iteratively update a popPK model to support pediatric development of vonoprazan. This model was designed to inform dose selection, characterize pharmacokinetics in children and adolescents, and confirm drug exposure levels across age groups.

2. Materials and Methods

All studies included in this analysis were conducted in accordance with relevant national regulations, International Conference on Harmonization Good Clinical Practice guidelines, and the principles of the Declaration of Helsinki [14, 15]. All studies received ethical approval. Informed consent was obtained from all participants or their representatives as outlined in each study protocol. This analysis was conducted in accordance with guidance from the US Food and Drug Administration and the European Medicines Agency [16, 17, 18].

2.1. Clinical Studies and Samples

We used vonoprazan pharmacokinetic data from studies in healthy subjects and patients with GERD in this analysis. Pharmacokinetics are not expected to differ substantively between patients with GERD and healthy individuals, as disease status was not a significant covariate in prior popPK analyses [19]. A previous popPK model included data from healthy adult, special population, and food effect studies [19]. For this analysis, the dataset was restricted to healthy adults for development of the initial model (Model 1) in order to focus on developmental changes.

Our complete dataset included information from eight studies in adults aged 18 to 54 years, one study in adolescents (≥ 12 to 17 years) with GERD (NCT05343364), and one study in children (≥ 6 to < 12 years) with GERD (NCT06106022). Details on individual studies are presented in Table S1. Summary statistics of continuous patient characteristics are located in Table S2. Body weight distribution is shown in Figure S2; race‐stratified individual random effects are shown in Figure S3. Model dataset summary of the pediatric studies included is described in Table S3. Detailed pharmacokinetic and safety results from the pediatric studies have been reported separately [20].

Any missing pharmacokinetic observations, including actual time, doses, or concentrations, were removed from the analysis dataset. Any samples collected before the first dose were designated time 0.

2.2. Population Pharmacokinetic Model Development

2.2.1. General Approach

We developed a popPK model for adults (Model 1), which was then used to support study planning for VPED‐102, the study in adolescents (12 to 17 years), by simulating exposure in adolescents to guide dose selection. After completion of the adolescent study, we updated Model 1 with the newly collected data (Model 2) and used it to evaluate the pharmacokinetic data in adolescents. We also used the updated model to support study planning for VPED‐103 by simulating exposure in children aged ≥ 6 to < 12 to guide dose selection. As in the previous iteration, we subsequently updated the model with the newly collected data from children (Model 3) and used this model to predict exposure levels in infants and children aged 1 month to < 6 years (VPED‐104). (Figure 1).

FIGURE 1.

Timeline of the pediatric development for vonoprazan.

2.2.2. Modeling Methodology

Based on previous experience and graphical data exploration, we started the nonlinear mixed effects model development with a linear two‐compartment model with a delayed oral absorption. We tested several absorption and distribution models as structural model alternatives. The variability model consisted initially of between‐subject variability terms on all major model parameters without correlation (i.e., we used a diagonal Omega matrix) and a combined additive and proportional error model to characterize the within‐subject or residual variability. When indicated by correlated individual parameter estimates, we tested estimation of off‐diagonal elements of the Omega matrix. For parameters with very low between‐subject variability or substantial shrinkage, we tested removing the respective random effect by setting the respective diagonal element of the Omega matrix to 0.

All model parameters were implemented, assuming a log‐normal distribution by means of the typical value and the variance of the (exp‐transformed) random effect $X={\mathit{TV}}_X·\exp \left({\mathit{ETA}}_X\right)$.

Because there were numerous pharmacokinetic samples with concentrations below the limit of quantification, we used the so‐called M3 method to use this censored information for the parameter estimation [21]. A BLQ sensitivity analysis is shown in Table S4.

After a base model was developed, we tested extrinsic and intrinsic covariate candidates: dose (to characterize nonlinear pharmacokinetics), time (for temporal changes in pharmacokinetic parameters), sex, weight, and age. We implemented continuous covariate effects as power models centered at 20 mg for dose, 70 kg for weight, and 18 years for age. For example, for testing a body weight effect on parameter $X$, we modified the fixed typical value ${\mathit{TV}}_X$ to become weight‐dependent ${\mathit{TV}}_X·{\left(\mathit{\text{weight}}/70\mathrm{kg}\right)}^{{\mathit{WGT}}_X}$, where ${\mathit{WGT}}_X$ is the to be estimated power coefficient. For categorical covariates, we implemented a percent‐change model. For example, for testing a gender effect (for females) on parameter $X$, ${\mathit{TV}}_X$ was modified to ${\mathit{TV}}_X·\left(1+\frac{{\mathit{FEM}}_X}{100\%}\right),$ where ${\mathit{FEM}}_X$ is the estimated gender effect for females in percent. We also tested other covariate models when appropriate. We added covariate effects to the model based on statistical (e.g., likelihood ratio test indicates a significant improvement at a 0.05 level, standard error of the estimated parameter was reasonably small, and no substantial convergence problems) and biological considerations (covariate effect results in a clinically meaningful change in the pharmacokinetic parameter, biological plausibility). A particular focus was on identifying potential age or weight effects that may suggest exposure changes for the next younger age cohort. No formal, strict covariate screening procedure was followed. The model code is provided in the Supporting Information Model Code.

2.3. Model Assessments

We critically assessed key models (base and final) in terms of parameter precision and shrinkage, goodness of fit plots, assessment of random effects, and their ability to reproduce the data used by means of visual predictive checks.

2.3.1. Posterior Predictive Checks

We used the previously established final model to evaluate alignment with the data from each subsequent study (adolescents and then children). For this purpose, we performed a visual predictive check using the previously established final model to simulate the new study data. These simulations were then compared with the observed data from the new study to identify any deviations. If discrepancies in predictive performance assessments were identified, model updates were performed. These included re‐estimation of fixed and random effect parameters, refinement of the covariate model (adding, removing, or changing covariate effects), and modification of the variability model (adding, removing, or changing components of the between and/or within‐subject variability model). At each step, we investigated if potential age‐ or weight‐dependent changes in pharmacokinetic parameters were needed to capture trends in the data, particularly in the pediatric subpopulation. We fitted the model to the pooled dataset to update the parameter estimates.

2.4. Predicting Pediatric Exposure

2.4.1. Adolescents and Children (6 to 17 Years)

We predicted exposure in adolescents using simulations from adult data (Model 1). Subsequently, we predicted exposure in children aged ≥ 6 to < 12 years using simulations from adult and adolescent data (Model 2). For each simulation, we created a virtual population of pediatric subjects (50:50 boys and girls) matching the target age for each study (VPED‐102 and VPED‐103). Body weight was randomly sampled from the CDC normal growth statistics characterizing sex and age‐matched body weight distributions.

2.4.2. Infants and Children (1 Month to < 6 Years)

Predictions for infants and young children should account for expected effects of organ maturation on the exposure to vonoprazan. For the purpose of predicting exposure in this population, we extended Model 3 to incorporate an age‐dependent maturation factor F _mat of organs [22]. The formula below depicts a generic maturation function as previously described [23].

$$F_{\mathrm{mat}}=\frac{\mathit{pm}{a}^{3.4}}{{47.3}^{3.4}+\mathit{pm}{a}^{3.4}}$$

where F _mat is the organ maturation factor, pma is postmenstrual age (in weeks), 47.3 represents the postmenstrual age at which 50% of maturation is achieved, and 3.4 is the Hill coefficient describing the steepness of the maturation function. The postmenstrual age (pma) in weeks was derived from the chronological age in years (age) as pma = age $·$ 365.25/7 + 42. From this, the age‐dependent clearance was derived as CL_mat = F _mat $·$ CL.

For these simulations, we created a virtual study population of 50% boys and 50% girls aged 1 month to < 6 years in steps of 1 month with 100,000 subjects each, for a total of 14,200,000 virtual subjects. We sampled appropriate age and gender matched body weight from the WHO/CDC normal growth statistic. Subjects were subsequently grouped into age ranges (1 to < 3 months, 3 to < 6 months, 6 to < 9 months, 9 to < 12 months, 12 to < 18 months, 18 months to < 2 years, 2 to < 3 years, 3 to < 4 years, 4 to < 5 years, and 5 to < 6 years).

2.4.3. Predicting Exposure Matched Pediatric Doses

The area under the curve (AUC) closely characterized the effects of vonoprazan, so we chose pediatric doses to match the adult reference AUC on average [24]. We started with a set of candidate pediatric doses ranging from 1 mg to 25 mg to predict ${\mathit{AUC}}_{\mathit{ss}}$. The dose where the median predicted pediatric exposure was closest to the median predicted adult exposure was defined as equivalent pediatric dose. For this purpose, ${\mathit{AUC}}_{\mathit{ss}}$ was derived as $\mathit{\text{dose}}·F_{\mathit{rel}}/\mathit{CL},$ where $F_{\mathit{rel}}$ was the model‐predicted relative bioavailability and $\mathit{CL}$ was the model‐predicted elimination clearance.

2.5. Software

Nonlinear mixed effects models were fitted to data using the nonlinear mixed effects model program (NONMEM) version 7.5.1 (ICON Development Solutions, Gaithersburg, Maryland) [25], compiled with GNU Fortran 12.2.0. Figures and tabulations were produced using R version 4.2.2 [26].

3. Results

3.1. Final Adult Model (Model 1)

The final adult model was a two‐compartment linear model with delayed oral absorption. (Figure 2).

FIGURE 2.

Schematic of the final adult model (Model 1). Compartments are shown as gray boxes. The line‐type indicates unobserved (dashed) or observed (solid) compartments. Model parameters are shown in orange, with asterisk (*) when estimated on patient level (with BSV). Covariate effects are shown in blue.

We implemented the absorption delay as a series of three delay compartments parameterized by a mean transit time (MTT), while all three transfer rates were derived as 4/MTT. A first‐order absorption from the third delay into the central compartment was characterized by an absorption rate constant (k_a). A relative bioavailability parameter F_rel was estimated to account for relative differences in the absorbed drug amount. We fixed the typical value to 1 because no intravenous data were available, which would have allowed the estimation of absolute bioavailability. Drug elimination was characterized using a linear elimination clearance (CL), drug distribution using a central volume of distribution (V_c), distribution clearance was between the central and peripheral compartments (Q), and a peripheral volume of distribution (V_p) was also assessed. Between‐subject variability (BSV, random effects) was estimated for CL and V_c (with correlation), MTT, and k_a. The within‐subject (residual) variability was characterized using a mixed additive and proportional error model.

We identified the following covariate effects to statistically improve the model fit, with numerical estimates referring to the latest available model (Model 3) as shown in the rightmost column of Table 1: a slight over‐proportional bioavailability was implemented as a dose‐effect on F_rel using a power model with an exponent of 0.29; the relationship between weight and the volume of distribution was implemented as a power model on V_c with an exponent of 0.668; gender effects were implemented as a 49% slower k_a and a 35.1% higher fraction absorbed (F_rel) for females; the absorption was found to be slower in younger subjects, which was implemented as a covariate effect of age on k_a using a power model with an exponent of −0.693; and time‐dependent pharmacokinetic changes were implemented as effects of study day on CL, V_c, and k_a. Because plasma concentrations were mainly observed on Day 1 and again on Day 7, we could not identify a continuous relationship but instead used a binary change for day > 1. Final parameter estimates for this model are in Table 1 (Adults).

TABLE 1.

PopPK parameter estimates for all three models developed.

Parameter	Role	Adults (18 to 54 years)		Adolescents (12 to 17 years)		Children (≥ 6 to < 12 years)
		Estimate	RSE (%)	Estimate	RSE (%)	Estimate	RSE (%)
Relative bioavailability (Frel)	TV	100	(fixed)	100	(fixed)	100	(fixed)
	Dose‐effect (power)	0.298	4.6	0.299	4.6	0.290	4.8
	Female‐effect (%)	41.0	10.4	39.8	10.4	35.1	10.8
Mean transit time (MTT)	TV (h)	0.774	3.9	0.774	4.2	0.762	3.3
	BSV a	0.542	4.8	0.542	10.4	0.526	4.1
	BSV b	58.4	—	58.4	—	56.4	—
Absorption rate (ka)	TV (1/h)	3.18	8.5	3.18	8.4	3.08	7.7
	Female‐effect (%)	−49.0	15.2	−49.2	14.7	−49.0	12.9
	Age‐effect (power)	−1.02	18.2	−1.02	17.7	−0.693	10.6
	Day > 1‐effect (%)	−18.4	12.0	−18.4	11.7	−20.8	9.9
	BSV a	0.601	6.3	0.601	7.9	0.643	7.7
	BSV b	66.0	—	66.0	—	71.5	—
Elimination clearance (CL)	TV (L/h)	117	2.4	117	2.3	118	2.2
	Day > 1‐effect (%)	−11.3	3.6	−11.3	3.5	−11.5	3.5
	BSV a	0.349	4.3	0.349	3.5	0.366	4.0
	BSV b	36.0	—	36.0	—	37.8	—
	Cor (central volume) a	0.932	1.0	0.932	0.9	0.913	1.3
Central volume (Vc)	TV (L)	747	3.2	746	3.0	751	2.8
	Weight‐effect (power)	0.577	11.4	0.576	9.9	0.668	6.6
	Day > 1‐effect (%)	−10.7	13.0	−10.7	12.5	−11.6	11.4
	BSV a	0.392	5.4	0.391	4.2	0.389	5.0
	BSV b	40.7	—	40.7	—	40.4	—
Distribution clearance (Q)	TV (L/h)	50.2	4.3	50.2	4.8	49.8	4.8
Peripheral volume (Vp)	TV (L)	275	2.1	275	2.2	271	2.2
Residual variability	Prop (%)	24.3	0.4	24.3	0.4	24.7	0.4
	Add (ng/mL)	0.001	(fixed)	0.001	(fixed)	0.001	(fixed)

^a Shown as standard deviations/correlations as reported by NONMEM.

^b Shown as coefficient of variation derived using $100\%·\sqrt{e^{w^2}-1}$ (without RSE because this is a derived parameter). Although the model structure remained consistent across updates, all model parameters were re‐estimated after each extension of the analysis dataset and compared across all stages of model development.

Scatter plots of observed concentrations versus predicted concentrations from the adult model (Model 1) are presented in Figure 3a.

FIGURE 3.

Observations vs. predictions. Scatterplots of observed vs. predicted pharmacokinetic concentrations in all three models developed. The left panels show population predictions (PRED) and the right panels show individual predictions (IPRED). Dots are color coded to indicate the respective study population.

Figure S1 shows prediction corrected visual predictive checks (pcVPCs) for Model 1, stratified by age and weight quartiles.

Model 1 included data across the age and weight range listed in Table S2, and model predictions across the 4 quartiles did not show a trend for either age or weight and therefore confirmed the model was suitable to be used for a moderate extrapolation towards lower aged adolescents with lower body weights.

3.2. Adolescents

3.2.1. Simulating Exposure in Adolescents

We used the adult model to simulate exposure in adolescents aged 12 to 17 years. No age‐ or weight‐dependent changes in clearance were identified in Model 1. As expected, the simulations showed that no dose adjustments were required to match the predicted exposure in adolescents with the target exposure in adults at both intended dose levels (10 mg and 20 mg).

3.2.2. Model 2 Development

After the adolescent study (VPED‐102) was completed, we performed a posterior predictive check (PPC) to assess if observed PK concentrations in adolescents would align with the predictions for adolescents using the adult model. For this purpose, we used the adult model and the observed subject characteristics to simulate PK concentrations for each observation in the adolescent study. The resulting simulations were graphically compared with observed pharmacokinetic concentrations. Results are shown in Figure 4a.

FIGURE 4.

Posterior predictive check (PPC) for the respective new data. The PPCs show summary statistics (5th, 50th [median] and 95th percentiles) for both the observed and simulated pharmacokinetic concentrations. Observations are shown as black dots and their summary statistics as lines (dashed black for the outer percentiles and solid orange or blue for the medians). The 90% confidence intervals for the simulation percentiles are shown as shaded areas in the same color. A prediction correction method was applied as described in Bergstrand et al. [27].

No systematic deviations between the model predictions and observed data were identified, which is why the model was left unchanged. All parameters were re‐estimated based on the pooled dataset of adults and adolescents. The resulting model was referred to as Model 2. A scatter plot of observations versus predictions for adolescents (orange dots) and adults (black dots) is included in Figure 3b. Final parameter estimates for this model are included in Table 1 (Adolescents).

3.3. Children

3.3.1. Simulating Exposure in Children

We used the adult and adolescent model to simulate exposure in children aged ≥ 6 to < 12 years. No age‐ or weight‐dependent changes in clearance were identified in Model 2. As expected, the simulations confirmed that no dose adjustments were required to match the predicted exposure in children aged ≥ 6 to < 12 years with the target exposure in adults at both intended dose levels (10 mg and 20 mg).

3.3.2. Model 3 Development

After the study in children aged ≥ 6 to < 12 years (VPED‐102) was completed, we performed a posterior predictive check to assess if observed pharmacokinetics in children of this age would align with the predictions for children using the adult and adolescent model (Model 2). We used the observed subject characteristics to simulate PK concentrations for each collected pharmacokinetic sample and we then graphically compared the resulting predictions with observed pharmacokinetic concentrations. Results are shown in Figure 4b, which shows a pcVPC that overlays predictions from Model 2 with observations from children (VPED‐103).

As before, no systematic deviations between the model predictions and observed data were identified, which is why the model was left unchanged. All parameters were re‐estimated based on the pooled dataset of adults, adolescents, and children. The resulting model was referred to as Model 3. A scatter plot of observations versus predictions for children (blue dots) along with those for adolescents (orange dots) and adults (black dots) is included in Figure 3c. Final parameter estimates for this model are included in Table 1 (Children).

3.3.3. Exploration of Age and Weight Affects

To explore potential differences in exposure between adults, adolescents, and children, predicted dose‐normalized exposures (AUC_ss,DN20) were plotted versus age and weight (Figure 5). A potential trend of decreasing AUC_ss,DN20 with decreasing age for the adolescent (VPED‐102) population was observed; however, for children, the opposite was observed, with a potential trend of increasing AUC_ss,DN20 with decreasing age. Nevertheless, the wide 95% prediction interval of the slopes indicated high uncertainty for both of these slopes, and there was no obvious trend with age across adolescents and children. For body weight, there was no obvious trend with AUC_ss,DN20 for adolescents (VPED‐102) either. However, for children, a potential trend was observed, with AUC_ss,DN20 increasing with decreasing body weight. As with the potential trends observed with age, the wide 95% prediction interval of the slope indicated high uncertainty for this slope, and there was no obvious trend with weight across the entire population. Consequently, we chose not to update the model to include age or weight effects on vonoprazan clearance.

FIGURE 5.

Model‐predicted individual steady state AUC at 20 mg. Scatterplot of model‐predicted individual steady state AUC at 20 mg versus age (upper panel) or body weight (lower panel). Each dot represents a subject, color coded by study population. For each population, we fitted a linear regression model which is illustrated as a solid line with 95% confidence intervals shown as shaded areas.

3.3.4. Predicting Exposure for Infants and Children Aged 1 Month to < 6 Years

Similar to the previous models, we did not identify any age‐ or weight‐dependent changes in clearance in Model 3, which included children aged ≥ 6 to < 12 years along with the adults and adolescents. Given that organ maturation is expected to affect vonoprazan exposure, predicting exposures in infants and young children required a model extension, as detailed in the Methods section. We simulated steady state exposures for a set of pediatric candidate doses and compared them with steady state adult reference exposures. For subjects aged ≥ 1 to < 6 years, vonoprazan doses of 10 mg and 20 mg were predicted to result in exposures nearly equivalent or just slightly higher than those in adults, with median predictions nearly the same for each age group in this range (Figure 6). For subjects aged 1 month to < 1 year, the simulations suggested that 10 mg and 20 mg doses would likely provide exposures higher than the target adult exposures.

FIGURE 6.

Predicted pediatric exposure to match adult reference doses (a) 10 mg and (b) 20 mg. Steady state exposures were simulated for a set of pediatric candidate doses and compared with steady state adult reference exposures (AUC_ss = 92 h $·$ ng/mL [50.4–168] for vonoprazan 10 mg and 225 h $·$ ng/mL [123–411] for vonoprazan 20 mg). Simulation results were grouped into age ranges and summarized as medians (red dots) and 90% prediction intervals (red error bars). The respective adult reference exposure ranges are shown as black line (median) and 90% prediction intervals (gray shaded area).

4. Discussion

Best practices in pediatric pharmacokinetic modeling include an iterative “learn and confirm” approach, in which models are continuously refined as new data become available [28]. Model‐informed drug development involves leveraging adult data to predict pediatric responses [29]. In the present analysis, this approach was implemented by first developing a model based on adult data, followed by two subsequent updates incorporating pediatric data. The pediatric popPK model described here provides a solid framework for analyzing future pediatric clinical trial data and validating insights from existing pediatric studies.

The final model adequately described the dataset, based on consideration of physiological plausibility, goodness‐of‐fit criteria, and precise parameter estimation. Including MTT and accounting for time‐dependent bioavailability changes enhanced the model's fit and parameter accuracy. The diverse age and body weight ranges also contributed to better pediatric model predictions. Although extrapolation of efficacy data from adults to children and adolescents is acceptable when disease pathophysiology and exposure response are comparable—such as for antisecretory agents for GERD—important age‐related differences must still be considered. This is particularly relevant when comparing vonoprazan to PPIs. Children have consistently shown higher weight‐normalized clearance and reduced systemic exposure (AUC) for multiple PPIs including omeprazole, lansoprazole and pantoprazole at equivalent mg/kg doses as adults, due to age‐dependent maturation of hepatic enzymes such as CYP2C19 [30, 31]. Consequently, children often require higher per kg doses than adults to achieve comparable plasma exposures. The same trend was not observed with vonoprazan: observed data in subjects aged ≥ 6 years to adults demonstrates no apparent age‐ or weight‐related differences in systemic exposure, and model‐based simulations suggest that fixed doses of 10 mg and 20 mg provide systemic exposures in children aged ≥ 1 to < 6 years that are similar to those observed in adults. While gastric acid secretion reaches adult levels by approximately 3 years of age, the underlying mechanisms of reflux can differ in younger children [32]. Gastroesophageal reflux (as opposed to GERD) is a physiologic event during infancy that is associated with developmental immaturity of the gastroesophageal junction; it typically resolves spontaneously by 1 to 2 years of age. The described simulations suggest that vonoprazan doses of 10 and 20 mg will be appropriate in children aged ≥ 1 to < 6 years, while lower doses may be appropriate for infants aged 1 month to < 1 year as a result of the CYP enzyme maturation factors, which were added to the model. Updating the popPK model when pharmacokinetic data from subjects aged ≥ 1 to < 6 years is available and subsequently simulating steady state exposures in subjects aged 1 month to < 1 year will provide additional confidence in selecting the appropriate doses for infants aged 1 month to < 1 year.

These findings highlight the value of a model‐based approach to pediatric drug development. This strategy balances scientific rigor and ethical responsibility. Utilizing a stepwise modeling framework, we minimized pediatric exposure to any unnecessary risk while still enabling gradual extrapolation from adult to pediatric populations. Approaching popPK modeling in an iterative manner improves data utilization and model stability; leveraging real‐world data will allow for more accurate predictions and more confident pediatric dosing recommendations.

Despite this, our study has several limitations. Our modest pediatric sample size reflects the ethical and practical realities of pediatric research. The pediatric data were not intended to independently define model structure but to inform and evaluate extrapolation from adults within an integrated framework. While some subjects underwent dense plasma sampling, we pivoted to sparse sampling to reduce patient burden and ensure retention. As these studies are the first to examine vonoprazan pharmacokinetics in children and adolescents, a greater age distribution will be needed in order to align with regulatory guidance to capture developmental changes, address recruitment challenges, and enhance the model's ability to predict dosing across younger age groups.

5. Conclusion

The final two‐compartment model provides a robust foundation for characterizing the pharmacokinetics of vonoprazan in subjects aged 6 to 17 years and adults. The iterative model‐building processes demonstrated that exposures in adolescents or children receiving 10 or 20 mg of vonoprazan were comparable to those observed in adults. Extending this popPK model to predict pharmacokinetics in younger children and infants represents a logical and necessary next step. This model provides a strong platform for future development to include patients aged 1 month to < 6 years, which would enable the comprehensive characterization of vonoprazan exposure across an even wider age range.

Author Contributions

All authors wrote the manuscript. D.J.M., A.F., G.W., and T.W. designed and performed the research. D.J.M., A.F., G.W., T.W., E.L., and C.S. analyzed the data.

Funding

This study was supported by Phathom Pharmaceuticals.

Conflicts of Interest

D.J.M., G.W., and E.L. are employees of Phathom Pharmaceuticals; A.F. and T.W. of thinkQ² were funded by Phathom Pharmaceuticals to undertake this work; C.W.H. is a consultant and speaker for Phathom Pharmaceuticals and a consultant for Sebela; C.S. has served as a consultant, an advisory board member, and a speaker for Phathom Pharmaceuticals, and as a speaker for Takeda.

Supporting information

Table S1: Studies used in the population pharmacokinetic analysis.

Table S2: Summary statistics of continuous subject characteristics.

Table S3: Model dataset summary of pediatric studies included in the population pharmacokinetic analysis.

Table S4: Comparison of parameter estimates from the final population pharmacokinetic model using the M3 method to incorporate observations below the lower limit of quantification (BLQ) versus a sensitivity analysis excluding BLQ data (“W/o M3”). Approximately 5% of pharmacokinetic observations were BLQ. Absolute and percent differences are shown. Structural parameters (e.g., elimination clearance, central volume of distribution, distribution clearance, peripheral volume) are largely unchanged between approaches, while absorption‐related parameters show larger differences, as expected because BLQ observations occur predominantly during absorption and terminal elimination phases. Exposure predictions and covariate conclusions were not meaningfully affected.

Figure S1: Prediction corrected visual predictive checks in adults stratified by age and weight quartiles. The visual predictive checks (pcVPCs) show summary statistics (5th, 50th [median] and 95th percentiles) for both the observed and predicted PK concentrations. Summary statistics for observations are shown as lines (dashed black for the outer percentiles and solid red for the medians), and 90% confidence intervals for the summary statistics for predictions are shown as shaded areas in the same color. A prediction correction method was applied as described in Bergstrand et al. [32] pcVPCs were stratified into age (upper panels) or weight (lower panels) quartiles with approximately equal number of subjects.

Figure S2: Distribution of body weight across age groups (adults, adolescents, and children) included in the population pharmacokinetic analysis. Each dot represents an individual subject; box plots show the median and interquartile range. A single adolescent 15 years of age with a body weight of 132 kg is at the upper extreme of the adolescent distribution and contributed to the higher mean weight observed in this age group. Median body weights were 66 kg in adults, 64 kg in adolescents, and 32 kg in children.

Figure S3: Distributions of individual random effects (ETAs) for elimination clearance (CL), central volume of distribution (V_c), mean transit time (MTT), and absorption rate constant (k_a) stratified by race (Asian, Black, Other, White). Box plots show median and interquartile range; transformed ETAs (exponentiated) are shown on the y‐axis. Overlapping central tendencies and variability across regions indicate that race was not associated with clinically meaningful differences in pharmacokinetic parameters. Race was formally tested as a covariate during prior adult population pharmacokinetic model development and was neither statistically significant nor clinically meaningful; therefore, it was not retained in the present analysis.

Data S1: psp470291‐sup‐0002‐supinfo.docx.

Data S2: psp470291‐sup‐0003‐supinfo.pdf.

Data Availability Statement

Individual participant data will not be shared.