Background: Immunosuppressants like tacrolimus prevent transplant rejection but show high variability in pediatric populations. Traditional bioequivalence (BE) assessments rely on total AUC, but partial AUC (pAUC) may better evaluate early drug absorption.

Aims: This study aimed to assess BE using pAUC for tacrolimus in pediatric transplant recipients through pharmacometric modeling and simulation.

Methods: A prospective observational study included 25 pediatric solid organ and 10 hematopoietic stem cell transplant recipients. Pharmacokinetic (PK) data were collected during the absorption phase (0-4 hours post-dose). Population PK modeling using NONMEM 7.4.2 applied a one-compartment model with first-order absorption and elimination. Simulations evaluated inter-individual variability, CYP3A5 polymorphisms, and dosing strategies. pAUC4 and pAUC8 were compared with Cmax and AUCT last using Phoenix WinNonlin.

Results: The PK model described tacrolimus disposition, with absorption rate constant (Ka) of 4.23 hr⁻¹ (RSE 78.5%) and bioavailability of 23.3% (RSE 37.9%). Model simulations demonstrated that CYP3A5 expressers exhibited significantly higher variability in pAUC compared to non-expressers. Dose-normalized median pAUC4 was 7.44 hr*ng/mL/mg, and pAUC8 was 7.95 hr*ng/mL/mg. pAUC metrics exhibited strong correlation (r=0.99, p<0.0001) but showed distinct patterns compared to Cmax and AUCT last. Adjusted dosing for CYP3A5 expressers reduced variability and improved BE assessment.

Conclusions: This study supports pAUC as an additional BE metric in pediatric immunosuppressant therapy. Simulations highlight CYP3A5 polymorphisms' impact on BE, suggesting dosing adjustments. Findings support further validation and potential regulatory integration of pAUC. Pharmacometric modeling aids drug assessment and therapeutic equivalence in vulnerable populations.

Key Words: bioequivalence, pharmacometrics, partial AUC, pediatric transplants, CYP3A5, tacrolimus

Background
Model-informed precision dosing (MIPD) optimizes the initial dosing regimens to achieve therapeutic targets by integrating patient-specific covariates. While this process involves simulations and manual dose selections, reinforcement learning (RL), specifically Deep Q-Networks (DQN), offers the potential to automate and standardize this process by learning optimal strategies across numerous scenarios.
Aims
This study aimed to evaluate the feasibility of applying DQN to automate the selection of initial dosing regimens to achieve predefined PK targets. The study focuses on infliximab as a case example, training the algorithm using model-based synthetic data.
Methods
A pediatric population PK model (Samuels, CTS 2025) was used to simulate infliximab PK. A virtual cohort of 10,000 pediatric patients was generated with random assignments of four key covariates: weight, albumin, erythrocyte sedimentation rate, and neutrophil CD64. The Q-learning algorithm was trained with a reward function designed to achieve infliximab concentrations of 5–10 μg/mL at 14 weeks within a dosing range of 1–20 mg/kg. A DQN model was developed to approximate the relationship between reward and covariate data across different dose levels.
Results
With the standard weight-based dosing (5 mg/kg), only 29.4% of virtual patients achieved the target concentrations. By contrast, using the DQN model, 96.1% of patients reached the target, while 3.9% of patients who did not reach the target concentration even at the highest dose (20 mg/kg).
Conclusion
This study demonstrated that DQN can be used to optimize initial dosing regimens in MIPD. Further model refinement and validation in real-world settings are warranted.

Background
Infliximab is a first-line therapy for pediatric Crohn’s disease (CD). While pediatric-specific population pharmacokinetic (PK) models have been developed for model-informed precision dosing (MIPD), no pharmacokinetic/pharmacodynamic (PK/PD) model has been published to describe the infliximab exposure-response relationship in pediatric CD.
Aims
This study aimed to develop a PK/PD model incorporating fecal calprotectin (FCAL) as a biomarker of gut inflammation to characterize the exposure-response relationship of infliximab.

Methods
A total of 186 pediatric and young adult patients (median age 13.8 years, range 1.4–21.7) were included. A total of 1,091 infliximab concentrations and 497 FCAL measurements were available for PK/PD modeling. Area under the concentration-time curve (AUC) was estimated using the established pediatric PK model (Xiong et al., CPT 2021). An indirect response PK/PD model was developed to describe the relationship between infliximab AUC and FCAL levels using NONMEM.

Results
The PK/PD model, parameterized with baseline FCAL (BASE), Imax, EC50, Kin, and Kout, described FCAL dynamics during infliximab treatment. C-reactive protein was identified as a predictor of FCAL levels for both baseline and time-varying manners (dOFV = - 49.787, p-value < 0.001). Large interindividual variability in EC50 (>170 CV%) suggested substantial variability in exposure-response relationship. Diagnostic plots including visual predictive check (VPC) indicate that the final model has no systemic bias and specifications.

Conclusions
The PK/PD model successfully characterized the dynamic interactions between infliximab exposure and FCAL, demonstrating its potential to support MIPD based on not only PK but also PD targets in pediatric patients with CD.

Keyword
Pediatrics

Background: Accurate initial dosing of intravenous vancomycin requires careful consideration of weight and renal function. In obese patients, body mass index (BMI) further complicates dosing. Although web and mobile tools are available, many Thai clinicians remain unaware of these resources or face barriers to their use. LINE, a widely used messaging platform in Thailand, offers a user-friendly solution for vancomycin dosing support. However, its accuracy requires validation.

Objective: This study aimed to develop and evaluate a LINE-based chatbot for initial vancomycin dosing in Thai adults.

Methods: The study consisted of three phases: (1) chatbot development using the LINE Messaging API and UpToDate guidelines to generate dosing recommendations based on patient demographics, including sex, age, weight, height, and serum creatinine, while also calculating pharmacokinetic parameters; (2) data collection from patients aged ≥18 years who received intravenous vancomycin, including a loading dose and at least two maintenance doses, excluding those undergoing renal replacement therapy; and (3) evaluation of dosing accuracy by comparing predicted and actual trough levels using root mean square error (RMSE) and mean absolute error (MAE).

Results: The chatbot’s dosing recommendations were evaluated in 34 patients, with an average processing time of 31.3 ± 4.4 seconds per case. The predicted trough concentrations demonstrated an RMSE of 2.13 (95% CI = 1.39–2.87) and an MAE of 1.58 (95% CI = 1.10–2.07).

Conclusion: The LINE chatbot provides an accurate and efficient tool for initial vancomycin dosing with minimal prediction error. As a user-friendly tool, it has the potential to improve clinical decision-making.

Background: Model-informed precision dosing (MIPD) is crucial for optimizing busulfan therapy. We previously developed a population pharmacokinetic model using data from 188 children and young adults (Shukla et al. 2020). After implementation, some biases were observed which were potentially related to the covariate effects identified in the original model.

Aims: 1. Assess target attainment of busulfan MIPD in children and adults. 2. Refit the Shukla model with an expanded, multicenter dataset including both children and adults.

Methods: Data were collected from 1062 individuals and 9123 busulfan concentration samples across 11 US medical centers using the InsightRX Nova platform from 2017-2024. The data were modeled using NONMEM and R, comparing performance with the previously published model.

Results: For individuals undergoing busulfan MIPD, 84.4% achieved a cAUC within 80 to 120% of the target. Model refinement included adding a peripheral compartment and an exponential decay function for clearance, while the previously reported regimen effect on clearance was found insignificant. The refined model outperformed the original model in forecasting prediction (mean percent error: -2.9% vs -6.8%; normalized root mean square error: 23.2% vs 27.8%).

Conclusions: MIPD of busulfan improves therapeutic target attainment and continuous refinement of models is essential to improve predictive performance. The refined model represents an improvement in forecasting performance and generalizability across centers and ages.

Background: Therapeutic drug monitoring of adalimumab has been recommended for the management of inflammatory bowel disease.

Aims: This study aimed to investigate the exposure – response relationship of adalimumab in patients with ankylosing spondylitis.

Methods: A prospective cohort study was conducted in patients with ankylosing spondylitis who were initiating adalimumab treatment. Plasma samples were collected prior to the next drug adaministration. Adalimumab concentrations were quantified using an in-house ELISA, and anti-drug antibodies (ADA) were measured using Meso Scale Discovery. Exposure – response modeling was performed using logistic regression.

Results: Ninety-eight patients with 12-24 weeks of follow-up were included, of whom 39 were classified as primary non-responders. Steady-state trough concentrations (Css-min) were significantly higher in primary responders compared to non-responders (6.27 vs. 1.17 μg/mL, P<0.0001). The exposure – response model indicated that a Css-min target of 5.4 μg/mL was associated with an 80% probability of response. Additionally, week 4 trough concentrations (Ctrough-wk4) were higher in primary responders, and a target of 2.5 μg/mL was predictive of primary response (sensitivity 0.81, specificity 0.88). ADA development was observed in 8/41 (19.5%), 38/58 (65.5%), and 90/98 (91.8%) patients at week 2, week4, and during follow-up, respectively. ADA positivity at week 4 was significantly associated with primary non-response (odds ratio=5.5).

Conclusions: A statistically significant exposure – response relationship and an optimal adalimumab concentration target were identified. ADA formation was the primary determinant of low drug concentrations and primary non-response.

Keywords: adalimumab, exposure – response relationship, therapeutic drug monitoring

Background & Aims: In Japan, promoting self-medication is vital for reducing medical costs and extending healthy life expectancy. A broader public understanding of pharmaceuticals is necessary. This study aims to develop an accurate and generalizable pharmacokinetic (PK) model and create a visualization tool for drug kinetics.

Methods: Literature on PK/PD models beyond TDM drugs is limited. We extracted data from the Interview Form graphs to develop PK Model 1 and reconstructed PK Model 2 based on previous research. A PBPK simulation using Simcyp® simulator was conducted with fexofenadine's physicochemical properties. Simcyp® simulator generated plasma concentration data were incorporated into both models, and RMSE was compared to assess PK Model 1’s accuracy and generalizability. Additionally, a web-based visualization tool was developed for PK Model 1.

Results: The RMSE between the simulated values obtained from Simcyp® simulator and PK Model 1 was 0.106, while that for PK Model 2 was 0.098. Based on these RMSE values, PK Model 1 demonstrated comparable accuracy to the established PK Model 2 in performing PK simulations. Furthermore, PK Model 1 was visualized using Python and Matplotlib.

Conclusion: The results indicate that an accurate and generalizable PK model can be developed using data from the interview form and Simcyp® simulator. Additionally, while this study focused on visualizing pharmacokinetics, we aim to integrate it with a PD model to develop a tool that allows individuals without pharmaceutical knowledge to understand when a drug takes effect and how long its effects last.

Keywords: Self-medication, Pharmacokinetic Simulation, Visualization

Background: Over 14,000 patients are admitted to intensive care each year in New Zealand. About half have or develop infections. Differentiating bacterial infections which require antibiotics from viral or other causes of inflammation is a longstanding challenge. Procalcitonin is a biomarker that may help diagnose bacterial infections, typically using a single threshold approach but its utility in intensive care is unclear. A population model describing procalcitonin dynamics during sepsis and infection in adults (1) may provide an alternative.
Aims: To determine whether reference ranges for procalcitonin, as determined by a population model, capture concentrations measured in adult ICU patients.
Methods: 100 adult patients, either admitted to the ICU with clinical infection or at increased risk of developing infection (receiving mechanical ventilation for > 48 h), were observed. Clinical data, including infection status over time, were collected. Procalcitonin concentrations were quantified from residual clinical samples and assessed against the reference ranges generated by a model incorporating common covariates including patient characteristics, disease severity and time following admission, for patients with infection, sepsis, or severe sepsis.
Results: Patients were 67% male, with average age 53.9 ± 15.1 years, and weight of 91.1 ± 30.4 kg. 33% were infected at admission, with 80% of those diagnosed with sepsis. Expected procalcitonin concentration reference ranges (10th and 90th percentile) generated by the model are assessed against 707 observed concentrations.
Conclusions: Model-predicted reference ranges of expected procalcitonin concentrations can be generated to support interpretation of procalcitonin concentrations.
1. Bokor et al. [Abstract] PAGANZ 2023; Australia.

Background
Rifampicin is essential for tuberculosis treatment, but its efficacy is often limited by high inter-individual variability (IIV) subject to demographic, clinical and environmental factors potentially leading to subtherapeutic exposure, emphasizing the need for a model‐assisted therapeutic drug monitoring (TDM).

Aims
This study aimed to develop a population pharmacokinetic model to support rifampicin TDM.

Methods
Plasma concentrations were prospectively collected from tuberculosis patients in Tanzania at day 1-3 and 12-16 after treatment initiation and retrospectively from a South African cohort sampled at day 7 and 14. A classical stepwise approach in MonolixSuite™ was used to study rifampicin pharmacokinetics (PK) and explore differences due to populations characteristics.

Results
Rifampicin PK was best described by a one-compartment model with linear absorption (rate of 2.64 h-1), delayed by a lag time of 0.32 h, a volume of distribution of 41 L and a saturable Michaelis-Menten elimination with a Km of 27.5 mg/L and Vm values of 124 mg/h and 419 mg/h for the South-African and Tanzanian populations, respectively. The autoinduction was depicted by an Emax time-dependent function, with a steady state reached after a month of treatment (autoinduction half-life of 157 h), and a corresponding elimination increase of 1.38-fold in Tanzanians and 3.76-fold in South Africans. A large IIV was associated with all PK parameters.

Conclusions
Substantial IIV with population differences in autoinduction amplitude and maximum elimination rates associated to ethnic specific metabolic pathways call for a personalized tuberculosis management, such as model-assisted TDM.

Keywords
rifampicin, tuberculosis, population pharmacokinetics, therapeutic drug monitoring