Age and aging are important predictors of health status, disease progression, drug kinetics, and effects. The purpose was to develop ensemble learning-based physiological age (PA) models for evaluating drug metabolism. National Health and Nutrition Examination Survey (NHANES) data were modeled with ensemble learning to obtain two PA models, PA-M1 and PA-M2. PA-M1 included body composition, blood and urine biomarkers, and disease variables as predictors. PA-M2 had blood and urine-derived variables as predictors. Activity phenotypes for cytochrome-P450 (CYP) CYP2E1, CYP1A2, CYP2A6, xanthine oxidase (XO), and N-acetyltransferase-2 (NAT-2) and telomere attrition were assessed. Bayesian networks were used to obtain mechanistic systems pharmacology model structures for PA. The study included n = 22,307 NHANES participants (51.5% female, mean age 46.0 years, range: 18-79 years). The PA-M1 and PA-M2 distributions had greater dispersion across age strata with a right skew for younger age strata and a left skew for older age strata. There was no evidence of algorithmic bias based on sex or race/ethnicity. Klotho, lean body mass, glycohemoglobin, and systolic blood pressure were the top four predictors for PA-M1. Glycohemoglobin, serum creatinine, total cholesterol, and urine creatinine were the top four predictors for PA-M2. The models also performed satisfactorily in independent validation. Model-predicted PA was associated with CYP2E1, CYP1A2, CYP2A6, XO, and NAT-2 activity. Telomere attrition was associated with greater PA-M1 and PA-M2. Ensemble learning models provide robust assessments of PA from easily obtained blood and urine biomarkers. PA is associated with Phase I drug-metabolizing enzyme phenotypes.

Breakthrough bleeding (BTB) is a common side effect of hormonal contraception and is thought to impact adherence to combined oral contraceptives (COCs) but respective dose-response relationships are not yet fully understood. Therefore, the objective of this model-based meta-analysis (MBMA) was to establish dose-response for COCs containing different progestin/EE combinations using BTB as the pharmacodynamic endpoint. Data from 25 studies containing BTB information of 4 progestins (desogestrel, drospirenone, gestodene, and levonorgestrel) in combination with ethinyl estradiol (EE) at various dose levels was used for this analysis. The results of our MBMA show that BTB is significantly increased upon initiation of COC use but subsides over time. The time needed for BTB to return to baseline depends on the EE dose and differs marginally between progestins during the initial months of use at the same EE dose. BTB typically returns to baseline within 3 months at the highest (30 μg) dose, whereas it can take significantly longer to reestablish a regular bleeding pattern at lower EE doses (15 and 20 μg), irrespective of the progestin used. The dose-response relationships established for BTB across different progestin/EE combinations can now be used to support the selection of optimal COC dosing/treatment regimens and serve as the scientific basis for evaluating the impact of clinically relevant factors, including drug-drug interactions and demographics, on BTB.

Neural ordinary differential equations (NODEs) are an emerging machine learning (ML) method to model pharmacometric (PMX) data. Combining mechanism-based components to describe "known parts" and neural networks to learn "unknown parts" is a promising ML-based PMX approach. In this work, the implementation of low-dimensional NODEs in two widely applied PMX software packages (Monolix and NONMEM) is explained. Inter-individual variability is introduced to NODEs and proposals for the practical implementation of NODEs in such software are presented. The potential of such implementations is shown on various demonstrational datasets available in the Monolix model library, including pharmacokinetic (PK), pharmacodynamic (PD), target-mediated drug disposition (TMDD), and survival analyses. All datasets were fitted with NODEs in Monolix and NONMEM and showed comparable results to classical modeling approaches. Model codes for demonstrated PK, PKPD, TMDD applications are made available, allowing a reproducible and straight-forward implementation of NODEs in available PMX software packages.

This study aimed at leveraging data from phase I/II clinical trials to build a nonlinear joint model of serum M-protein kinetics and progression-free survival (PFS) accounting for the effects of isatuximab (Isa), pomalidomide (Pom), and dexamethasone (Dex) in patients with relapsed and/or refractory multiple myeloma. Serum M-protein levels and PFS data from 203 evaluable patients, included either in a phase I/II study (n = 173) or in a phase I study (n = 30), were used to build the model. First, we independently developed a longitudinal model and a PFS model. Then, we linked them in a nonlinear joint model by selecting the link function that best captured the association between serum M-protein kinetics and PFS. A Claret tumor growth-inhibition model accounting for the additive effects of Isa, with an E function, Pom, and Dex on serum M-protein elimination was selected to describe serum M-protein kinetics. PFS was best described with a log-logistic model and associations with baseline beta-2 microglobulin level, age, and coadministration of Dex were identified. The instantaneous change in serum M-protein level was found to be associated with PFS in the final joint model. Using model simulations, we retrospectively supported the Isa 10 mg/kg weekly for 4 weeks, then biweekly (QW/Q2W) dosing regimen of the ICARIA-MM phase III pivotal study, and validated it using the same phase III pivotal study data.

Despite the initial success of single-targeted chimeric-antigen receptor (CAR) T-cell therapy in hematological malignancies, its long-term effectiveness is often hindered by antigen heterogeneity and escape. As a result, there is a growing interest in cell therapies targeting multiple antigens (≥2). However, the dose-exposure-response relationship and specific factors influencing the pharmacology of dual-targeted CAR-T-cell therapy remain unclear. In this study, we have developed a multiscale cellular kinetic-pharmacodynamic (CK-PD) model using case studies from CD19/CD22 and GPRC5D/BCMA autologous CAR-Ts. Initially, an in vitro tumor-killing model characterized the impact of individual binder affinities and their contribution to overall potency across varying (1) effector: target (ET) ratios and (2) tumor-associated antigen (TAA) expressing cell lines. Subsequently, an integrated CK-PD model was developed in pediatric acute lymphoblastic leukemia (ALL) patients, which accounted for CAR-T-cell product composition and relative antigen abundance in patients' tumor burden to characterize patient-level multiphasic cellular kinetics using multiple bioanalytical assays (e.g., flow and qPCR-based readouts). Global sensitivity analysis highlighted relative antigen expression, maximum killing rate constant, and CAR-T expansion rate constant as major determinants for observed exposure of dual-targeted CAR-T-cell therapy. This modeling framework could facilitate dose-optimization and construct refinement for dual-targeted bicistronic CAR-T-cell therapies, serving as a valuable tool for both forward and reverse translation in drug development.

Ritonavir-boosted atazanavir (ATV/r) and rifampicin are mainstays of second-line antiretroviral and multiple anti-TB regimens, respectively. Rifampicin induces CYP3A4, a major enzyme involved in atazanavir metabolism, causing a drug-drug interaction (DDI) which might be exaggerated in pregnancy. Having demonstrated that increasing the dose of ATV/r from once daily (OD) to twice daily (BD) in non-pregnant adults can safely overcome this DDI, we developed a pregnancy physiologically based pharmacokinetic (PBPK) model to explore the impact of pregnancy. Predicted pharmacokinetic parameters were validated with separate clinical datasets of ATV/r alone (NCT03923231) and rifampicin alone in pregnant women. The pregnancy model was considered validated when the absolute average fold error (AAFE) for C and AUC of both drugs were <2 when comparing predicted vs. observed data. Thereafter, predicted atazanavir C was compared against its protein-adjusted IC (14 ng/mL) when simulating the co-administration of ATV/r 300/100 mg OD and rifampicin 600 mg OD. Pregnancy was predicted to increase the rifampicin DDI effect on atazanavir. For the dosing regimens of ATV/r 300/100 mg OD, ATV/r 300/200 mg OD, and ATV/r 300/100 mg BD (all with rifampicin 600 mg OD), predicted atazanavir C was above 14 ng/mL in 29%, 71%, and 100%; and 32%, 73% and 100% of the population in second and third trimesters, respectively. Thus, PBPK modeling suggests ATV/r 300/100 mg BD could maintain antiviral efficacy when co-administered with rifampicin 600 mg OD in pregnancy. Clinical studies are warranted to confirm safety and efficacy in pregnancy.

Selexipag is an oral selective prostacyclin receptor agonist approved for the treatment of pulmonary arterial hypertension (PAH) in adults. To date, no treatment targeting the prostacyclin pathway is approved for pediatric patients. Our goal is to identify a pediatric dose regimen that results in comparable exposures to selexipag and its active metabolite JNJ-68006861 as those shown to be efficacious in adult PAH patients. Extrapolation from the population pharmacokinetic (PK) model developed in adults (GRIPHON study; NCT01106014) resulted in the definition of three different pediatric body weight groups (≥9 to <25 kg, ≥25 to <50 kg, and ≥50 kg) with corresponding starting doses (100, 150, and 200 μg twice daily) and maximum allowed doses (800, 1200, and 1600 μg twice daily). The proposed pediatric dose regimen was subsequently tested in a clinical study (NCT03492177), including 63 pediatric PAH patients ≥2 to <18 years of age and a body weight range of 9.9-93.5 kg. The body weight-adjusted dose regimen for selexipag resulted in comparable systemic exposures to selexipag and its active metabolite in pediatric patients as previously observed in adult PAH patients. Updating the adult selexipag population PK model provided overall consistent parameters and confirmed that the PK characteristics of selexipag and its active metabolite were comparable between pediatric and adult patients. The presented selexipag dose regimen for pediatric PAH patients is considered appropriate for continuing the clinical evaluation of the safety and efficacy of selexipag in pediatric patients ≥2 years of age.

Batch-to-batch variability in inhalation powder has been identified as a potential challenge in the development of generic versions. This study explored the impact of batch-to-batch variability on the probability of establishing pharmacokinetic (PK) bioequivalence (BE) in a two-sequence, two-period (2 × 2) crossover study. A model-based parametric simulation approach was employed, incorporating batch-to-batch variability through the relative bioavailability (RBA) ratio. In the absence of batch variability, recruiting a total of 48 subjects in a 2 × 2 crossover study with the reference formulation resulted in a 95% probability of concluding BE. However, this probability decreased to 80% with a 5% batch difference in RBA and further declined to 30% with a 10% batch difference. With a 10% batch difference, the required number of subjects to achieve an 80% probability of concluding BE increased to 84. When considering product differences between the reference and the test formulations, an additional 10% batch difference reduced the study power from 97% to 30% for a T/R bioavailability ratio of 100% in a 2 × 2 crossover study with 48 subjects. As a result, the substantial impact of batch-to-batch variability on the study power and type I error of the PK BE study may pose significant challenges for the development of generic Advair Diskus due to its degree of PK batch-to-batch variability. Therefore, alternative PK BE study designs and guidelines are needed to adequately address the influence of batch-to-batch variability in products like Advair Diskus.

Time-dependent pharmacokinetics (TDPK) is a frequent confounding factor that misleads exposure-response (ER) analysis of therapeutic antibodies, where a decline in clearance results in increased drug exposure over time in patients who respond to therapy, causing a false-positive ER finding. The object of our simulation study was to explore the influence of clinical trial designs on the frequency of false-positive ER findings. Two previously published population PK models representative of slow- (pembrolizumab) and fast-onset (rituximab) TDPK were used to simulate virtual patient cohorts with time-dependent clearance and the frequency of false-positive ER findings. The impact of varying the number of dose groups, dose range, and sample size was evaluated over time. Study designs with a single tested dose level showed a high probability of showing a false-positive ER finding. When TDPK has a slow onset, use of exposure measures from early timepoints in ER analysis significantly reduces the risk of a false-positive, while with fast onset it did not. Randomization of patients to two dose levels greatly reduced the risk, with a threefold or greater dose range offering the greatest benefit. The likelihood of false-positive increases with a larger sample size, where greater care should be taken to identify confounding factors. Clinical trial simulation supports that appropriate clinical study design and analysis with adequate dose exploration can reduce but cannot entirely eliminate the risk of misleading ER findings.

Midomafetamine (3,4-methylenedioxymethamphetamine [MDMA]) is under the U.S. Food and Drug Administration review for treatment of post-traumatic stress disorder in adults. MDMA is metabolized by CYP2D6 and is a strong inhibitor of CYP2D6, as well as a weak inhibitor of renal transporters MATE1, OCT1, and OCT2. A pharmacokinetic phase I study was conducted to evaluate the effects of food on MDMA pharmacokinetics. The results of this study, previously published pharmacokinetic data, and in vitro data were combined to develop and verify MDMA population pharmacokinetic and physiologically based pharmacokinetic models. The food effect study demonstrated that a high-fat/high-calorie meal did not alter MDMA plasma concentrations, but delayed T. The population pharmacokinetic model did not identify any clinically meaningful covariates, including age, weight, sex, race, and fed status. The physiologically based pharmacokinetic model simulated pharmacokinetics for the proposed 120 and 180 mg MDMA HCl clinical doses under single- and split-dose (2 h apart) conditions, indicating minor differences in overall exposure, but lower AUC within the first 4 h and delayed T when administered as a split dose compared to a single dose. The physiologically based pharmacokinetic model also investigated the drug-drug interaction magnitude by varying the fraction metabolized by a representative CYP2D6 substrate (atomoxetine) and evaluated inhibition of renal transporters. The simulations confirm MDMA is a potent CYP2D6 inhibitor, but likely has no meaningful impact on the pharmacokinetics of drugs sensitive to renal transport. This model-informed drug development approach was employed to inform drug-drug interaction potential and predict pharmacokinetics of clinically relevant dosing regimens of MDMA.

Vancomycin is an antibiotic used for severe infections. To ensure microbiological efficacy, a ratio of AUC/MIC ≥400 is recommended. However, there is significant interindividual variability in its pharmacokinetic parameters, necessitating therapeutic drug monitoring to adjust dosing regimens and ensure efficacy while avoiding toxicity. Population pharmacokinetic (PopPK) models enable dose personalization, but the challenge lies in the choice of the model to use among the multitude of models in the literature. We compared 18 PopPK models created from populations with the same sociodemographic and clinicobiological characteristics. Simulations were performed for a 47 years old man, weighing 70 kg, with an albumin level of 35.5 g/L, a creatinine clearance of 100 mL/min, an eGFR of 106 mL/min/1.73 m, and receiving an intravenous infusion of 1 g × 2/day of VCM over 1 h for 48 h. Simulations of time-concentration profiles revealed differences, leading us to determine the probability of achieving microbiological efficacy (AUC/MIC ≥ 400) with each model. Depending on some models, a dose of 1 g × 2/day is required to ensure microbiological efficacy in over 90% of the population, while with the same dose other models do not exceed 10% of the population. To ensure that 90% of the patients are correctly exposed, a dose of vancomycin ranging from 0.9 g × 2/day to 2.2 g × 2/day is necessary a priori depending on the chosen model. These differences raise an issue in choosing a model for performing therapeutic drug monitoring using a PopPK model with or without Bayesian approach. Thus, it is fundamental to evaluate the impact of these differences on both efficacy/toxicity.

Duchenne muscular dystrophy (DMD), a rare pediatric disease, presents numerous challenges when designing clinical trials, mainly due to the scarcity of available trial participants and the heterogeneity of disease progression. A quantitative clinical trial simulator (CTS) has been developed based on previously published five disease progression models describing each of the longitudinal changes in the velocity at which individuals can complete specified timed functional tests, frequently used as clinical trial efficacy endpoints (supine-stand, 4-stair climb, and 10 m walk/run test or 30-foot walk/run test), as well as each of the longitudinal changes in forced vital capacity and North Star Ambulatory Assessment total score. The model-based CTS allows researchers to optimize the selection of numerous trial parameters for designing trials for the five functional measures commonly used as endpoints in DMD clinical trials. This case report serves as a demonstration of the tool's functionality while providing an easy-to-follow guide for users to reference when preparing simulations of their own design. Two case studies, using input selection based on previous DMD clinical trials, provide realistic examples of how the tool can help optimize clinical trial design without the risk of decreasing statistical significance. This optimization allows researchers to mitigate the risk of designing trials that may be longer, larger, or more inclusive/exclusive than necessary.

Drug-target binding determines a drug's pharmacodynamics but can also have a profound impact on a drug's pharmacokinetics, known as target-mediated drug disposition (TMDD). TMDD models describe the influence of drug-target binding and target turnover on unbound drug concentrations and are frequently used for biologics and drugs with nonlinear plasma pharmacokinetics. For drug targets expressed in tissues, the effect of TMDD may not be detected when analyzing plasma concentration curves, but it might still affect tissue concentrations and occupancy. This review aimed to investigate the likeliness of such a scenario by reviewing the literature for a typical range of TMDD parameter values and their impact on local drug concentrations and target occupancy in a whole-body PBPK model with TMDD. Our analysis demonstrated that tissue drug concentrations are impacted and significantly depleted in many physiological scenarios. In contrast, the effect on plasma concentrations is much lower, specifically for smaller organs with lower perfusion. Moreover, in scenarios with fast internalization of the drug-target complex, the distribution of large molecules from plasma to tissue interstitial space emerges as a rate-limiting step for the drug-target interaction. These factors may lead to overpredicting local drug concentrations when considering only plasma pharmacokinetics. A sensitivity analysis revealed the high and not always intuitive impact of drug-specific parameters, including the drug molecule hydrodynamic radius, dissociation constant (K), drug-target complex internalization rate constant (k), and target dissociation rate constant (k), on the drug's pharmacokinetics. Our analysis demonstrated that tissue TMDD needs to be considered even if plasma pharmacokinetics are linear.

Edoxaban is an orally active inhibitor of activated factor X (FXa). Population pharmacokinetic (PK) and pharmacodynamic (PD) analyses were performed to characterize the PK and PK-PD relationships of edoxaban in pediatric patients to identify the covariates that may contribute to inter-subject variability in PK and PD of edoxaban in pediatric patients, and to compare the PK and PD data between pediatric and adult patients. The pediatric PK of edoxaban was best described by a two-compartment model with transit compartments, first-order oral absorption, and linear elimination. The estimated glomerular filtration rate (eGFR), body weight, and post-menstrual age were the significant covariates explaining variability in edoxaban PK among pediatric patients. A function based on renal maturation was applied to edoxaban clearance. The clearance for a 70 kg patient with an eGFR of 110 mL/min/1.73 m was estimated to be 42.9 L/h (CV ~ 31.8%). PK simulation showed that exposures across five pediatric age groups were comparable to that in adult patients receiving 60 mg once daily dose. The PK-PD relationship for anti-factor Xa was best fit with an E (8.65 IU/mL) model with an EC of 631 ng/mL. The PK-PD relationships for activated partial thromboplastin time and prothrombin time were best fit with linear models (slopes of 0.0467, and 0.0415 s mL/ng, respectively). In addition, due to the small number of efficacy and safety events, an exploratory analysis did not detect a correlation between efficacy events (recurrent venous thromboembolism) or safety events (clinically relevant bleeding) and edoxaban exposure.

Non-alcoholic fatty liver disease and non-alcoholic steatohepatitis describe a collection of liver conditions characterized by the accumulation of liver fat. Despite biopsy being the reference standard for determining the severity of disease, non-invasive measures such as magnetic resonance imaging proton density fat fraction (MRI-PDFF) and FibroScan® controlled attenuation parameter (CAP™) can be used to understand longitudinal changes in steatosis. The aim of this work was to describe the exposure-response relationship of ervogastat with or without clesacostat on steatosis, through population pharmacokinetic/pharmacodynamic (PK/PD) modeling of both liver fat measurements simultaneously. Population pharmacokinetic and exposure-response models using individual predictions of average concentrations were used to describe ervogastat/clesacostat PKPD. Due to both liver fat endpoints being continuous-bounded outcomes on different scales, a dynamic transform-both-sides approach was used to link a common latent factor representing liver fat to each endpoint. Simultaneous modeling of both MRI-PDFF and CAP™ was successful with both measurements being adequately described by the model. The clinical trial simulation was able to adequately predict the results of a recent Phase 2 study, where subjects given ervogastat/clesacostat 300/10 mg BID for 6 weeks had a LS means and model-predicted median (95% confidence intervals) percent change from baseline MRI-PDFF of -45.8% and -45.6% (-61.6% to -31.8%), respectively. Simultaneous modeling of both MRI-PDFF and CAP™ was successful with both measurements being adequately described. By describing the underlying changes of steatosis with a latent variable, this model may be extended to describe biopsy results from future studies.

Around 1.2 million women living with HIV give birth annually, majority of whom will breastfeed their infants while receiving antiretroviral therapy (ART). Lamivudine, a component of first-line ART regimens crosses from maternal plasma to breast milk, with measurable concentrations in some breastfed infants. Wide variability in plasma-to-breast milk transfer has been reported within- or across studies, probably due to differences in sampling framework. This work sought to characterize the milk-to-plasma transfer of lamivudine, quantify inter-patient variability and associated factors, and predict exposure of a breastfed infant. We explored data from an observational pharmacokinetic study that included 35 Ugandan mothers and their infants. Mothers received lamivudine doses of 150 mg twice daily or 300 mg once daily as part of their antiretroviral regimen. Pharmacokinetic sampling was undertaken across two visits approximately 8 weeks apart, providing 248 maternal plasma, 256 breast milk-, and 151 infant blood concentrations, measured across a 24-h sampling interval. A one-compartmental model best described the plasma disposition of lamivudine, with first-order absorption, interindividual variability on clearance and volume of distribution, and a proportional residual error model. A lag in time of plasma-to-breast milk drug accumulation was described using an effect compartment model with a milk-to-plasma ratio of 1.77. An estimated daily infant dose of 179.3 μg/kg (range: 125.8, 282.3) closely predicted the observed infant steady-state concentrations and translated into 3.34% (2.13, 7.20) and 3.35% (1.10, 7.15) of the standard daily maternal dose in visits 1 and 2, respectively. The established modeling framework can be extended to other drugs.

The COVID-19 pandemic, caused by SARS-CoV-2, has underscored the urgent need for innovative therapeutic approaches. Recent studies have revealed a complex interplay between the circadian clock and SARS-CoV-2 infection in lung cells, opening new avenues for targeted interventions. This systems pharmacology study investigates this intricate relationship, focusing on the circadian protein BMAL1. BMAL1 plays a dual role in viral dynamics, driving the expression of the viral entry receptor ACE2 while suppressing interferon-stimulated antiviral genes. Its critical position at the host-pathogen interface suggests potential as a therapeutic target, albeit requiring a nuanced approach to avoid disrupting essential circadian regulation. To enable precise tuning of potential interventions, we constructed a computational model integrating the lung cellular clock with viral infection components. We validated this model against literature data to establish a platform for drug administration simulation studies using the REV-ERB agonist SR9009. Our simulations of optimized SR9009 dosing reveal circadian-based strategies that potentially suppress viral infection while minimizing clock disruption. This quantitative framework offers insights into the viral-circadian interface, aiming to guide the development of chronotherapy-based antivirals. More broadly, it underscores the importance of understanding the connections between circadian timing, respiratory viral infections, and therapeutic responses for advancing precision medicine. Such approaches are vital for responding effectively to the rapid spread of coronaviruses like SARS-CoV-2.

In drug development, quantitative systems pharmacology (QSP) models are becoming an increasingly important mathematical tool for understanding response variability and for generating predictions to inform development decisions. Virtual populations are essential for sampling uncertainty and potential variability in QSP model predictions, but many clinical efficacy endpoints can be difficult to capture with QSP models that typically rely on mechanistic biomarkers. In oncology, challenges are particularly significant when connecting tumor size with time-to-event endpoints like progression-free survival while also accounting for censoring due to consent withdrawal, loss in follow-up, or safety criteria. Here, we expand on our prior work and propose an extended virtual population selection algorithm that can jointly match tumor burden dynamics and progression-free survival times in the presence of censoring. We illustrate the core components of our algorithm through simulation and calibration of a signaling pathway model that was fitted to clinical data for a small molecule targeted inhibitor. This methodology provides an approach that can be tailored to other virtual population simulations aiming to match survival endpoints for solid-tumor clinical datasets.

Due to limited non-clinical and clinical data, European guidance recommends to discontinue breastfeeding when taking albendazole. The aim of this study was to consider the use of PBPK modeling to support the expected exposure in breastfed infants. A fully mechanistic PBPK approach was used to provide quantitative predictions of albendazole and its main active metabolite, albendazole sulfoxide, concentrations in plasma and breast milk of lactating women. The model predicted the exposure in adults and the large food effect, however, it does not predict all the clinical data for the exposure in children. Milk/plasma ratio predictions were also largely over-predicted for this lipophilic compound, but not for the less lipophilic metabolite. Predictions using the observed ratio and a worse-case exposure based on C predictions, suggest doses to children through milk will be low. However, more clinical data are required before full exposure predictions can be made to breastfed infants.