Physiologically Based Pharmacokinetic Modeling to Assess the Impact of Pathophysiological Changes in Neonates: Strengths, Weaknesses, and Next Steps

Abstract

In neonates, there are many unmet needs to assure safe and effective therapeutics for their conditions. This is also reflected in the still commonly used off-label practices in this population. There are several reasons why drug development as well as licensing or labeling remains limited in newborns, even when weighted to other pediatric subpopulations. Among others, these reasons relate to economic sustainability (market size and difficulty in pricing), as well as to efficacy and safety assessment (clinical outcome assessment and endpoints), poorly understood mechanisms of disease, or challenges in trial design (time-dependent physiology, driven by [non]-maturational factors).¹

Effective and safe pharmacotherapy in neonates necessitates understanding of the pharmacokinetics (PK) and pharmacodynamics (PD) of drugs and doses selected to treat their specific diseases. Differences in gestational and postnatal age or weight (birth weight and current weight) are the major drivers of the observed intra- and inter-variability in drug disposition and effects: the key characteristic of neonatal pharmacology and physiology is fast maturation.² This variability is further extended due to non-maturational factors, like co-morbidity or disease characteristics.

To mitigate these burdens and characteristics, new approaches emerged to support orphan, pediatric, or neonatal drug development. These mitigation strategies include the use of real-world data and evidence, and the development of tools to support extrapolation. When focusing on extrapolation tools, there are obvious strengths, as well as weaknesses and next steps are necessary to further improve the applicability and confidence in these tools.^3-5

Extrapolation to pediatric patients, including to neonates is getting increasingly important. The extrapolation concept is based on a well-characterized source population (like adults or older children, treated for a specific condition) and a well-described target population (like neonates). When the condition is similar between the target and source population, source population-related information can be applied to the target population. For example, if a bacterial infection has similar aspects in adults and neonates, antibiotic efficacy can be “extrapolated” to newborns. Even in a setting of conditions unique to neonates, leveraging prior information available from preclinical or clinical (adult and other pediatric studies) coupled with novel quantitative approaches can be instrumental to predict neonatal doses and optimize trial design.

The International Council for Harmonization (ICH) only very recently (August 21, 2024) adopted a guideline on pediatric extrapolation (E11A), providing a framework, a concept, and a plan on how to apply pediatric extrapolation.⁴ The ICH hereby clearly mentions that extrapolation to younger pediatric populations, particularly neonates, may be challenging due to rapid physiologic changes and organ maturation, while the general principles in this pediatric extrapolation framework still apply.⁴ The latest Food and Drug Agency (FDA) guidance document on clinical pharmacology considerations for neonatal studies also highly recommends using quantitative approaches such as population pharmacokinetics and physiologically based pharmacokinetic (PBPK) modeling to inform neonatal drug development.⁵

Population pharmacokinetic modeling (popPK) is a data (concentration–time profiles) driven tool (“top-down”) to estimate population-level pharmacokinetic parameters, while identifying factors contributing to intra- or inter-individual variability. Mathematically advanced nonlinear mixed-effects models are hereby commonly applied. “Mixed effects” hereby encompasses a combination of fixed parameters, variables that describe the behavior of a “typical” individual, and random effects parameters. In such top-down studies, drug concentrations from a number of individuals are aggregated into one dataset. Nonlinear mixed-effects regression approaches are subsequently applied to analyze both central tendencies for the population and variations between individuals and time points.^{6, 7}

In contrast, PBPK models are mechanistic models, constructed based on a multitude of differential equations that deterministically estimate or simulate time–concentration drug profiles within a physiologically realistic structure for a given scenario or specific (sub)population (“bottom-up”). Within such a framework, organs and tissues are compartmentalized, based on physiologic composition and size, while they are interconnected through organ-specific regional blood flows in a parallel circuit, determined by the cardiac output.^{6, 7}

The recent publication in this journal on a PBPK model that captures vancomycin pharmacokinetics following incorporation of pathophysiological changes in neonates during intensive care is an example on potential applicability, since the model allows dosing optimization at initiation of treatment, before subsequent therapeutic drug monitoring is applied.⁸

While promising, we should neither be naïve. PBPK modeling and simulation is an established tool for drug development with a proven track record, up to regulatory acceptance.^{4, 5} However, the currently available PBPK models overall still have poor predictive performance when applied to neonates, related to knowledge gaps concerning (patho)physiology, as well as limitations related to the currently applied acceptance criteria for PBPK efforts.^9-11

These knowledge gaps give rise to various levels of model uncertainty. Dinh et al recently reported on these sources of variability in the data currently available on neonatal ontogeny, while developing a neonatal PBPK model.¹⁰ Sources of uncertainty include—among others—pathophysiology (like asphyxia, sepsis, or poor cardiac output), growth differences (like small vs appropriate for gestational age), age characteristics (like gestational, postnatal, or postmenstrual age), drug target ontogeny, hepatic and renal disposition ontogeny, absorption, or protein binding. The gaps related to time-dependent physiology encompass both renal and hepatic transporter ontogeny and phase II drug metabolizing enzymes ontogeny, as well as physiological parameters like regional hepatic blood flow, small intestinal transit time, and intestinal enzyme ontogeny or tissue composition.¹⁰ Further extending the theme to neonates with relevant co-morbidity characteristics, Zhang et al summarized the efforts on changes in neonatal (patho)physiology as integrated in published PBPK models.¹¹ While sepsis, patent ductus arteriosus, acute kidney injury, and asphyxia were suggested as relevant scenarios, pathophysiology-related PBPK models were only retrieved for decreased cardiac output (acetaminophen and propofol), and renal impairment (aminophylline).¹¹ The published paper on vancomycin PK in this issue of the journal hereby adds to this still limited literature.⁸

Finally, and related to clinical applications, acceptance criteria of PBPK simulations are traditionally based on a 2-fold range of prediction, while a higher level of accuracy (like 0.8- to 1.25-fold or evaluating 95% confidence intervals instead of mean values) is likely more reasonable, especially for drugs with a narrower therapeutic range.¹²

Since extrapolation and its tools provide a relevant pathway to create impact by improving and facilitating neonatal drug development and pharmacotherapy, cross-talk between clinical researchers and modelers is urgently needed. We hereby should integrate as good as possible the already available knowledge (PK datasets, system knowledge, maturational physiology, and pathophysiology), to subsequently identify and fill the existing gaps to refine PBPK models.⁹ We hereby should not only consider small molecules, as the increasing availability of a diversity of therapeutic proteins may also hold promises for major advances in neonatal care and outcome.

Related to the collection of available knowledge, we are still surprised by the commonly used comprehensive literature approaches, since systematic assessment of a review is the gold standard in meta-analysis to retrieve all data. Along the same line, a common shortage explicitly mentioned during the development of PBPK tools is the shortage of “system parameters” on longitudinal intra-patient patterns (like body composition or weight over age).^9-11 Finally, any PBPK effort needs exploration on its performance, so that shared data are another important resource to get access to. This will necessitate contributions of clinical researchers and data scientists, in collaboration with modelers. As a case example of such an effort, we refer to the initiatives to develop PBPK models for neonates undergoing therapeutic hypothermia because of moderate-to-severe encephalopathy, or to quantify lactation-related drug exposure in infants following maternal pharmacotherapy.^{13, 14}

The author declares that he has no conflicts of interest to disclose.

No funding was received for this work.