Enhancing inclusivity in clinical trials: Model‐informed drug development for pregnant individuals in the era of personalized medicine

The underrepresentation of pregnant individuals in clinical trials and drug development programs causes a lack of safety and efficacy data for this population and poses a significant public health challenge. This article outlines various stakeholder efforts that aim to address this disparity. It emphasizes the largely untapped potential of model-informed drug development (MIDD) tools and underscores the importance of early engagement between pharmaceutical companies and regulatory agencies during drug development. For decades, the administration of medication to pregnant and lactating individuals has occurred and the majority of pregnant individuals commonly receive medication during pregnancy. However, the inclusion of pregnant individuals is limited or is significantly underrepresented in global clinical trial research. Factors that may contribute to this gap include hesitancy of healthcare providers and patients, complex trial designs, ethical concerns, and the potential risk to the pregnant individual and fetus.1 Consequently, pregnant and lactating individuals are prescribed potentially beneficial medicines with limited safety and efficacy information or guidance on optimal dosing for this patient population. Thus, it is vital to include pregnant individuals in the drug development process and engage early with global regulatory agencies. Model-informed drug development (MIDD) methods are a selection of various quantitative methods that help to balance the risks and benefits of drug products in development. As such, these techniques are paramount to maximize the number of safe and effective medicines for pregnant individuals. Here, we discuss a roadmap of how each MIDD method (Figure 1) can be used to address the various challenges faced in this vulnerable patient population. Over the past decades, physiologically-based pharmacokinetic (PBPK) modeling in pregnant individuals has advanced from a niche technique to a powerful tool in environmental toxicology and pharmaceutical research. PBPK modeling holds significant potential for advancing research in maternal-fetal pharmacology. Key benefits include a mechanistic understanding of pharmacokinetic alterations throughout pregnancy, the integration of patient covariates (e.g., disease states, co-medications, and genotypes), and the capability to predict maternal and fetal drug exposure in clinical scenarios that are untested or untestable. These qualities of PBPK modeling can help identify pertinent inquiries necessary to advance research, reduce the need for extensive clinical trials, and lead to optimized dosing regimens and personalized treatment plans in pregnant patients. While documented examples of pregnancy PBPK models in drug development and regulatory submissions are scarce, medications approved by the FDA and EMA often include clinically relevant interventions during pregnancy in their labeling, such as dose adjustment due to pharmacokinetic reasons.2 Although it is unclear whether PBPK modeling guided these interventions, its potential to anticipate altered pharmacokinetics primes it for such applications. Integrating PBPK models with other MIDD tools might further leverage their use in drug development for pregnant patients. Coupling PBPK with pharmacodynamic models provides a comprehensive understanding of the exposure–efficacy profile while integrating it with toxicology models improves risk assessment. Additionally, combining PBPK with quantitative systems pharmacology (QSP) models can elucidate complex biological pathways in maternal and fetal organs. In summary, PBPK modeling holds untapped promise in supporting regulatory interactions and establishing evidence-based labeling recommendations for the pregnant population. Although the scope of MIDD tools for drug developmental safety/toxicology in pregnancy is large, its application has been limited largely to environmental and endogenous chemicals to date. The current paradigm of developmental and reproductive toxicology (DART) safety assessment for drugs is to evaluate the drug in pharmacologically relevant species (nonrodent and/or rodent) during critical developmental windows. It is challenging to extrapolate preclinical adverse findings to the clinical setting since (1) clinical exposure in pregnancy is often unknown and (2) there are physiological differences in placenta between animals and humans. PBPK modeling is an MIDD method which can incorporate in vitro and in vivo data to predict exposures in pregnancy. Linking PBPK or other exposure models to toxicity data can go a step further to identify the most relevant exposure metric (AUC vs. Cmax, blood vs. tissue concentration) for safety margin calculation between clinical and nonclinical exposure. For example, the framework developed by Martin et al.3 included PBPK models for rats and mice for ethanol which was extended to humans in nonpregnancy and pregnancy. In vitro to in vivo extrapolation (IVIVE) of embryotoxicity data was used to predict reasonable exposure scenarios to produce the lowest observed adverse effect concentration (LOAEC) across species. Such a framework can be adapted to enhance drug safety testing in pregnancy. Placental transfer rates, generated from in vitro models, can also be compared across multiple species to predict in vivo tissue concentrations associated with embryofetal developmental toxicity. Enhanced use of MIDD tools in the preclinical setting will improve nonclinical study interpretation and help inform drug safety labeling for the pregnant population. Gaining substantial regulatory acceptance, PBPK models have proven to be invaluable tools to address critical questions in drug development for pregnant individuals and fetal exposure. However, the same cannot be said for the broader field of QSP models, despite their proven value in discovering targets and elucidating drug mechanisms. This disparity may stem from the inherent diversity among QSP models, presenting challenges in standardization and comparability. To unlock the QSP model potential in pregnancy drug development, a consistent risk-based credibility framework should be collaboratively established across industries and regulatory agencies. This framework should agree on standards, consistent terminology, and documentation to ensure reproducibility. Constructing QSP models requires high-quality data, a particularly challenging undertaking during early pregnancy. Effective cross-functional collaboration is needed to ensure a comprehensive understanding of both model applications and limitations. Lastly, a benchmark toolset is needed for model analysis and simplification, such as sensitivity analysis and model reduction techniques. An important advancement is integrating various omics data sources, highlighted by Quinney et al.,4 to understand the complex interplay among maternal, fetal, and placental organs. This integration may be aided by artificial intelligence/machine learning (AI/ML) approaches that allow to uncover underlying correlations and the identification of the most relevant features. Furthermore, advancements in tool development provide further insights into QSP bridging the gap between empirical and mechanistic modeling.5 As efforts progress, the utilization of PBPK in conjunction with QSP models should improve the identification of optimized dosing regimens in pregnancy. Studying the effects of drugs during pregnancy is difficult for many reasons. Real-world data, including electronic health records, medical claims, and patient-generated data like wearables, can potentially study pregnancy's effect indirectly. Patients taking stable doses of drugs that have dose changes during pregnancy are evidence of the need for dose modifications during pregnancy. This could be monitored using pharmacy databases and insurance claims for new prescriptions. Along these lines, real-world databases can be used to detect an increased risk of drug interactions through monitoring of dose discontinuations or reductions that occur during pregnancy. Wearables are particularly interesting because they allow continuous health status monitoring, like blood glucose concentration. Combined with machine learning, real-time monitoring with wearables could be used to maintain better glycemic control during pregnancy. Machine learning could also be used in a broader context to identify potentially unsafe drugs for the fetus. Using chemoinformatics and electronic health records to identify potentially unsafe drugs, Boland, Polubriaginof, and Tatonetti6 developed a random forest machine learning algorithm with an accuracy rate of 91% and 87% for predicting fetal loss and congenital abnormalities, respectively. A total of 57 and 11 drugs previously unidentified were associated with fetal loss and congenital abnormalities, respectively. Real-world data and machine learning are such nascent fields that more research is needed as it relates to pregnancy and clinical pharmacology. As stated earlier, pregnant individuals have historically been excluded from clinical therapeutics development trials and continue to be underrepresented in research. Importantly, failure to establish the correct dose/dosing regimen and the safety of treatments used during pregnancy may compromise the health of pregnant individuals and their fetuses. Under certain circumstances, it is ethically justifiable to include pregnant individuals in clinical trials in both the premarketing and postmarketing setting.7 Additionally, it may also be ethically justifiable to obtain information on individuals who become pregnant while enrolled in a clinical trial. For example, if an individual becomes pregnant while on an investigational agent, they may consent to the collection of pharmacokinetic data that can be used to identify any changes in dosing that may be needed during pregnancy. However, at the time of marketing approval, there is generally little to no human data to inform the safety of drugs and biological products when used during pregnancy. Consequently, the FDA has the authority to issue postmarketing required (PMR) studies to collect information on the safety of medicines used during pregnancy. PMR studies are considered during the review of a marketing application and may be issued for treatments that will be used in females of reproductive potential when there is a need for data to inform on the safety of the use of the treatment during pregnancy.8 In a recent review, only 16% of drugs that may be used in females of reproductive potential were issued PMRs for pregnancy (and/or lactation) studies.9 However, there has been increasing stakeholder interest in the importance of collecting data in pregnant individuals as illustrated by the recommendations of the Task Force on Research Specific to Pregnant Women and Lactating Women (PRGLAC). The PRGLAC report included 15 recommendations to increase the availability of safe and effective therapies specific to pregnant and lactating individuals.10 Although diversity and inclusion in clinical trials and personalized medicine are current paradigms, there is a lack of information about drug safety and efficacy in pregnant individuals. Various stakeholder efforts, further outlined in Table 1, strive to bridge this gap. While incentives for conducting clinical trials in pregnant individuals remain limited, regulatory agencies are increasingly emphasizing the need for PMR studies to gather essential information. To advance drug development in this population, it is crucial to engage early with regulatory agencies and explore MIDD tools which present promising avenues to broaden the access of pregnant individuals to the benefits of clinical research. No funding was received for this work. J.K. is an employee of AstraZeneca and owns stock. All other authors declared no competing interests for this work. The views expressed in this perspective are the authors' views alone. They do not reflect the official views or guidance from the US Food and Drug Administration, nor should they be construed as such. As Editor-in-Training of CPT: Pharmacometrics & Systems Pharmacology Jane Knöchel was not involved in the review or decision process for this paper.