Model-Informed Drug Development in Early-Phase Development: Navigating Complexity With Quantitative Clarity

Model-informed drug development (MIDD) has evolved from a promising innovation to a regulatory imperative in drug development. Over the past year, the regulatory landscape shifted. The Food and Drug Administration (FDA) finalized guidance on oncology dose optimization under Project Optimus^{1, 2} and institutionalized its MIDD Paired Meeting Program,³ while ICH issued E11A for pediatric extrapolation⁴ and released a draft guideline M15, developed under ICH auspices, on generalized MIDD principles.⁵ These developments collectively transform early drug development: the goal is no longer just identifying a tolerated dose but quantitatively justifying optimal dosing for pivotal trials. By integrating pharmacokinetics (PK), pharmacodynamics (PD), systems pharmacology, and real-world data, MIDD reduces development risks, accelerates first-in-human studies, and enables rational dose selection. Its expanding influence on decision-making underscores its role not only as a technical methodology but as a paradigm shift in early-phase development strategy.

In the last decade, the utility of MIDD has expanded significantly, driven by advances in computational modeling and an increasingly favorable regulatory environment. The FDA and the European Medicines Agency (EMA) have articulated frameworks encouraging the use of MIDD in early development, with the FDA's MIDD Pilot Program serving as a cornerstone initiative fostering sponsor–regulator dialogue.^{3, 5-7} In 2025, such efforts have matured, with MIDD now recognized as central to regulatory submissions in oncology, rare diseases, and immunology.

Recent literature emphasizes that MIDD enables the translation of preclinical data into clinically relevant predictions with unprecedented precision. For example, Ren et al⁸ highlight how PK/PD modeling refines dose optimization strategies in discovery and development, allowing earlier identification of optimal therapeutic windows.³ This has relevance in first-in-human trials, where dose escalation strategies can now be informed by quantitative predictions rather than empirical trial-and-error. This shift reframes early-phase development from “first safe dose” thinking to “first scientifically informative dose.”

1. First-in-Human Dose Prediction. Physiologically based pharmacokinetic (PBPK) modeling has revolutionized first-in-human dose selection, particularly for biologics and small molecules with complex metabolism. PBPK models incorporate variability in absorption, metabolism, and clearance across virtual populations, offering mechanistic insights that go beyond traditional allometric scaling.⁹ In oncology and rare diseases, such models are increasingly used to support starting dose rationalization, minimizing both underdosing and toxicity risk.

2. Adaptive Trial Designs. MIDD facilitates adaptive trial designs in Phase I studies. By integrating Bayesian hierarchical models with real-time PK/PD data, investigators can adjust dose escalation schemes dynamically. This reduces trial duration and exposure of participants to subtherapeutic or unsafe doses. The FDA's recent endorsements of adaptive MIDD frameworks underscore their regulatory acceptance.^{1, 3, 5}

3. Drug–Drug Interaction (DDI) Risk Mitigation. PBPK models are now routinely leveraged to predict cytochrome-450 (CYP)-mediated and transporter-mediated DDIs early in development, reducing the need for extensive dedicated clinical DDI studies.¹⁰ This integration saves time and resources while providing mechanistic justifications to regulators.

4. Rare Disease and Precision Medicine. For rare diseases, where patient numbers are inherently small, MIDD supports extrapolation from sparse datasets to inform early-phase dosing strategies. Chen et al (2025) note that MIDD also provides a path for evaluating pharmacological modulation of post-translational modifications in cancer, where biomarker-driven modeling is critical for dose selection.¹¹

The growth of artificial intelligence (AI) and machine learning has further augmented MIDD. These technologies allow for real-time updating of population PK models with electronic health record data and wearable-derived biomarkers, enabling continuous model refinement during early-phase trials.¹² Hybrid approaches that combine mechanistic PBPK with data-driven AI models are becoming the standard, improving predictive accuracy across diverse populations.

Another frontier is quantitative systems pharmacology (QSP). QSP integrates network biology, omics data, and mechanistic modeling to understand drug–disease interactions. Its application in Phase I studies is expanding, particularly in immuno-oncology and metabolic disorders, where dynamic feedback loops complicate dose–response predictions.¹³

It is one thing to laud MIDD's conceptual elegance; it is another to show tangible returns. Analyses have shown considerable savings both in terms of time and costs. Importantly, these savings are not just monetary or temporal—they reshape strategic decision-making. MIDD facilitates early “No-Go” decisions, reallocation of resources, and heightened confidence in dosing strategy. Automation tools such as Automated Monitoring of Phase I; 2 (AMP) for Phase I data processing and Cardio Exposure–Response Modeling (CardioERM) for concentration–QT analysis further compressed timelines, turning weeks of report generation into minutes and saving hundreds of business days annually.¹⁴

The regulatory environment in 2025 actively embraces MIDD, publishing detailed guidance on the use of PBPK, population PK, and QSP in early development.^{3, 5, 6, 15, 16} Regulatory science is increasingly collaborative, with workshops and pre-IND consultations dedicated to discussing model credibility. Importantly, regulators now expect sponsors to demonstrate how modeling influenced dose selection and trial design.

Ethically, MIDD reduces unnecessary human exposure to unsafe doses, aligning with the principle of minimizing participant risk. In rare and pediatric diseases, MIDD provides an ethical pathway for extrapolating dosing regimens from adult or preclinical data, reducing the burden on vulnerable populations.¹⁷

Despite progress, challenges persist. Model credibility and reproducibility remain concerns, particularly when integrating AI-driven predictions. Transparency in modeling assumptions and rigorous qualification frameworks are essential to maintain trust among regulators and clinicians.¹² Moreover, the heterogeneity of data sources from preclinical models to real-world datasets poses integration challenges that require standardization.

The evolution of MIDD in early development parallels the growing adoption of AI and machine learning-enabled modeling approaches. As real-world data and in silico methods become increasingly integrated into preclinical-to-clinical translation, a shift toward what is now termed MID³ (model-informed drug discovery and development)²¹ is becoming evident. At this convergence point, investment in modeling expertise, infrastructure, and regulatory–industry collaboration is essential to sustain its impact.

MIDD represents not merely a technical toolkit but a paradigm shift for early-phase innovation. By integrating quantitative predictions, mechanistic modeling, and adaptive learning, MIDD de-risks early clinical development and establishes a stronger scientific foundation for subsequent phases. Regulatory willingness to embrace these approaches creates an opportunity to embed modeling as a standard, rather than exceptional, practice in first-in-human drug development.

The author declares no conflicts of interest.

No funding was obtained for this work.