Liang Zhao

The pharmaceutical industry is undergoing a pivotal transformation driven by advances in computing, deeper biological insight, and evolving regulatory expectations. Despite these innovations, drug development continues to suffer from “Eroom’s Law,” whereby the cost of developing a new medicine roughly doubles every 9 years. To counter this productivity crisis, a paradigm shift is emerging from traditional model-informed drug development (MIDD), in which models primarily support sponsor decisions and strategy toward model-integrated evidence (MIE). In the MIE paradigm, rigorously verified and validated computational models generate “digital evidence” with evidentiary weight comparable to empirical clinical data for regulatory purposes, including approvals, labeling, and lifecycle management. This approach reflects an ethical and economic imperative to maximize the value of every data point and to leverage accumulated knowledge to predict future outcomes. This paper reviews the evolving MIE landscape in light of recent regulatory initiatives, including the Food and Drug Administration (FDA) MIE pilot for generic drugs and the International Council for Harmonisation (ICH) M15 guideline. We will also provide an overview of some of the quantitative methodologies enabling MIE, and present high-impact case studies illustrating recent applications of MIE. Finally, we will propose measures that could be enacted by regulatory agencies, pharmaceutical companies, and academic institutions to further expand the adoption and impact of MIE. Should these measures prove successful, we envision a future in which MIE is treated as primary evidence of medical product safety and effectiveness, thereby accelerating biopharmaceutical research and development and ultimately improving treatment options for patients.

Typically, parent drug is measured for bioequivalence (BE) assessment because it’s more sensitive to detect formulation differences, compared to its

Optimal pharmacokinetic (PK) profile, including tissue distribution, is pivotal for a drug achieving success in clinical trials. Traditionally, PK estimation in early drug development has relied on extensive in vitro and in vivo testing to assess drug‐like properties, a process that is not only costly and time‐consuming but also limited in its ability to evaluate the synergistic effects of multiple properties. This study aims to develop an integrated artificial intelligence (AI) and physiologically based pharmacokinetic (PBPK) platform to rapidly estimate drug in vivo fate based solely on molecular structures. The AI models were trained to predict eight types of key properties (solubility, pKa values, crystal density, intrinsic dissolution rate, apparent permeability, protein unbound fraction, plasma clearance, and tissue partition coefficients for 15 organs), from which the PBPK model forecasted PK curves without further training. The AI‐PBPK approach was validated against human PK data of 71 intravenous and 606 oral administrations collected from the PK‐DB database. The results were robust, with most of the AUC predictions falling within two and threefold error ranges. The AI‐PBPK model also accurately predicted drug organ selectivity, and for drugs exhibiting high plasma clearance, predictions were optimized through an inter‐species extrapolation approach. This study illustrates that the developed modeling strategy adeptly addresses pivotal PK challenges in drug discovery and aligns with contemporary drug development processes. The modeling system can guide candidate selection, advancing more drugs with favorable PK profiles into clinical trials, thereby significantly enhancing the efficiency of drug development.

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.

The U.S. Food and Drug Administration (FDA) has defined narrow therapeutic index (NTI) drugs as “those drugs where small differences in dose or blood concentration may lead to serious therapeutic failures and/or adverse drug reactions that are life‐threatening or result in persistent or significant disability or incapacity.” FDA has undertaken efforts to develop NTI assessment criteria and enhance public confidence in generic NTI drugs through public workshops, research, and post‐marketing surveillance. In 2015, FDA formed the NTI Drug Working Group to develop a consistent approach to identify NTI drugs and resolve NTI‐related scientific and regulatory issues in a transparent and collaborative manner. One key objective of the NTI Drug Working Group is to evaluate potential NTI drugs based on five general characteristics of NTI drugs as highlighted in the case example for theophylline drug products. As of January 5, 2024, there are 33 drug products, with 14 distinct active ingredients, specified as NTI drugs in their respective product‐specific guidances (PSGs) for generic drug development. Future collaborative efforts with other agencies to harmonize the terms and definitions for NTI drugs may help enhance clarity and consistency during the drug development and the regulatory review process.

Conventional approaches for establishing bioequivalence (BE) between test and reference formulations using non‐compartmental analysis (NCA) may demonstrate low power in pharmacokinetic (PK) studies with sparse sampling. In this case, model‐integrated evidence (MIE) approaches for BE assessment have been shown to increase power, but may suffer from selection bias problems if models are built on the same data used for BE assessment. This work presents model averaging methods for BE evaluation and compares the power and type I error of these methods to conventional BE approaches for simulated studies of oral and ophthalmic formulations. Two model averaging methods were examined: bootstrap model selection and weight‐based model averaging with parameter uncertainty from three different sources, either from a sandwich covariance matrix, a bootstrap, or from sampling importance resampling (SIR). The proposed approaches increased power compared with conventional NCA‐based BE approaches, especially for the ophthalmic formulation scenarios, and were simultaneously able to adequately control type I error. In the rich sampling scenario considered for oral formulation, the weight‐based model averaging method with SIR uncertainty provided controlled type I error, that was closest to the target of 5%. In sparse‐sampling designs, especially the single sample ophthalmic scenarios, the type I error was best controlled by the bootstrap model selection method.

By applying nonlinear mixed‐effect (NLME) models, model‐integrated evidence (MIE) approaches are able to analyze bioequivalence (BE) data with pharmacokinetic end points that have sparse sampling, which is problematic for non‐compartmental analysis (NCA). However, MIE approaches may suffer from inflation of type I error due to underestimation of parameter uncertainty and to the assumption of asymptotic normality. In this study, we developed a MIE BE analysis method that is based on a pre‐defined model and consists of several steps including model fitting, uncertainty assessment, simulation, and BE determination. The presented MIE approach has several improvements compared with the previously reported model‐integrated methods: (1) treatment, sequence, and period effects are only added to absorption parameters (such as relative bioavailability and rate of absorption) instead of all PK parameters; (2) a simulation step is performed to generate confidence intervals of the pharmacokinetic metrics for BE assessment; and (3) in an effort to maintain type I error, two more advanced parameter uncertainty evaluation approaches are explored, a nonparametric (case resampling) bootstrap, and sampling importance resampling (SIR). To evaluate the developed method and compare the uncertainty assessment methods, simulation experiments were performed for BE studies using a two‐way crossover design with different amounts of information (sparse to rich designs) and levels of variability. Based on the simulation results, the method using SIR for parameter uncertainty quantification controls type I error at the nominal level of 0.05 (i.e., the significance level set for BE evaluation) even for studies with small sample size and/or sparse sampling. As expected, our MIE approach for BE assessment exhibited higher power than the NCA‐based method, especially as the data becomes sparser and/or more variable.

Forward addition/backward elimination (FABE) has been the standard for population pharmacokinetic model selection (PPK) since NONMEM® was introduced. We investigated five machine learning (ML) algorithms (Genetic algorithm [GA], Gaussian process [GP], random forest [RF], gradient boosted random tree [GBRT], and particle swarm optimization [PSO]) as alternatives to FABE. These algorithms were applied to PPK model selection with a focus on comparing the efficiency and robustness of each of them. All machine learning algorithms included the combination of ML algorithms with a local downhill search. The local downhill search consisted of systematically changing one or two “features” at a time (a one-bit or a two-bit local search), alternating with the ML methods. An exhaustive search (all possible combinations of model features, N = 1,572,864 models) was the gold standard for robustness, and the number of models examined leading prior to identification of the final model was the metric for efficiency.All algorithms identified the optimal model when combined with the two-bit local downhill search. GA, RF, GBRT, and GP identified the optimal model with only a one-bit local search. PSO required the two-bit local downhill search. In our analysis, GP was the most efficient algorithm as measured by the number of models examined prior to finding the optimal (495 models), and PSO exhibited the least efficiency, requiring 1710 unique models before finding the best solution. Additionally, GP was also the algorithm that needed the longest elapsed time of 2975.6 min, in comparison with GA, which only required 321.8 min.