Model‐Informed Drug Development for Long‐Acting Injectable Products: Summary of American College of Clinical Pharmacology Symposium

Poor adherence to prescribed medications significantly impacts the U.S. health care system.1-4 Given the human and financial consequences, the development of strategies for improving adherence to prescribed medications is imperative. Long-acting injectable (LAI) drug products are 1 of several interventions for improving patient adherence to prescription medications.5-8 LAIs are formulated to achieve extended-release drug action over days and months, when administered via intramuscular and subcutaneous routes. These products can help to improve adherence in patients who need medications to be administered frequently. Several LAIs have been approved by the U.S. Food and Drug Administration (FDA), for example, paliperidone palmitate, aripiprazole, naltrexone, and buprenorphine. LAI products have unique pharmacokinetic (PK) characteristics. Their pharmacokinetics generally are characterized by a rate of drug absorption that is slower than their rate of elimination; hence, they exhibit flip-flop kinetics. In these products, the terminal phase of the drug profile reflects the rate of absorption, rather than the rate of elimination, as is usually observed in classical linear drug PK.9-13 The long terminal phase of these products poses several challenges for the development of new versions, as well as generic copies. Model-informed drug development (MIDD) has the potential to make drug development of LAI more efficient by integrating mechanistic understanding and improving study designs. Therefore, gathering stakeholders to discuss the current status and application of MIDD in drug development and regulatory decision making for LAI products is both important and timely. A symposium titled, "Model-Informed Drug Development for Long-Acting Injectable Products" was organized and presented at the American College of Clinical Pharmacology meeting in Chicago, Illinois, on September 16, 2019.14 Goals and objectives of this symposium were to describe opportunities and challenges in the development of LAI products, to identify MIDD opportunities for new and generic LAI products, and to discuss challenges and opportunities from a regulatory perspective. This report summarizes the presentations and discussions that made up the symposium. Dr. Lanyan (Lucy) Fang presented cases that illustrated MIDD, both from new and generic LAI perspectives. The first case presented was the new drug development of paliperidone LAI, which is approved for schizophrenia. The extended-release (ER) tablet dosage form was approved as a once-daily regimen in 2006. Subsequently, a monthly LAI dosage form was developed. Different dosing regimens were studied in phase 3 trials. In the first regimen, fixed doses such as 25, 50, 100, or 150 mg, were given on days 1, 8, 36, and 64. The first 2 doses were given 1 week apart, with the following doses given monthly. In the second regimen, a high loading dose of 150 mg on day 1, followed by 25, 100, or 150 mg on days 8, 36, and 64 were administered. The final proposed regimen, however, was 150 mg on day 1, 100 mg on day 8, and 75 mg as the monthly maintenance dose, a regimen not directly studied in any of the late-phase trials. The rationales for the approved dosing regimen were based on results from phase 3 studies and on PK simulations (Table 1).15 In one 13-week parallel trial (Table 2), each arm included about 160 patients. The primary efficacy end point was the change in the total score on the Positive and Negative Syndrome Scale (PANSS) from baseline at the end of 13 weeks. All active arms of the above-mentioned second regimen showed a greater, statistically significant reduction of the PANSS and a dose-response relationship, providing evidence of the new regimen's efficacy. Regarding safety, 1 death occurred in the monthly 150-mg arm and was considered related to the drug. In addition, dose-dependent increase in body weight and serum prolactin levels, which are known safety concerns, had already been observed. Given this benefit/risk assessment, the applicant proposed 75 mg as the monthly maintenance dose, with 150 mg as the first dose on day 1 and 100 mg as the second dose on day 8. Additional support for 75 mg came from a PK bridging analysis with the approved ER tablet formulation. The ER formulation had been on the market for a few years, with an approved maintenance dose of 6 mg once daily. The PK model for the monthly injection formulation predicted that the concentration between 20 and 24 weeks under a 75-mg monthly regimen should be within the predicted steady-state concentration range following daily administration of a 6-mg ER tablet. Therefore, 75 mg was considered the dose with the best benefit/risk profile (Figure 1). Dr. Fang also stated that model-based bioequivalence (BE) approaches are essential to the development of complex generic LAI products, enabling abbreviated new drug application applicants to conduct BE studies efficiently.16-18 The modeling approaches can help applicants evaluate the study design and sensitivity quantitatively, as well as maximize the information gained from efficient BE studies. In designing adequately powered PK BE studies, the applicant can conduct virtual BE study simulations comparing alternative designs (such as optimal BE study design) based on the prior PK information of the reference product.19 Modeling approaches also can help the BE assessment. For instance, patients who switch treatment from the reference product to products with slower release rates or delayed release will take a few months to reach a new steady state. As simulation results illustrate, products with slower release rates will show greater PK differences at the new steady state after the switch, whereas those with delayed release will have greater PK differences during the transition period right after the switch. Model-based BE approaches can be used to tackle those challenges. For example, PK models can be fitted to the observed data to account for residual drug exposure from prior treatment, as well as for covariate effect on BE assessment. Subsequently, steady-state simulations can be performed to enable BE assessment in both the steady-state condition and the immediate transition stage. Dr. Fang concluded that modeling and simulation (M&S) approaches have been used throughout the life cycle of LAI products: in new drug development, regulatory approval, dosing management in clinical practice, and generic drug development.15, 20-28 Dr. Satish Sharan presented challenges and opportunities for M&S in the development of LAIs, focusing especially on generic drug development. He pointed out that LAI products have not seen much generic competition, which is important for increasing the U.S. public's access to affordable LAI drug products. LAIs by design have a long terminal half-life, leading to much longer times to reach steady state and to disappear from the body. As a result, life-cycle management of LAI drug products is challenging because of the longer duration of clinical studies for determining BE. This longer duration leads to higher subject dropout rates from clinical studies than from conventional BE studies. Because of safety considerations for some LAI products, their BE studies are recommended to be conducted in patients, and sometimes the washout of patients on treatment and burdensome prolongation of crossover studies are unethical. In addition, ensuring attainment of steady state in subjects for BE studies of LAI products can be challenging. Dr. Sharan pointed out that challenges in the development of LAI products present perfect opportunities for M&S. Because clinical trials are costly and time-consuming, M&S can be used to optimize BE study design, increasing its chances of success by adequately powering the study, simulating scenarios that can lead to BE, and designing and justifying innovative BE study designs that can be used to shorten studies or to develop efficient study designs with fewer subjects. Dr. Sharan provided an example of BE trial simulation using paliperidone 1-month injection to demonstrate the utility of M&S in designing a clinical BE study. He provided information on resources that can be used by generic drug developers as starting points in their M&S efforts, such as the publicly-available clinical pharmacology reviews of new drug applications on the FDA website Drugs@FDA and published articles.23 In addition, he commented that applicants can use BE trial simulation to simulate physiologically plausible scenarios likely to be observed in the development of generic products, for example, differences in the rate and extent of absorption, and to test its impact on the probability of BE determination (Figure 2). M&S also can be used to identify adequate PK sampling times, to determine sample size, and to support innovative and efficient BE study designs.16, 17, 26, 27, 29-33 Dr. Viera Lukacova presented on the application of the physiologically based pharmacokinetic (PBPK) modeling approach34-37 in the development of LAI injectable products. The motivation for developing LAI microsphere formulations is to provide a means for releasing a drug over a long period (weeks to months) to improve patient comfort and compliance while maintaining therapeutic efficacy and avoiding adverse effects because of fluctuations in drug concentrations. Unfortunately, developing bioequivalent generic LAI microsphere formulations is complicated by the lack of standardized in vitro dissolution/release experiments, the long release times involved, and the large variabilities in in vivo dosing tissue environments among patient populations. Thus, more predictive preclinical methods, including both better in vitro dissolution/release experiments and predictive mechanistic models are needed to reduce failed human BE trials, as well as to assess the sensitivities of critical quality attributes and their impact on in vitro and in vivo performance of LAI products. The presentation focused on poly(lactic-co-glycolic acid) (PLGA)-based microparticles and discussed: (1) the use of mechanistic modeling of in vitro dissolution experiments for LAI microspheres to evaluate the effect of formulation parameters on in vitro drug release, (2) the lessons learned from generating empirical in vitro-in vivo correlation (IVIVC) for these types of formulations, and (3) a PBPK model to simulate the performance of LAI microspheres in vivo. Two previously published models38, 39 for simulation of drug release from PLGA-based LAI microspheres were evaluated. The performance of both models was tested against literature data for LAI microspheres with several different drugs. The main conclusions from the initial evaluations were that: (1) both models failed to capture the initial lag time in the in vitro release profiles observed for some formulations, (2) different drugs followed different functions for scaling of the diffusion coefficient with changes in polymer molecular weight, and (3) the effect of particle size on the release rate was not predicted accurately by either model. The more complex model was14 selected as the better platform for the addition of new mechanisms and was extended by adding mechanisms for: (1) autocatalytic degradation of the polymer; (2) pH-dependent solubility of the drug within the particle; and (3) diffusion of water, drug, soluble oligomers, and free acid through the PLGA particle. This extended model showed the potential ability to account for differences in drug release rates from formulations with varying particle sizes and PLGA molecular weights. In addition, an exponential correlation between the PLGA degradation rate and the lactic acid/glycolic acid (LA/GA) ratio was derived based on literature data40-42 to help to extrapolate the drug release rates across formulations with varying LA/GA ratios. Finally, a function to account for nonhomogeneous distribution of the drug throughout the particle was added to the model. Deconvolution of the in vivo dissolution or release rate of the drug from LAI microspheres and subsequent correlation with the measured in vitro dissolution/release profile was incorporated in the PBPK model GastroPlus (Simulations Plus, Inc., Lancaster, California). The GastroPlus PBPK model describes partitioning of the released drug from the injection site into systemic circulation and subsequent systemic drug distribution and elimination and has been extensively used for a variety of applications,43-47 which show validation of drug distribution between plasma and tissues. Using the PBPK model for drug uptake and distribution in the systemic circulation enables deconvolution of the in vivo release profile rather than a profile of drug appearance in plasma obtained through traditional deconvolution methods. Although this is not a mechanistic prediction of the drug release from the particles in vivo, it enables a direct comparison of differences between the drug release in vitro and in vivo. Several data sets from the literature were used to deconvolute in vivo dissolution profiles and identify areas in which further development would be needed to increase chances for successful IVIVC. A parametric deconvolution approach, that is, an approach inn which the in vivo release is described by a known model, was used for the deconvolutions. Single- and double-Weibull functions are commonly used with the parametric deconvolution approach48 and have been shown appropriate to describe the in vivo release profiles of orally administered products35, 49 However, those functions did not provide enough flexibility to deconvolute the in vivo release for a number of tested LAI formulations and a more complex, triple-Weibull, function had to be introduced. The ability to form a successful IVIVC also may be affected by sampling density in the in vitro release profile. Several analyzed data sets showed a quick initial peak in the observed plasma concentration time (Cp-time) profile, suggesting a small burst of drug release shortly after injection, followed by a long, slow release of the remaining drug. It was possible to successfully deconvolute the in vivo release profile that would account for the small initial burst of release and find the correlation between the drug release in vitro and in vivo. However, because of insufficient in vitro sampling at the early times, the correlation was not able to predict the small initial peak in the Cp-time profile for a new formulation. The prediction of Cp-time profile for the new formulation was improved by adding interpolated points in the in vitro dissolution profile. Addressing these technical aspects helped to create and validate IVIVC for some, but not all, test scenarios, and additional possible aspects were explored. On injection in vivo, the rate of drug release from LAI microspheres and the rate of drug appearance in systemic circulation is dependent not only on the formulation, but also on biological factors. An initial literature review revealed possible in vivo factors that may be affecting the LAI formulation behavior,42, 50-52 for example, (1) inflammatory tissue response, which may affect the rate and mechanism of PLGA degradation and diffusion of released drug away from the microparticles; (2) the presence of enzymes that may affect polymer degradation; (3) the presence of lipids that may affect API diffusion; (4) the presence of other endogenous compounds that may affect the pH environment; and (5) fluid volume and convection. The effect of the immune cell layer (ICL) that may form around the formulation depot and affect the rate of drug appearance in systemic circulation after intramuscular injection was investigated more closely. A model that accounts for the drug diffusion through ICL with a time-varying thickness, and nonspecific tissue binding was developed based on literature data53 and incorporated in the GastroPlus PBPK model. The data used to develop the model is based on a rodent study, and it is possible that the values of the model parameters may need to be adjusted for other species (human or other animals). Test simulations showed the potential effect of this layer on the shape of the drug's plasma concentration-time profile, but the purely predictive ability of the model could not be evaluated at this stage because of insufficient data. In summary, a detailed in vitro dissolution model that enables the investigation of mechanisms and processes affecting the drug release from PLGA-based LAI microspheres was developed. However, the drug release and appearance in systemic circulation after injection in vivo may be significantly impacted by physiological response. Further research is needed in this area, as understanding and characterizing different phases of this physiological response is critical to predicting the in vivo behavior of these formulations. Dr. Mats Karlsson presented possible model-informed BE evaluation strategies to solve some challenges of BE studies for LAI products. One challenge of LAI BE studies is their prolonged duration because of the long half-life of LAIs. In addition, multiple-dose steady-state designs carried out on patients may require even longer studies to achieve steady state in cases in which drugs show long-term toxicity, for example, antipsychotic drugs.23, 54, 55 Dr. Karlsson suggested investigating multiple-dose crossover switch studies for problematic LAI products to shorten BE studies, that is, assessing BE based on samples from the steady-state PK of a reference drug (before switch) and the first dose interval of a test drug (after switch). To bridge between BE criteria for a steady-state design and that for a crossover switch study, M&S serves as a promising tool. Through simulation studies based on a well-developed LAI model, surrogate criteria (eg, surrogate BE limits corresponding to 80%-125% of the BE limits for steady-state designs) for crossover switch studies may be determined for the metric ratio (AUC and Cmax) of the steady state of a reference drug and the first dose interval of a test drug. For those drugs with no long-term toxicity (eg, for birth control), the FDA suggests a single-dose parallel study design. A challenge of parallel BE studies for LAI products is low power because of high between-subject variation caused by multiple physiological factors affecting drug PK (such as sex and injection site), especially concerning absorption.56, 57 Standard BE analysis only considers treatment effect (ie, reference vs test drug), which tends to have large variance, leading to lower power. Dr. Karlsson proposed performing simulation studies based on a well-developed PK model with covariate effects (ie, PK-related affecting factors) to determine the (fixed) effect size of covariates in linear regression models used during noncompartmental analysis (NCA)-based BE analysis. By adding the related covariate effects in the standard BE analysis, less unexplained between-subject variation will occur, thus increasing BE study power and, accordingly, reducing the sample sizes of parallel BE studies for LAI products. The strategy for both problems is to apply M&S to determine potential modifications of current BE study designs, analysis methods, and/or criteria, that is, model-informed approaches. Dr. Karlsson also proposed model-integrated approaches (Figure 3) to solve the above problems, in other words, to directly apply M&S in the BE analysis procedure. First, BE data from a multiple-dose crossover switch study or a single-dose parallel study are analyzed using a previously developed PK model with treatment effects on all absorption parameters. From the resulting estimates, along with their uncertainty, researchers can perform population simulations of a single-dose crossover BE study, the standard design for BE trials, without any covariate effects. Across all population simulations, 90% confidence intervals of the geometric mean of metrics of interest can be calculated and used for a final BE conclusion. The described model-integrated BE analysis method has been shown to have acceptable type I error and higher power than an NCA-based method in sparse sampling scenarios. In summary, Dr. Karlsson proposed 2 approaches for BE analysis of LAI products: (1) a model-informed approach, in which M&S is used to inform and evaluate potential modifications of study designs, analysis methods, and/or criteria; and (2) a model-integrated approach, in which model parameter estimation is performed on BE study data and simulated results will be used as model-integrated evidence to support LAI BE approval. Both strategies aim to reduce study duration and/or sample size, leading to more feasible BE studies for LAI products. The symposium provided a good overview of the opportunities and challenges involved in the development of LAI products, identified opportunities for MIDD for new and generic LAI products, and discussed avenues available to both new and generic drug developers for obtaining feedback from the FDA on applications of M&S in drug development. The authors thank Yaning Wang, PhD, for sharing his valuable inputs on the MIDD of LAI from a new drug development perspective and Joanne Berger (FDA Library) and Sara Lomonaco for manuscript editing assistance. Satish Sharan, Lanyan Fang, and Xiaomei Chen have no conflicts of interest. Mats O. Karlsson and Andrew C. Hooker own stock in and receive consulting fees from Pharmetheus AB. Viera Lukacova is employee and holds stock options of Simulations Plus, Inc, the company that develops GastroPlus® software tools mentioned in the presentation and used for the presented simulations results. This work was partially supported by FDA contract (75F40119C10018) awarded to the University of Uppsala, Sweden (Dr. Mats O. Karlsson) and FDA grant (U01FD005463) awarded to Simulation Plus Inc. (Dr. Viera Lukacova). This article reflects the views of the authors and should not be construed to represent FDA's views or policies. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.