(T1130-05-31) Combining Biorelevant Dissolution, Machine Learning, and PBBM to Explore Drug Exposure Variability due to Gastric Contractions and pH: A Case Study of Dabigatran Etexilate Capsules

Purpose: Dabigatran etexilate (DABE) is a poorly-soluble, weakly basic prodrug that undergoes metabolism to dabigatran (DAB), a direct thrombin inhibitor used for stroke prevention and venous thromboembolism treatment. The marketed formulations, including the original product of Pradaxa capsule, contain DABE mesylate coated onto tartaric acid-containing pellets inside a hard gelatin capsule. This formulation aims to promote intestinal dissolution by creating an acidic microenvironment, thereby improving intestinal drug solubility. Despite this design, the bioavailability of DAB is low (3 – 7% in healthy subjects) and significant interindividual variability in plasma concentrations persists in clinical use [1,2]. Current physiologically-based biopharmaceutics modeling (PBBM) implemented in commercial platforms neglect the trigger of drug release from solid dosage forms caused by intragastric pressure events and a housekeeping wave. These effects cannot be predicted using bottom-up methods and therefore need nonstandard experimental data. Biorelevant dissolution tests in the PhysioCell apparatus have recently shown that DABE release from the Pradaxa capsule is highly sensitive to simulated fasted-state gastric stress events and pH [3]. This study sought to bridge that gap by integrating experimental and machine learning-generated dissolution data with a novel pharmacokinetic (PK) approach to simulate plasma DAB concentrations under realistic gastrointestinal conditions in a population after single and multiple dosing. The part of this work concerning the single dosing has been previously published (https://doi.org/10.1016/j.ijpharm.2025.125464).

Methods: All PK modeling and simulations were performed in Spyder and Google Colaboratory environments using the Python (v. 3.10.12) language. A basis for them was twenty in vitro biorelevant dissolution tests of the Pradaxa capsule (110 mg of DABE) performed in the PhysioCell apparatus under fasted-state gastric conditions, according to a Design of Experiments approach [3]. A machine learning model was trained on this dataset (in Weka software) to generate 1,000 additional dissolution profiles, reflecting plausible physiological combinations of housekeeping wave time (GET, 0 – 60 min), intragastric stress time (IST, 24 – 100% GET), and pH (1.2 – 4.0). To parameterize the obtained diverse and irregular DABE dissolution profiles, we used a novel timewise first-order dissolution model, which segmented the entire dissolution curve into consecutive time intervals and applied individual rate constants to each segment. The dissolution estimates were used as inputs for PBBM-based simulations. They were conducted for a typical individual and a population of 1,000 virtual Caucasian subjects. Population scenarios included variable gastric motility and fixed intestinal permeability (P_eff), fixed gastric motility and variable P<sub>eff, and the variability of both. Multiple dosing simulations involved thirteen doses of 110 mg DABE given every 12 h and leading to a steady state. The PK parameters associated with random interindividual variability (P_eff and V1) were assumed to have 15% interoccasion variability. The gastric conditions were randomized between the dosing events by withdrawing with replacement from the dataset of 1,000 dissolution profiles used in the single dosing simulation.

Results: The timewise first-order model demonstrated high accuracy across all dissolution profiles, with a median R² ≥ 0.9979. It successfully captured diverse dissolution kinetics triggered by gastric pressure events. PK simulations of variable gastric motility in a subject with typical PK parameters and in a population with fixed P<sub>eff revealed a clear sigmoid relationship between DAB bioavailability and the time interval between capsule opening (T_lag) and GET (sigmoid midpoint at ~7 min, maximum at 6.6 – 7.8% at pH 1.2 – 4.0). Early capsule opening, especially before or at IST, followed by delayed GET, created favorable conditions for gastric dissolution, resulting in maximal DAB bioavailability under typical P<sub>eff. When the capsule opened shortly before GET, DABE had insufficient time to dissolve in the stomach and early precipitated in the intestine, which led to minimal bioavailability of ~1%. Intragastric pH was a milder driver of DAB bioavailability. The population simulation scenario with variable gastric motility and fixed P<sub>eff closely matched the clinical variability of DEB exposure. For single dosing, 90% prediction interval of simulated bioavailability (BA-90% PI) was 1.5 – 8.0% vs. observed range of 3 – 7% [2]. For multiple dosing, the simulated coefficient of variation of steady-state peak and trough concentrations (Figure 1) fell within the range of clinical data (Table 1). It indicated that variable gastric motility together with variable drug disposition explains much of the observed drug interindividual DAB exposure differences. The variable P<sub>eff scenario, alone or in combination with variable gastric  motility, overpredicted the drug exposure variability (BA-90% PI: 2 – 16%; Table 1).

Conclusion: This study introduced a novel PK simulation combining biorelevant experiments, machine learning-generated dissolution profiles, a timewise first-order dissolution model, and PBBM to comprehensively evaluate the impact of gastric contractions variability on drug exposure. We demonstrated that gastric motility, particularly the timing between capsule opening and GET, are critical determinants of DAB exposure. The proposed methodology provides a tool for capturing irregular dissolution behavior in biorelevant dissolution systems and can be applied to other oral products sensitive to mechanical stress.

References: 1. Hindley B, Lip GYH, McCloskey AP, Penson PE. Pharmacokinetics and pharmacodynamics of direct oral anticoagulants. Expert Opin Drug Metab Toxicol 2023;19:911–923.
2. Blech S, Ebner T, Ludwig-Schwellinger E, Stangier J, Roth W. The metabolism and disposition of the oral direct thrombin inhibitor, dabigatran, in humans. Drug Metab Dispos 2008;36:386–399.
3. Staniszewska M, Myslitska D, Romański M, Polak S, Garbacz G, Dobosz J, et al. In Vitro Simulation of the Fasted Gastric Conditions and Their Variability to Elucidate Contrasting Properties of the Marketed Dabigatran Etexilate Pellet-Filled Capsules and Loose Pellets. Mol Pharm. 2024;21:2456 – 2472.
4. Stangier J, Stähle H, Rathgen K, Roth W, Reseski K, Körnicke T. Pharmacokinetics and pharmacodynamics of dabigatran etexilate, an oral direct thrombin inhibitor, with coadministration of digoxin. J Clin Pharmacol 2012;52:243–250.
5. Stangier J, Rathgen K, Stähle H, Reseski K, Körnicke T, Roth W. Coadministration of dabigatran etexilate and atorvastatin: assessment of potential impact on pharmacokinetics and pharmacodynamics. Am J Cardiovasc Drugs Drugs Devices Interv 2009;9:59–68.

Acknowledgements: The authors declare no competing interest.

Figure 1. Plasma concentrations of total dabigatran at steady state after multiple dosing of 110 mg dabigatran etexilate in 1,000 virtual subjects with different sources of drug absorption variability. The grey points mark the individual concentrations. The solid lines show the geometric means, while the dashed lines mark the 90% prediction interval.