Prediction of food effects on the absorption of celecoxib based on biorelevant dissolution testing coupled with physiologically based pharmacokinetic modeling

Since the rate-determining step to the intestinal absorption of poorly soluble drugs is the dissolution in the gastrointestinal (GI) tract, postprandial changes in GI physiology, in addition to any specific interactions between drug and food, are expected to affect the pharmacokinetics and bioavailability of such drugs. In this study, in vitro dissolution testing using biorelevant media coupled with in silico physiologically based pharmacokinetic (PBPK) modeling was applied to the prediction of food effects on the absorption of a poorly soluble drug, celecoxib, from 200 mg capsules. A PBPK model was developed based on STELLA^® software using dissolution kinetics, solubility, standard GI parameters and post-absorptive disposition parameters. Solubility, dissolution profiles and initial dissolution rate from celecoxib 200 mg capsules were measured in biorelevant and compendial media. Standard GI parameters (gastric emptying rate and fluid volume) were varied according to the dosing conditions. Disposition parameters were estimated by fitting compartmental models to the oral PK data, since intravenous data are not available for celecoxib. Predictions of food effects and average plasma profiles were evaluated using the AUC and C_max and the difference factor (f₁). An approximately 7-fold difference in the maximum percentage dissolved was observed in in vitro dissolution tests designed to represent the fed and fasted states. By contrast, the food effect estimated by simulating the plasma profiles with the PBPK model predicted only a slight delay in the peak plasma level (1 h), and modest increases in the C_max and AUC of 1.9-fold and 1.3-fold in the fed state, respectively. The PBPK approach, combining in silico simulation coupled with biorelevant dissolution test results, thus corresponds much better to the food effect observed for celecoxib in vivo. Additionally, point estimates of AUC and C_max as well as f₁ calculations demonstrated clear advantages of using results in biorelevant rather than compendial media in the PBPK model.