In-vitro in-vivo correlation models for glibenclamide after administration of metformin/glibenclamide tablets to healthy human volunteers

In this study, level C and A in-vitro in-vivo correlation (IVIVC) models were developed for glibenclamide. In-vitro dissolution data were collected for the glibenclamide component of three metformin/glibenclamide tablets using a USP Type II apparatus. In-vivo plasma concentration data were obtained after administration of the prototype formulations to 24 healthy volunteers and subject to deconvolution analysis to obtain percentage in-vivo absorbed profiles. Multiple linear level C models were developed for CMAX and AUC(0-48) using percentage in-vitro dissolved data at 10, 45 and 120 min. Initially, the level A model was constructed for the first 2 h only, based on availability of in-vitro data. Another level A model was attempted using a time-scaled approach, with percentage in-vivo absorbed at time t and percentage in-vitro dissolved at time t/I as the correlating data. Internal predictability was evaluated for the level C and time-scaled level A models. For all level C approaches, linear regression models with r(2) > 0.99 were determined. The prediction errors (% PE) for C-max and AUC((0-48)) were less than 1% for all formulations at all three chosen time points. The deconvolution analysis indicated biphasic absorption for glibenclamide, with one phase occurring at 2-3 h and another at 6-12 h after dose administration. The level A model using 2-h data was not unique for all formulations and was therefore not developed. The time-scaling factor I correlated highly (r(2) = 0.99) with in vitro mean dissolution time (MDT). A linear regression time scaled model (r(2) = 0.97) was successfully developed using in-vitro and in-vivo data from all 3 formulations. However, the internal predictability of the time-scaled model was poor, with % PE values for C-max and AUC((0-48)) being as much as 30.5% and 18.7%, respectively. The results indicate that level C models have good internal predictability. Though a time-scaled level A IVIVC model was successfully developed, the model was found to have poor internal predictability.