I have developed a correlation model for a BCS class 2, IR drug formulation (taking one test (T) & one ref. product ® for internal validation). T is BE to R in a biostudy.
The developed ivivc model gives me an under-prediction (in terms of %PE) of Cmax by 8% and over-prediction of AUC by 4-5% for both test & ref. products. On average, I get the %PE of the model, for Cmax - 3% and for AUC - 9%.
The %PE for both the products for Cmax & AUC are well within 10% and the average %PE are also within 10% (as I mentioned).
Can I count the model as "good" for predicting in vivo profiles ?
I also like to understand when Tcut-off can be used and the rationale.
Going by the knowledge of Tcut-off, I chose a Tcut-off of 3.5 hrs (in vivo median Tmax for both the products are 2.33 hrs and range is 1-4.5 hrs), this reduced the error to 2% & 4% for Cmax & AUC respectively, of the model.
But, should I go by a Tcut-off like that to simply reduce PE ?