Evaluation of a Method to Better Predict Human Absorption from Non-Clinical Data; Comparison of an in silico approach with population modelling of in vivo data
Eli Lilly & Company Ltd
Background: The accurate prediction of oral absorption and its associated variability is necessary to guide dose selection for phase I studies of novel drugs. Common practice is to calculate the bioavailability in preclinical species and take the average as an indicator of human bioavailability. This approach does not provide an estimate of population variability in absorption and is problematic for poorly soluble compounds where differences in formulations may have a large impact on bioavailability. In the current work, the success of an in silico approach predicting human data will be compared with current practice and the observed data.
Methods: In silico models for the prediction of bioavailability take into account various physiological factors, in combination with physicochemical and in vitro drug information. A valuable feature of such models is the assessment of population variability in absorption. One such model, the ADAM model (a compartmental model of drug absorption implemented within Simcyp®) was used for the current work. ADAM outputs were compared with estimates of bioavailability based on pre-clinical species and also with observed clinical data.
Results: As an example, pre-clinical experiments for a compound undergoing development demonstrated consistently high bioavailability. Therefore, the human prediction of bioavailability was ‘high’ (average of animal bioavailability ≈ 60%). However, a different formulation was used for the clinic and data from the first in human study demonstrated a 10-20 fold over-prediction of the oral exposure. The compound was poorly soluble and ADAM simulations indicated a more conservative estimate of bioavailability of ≈ 2 to 4% (assuming no first pass metabolism), which could have accounted for the over-prediction. Several further examples will be provided as well as an assessment of the predictability of the ADAM model for the absorption rate constant, compared to that estimated by NONMEM from human data.
Conclusions: Using the average of bioavailability in pre-clinical species to estimate human absorption prior to the clinic is not optimal since it ignores physiological differences between animal and human and does not accommodate the prediction of variability. Furthermore, such an approach should not be used when variability between animal species is high and/or where there are differences in formulations between the pre-clinical experiments and the clinic. In such cases, the use of in silico models, can provide more refined estimates of drug absorption and offers the added value of the prediction of variability in absorption parameters.