Hans Peter Grimm (1), Siân Lennon-Chrimes (2), Peter N. Morcos (3)
Roche Pharmaceutical Research and Early Development, (1) Roche Innovation Center Basel, Switzerland; (2) Roche Innovation Center Welwyn, UK; (3) Roche Innovation Center New York, USA
Objectives:
Defining safe and pharmacologically meaningful doses for clinical studies at Entry Into Human (EIH) strongly relies on human PK projections based on pre-clinical PK and other information for the molecule and to some extent inference from similar molecules. Here, we present the results of a review of the methodology and the success of model-based prediction of exposures for EIH studies in Roche pRED (pharma Research Early Development) between 2004 and 2016.
Methods:
Information on models for human PK prediction was systematically collected for all 21 Roche pRED-sponsored monotherapy projects with therapeutic proteins achieving EIH between 2004 and 2016. Where a prediction model was available prior to EIH, its quality was assessed based on Cmax and AUC. For this, the reported models were implemented in Berkeley-Madonna and simulations run reflecting the design of the clinical studies. Predictions were considered adequate when they were within 2-fold of the averaged observed values (NCA) for >90% of the size-weighted cohorts. Where several prediction scenarios were used in parallel, the model predicting the higher exposure was used for comparison with observations.
Results:
Models for human PK prediction were available in 18 of the 21 projects investigated. In all these cases, the models were built on PK data from non-human primates (cynomolgus monkeys) and allometrically scaled using fixed exponents: exponent of 1 for volumes and ranging from 0.75 to 1 for clearance. In several cases (5/18) two parallel prediction scenarios were given with the intention of providing upper and lower bounds for the prediction; this included one case in which the PK of a typical monoclonal antibody was used for the prediction. 10 out of the 18 prediction models were linear 2-comparment models. Another 5 models captured the impact of target-mediated drug disposition (TMDD) by a non-linear clearance function to which scaling was equally applied. In the remaining cases more complex models were used.
More than 80% of the predictions of Cmax and roughly 55% of predictions of AUC were found within 2-fold of observations. In seven of the 10 cases where a linear model was used this was found to be adequate. Where TMDD was predicted, this was found adequate in 6 of the 8 cases. In the remaining cases, TMDD was either not anticipated at all or under-estimated by these models.
Conclusions:
This systematic review of model predictions for human PK predictions for therapeutic proteins shows that the relatively simple allometric scaling from the PK in non-human primates provides useful and robust projections for the planning of EIH studies. Challenges, where they appeared, were most often related to the TMDD and its projection, exacerbated by the very low doses used in some studies with compounds for cancer immunotherapy. This underlines the need in developing robust methods to improve confidence in mechanistic and/or physiologically based models for human PK prediction.
Reference: PAGE 27 (2018) Abstr 8744 [www.page-meeting.org/?abstract=8744]
Poster: Drug/Disease Modelling - Other Topics