Martin Fink (1), Mark N. Milton (2), and Philip J. Lowe (1)
(1) Novartis Pharma AG, Pharmacometrics, Basel, Switzerland; (2) Novartis Institutes for BioMedical Research, Cambridge, Massachusetts, USA.
Objectives: Drug development aims for most informative experimental designs while minimizing exposure of animals and patients to invasive procedures and potentially harmful drugs. Dose-concentration-response information can be generated efficiently by investigating a range of exposures within individuals rather than relying on between subject analyses, either by following individuals’ responses as their drug concentrations decline or by applying within-individual dose escalation. Here, we evaluate the robustness of a within-individual dose escalation design (“Espresso design”) in a non-human primate study to inform the first-in-human dose for monoclonal antibodies.
Methods: Optimal design software PopED [1] was used to evaluate 3 different study designs with 4 animals each: (a) “Standard design”: all administered a high dose, (b) “Dose-spread design”: spanning a dose range with 1 animal per dose, (c) “Espresso design”: within-individual dose escalation with increasing dose-amounts every other day [2] (a similar approach has been published for an oncology Ph2-trial [3]).
A 1-compartment (cmt) taarget mediated drug disposition (TMDD) and a 2-cmt TMDD model with a mathematical approximation for immediate binding to a soluble target were investigated. The parameters were set to typical values for monoclonal antibodies. To assess the robustness of the study designs against the unknown binding coefficient in-vivo the Kd parameter was modified up and down 30-fold (thus a 900-fold range) and the relative standard errors (RSE) were derived with PopED. In the Espresso design, the time interval between up-titrations was also varied. When considering the appearance of anti-drug-antibodies (ADAs) after 10 days in all animals the data after their appearance was set to missing (as being unreliable) thus all sampling time points after day 10 were ignored.
Results: For the 1-cmt TMDD model, the Espresso design with exponentially increasing dose-levels up-titrated every other day provided only minor improvement over the other designs for the clearance of the drug, but was substantially better for estimating the clearance of the target (50% RSE versus more than 100% RSE for most Kd values investigated). Estimation of Kd and its inter-individual variability was limited by an additive error operating predominantly on the lower concentrations but proved also to be at least 2-fold better for the Espresso design. Of note, the Dose-spread design also out-performed the Standard design.
The superiority of the Espresso-design was more pronounced with the 2-cmt TMDD model. The standard design performed very poorly (only clearance of the drug was estimated with less than 100% RSE). The Dose-spread design gave good precision of parameters, except for the inter-individual variability of Kd. The Espresso-design out-performed the two other designs again with overall robust estimates of most parameters (<100% RSE). When including the 2nd compartment the up-titration interval of 2 days was slightly less optimal than an interval of 4 days.
When anti-drug-antibodies appear at Day 10 then none of the clearance parameters for drug and target were identifiable with precision in any of the designs. The remaining parameters were however estimated with good precision with the Espresso design, as well as the Dose-spread design, but not the Standard design with only a high-dose.
Conclusions: The Espresso design with within-subject dose escalation proved more robust and efficient than Standard designs. It also allowed high exposure levels to be achieved in all subjects (for safety investigations) compared with a design which spread subjects across dose-levels. It provided good estimates of all parameters (except for clearances) in the first 10 days (e.g., in case of appearance of anti-drug-antibodies) and thus allowed for assessment of possible time-dependent changes of the underlying physiological system. In future practice, a mix of different design types will be applied to assess also additional study objectives, such as, for example, injection site reactions with high doses at the first injection.
References:
[1] Nyberg, J et al. PopED: An extended, parallelized, nonlinear mixed effects models optimal design tool. Comput Methods Programs Biomed. 108(2), 789-805 (2012).
[2] Lowe PJ, Fink M, Milton MN. Two Case Studies on How Study Designs Can Be Made More Informative Using Modeling and Simulation Approaches. Clin Pharmacol Ther. 102(6), 908-911 (2017).
[3] Patnaik, A et al. Phase I study of pembrolizumab (MK-3475; anti-PD-1 monoclonal antibody) in patients with advanced solid tumors. Clin Cancer Res. 21, 4286-4293 (2015).
Reference: PAGE 27 (2018) Abstr 8463 [www.page-meeting.org/?abstract=8463]
Poster: Methodology - Study Design