S. Weber, V. Stalbovskaya, C. Darstein, G. Heimann, P. Gopalakrishna, S. Roychoudhury
Novartis Pharma, Basel
Objectives: In phase I oncology studies a recommended phase II dose (RPIID) is declared at an early stage. With prolonged exposure a dose optimization of the RPIID becomes necessary. As an example, this work
presents an evaluation of the RPIID based on recurrent safety adverse events (AEs) of a phase Ib study [1] of orally administered Ruxolitinib (RUX) and Panobinostat (PAN) for the treatment of patients with Myelofibrosis.
Methods: To understand the full safety profile the clinically relevant AEs are considered for analysis (e.g. thrombocytopenia CTCAE grade G3-4, anemia G3-4, diarrhoea G2-4 and asthenia G2-4). This analysis considers all occurrences of AEs in the dose escalation and the expansion phase. A one-compartment pharmacokinetic (PK) model with an effect compartment (EC) is used. The PK parameters are fixed using previous PK analyses. The recurrent AEs are modeled with a time-varying Poisson process [2]. We consider independent contributions to the intensity h(t) from the natural disease progression, h_0 (t), and each drug h_(1/2)(t). The events are considered as the result of the joint Poisson process such that h(t) is the sum over each component. As h_0(t) we use a Weibull form while we set h_(1/2)(t) proportional to the EC concentration. The model is fitted in a Bayesian framework using Stan [4] for each AE
Results: We evaluated the models using posterior predictive checks. These show that the AEs thrombocytopenia and anemia are not described satisfactory, while diarrhoea and asthenia AEs are well described. The baseline intensity had a large contribution (>50%). The key decision metrics calculated for regimens other than the RPIID with the model are (i) the predicted probability, P(≥1 AE), for at least one AE during 12 weeks and (ii) the relative reduction in cumulative hazard wrt. to the RPIID.
The limitations of the analysis are: (i) no individual PK data, (ii) no frailty term, (iii) assumption of linear PK and no DDI, (iv) more than 50% of the data at the RPIID.
Conclusions: The time-varying Poisson process with an additive intensity function was key to model independent contributions (in contrast to proportional approaches [3]).
The analysis established a dose-exposure-response relationship and enabled derivation of key decision metrics for alternative regimens wrt. to the RPIID. The key metrics derived from the model were communicated to the project team and were instrumental in the decision process.
References:
[1] Ribrag, V. et al., 55th ASH meeting, Dec. 7-10, 2013, https://ash.confex.com/ash/2013/webprogram/Paper56214.html
[2] Kalbfleisch, J.D. and Prentice, R.L., 2002, The Statistical Analysis of Failure Time Data, New Jersey, John Wiley & Sons
[3] Cox, Eugene H. et al., J. Pharmaco. and Biopharm., Dec 1999, Vol. 27, 6, pp 625-644
[4] Stan Development Team (2016). Stan: A C++ library for probability and sampling.
Reference: PAGE 25 (2016) Abstr 5911 [www.page-meeting.org/?abstract=5911]
Poster: Drug/Disease modeling - Safety