Muriel Boulton (1), Nele Goeyvaerts (1), Nahor Haddish-Berhane (2), Daniele Ouellet (2), Justin Li (3), Oliver Ackaert (1), Alex Yu (2), Juan Jose Perez Ruixo (1)
(1) Janssen Research and Development, Beerse, Belgium, (2) Janssen Research and Development, Spring House, US, (3) Janssen Research and Development, Raritan, US
Introduction: Prostate cancer, and more specifically metastatic castration-resistant prostate cancer (mCRPC), is a common cancer in men worldwide. Abiraterone acetate (AA) plus prednisone (P) is currently one of the standard regimens for treatment of patients with mCRPC. Prostate‐specific antigen (PSA) plays an important role in the diagnosis, monitoring and management of prostate cancer. It was previously reported that PSA metrics (e.g., PSA doubling time, time to PSA progression) were highly associated with overall survival following AAP treatment [1].
Objectives: We aimed to evaluate the predictive performance of PSA kinetics in explaining the AA effect on radiographic Progression-Free-Survival (rPFS) in mCRPC subjects.
Methods: Data from two phase 3 studies investigating the efficacy of AAP in mCRPC patients pre-treated with chemotherapy (AA-COU-301 study, N=789) or chemotherapy naïve (AA-COU-302 study, N=750) were combined. In both studies, patients were randomized to AAP or placebo plus prednisone. As an initial step, rPFS data from the placebo subjects were analysed to identify the parametric form of the time to event model as well as key prognostic factors allowing to differentiate the background rPFS risk between naïve and pre-treated populations [2]. In a second step, a joint modelling approach was applied to characterize the relationship between longitudinal PSA levels (referred to as current PSA) and rPFS [3]. A mechanistic model with treatment-sensitive and treatment-resistant cells was used to describe the PSA levels. The mechanistic PSA model was then combined with the rPFS model with current PSA, using an exponential function, as link on the baseline hazard. The SAEM algorithm as implemented in NONMEM 7.3 was used for parameter estimation [4]. The predictive performance of the joint model was assessed through stochastic simulations of rPFS [5]. The final model was used to estimate the proportion of treatment effect explained by PSA kinetics [6].
Results: An accelerated failure time model with a log-normal distribution for rPFS described the placebo data best. The covariate analysis identified four prognostic factors (baseline lactose dehydrogenase (LDH), number of prior cytotoxic chemotherapy, bone metastases only at entry and baseline albumin) allowing to fully characterize the rPFS risk difference between chemotherapy naïve and pre-treated populations in the placebo group. For each unit increase of baseline LDH (log10) or number of prior cytotoxic chemotherapy (0-2), the survival time decreased by 65% and 15%, respectively. If only bone metastases were present and for each unit increase of baseline albumin, the survival time increased by 30% and 35%, respectively.
The mechanistic model for PSA, assuming the treatments inhibit treatment-sensitive cell proliferation, characterized the PSA levels adequately for both treatments and patient populations. Current PSA improved the rPFS predictive performance for the placebo subjects in both studies. The joint model reasonably captured rPFS in the chemotherapy pre-treated patients randomized to AAP, while it underestimated rPFS in the chemotherapy naïve patients allocated to AAP. Other functions linking PSA with rPFS hazard, such as combined baseline PSA and PSA change from baseline, were investigated but they failed to improve model performance for the AAP group in both studies. The proportion of the overall AA effect on rPFS explained by current PSA was estimated to be 60% and 30% in the chemotherapy pre-treated and naïve populations respectively.
Conclusions: PSA kinetics explained a significant proportion of overall AA effect on rPFS in mCRPC patients treated with AAP. The predictive performance of PSA kinetics on rPFS in mCRCP patients treated with AAP was dependent on the previous administration of chemotherapy. Our conclusions are limited by the retrospective nature of our analysis and the absence of an AA-only arm in the AA-COU-301 and AA-COU-302 studies.
References:
[1] Xu S, Ryan C, Stuyckens K, Smith M, Saad F, Griffin T, Park Y, Yu M, Vermeulen A, Poggesi I, Nandy P. Correlation between prostate-specific antigen kinetics and overall survival in abiraterone acetate-treated castration-resistant prostate cancer patients. Clin Cancer Res; 21(14) July 15, 2015
[2] Jackson C. flexsurv: A platform for parametric survival modeling in R. Journal of statistical software, Vol 70, May 2016
[3] Desmée S, Mentré F, Veyrat-Follet C, Sébastien B, and Guedj J. Using the SAEM algorithm for mechanistic joint models characterizing the relationship between nonlinear PSA kinetics and survival in prostate cancer patients. Biometrics 73, March 2017.
[4] Beal SL, Sheiner LB, Boeckmann AJ, Bauer RJ (Eds.) NONMEM User Guides. 1989-2015. Icon Development Solutions, Ellicott City, Maryland, USA.
[5] Karlsson K, Plan E, Karlsson M. Performance of three estimation methods in repeated time-to-event modelling. The AAPS Journal, Vol. 13, March 2011
[6] Li Z, Meredith M and Hoseyni M. A method to assess the proportion of treatment effect explained by a surrogate endpoint. Statist. Med. 2001; 20:3175–3188
Reference: PAGE 28 (2019) Abstr 8924 [www.page-meeting.org/?abstract=8924]
Poster: Drug/Disease Modelling - Oncology