A drug-independent model predicting Progression-Free Survival to support early drug development in recurrent ovarian cancer
Mélanie Wilbaux (1), Emilie Hénin (1), Olivier Colomban (1), Amit Oza (2), Eric Pujade-Lauraine (3), Gilles Freyer (1,4), Benoit You (1,4) , Michel Tod (1)
(1) EMR 3738, Ciblage Thérapeutique en Oncologie, Faculté de Médecine et de Maïeutique Lyon-Sud Charles Mérieux, Université Claude Bernard, Oullins, France ; Université de Lyon, Lyon, France, (2) Department of Medical Oncology and Hematology, Princess Margaret Hospital, University Health Network, Toronto, Ontario, Canada, (3) Hôpital Hôtel Dieu, Place du Parvis Notre-Dame, 75004 Paris, France, (4) Service d’Oncologie Médicale, Investigational Center for Treatments in Oncology and Hematology of Lyon, Centre Hospitalier Lyon-Sud, Hospices Civils de Lyon, F-69310, Pierre-Bénite, France
Objectives: Early prediction of the expected benefit, based on change in CA125 change, in treated patient with recurrent ovarian cancer (ROC) may help for early selection of the best drug candidates during drug development. The aim of the present study was to quantify and to validate the links between CA125 kinetics and Progression-Free Survival (PFS) in ROC patients.
Methods: Patients from the CALYPSO randomized phase III trial, comparing 2 platinum-based regimens in ROC patients were considered. The cohort was randomly split into a "learning dataset" (N=356) to estimate model parameters and a "validation dataset" (N=178) to validate model performances. Screening for consistent significant factors was performed using Kaplan-Meier plots and semi-parametric Cox regression analyses. A K-PD semi-mechanistic joint model for tumor size and CA125 [1] was used to estimate their values at week 6. Fractional changes in CA125 (ΔCA125) and in tumor size (ΔTS) from baseline at week 6 were then calculated. A full parametric survival model was developed to quantify the links between ΔCA125, ΔTS, significant prognostic factors and PFS. This model was reduced in 2 separate models to compare the predictive ability of ΔTS versus ΔCA125 on PFS. The respective predictive performances were evaluated through simulations on the validation dataset.
Results: PFS from 534 ROC patients were properly characterized by a parametric model with log-logistic distribution. The factors significantly linked to PFS in the full parametric model were ΔCA125, ΔTS, baseline CA125 (CA125BL) and therapy-free interval. By reducing this model, according to the Akaike criterion, ΔCA125+CA125BL was a better predictor of PFS than ΔTS. Simulations confirmed the predictive performance of this model. Patients should achieve at least 49% ΔCA125 decline during the first 6 weeks of treatment to observe 50% PFS improvement. This effect was independent of treatment arm. On the basis of individual ΔCA125, patients could be categorized across 2 groups: responder and non-responder.
Conclusion: This is the first drug-independent parametric survival model quantifying links between PFS and CA125 kinetics in ROC. The modeled CA125 decline required to observe a 50% improvement in PFS in treated ROC patients was defined. It may be a surrogate marker of the expected gain in PFS, and may embody an early predictive tool for go/no go drug development decisions.
References:
[1] Wilbaux M, et al: Population K-PD joint modeling of tumor size and CA-125 kinetics after chemotherapy in relapsed ovarian cancer (ROC) patients: PAGE 2012 Abstr 2587 [www.page-meeting.org/?abstract=2587]
