Population K-PD joint modeling of tumor size and CA 125 kinetics after chemotherapy in relapsed ovarian cancer (ROC) patients
Mélanie Wilbaux (1), Benoit You (1), Emilie Hénin (1), Olivier Colomban (1), Gilles Freyer (1), Michel Tod (1)
(1) EMR 3738 CTO, UCBL - HCL Faculté de Médecine Lyon-Sud, Université Lyon 1
Objectives: Ovarian cancer remains the leading cause of gynecologic cancer deaths. CA125 is used as a serum marker of epithelial ovarian cancer. Although lacking of specificity, it may be used to predict tumor burden after chemotherapy and before surgery. The aim of this work is to externally validate a population semi-mechanistic model of CA125 and tumor size kinetics.
Methods: Patients: 535 ROC patients from the CALYPSO trial, a randomized phase III study comparing 2 platine-based regimens (Carboplatin-Paclitaxel vs Carboplatin-Pegylated Liposomal Doxorubicin) were analyzed. Median of 10 CA125 concentration values and 4 tumor size observations per subject were available. 2/3 of patients were randomized to a learning dataset for model building; 1/3 to validation dataset for the external validation.
Model: A semi-mechanistic model was built to describe CA125 and tumor size kinetics after chemotherapy administration. The population analysis was performed with a nonlinear mixed effects model using Monolix 4.1.1. Selection of best model was achieved using criteria based on the likelihood, GOF plots and simulation-based diagnostics. External validation was done using the normalized prediction distribution errors (NPDE) from 2000 replications of the validation set.
Results: Since no drug concentration data were available, a KPD approach has been used for the kinetics of the drug effect. The KPD was described by 2 virtual compartments: 1 central compartment receiving the dose, and 1 transit compartment allowing for a delayed drug effect. Tumor kinetics was dependent on the treatment effect, acting as an inhibitor of tumor growth. CA125 production rate was linked to tumor size variations.CA125 and tumor size kinetics in ROC patients after chemotherapy were properly fit over a 500 days period.NPDEs, calculated on validation dataset, did not deviate from a standard normal distribution, which lead to conclude that the model and population parameter distributions are correct. On the validation dataset tumor size could be adequately predicted using only CA125 levels and model parameters estimated on the learning dataset.
Conclusion: Our semi-mechanistic model is the first to link tumor size and CA125 kinetics to cytotoxic treatment in ROC patients receiving chemotherapy. External validation showed the predictive ability of this model. Modeled CA125 kinetics will be used to compare treatments and to derive predictors of tumor burden dynamics and tumor resectability.
