Predictive Model for Identification of Responders/Non Responders in Metastatic Breast Cancer Patients Treated with Docetaxel
O. Amir(1), Y. Kheifetz(1), A. Zohar(1), M. Kleiman (1)
Optimata Ltd., Israel
Objectives: This study aimed to create a predictive personalized tool for metastatic breast cancer (MBC) patients treated with Docetaxel given histopathology data obtained retrospectively from paraffin preserved samples of patient’s primary tumors; ultimately providing an estimate for a patient’s benefit from treatment, and thus avoiding possibly ineffective and potentially toxic therapy.
Methods: Using the Optimata Virtual Patient (OVP)[1] breast cancer model with a population pharmacokinetic model for Docetaxel [2], a two stage analysis of the clinical trial data was carried out. The analysis was performed using Matlab and R based on tumor diameter measurements over time from MBC patients receiving tri-weekly Docetaxel. Missing data for covariates from histopathology were generated using multiple data imputation methods. Given individual parameter estimates, the correlation between individual model parameters and patient histopathology data were examined and significant linear relationships were found and used in a leave one out cross validation analysis to predict individual parameters.
Results: 25 patients contributed to 287 tumor measurements from 64 metastases. Estimated model parameters were maximal tumor cell proliferation rate, maximal tumor cell death rate, sensitivity of HUVEC cell proliferation to VEGF, and lastly the EC50 of the modeled drug effect. Given the covariate relationships and baseline patient data, the cross-validation analysis yielded individual tumor dynamics predictions over the entire treatment and follow-up period. Excluding the first measurement points used for calibration, the predicted dynamics showed an overall R2 of 0.73 and a bias of -0.6mm. Predicted tumor dynamics were then used to calculate the best overall response for each patient with regards to the RECIST criteria yielding a significant, albeit moderate kappa of concordance using linear weighting of 0.42 (CI: 0.19-0.65) when compared to the RECIST calculated from the clinical data.
Conclusion: Many failures in drug development and treatment in Oncology result from large patient variability and the lack of tools to identify responders. This presents a growing need for treatment individualization tools. Using Optimata's tumor growth model for Breast cancer treated with Docetaxel, clinical response was estimated based on one pre-treatment baseline measurement, demonstrating the possibility of identifying non-responders before the initiation of treatment.
References
[1]Agur Z. From the Evolution of Toxin Resistance to Virtual Clinical Trials – the Role of Mathematical Models in Oncology, Future Oncology, (2010) 6(6), 917–927.
[2] Bruno R. et al., A population pharmacokinetic model for docetaxel (Taxotere): Model building and validation. J Pharmacokinet Biopharm. 1996 Apr;24(2):153-72.