Orna Amir(1*), Assaf Zohar(1), Yuri Kheifetz(1), Ori Inbar(1), Marina Kleiman(1)
(1) Optimata Ltd., Israel
Objectives: Predicting a cancer patient’s likelihood to respond to treatment a priori is one of the greatest challenges both in drug development and in personalized medicine. The aim of this study was to create a prognostic tool for NSCLC using a mechanistic angiogenesis based vascular tumor (VT) model and to assess the quality of data necessary to calibrate the model and identify predictive covariates.
Methods: A mechanistic VT model for NSCLC was created based on literature data: biological data for untreated patients, PK models for cisplatin and pemetrexed and maximal PD effects were estimated from in-vitro data.  PD EC50 and critical VT model parameters were fitted to population response data. Furthermore, a Baysian analysis was conducted with a synthetic data set in which a fictitious covariate to parameter relationship was used to generate population parameters yielding a synthetic data set. This set was then sparsely sampled and used to fit individual parameters given the prior population data.
Results: Model estimated response rates for cisplatin were 10.8% overall response rate (ORR), 55.2% stable disease (SD), and 34.0% progressive disease (PD) as compared to 12.8%(ORR), 49.1%(SD), and 38.1% (PD) [1]. Estimated response rates for pemetrexed were 9.8% (ORR), 48.7% (SD), and 41.5% (PD) as compared to 9.1%( ORR), 45.8%(SD), and 45.1% (PD) [2].  As observed clinically, patients with higher blood volume had a higher chance of being responders. The model was validated on the observed median time to progression (TTP), not used in the calibration phase. The predictions for placebo, cisplatin and pemetrexed were 1.7, 3.8, and 2.8 months, respectively, as compared to the observed 1.8[3], 3.7, and 3.4 months. Additionally, the Bayesian analysis was able to identify the fictitious covariate relationship given 4 tumor measurements taken once every other cycle.
Conclusions: The NSCLC VT model has been shown to be a robust model with the potential of being a prognostic tool. Bayesian analysis shows that the model can be calibrated on a limited and clinically feasible data set. As such we believe the model has a high potential as a clinically predictive tool both for identifying subpopulations that would not otherwise be obvious and eventually for use as a tool for treatment personalization.
References:
[1] Sandler AB, Nemunaitis J, Denham C, von Pawel J, Cormier Y, et al (2000). Phase III trial of gemcitabine plus cisplatin versus cisplatin alone in patients with locally advanced or metastatic non-small-cell lung cancer. J Clin Oncol. 18(1):122-30.
[2] Hanna N, Shepherd FA, Fossella FV, Pereira JR, De Marinis F, et al (2004). Randomized phase III trial of pemetrexed versus docetaxel in patients with non-small-cell lung cancer previously treated with chemotherapy. J Clin Oncol. 22(9):1589-97.
[3] Shepherd FA, Rodrigues Pereira J, Ciuleanu T, Tan EH, Hirsh V, et al (2005). Erlotinib in previously treated non-small-cell lung cancer. N Engl J Med. 353(2):123-32.
Reference: PAGE 21 (2012) Abstr 2560 [www.page-meeting.org/?abstract=2560]
Poster: Oncology