Ileana Theofili1, Elias Liolis2, Vangelis Karalis1, Gregory Sivolapenko3
1Department of Pharmacy, National and Kapodistrian University of Athens, 2School of Medicine, University of Patra, 3Department of Pharmacy, University of Patra
Introduction: Increased VEGF expression in the tumor microenvironment contributes to tumor progression and metastasis [1]. Bevacizumab, a recombinant humanized monoclonal antibody, is used as monotherapy or in combination with chemotherapy for the treatment of various types of cancers including breast cancer, cervical, ovarian cancer, non-small cell lung cancer (NSCLC) and metastatic colorectal (CRC) [2]. Bevacizumab suppresses tumor angiogenesis by preventing VEGF-A from interacting with its receptors, thus, inhibiting tumor growth [3]. Bevacizumab has shown remarkable versatility across a plethora of cancers, showing efficacy in frontline treatment, maintenance therapy, and recurrent disease settings [4]. Objectives: The aim of this study is to investigate the impact of demographic and chemotherapy factors as well as bevacizumab levels (peak, trough, and their difference) on VEGF levels in patients with NSCLC, CRC, breast, cervical and ovarian cancer. Methods: Several machine learning techniques and artificial neural networks (ANN) were used for the analysis. Initially, methods such as Categorical Principal Component Analysis (CATPCA), Multiple Component Analysis (MCA), Principal Component Analysis (PCA) and Factor Analysis of Mixed Data (FAMD) were applied. To model the relationship between the variables and the reduction in VEGF levels (i.e., VEGF_D) models such as Random Forest, Boosted Trees, Bagged Trees, and ANN were developed and evaluated. Results: Both CATPCA and PCA revealed the opposing trends between VEGF and bevacizumab levels, suggesting different contributions to variance. CATPCA also indicated that VEGF reduction may follow distinct trajectories based on treatment duration and cycles, highlighting individualized responses. The random forest analysis identified AV_D, weight group, and number of chemotherapy cycles as key predictors of VEGF reduction, with AV_D having the strongest dose-dependent effect. Boosted trees and ANN confirmed previous findings identifying AV_D as the most decisive factor, while weight and chemotherapy duration remained significant but less influential. Overall, AV_D consistently exhibits the strongest effect on VEGF reduction across all models, with weight and treatment duration also playing an important role. Conclusion: This study underlines the central role of increasing bevacizumab levels in reducing VEGF levels. However, it is not the absolute bevacizumab levels that matter, but rather the increase in those levels. The influence of weight and treatment duration suggests that individualized factors should be considered when optimizing therapy. This study clearly showed that machine learning techniques can uncover critical relationships between demographic characteristics and VEGF reduction, providing valuable insights for personalized treatment.
[1] Alidzanovic, L. et al. (2016). The VEGF rise in blood of bevacizumab patients is not based on tumor escape but a host-blockade of VEGF clearance. Oncotarget, 7(35), 57197–57212. [2] European Medicines Agency. (2023). Avastin EPAR product information. Retrieved from https://www.ema.europa.eu/en/documents/product-information/avastin-epar-product-information_en.pdf [3] Han, K. et al. (2016). Population pharmacokinetics of bevacizumab in cancer patients with external validation. Cancer chemotherapy and pharmacology, 78(2), 341–351. [4] Ghezelayagh, T. S. et al. (2023). Timing and duration of bevacizumab treatment and survival in patients with recurrent ovarian, fallopian tube, and peritoneal cancer: a multi-institution study. European journal of gynaecological oncology, 44(1), 17–25.
Reference: PAGE 33 (2025) Abstr 11699 [www.page-meeting.org/?abstract=11699]
Poster: Drug/Disease Modelling - Oncology