Exploratory unsupervised machine learning of angiogenesis biomarkers in a phase II advanced cervical cancer trial of radiochemotherapy with or without neoadjuvant chemotherapy.

Objective: In a prospective, randomized phase II study, exploratory data analysis was conducted to evaluate angiogenesis-associated plasma protein levels in patients with locally advanced cervical cancer METHODS: Participants were divided into two groups: Group A received neoadjuvant cisplatin and gemcitabine treatment (NAC) followed by chemoradiation with cisplatin and brachytherapy (CRT), while Group B received only CRT. Plasma samples were collected from patients in Group A at three time points: baseline, after NAC, and after CRT. Group B patients had samples taken at two time points: baseline and after CRT. The study analyzed an angiogenesis-associated panel of plasma proteins, including angiopoietin-2, G-CSF, endothelin-1, FGF-1, FGF-2, follistatin, IL-8, HGF, EGF, HB-EGF, PLGF, VEGF-A, VEGF-C, and VEGF- D. Receiver Operating Characteristic (ROC) analyses assessed the predictive value of baseline biomarkers for 12, 24, and 36-month survival outcomes. Additionally, Principal Component Analysis (PCA) was applied to post-CRT biomarker changes to identify coordinated modulation patterns. PCA was based on normalized delta values and eigenvector loadings, enabling identification of biomarkers aligned with Progression-Free Survival (PFS) and Overall Survival (OS).

Results: Significant differences were observed in the levels of HB-EGF, IL-8, PLGF, and VEGF-C between Groups A and B following CRT. Additionally, angiopoietin-2 levels showed a significant increase in Group B only. NAC treatment in Group A appeared to downregulate IL-8. CRT induced significant changes in HB-EGF, IL-8, PLGF, and VEGF-C levels in both groups. Patients in Group B demonstrated improved PFS and OS compared to those in Group A. Despite these differences in survival outcomes, the authors observed no significant intergroup differences in the tested biomarkers after completion of CRT. ROC analysis of baseline angiogenesis biomarkers demonstrated limited predictive sensitivity for survival outcomes. However, PCA of biomarker changes following CRT highlighted VEGF-A, HB-EGF, and angiopoietin-2 as variables associated with PFS and OS.

Conclusion: Baseline biomarker levels were not predictive of long-term outcomes. In contrast, CRT alone modulated key angiogenic biomarkers, and post-CRT biomarker changes were associated with improved survival. Such biomarker alterations were not observed following NAC, which was not associated with clinical benefit in this study. These findings underscore the value of dynamic biomarker evaluation and highlight how treatment strategies differentially impact biomarker profiles in advanced cervical cancer.