Time-varying association of first-line bevacizumab with progression-free survival in advanced ovarian cancer: A multicenter matched cohort study

Randomized trials have demonstrated that first-line bevacizumab extends progression-free survival (PFS) in advanced ovarian cancer. However, its long-term effectiveness in routine clinical practice remains unclear. This study investigated the time-varying association between first-line bevacizumab and PFS in patients newly diagnosed with advanced epithelial ovarian, fallopian tube, or primary peritoneal cancer. Our retrospective cohort, drawn from four Croatian centers, included 249 eligible patients. After coarsened exact matching (CEM), 175 patients were retained: 102 received first-line bevacizumab, and 73 received no bevacizumab in either the first or second line. We used CEM-weighted, center-stratified Cox models to estimate log-time-varying and piecewise associations. A 2,000-replication full-process bootstrap determined 36-month restricted mean survival time (RMST) differences. Overall survival (OS) was a key secondary outcome. The PFS association varied significantly over time (interaction p < 0.001). Hazard ratios (HRs) with 95% confidence intervals (CIs) were: 0.12 (0.04-0.34) at 3 months, 0.69 (0.47-1.02) at 12 months and 2.81 (1.22-6.47) at 36 months. However, the piecewise HR beyond 24 months was 1.36 (0.61-3.05), influenced by sparse late risk sets. The 36-month PFS RMST difference was 3.42 months (95% CI, -0.18 to 7.40), and the OS HR was 1.15 (0.74-1.76). In a sensitivity analysis, where exposure was defined solely by first-line treatment, 217 patients were matched. Here, the 3-month PFS HR was 0.33 (0.14-0.78), the 36-month PFS RMST difference was 0.50 months (-3.02 to 4.14), and the OS HR was 1.47 (1.02-2.10). The 36-month OS RMST difference remained consistent with no significant difference. First-line bevacizumab was linked to a lower early risk of progression or death. However, this association diminished over time and was sensitive to how exposure was defined. Limitations such as calendar-period separation, sparse late follow-up data, and potential residual confounding prevent causal interpretation and do not establish definitive late benefit or harm.