The study population included patients with recurrent malignant gliomas who were treated with bevacizumab alone or in combination with irinotecan between August 2006 and September 2012 at Seoul National University Hospital and Seoul National University Bundang Hospital. Histological diagnosis of a grade of III or IV glioma as defined by the World Health Organization was required for inclusion in the study. In addition, to avoid the risk of recording inaccurate data due to pseudo-progression, only patients who demonstrated objective radiographic progression at > 12 weeks post-radiotherapy or in whom the majority of progressive disease was occurring outside the radiation field were included [12]. Bevacizumab (Avastin, Roche Pharma SChweiz AG, Reinach, Switzerland) alone (10 mg/kg) was administered intravenously once every 2 weeks, or was administered in combination with irinotecan (125 mg/m² for patients not receiving enzyme-inducing antiepileptic drugs [EIAEDs] or 340 mg/m² for patients receiving EIAEDs). The patients were followed and evaluated for clinical findings during every cycle of therapy. Brain magnetic resonance (MR) imaging was performed every three to four cycles, and when neurological deterioration was suspected. Clinical records and MR images of the patients were reviewed retrospectively. Responses were determined according to the new Response Assessment in Neuro-Oncology (RANO) criteria based on the MR imaging, clinical findings, and steroid requirements [13]. This study was approved by the Institutional Review Boards of both hospitals.

Pearson’s chi-square test or Fisher exact test was used for clinical comparisons between the two groups. The two-sample t test or Mann-Whitney test was used for comparison of the two groups for interval variables, as appropriate. Progression-free survival (PFS) was defined as the time from treatment initiation to documentation of disease progression or death from any cause, and overall survival (OS) was defined as the time from treatment initiation to death from any cause. The Kaplan-Meier method was used to estimate the median durations of PFS and OS, and PFS and OS were compared using log-rank tests.

Patients were categorized according to two groups based on the timing of bevacizumab discontinuation. Patients who continued bevacizumab until tumor progression were categorized as the late discontinuation (LD) group, while those who stopped bevacizumab therapy before tumor progression were categorized as the early discontinuation (ED) group. Landmark analyses were performed for comparison of survival times between the LD and ED groups. The timing of bevacizumab discontinuation was an event that varied over time; therefore, simple plotting of the survival function in both the LD and ED groups was inappropriate [14], and comparison of survival between the two groups using log-rank test was also inappropriate [14]. Landmark analysis determines OS from a specific time point after initiation of treatment (i.e., landmark). Thus, in landmark analysis, the OS for each patient is defined as the time from the landmark to death, not as the time from initiation of treatment to death. A survival function can then be plotted for comparison of survival rates between groups according to treatment outcome variables (the timing of discontinuation in our study). Three landmarks were selected in this study (weeks 9, 18, and 26). Because patient response was evaluated every 3-4 cycles (i.e., 6-8 weeks) during treatment, we believed that response evaluations and decisions regarding treatment continuation could best be made at those time points. Assessing the effects of ED on survival in primary non-responders is not relevant; therefore, our landmark analyses included only patients with a response showing stable or better than stable disease (SD) at their first response assessment visit. For each landmark analysis, multivariable analysis was performed using the Cox proportional hazard regression model to adjust for important baseline characteristics including age, performance status, number of relapses, and histologic grades, as well as the response achieved at each landmark. The proportionality of hazards assumption was checked by plotting the log minus log (LML) of the survival functions and the Cox proportional hazards model with time varying coefficients. The curves of LML for each variable were parallel, and time varying coefficients were not statistically significant. All p-values are the two-tailed type, and p < 0.05 was considered statistically significant. Statistical analyses were performed using IBM SPSS Statistics ver. 20.0 (IBM Co., Armonk, NY). The statistical methods and analyses used in this study were reviewed by the Medical Research Collaborating Center of Seoul National University Hospital.

A total of 94 consecutive patients were enrolled in this study. The median patient follow-up period was 4.8 years (range in living patients, 0.2 to 5.9 years). At the time of their last follow-up visit, 88 patients (93.6%) had discontinued bevacizumab containing chemotherapy and one patient was still receiving treatment (on the 19th cycle). The five remaining patients were lost to follow-up during treatment and were excluded from further analyses. Sixty-one patients who had discontinued bevacizumab treatment at the time of tumor progression and one patient still receiving bevacizumab therapy were classified as the LD group. Twenty-seven patients who discontinued bevacizumab treatment before tumor progression were classified as the ED group. The reasons for treatment discontinuation in the ED group included treatment toxicities in two patients (7.4%) (pulmonary thromboembolism and anaphylaxis), physician’s choice in five patients (18.5%), and patient’s choice in 20 patients (74.1%) (financial burden in 12 patients, unknown reasons in eight patients). The baseline characteristics of all 89 patients are shown in Table 1. Bevacizumab regimens (combination vs. single-agent), median duration, and cycles of treatment were comparable between the groups (Table 2); however, a higher response rate was observed in the ED group than in the LD group, although the difference was not statistically significant (45.8% vs. 22.6%, p=0.055).

Survival times in the LD and ED groups were compared by landmark analyses. The median survival times from each landmark (weeks 9, 18, and 26) in the LD and ED groups were as follows: 27.3 weeks versus 25.4 weeks (week 9); 19.1 weeks versus 19.0 weeks (week 18); and 14.3 weeks versus 18.1 weeks (week 26), respectively (Table 3). In comparison using log-rank tests, the residual survival times from all three landmarks were not significantly different between groups (Fig. 1). In multivariable Cox analysis, the timing of bevacizumab discontinuation was not a significant factor affecting survival at any time point (adjusted hazard ratio for the ED group, 1.02 [95% confidence interval (CI), 0.40 to 2.60] at week 9; 1.13 [95% CI, 0.56 to 2.29] at week 18, and 0.72 [95% CI, 0.37 to 1.39] at week 26) (Table 3). Because patients who achieve an objective response to bevacizumab treatment are more likely to be included in the ED group than those with SD, we further compared the two groups among both responders and non-responders separately at each landmark (Table 3). Although residual survival times from each landmark were not significantly different according to timing of bevacizumab discontinuation, a trend toward increased survival time was observed in the LD group compared to the ED group among patients who achieved an objective response at the 18th week (Fig. 2). The impact of bevacizumab discontinuation was also analyzed by histologic grades (glioblastoma multiforme and anaplastic glioma) (Tables 4 and 5). Patients with glioblastoma and anaplastic glioma showed clinical outcomes comparable to those found in the overall population.

In the ED group, 22 patients had experienced disease progression at the time of the last follow-up and the median time from discontinuation to progression (DTP) was 11.4 weeks (95% CI, 8.0 to 14.9) (Table 6). Among the 11 responders, the median time from DTP was prolonged to 13.3 weeks (95% CI, 11.1 to 15.5). Among the seven cases with a DTP of ≤ 8 weeks, two patients were primary non-responders and four patients had previously shown signs of impending progressive disease (within the range of SD) as recorded by MR images taken while on treatment. In comparison of patterns of disease progression between groups, a non-enhancing progression pattern was observed less frequently in the ED group than in the LD group (13.6% vs. 21.3%), although the difference was not statistically significant (p=0.746). The median time from progression to death was similar between the two groups (14.4 weeks [95% CI, 12.5 to 16.4] for the LD group vs. 15.7 weeks [95% CI, 12.3 to 19.1] for the ED group, p=0.251). The median time from discontinuation of bevacizumab to death was 28.6 weeks (95% CI, 25.0 to 32.1) in the ED group.

Among 83 patients (61 in the LD group and 22 in the ED group, respectively) who experienced disease progression at the time of the last follow-up, 38 patients (45.8%) underwent further treatment (Table 7). Fewer patients in the ED group underwent further treatment after disease progression (22.7% vs. 54.1%, p=0.013). Types of therapy and chemotherapy regimens were not significantly different between the two groups. Patients who underwent further therapy showed a prolonged median survival time following bevacizumab failure (17.7 weeks vs. 12.9 weeks, p=0.016). Patients in both the ED and LD groups who received further therapy showed longer survival times (Fig. 3). Among 28 patients who received chemotherapy, only one patient achieved an objective response. The median PFS and OS times were 6.0 weeks (95% CI, 4.6 to 7.4) and 17.0 weeks (95% CI, 6.9 to 27.1), respectively. Notably, three patients (two in the ED group and one in the LD group) were treated again with bevacizumab containing therapy, which was the most common form of post-bevacizumab treatment (40.0%) in the ED group. Re-introduction of bevacizumab resulted in a response rate of 33.3% and a PFS and OS of 6.1 weeks (95% CI, 3.9 to 8.4) and 38.4 weeks (95% CI, 4.1 to 72.7), respectively.