Despite a high radiographic response rate in patients with recurrent glioblastoma following bevacizumab therapy, survival benefit has been relatively modest. We assess whether tumor volume measurements based on baseline and early posttreatment MRI can stratify patients in terms of progression-free survival (PFS) and overall survival (OS).
Baseline (−4 +/- 4 days) and posttreatment (30 +/- 6 days) MRI exams of 91 patients with recurrent glioblastoma treated with bevacizumab were retrospectively evaluated for volume of enhancing tumor as well as volume of the T2/FLAIR hyperintensity. Overall survival (OS) and progression-free survival (PFS) were assessed using volume parameters in a Cox regression model adjusted for significant clinical parameters.
In univariable analysis, residual tumor volume, percentage change in tumor volume, steroid change from baseline to posttreatment scan, and number of recurrences were associated with both OS and PFS. With dichotomization by sample median of 52% change of enhancing volume can stratify OS (52 weeks vs. 31 weeks, P = .013) and PFS (21 weeks vs. 12 weeks, P = .009). Residual enhancing volume, dichotomized by sample median of 7.8 cm3, can also stratify for OS (64 weeks vs. 28 weeks, P < .001) and PFS (21 weeks vs. 12 weeks, P = .036).
Glioblastoma is the most common primary malignant brain tumor in adults, with an incidence of 3.1 per 100,000 in the United States. Survival in patients diagnosed with glioblastoma remains dismal despite the best available standard therapy including surgery, radiation, and chemotherapy. In patients with recurrent glioblastoma, the median overall survival is typically 25 to 40 weeks.[2-4] With advances in our understanding of the molecular mechanisms driving tumor growth and invasion, an increasing number of targeted therapeutic agents are being developed and evaluated.
Vascular endothelial growth factor (VEGF) plays a key role in tumor angiogenesis, and has been a therapeutic target in several clinical trials for various human cancers, including glioblastomas.[5-7] In 2009, the humanized anti-VEGF antibody bevacizumab received US Food and Drug Administration approval for the treatment of recurrent glioblastoma due to the observed high response rates and prolonged progression-free survival (PFS) compared to historical controls.[3, 8-10] More than half of patients, however, fail to respond and many of the responders inevitably develop resistance to therapy.[11, 12] Therefore, identification of patient subgroups that will preferentially benefit from bevacizumab therapy may help guide therapeutic management decisions.
Although prior studies suggest that postoperative tumor volume, as well as extent of resection, have prognostic significance in newly diagnosed glioblastomas,[13, 14] the importance of tumor volume in patients with recurrent tumor receiving anti-angiogenic therapy is unclear. In patients with recurrent tumor receiving bevacizumab, results from a phase 2 clinical trial have shown that 4-week posttreatment response based on 2-dimensional criteria correlates with PFS. However, neither this study nor others, to our knowledge, have demonstrated the utility of using early posttreatment tumor volume in predicting survival.
Furthermore, recurrent glioblastomas often have imaging components that make traditional measurements challenging including irregular geometry, multifocality, and cystic or necrotic regions. Volumetric methods have the advantage of more reproducibly and precisely measuring the size of tumor,[15-18] and they are being increasingly used in this setting. For example, volumetric measurement of both the enhancing and nonenhancing tumor have been evaluated as imaging markers for treatment response, PFS, and OS.[15, 20]
In this study, we used a volumetric approach to evaluate tumor size in patients with recurrent glioblastoma receiving bevacizumab therapy at our institution. We assessed whether measurement of enhancing tumor volume based on MRI at baseline and 3 to 6 weeks after treatment initiation are useful predictors for PFS and OS.
MATERIALS AND METHODS
This retrospective study was approved by the institutional review board. Using a pharmacy database, we retrospectively identified 252 patients with pathologically confirmed glioblastoma (WHO IV) who initiated bevacizumab at our institution between December 2005 and July 2012 with recurrence based on clinical and imaging data. Recurrence was defined by new or increased size of enhancing tumor (> 25% bidimensional products) based on MRI prior to bevacizumab initiation. For patients who developed radiological progression within 12 weeks of completing radiation treatment, only patients with progressive clinical deterioration despite steroid treatment were included. Patients in the current study received bevacizumab with or without concurrent chemotherapy after failing no more than 3 prior treatment regimens, including standard radiation and temozolomide therapy, and underwent pretreatment MRI within 2 weeks of bevacizumab initiation and a follow-up MRI 3 to 6 weeks following bevacizumab initiation. Interpretable FLAIR and post-gadolinium T1-weighted (T1W) sequences performed on either 1.5-T or 3-T MRI systems were also required for both imaging time points.
Disease Progression Assessment
Previously published RANO criteria, including parameters for changes in T1-weighted gadolinium-enhancing lesion, as well as nonenhancing T2/FLAIR progression, were used to assess treatment response and disease progression. The dates of best response scan and progression scans were determined by a board-certified radiologist (R.H.) blinded to patients' clinical information. As defined by the RANO criteria, the 2-dimensional measurements were the sum of the products of the largest diameters and their maximum perpendicular diameters in the axial plane. In order to quantitatively assess abnormal T2/FLAIR signal intensity, this was also measured as the sum of the products of the largest diameter(s) and their maximum perpendicular diameter(s). PFS and OS were calculated with respect to the date of bevacizumab therapy initiation.
Tumor Segmentation and Volume Estimation
Whole-brain post-contrast T1-weighted and T2/FLAIR images from MRI obtained at baseline and after initiating bevacizumab were used for tumor volume segmentation. All tumor segmentations were done using 3D Slicer software, version 4.1 (Brigham and Women's Hospital, Boston, Mass)[22, 23] by an investigator (R.R.) who was blinded to clinical outcomes. User-driven manual active contour segmentation was used to acquire volume quantification for the regions of interest in the enhancing tumor target(s) as well as T2/FLAIR abnormalities. Manual editing of tumor contour was also performed to exclude nontumor regions such as areas of intrinsic T1 shortening, necrosis, or surgical cavity. The edited tumor contours were overlaid with source images and were evaluated in random order by a radiologist (R.H.) who was blinded to patient information to determine adequacy of tumor segmentation. Image volume was calculated by adding the number of pixels within volume contour and multiplying by pixel area and slice spacing. For multifocal tumors, the volume of separate lesions was summed together. Hence, for each patient, a baseline and posttreatment enhancing and “T2/FLAIR” volume was calculated. The percentage change in volume was calculated for both the enhancing and T2/FLAIR volume measurements for each patient. Also, the relative nonenhancing tumor ratio (rNTR), originally described by Norden et al was calculated for both the baseline and posttreatment scans, respectively.
The primary outcome measures for this study were PFS and OS. The Kaplan-Meier method was used to provide median point estimates and time specific rates. The Cox proportional hazards model was used in univariable and multivariable settings to identify imaging markers for PFS and OS. Following analysis of clinical and volume parameters in this fashion, the sample was dichotomized by the median of the sample for each volume parameter. Appropriate subanalyses were planned to account for significant clinical variables found in the initial analysis. All of the statistical analyses were performed using STATA, version 12.0 (Stata, College Station, Tex).
1. Patient clinical characteristics
A total of 91 patients (51 males) were selected using the screening criteria. Ninety patients had initial histological diagnosis of glioblastoma and 1 patient had a diagnosis of gliosarcoma. Among 11 of 91 patients who initiated bevacizumab within 12 weeks of the completion of concurrent chemoradiotherapy, tumor recurrence was supported by pathological confirmation on repeat resection (1 patient), detection of new satellite lesion (1 patient), PET imaging (2 patients), or marked clinical deterioration (7 patients). There were 12 patients with multifocal disease. Forty-seven patients were initiated on bevacizumab monotherapy, whereas 44 initiated bevacizumab with concurrent therapy. The concurrent therapies utilized in conjunction with bevacizumab included irinotecan (29 patients), temozolomide (6 patients), carboplatin (2 patients), carmustine or lumustine (4 patients), plerixafor (1 patient) and panobinostat (1 patient).
At the time of initial diagnosis, all patients were treated with temozolomide and radiotherapy following maximal tumor resection. The mean age of patients was 56.3 years (range, 23-88 years). Of 91 total patients, 47 patients (51%) were initiated on bevacizumab on first recurrence, 29 (32%) on second recurrence, 11 (12%) on third recurrence and 4 (4%) on fourth recurrence. The baseline MRI was obtained a mean of 4.0 days (±4.0) before bevacizumab initiation, and the follow-up MRI was obtained a mean 30.0 days (±6.0) after bevacizumab initiation. In terms of steroid usage, 51 patients were receiving dexamethasone (dose range, 0.25-24 mg; median, 4 mg) at the time of the baseline imaging. Forty-nine patients were receiving dexamethasone at time of immediate posttreatment scan (dose range, 0.5-30 mg, median 4mg). There were 16 patients who had an increased steroid dose at time of immediate posttreatment scan, while there were 49 and 26 patients with stable and decreased dexamethasone dosing, respectively. At the time of analysis, 70 patients had died, and 85 patients had progressed per RANO criteria.
The baseline clinical variables with respect to OS and PFS are summarized in Table 1. Among clinical variables, number of recurrences and change in steroid dose from baseline to posttreatment scan were associated with OS and PFS.
Change in steroid dose (from baseline to posttreatment)
Increase (n = 16)
Stable or decrease (n = 75)
Bevacizumab treatment regimen
Monotherapy (n = 47)
With concurrent chemotherapy (n = 44)
Yes (n = 12)
No (n = 79)
Time to bevacizumab initiation
Univariable and Multivariable Analyses of Imaging Parameters as Continuous Variables Adjusted for Clinical Parameters
Age, Karnofsky performance status (KPS), number of recurrences, baseline steroid dose, change in steroid dose, treatment regimen (bevacizumab monotherapy vs. concurrent chemotherapy), and re-resection before treatment were included in a Cox proportional hazards model for evaluation with each imaging-based volumetric parameter (Table 2). Among radiologic variables, baseline enhancing volume, posttreatment enhancing volume, percent change in enhancing volume, and posttreatment FLAIR volume were associated with OS and PFS. The rNTR was not predictive of OS or PFS. In multivariable analysis investigating baseline and posttreatment volume parameters, posttreatment enhancing volume remained significantly associated with both OS and PFS (Table 3).
Table 2. Volumetric Parameters as Continuous Variables Adjusted for Clinical Variables
Adjusted for age, Karnofsky performance status, number of recurrences, re-resection before treatment (yes or no), baseline steroid dose, change in steroids (yes or no) and treatment regimen (bevacizumab monotherapy versus concurrent chemotherapy) within Cox proportional hazards model.
Dichotomization of Volumetric Parameters by Using Sample Median
Enhancing volume parameters versus OS and PFS
For each volume parameter, we dichotomized values by the sample median and then calculated Kaplan-Meier estimates of overall survival. The results for dichotomized enhancing volume parameters with hazard ratios (HR) from Cox proportional hazards model adjusted for age, KPS, baseline steroid use, change of steroid dose (yes or no), number of recurrences, re-resection before treatment, and treatment regimen (bevacizumab monotherapy vs. bevacizumab with concurrent chemotherapies) are summarized in Table 4. When dichotomizing our sample by median baseline enhancing volume (19.46 cm3), the baseline enhancing volume was predictive of OS (P = .001) and PFS (P = .03, Fig. 1A). When dichotomizing our sample by median posttreatment enhancing volume (7.8 cm3), the posttreatment enhancing volume was predictive of OS (P < .001), as well as PFS (P = .036, Fig. 1B). When dichotomizing our sample using median percentage change in enhancing volume (52%), the percentage change was also predictive of OS and PFS (p=0.013 and p=0.009, respectively; Fig. 1C).
Table 4. Dichotomized Enhancing Volume Parameters Versus OS and PFS
Based on Cox proportional hazards model adjusted for age, Karnofsky performance status, number of recurrences, re-resection before treatment (yes or no), treatment (monotherapy vs concurrent chemotherapy), baseline steroid (yes or no), and steroid change (stable/decrease or increase).
Because a 64% reduction in enhancing volume numerically extrapolates to a 50% reduction in bidirectional product and therefore is equivalent as a partial or complete response per RANO criteria, we also examined this threshold in a similar analysis. Accordingly, patients with a greater than 64% reduction in enhancing volume associated with longer OS but not PFS (p=0.011 and p=0.37, respectively).
T2/FLAIR volume parameters versus OS and PFS
Baseline, posttreatment, and percentage change of T2/FLAIR volume was not predictive of OS or PFS when dichotomizing our sample by median value for each parameter (P > .05).
Combined Stratifications Using Percentage Volume Change and Residual Enhancing Tumor Volume
To examine the value of using both percentage volume change and residual enhancing volume, we first dichotomized the total patients group using 52% percentage change in enhancing volume, followed by dichotomization of each subgroup based on median residual enhancing of 7.8 cm3. Kaplan-Meier estimates of overall survival were calculated for each of these groups (Fig. 2). Based on the Cox proportional hazards model, residual enhancing volume was still a significant predictor of OS within the subgroups of both responders and nonresponders, although not a predictor of PFS. Of these 4 groups, patients who were responders per enhancing volume reduction, defined as a ≥ 52% response with a posttreatment enhancing volume of < 7.8 cm3 had the highest median overall survival, 70.3 weeks. The lowest median overall survival was of patients who were nonresponders with a posttreatment volume of > 7.8 cm3, because the median overall survival for this group was 21.1 weeks.
5. Subgroup analyses
Given that the number of recurrences was a significant predictor of overall survival in univariable and multivariable analyses, we examined volume parameters in subsets of patients at first and second recurrence, respectively.
For patients at first recurrence (n = 47), neither baseline enhancing volume nor percentage change were able to stratify the sample for survival using the previously listed median values for each parameter (P = .294 and P = .068, respectively). Dichotomization by posttreatment enhancing volume, however, remained statistically significant (HR 1.98, P = .044).
For patients at second recurrence (n=29), neither baseline enhancing volume nor percentage change were able to stratify the sample for survival (P = .168 and P = .289, respectively). Dichotomization by posttreatment enhancing volume, however, again remained statistically significant in patients (HR=2.83, P = .035).
5b. Bevacizumab monotherapy vs. bevacizumab with concurrent therapies
In order to account for effects of heterogeneous concurrent therapies with bevacizumab given to some patients, we examined volume parameters among the subset of patients who received only bevacizumab monotherapy and those who received concurrent chemotherapy with Kaplan-Meier estimates. For patients with bevacizumab monotherapy (n=47), baseline enhancing volume was not able to stratify the sample for survival (P = .118). Patients with > 52% change or greater than 7.8 cm3 at posttreatment were associated with longer overall survival (P = .024 and P < .001, respectively) in this subset.
For patients with concurrent therapy in combination with bevacizumab (n=45), only baseline enhancing volume was able to stratify the sample for survival (HR = 2.14, P = .033). Neither percentage change of enhancing nor posttreatment enhancing volume parameters, dichotomized by median, were able to stratify by survival (P = .177 and P = .059, respectively).
In this retrospective study, we assessed whether tumor volume parameters on pre-treatment and early posttreatment MRI are useful predictors of survival in patients with recurrent glioblastoma receiving bevacizumab therapy. Our results show that residual enhancing tumor volume and percentage change of enhancing tumor volume from baseline are both associated with OS and PFS (Table 4). Specifically, patients with residual enhancing tumor volume < 7.8 cm3 have a longer median PFS (20.9 weeks versus 12.0 weeks) and OS (64.1 weeks vs. 27.7 weeks). Similarly, patients with > 52% percentage reduction of enhancing volume on posttreatment scan compared to baseline also have a longer median PFS (20.9 weeks vs. 11.9 weeks) and OS (52.3 weeks vs. 31.0 weeks). Our results support a potential advantage for quantitative volumetric analysis early during bevacizumab treatment to identify patients who may benefit more durably from bevacizumab therapy.
Objective response, defined as > 50% reduction in enhancing lesion in 2D measurements following bevacizumab, has been shown to be a predictor of OS at 9, 18 and 26 weeks using data from the phase 2 BRAIN Trial, although PFS was not found to be associated.[9, 24] Another phase 2 trial of bevacizumab among heavily pretreated patients with recurrent glioblastoma indicated that early response at 4 weeks based on the Levin criteria is associated with a longer PFS, whereas early response based on objective Macdonald criteria did not demonstrate the same association.[3, 25] Our study indicates that early treatment response, assessed by percentage change in enhancing tumor volume before and after treatment, may serve as a predictor for both OS and PFS. Extrapolating the percentage threshold from 3D to 2D, a 64% change in volume approximates a 50% change in 2D diameter product and can also stratify patients with respect to PFS and OS in our analysis. Our volumetric approach has the advantage of avoiding areas of necrosis, cysts and surgical cavity, and may potentially estimate tumor size more accurately than linear methods.[15-17] With such measurements, volumetric analysis allows for a quantitative method of assessing early tumor change in size that appears to be associated with clinical benefit from bevacizumab treatment.
Surprisingly, the same predictive significance of percentage change of enhancing tumor was not observed in a previous study using a volumetric approach. The difference in the timing of posttreatment scan may help account for this discrepancy. Our study included patients with a posttreatment MRI scan performed between 3 to 6 weeks (mean of 4 weeks); in contrast, the posttreatment scan was obtained after 6 to 8 weeks in the Ellingson et al study. It is possible that the anti-permeability effect of bevacizumab may have had a greater impact on the measurement of enhancing tumor volume. If so, the percentage change in volume of enhancing lesion over a longer duration may less accurately reflect underlying tumor activity.
Residual enhancing volume appears to be associated with OS and PFS, but it was not clear if this association was indirectly related to percentage change in enhancing volume. Prior volumetric analysis has indicated that residual enhancing volume at 6 to 8 weeks is associated with PFS but not OS. To investigate potential interactions between posttreatment and percentage volume change, we used residual enhancing volume (< or ≥ 7.8 cm3) to stratify patients who had an early response (defined as > 52% reduction of enhancing volume) and patients without an early response into a total of 4 subgroups (Fig. 2). There is significant difference in median OS between the subgroups. The difference in PFS was not significant between the subgroups, possibly due to smaller sample size following combined stratifications. These findings suggest that even if there is interaction between residual enhancing tumor volume and percentage volume change, the combined use of both parameters appears to stratify patients with respect to OS. Specifically, patients with > 52% reduction in enhancing volume from baseline and residual enhancing tumor volume smaller then 7.8 cm3 have the longest median survival (70 weeks). Conversely, patients with less than 52% change in enhancing volume and residual enhancing tumor volume larger then 7.8 cm3 have the shortest median survival (21 weeks). Patients with > 52% change in enhancing volume and residual enhancing tumor volume larger than 7.8 cm3 and patients with < 52% change in enhancing volume and residual enhancing tumor volume smaller than 7.8 cm3 have an intermediate median survival (28 weeks and 45 weeks, respectively).
Given the current controversy about appropriate timing of bevacizumab treatment in patients with recurrent glioblastoma, it is important to assess volumetric parameters as a function of the number of recurrences to better clarify their contribution to overall outcome. As the number of recurrences was significantly associated with OS and PFS in both univariable and multivariable settings, we conducted an analysis stratified by this variable. In the patient subgroup with one recurrence (n = 47), residual enhancing tumor volume and percentage change in enhancing volume remained predictive of OS. In patients with exactly 2 recurrences (n = 29), only residual enhancing volume was predictive of OS, whereas percentage change was not. These parameters were also mostly not significant for PFS, which may be attributable to the relatively small sample size in each subgroup.
We also determined that baseline enhancing volume was not predictive of OS or PFS when examining individual patient groups with different number of prior recurrences, suggesting that the observed association between baseline enhancing volume and survival may be partly attributable to the number of prior recurrences. In addition, while baseline enhancing volume was associated with PFS and OS after adjusting for clinical variables, its contribution appears to be less important compared to residual enhancing volume in multivariable analysis (Table 3). Although baseline enhancing tumor volume may have some prognostic value, therapy-related effect, represented by residual and percentage change of enhancing volume, appears to contribute more to the observed survival differences in our study. Hence, baseline enhancing tumor volume alone is likely not sufficient for selection of patients for anti-angiogenic treatment.
Although univariable analysis indicated that posttreatment T2/FLAIR volume correlated with PFS and OS, multivariable analysis indicated that T2/FLAIR is not significant following adjustment of clinical and other volumetric parameters (Table 3). In this sample of heavily pretreated patients, T2/FLAIR changes related to prior surgery, radiation and chemotherapy rather than underlying tumor activity may have contributed to measured T2/FLAIR volume. Advanced imaging techniques such as diffusion imaging may potentially increase our ability to measure actual nonenhancing tumor volume. With the methods used in this study, however, we were unable to show predictive ability of T2/FLAIR volume for neither OS nor PFS.
Our volumetric analysis supports the potential use of early posttreatment imaging data in making therapy decisions for patients with recurrent glioblastoma receiving bevacizumab therapy. If our volumetric approach to stratification of patient survival can be validated in a prospective trial, the lack of percentage volume change or presence of large residual posttreatment tumor volume could be incorporated into clinical decision making. Patients who do not demonstrate early percentage change of volume or have large residual tumor volume may require modification of therapy regimen.
First, as a retrospective analysis, our study findings require prospective evaluation. In this study, we cannot determine whether the predictive nature of residual and percentage change of enhancing volume would be applicable for patients with recurrent glioblastoma who do not receive antiangiogenic therapy. To determine whether the volumetric imaging markers are specific to antiangiogenic therapy, the same analysis should be investigated in recurrent glioblastoma patients treated with non-antiangiogenic therapy. Although every patient in this study was initiated on a treatment regimen including bevacizumab, nearly half were also given concurrent chemotherapy. Subgroup analysis of patients who only received bevacizumab monotherapy indicated that dichotomization using residual and percentage change of enhancing volume were still significant predictors of OS and PFS. The significance of tumor volume parameters as imaging predictors in the setting of different therapy regimens should be confirmed in prospective trials.
The distinction of pesudoprogression and progression is important, as the misclassification of pseudoprogression as recurrence before bevacizumab treatment can affect outcome assessment. Although we cannot definitively exclude the possibility of including cases pseudoprogression based on our selection criteria, we believe the impact is relatively small because the majority (84/91) of patients demonstrated radiological progression at least 12 weeks after completion of radiation treatment or evidence of increased metabolic activity on PET imaging and distant enhancing disease outside the irradiated region.
The MRI sequence used to assess tumor volume in our study has an inter-slice thickness of 5 mm, potentially introducing partial volume averaging during the tumor segmentation. Although high-resolution 3D T1-weighted MR images would be expected to allow for a more precise volume measurement, a prior study indicated that performing volumetric analysis on traditional 2D imaging, as done in this study, leads to comparable measurements. Nonetheless, if quantitative volumetric analysis is to be implemented clinically to assist individual outcome prediction, tumor volume measurement based on 3D source imaging will likely be preferred. With advances in imaging acquisition speed, 3D T1- and T2-weighted whole brain image sequences are increasingly acquired in both trial and clinical settings and will be helpful in allowing for more accurate volume measurements.
We assessed baseline and early posttreatment volumetric parameters from T1-weighted and T2-weighted MRI imaging to evaluate their ability to predict OS and PFS in patients with recurrent glioblastoma being initiated on bevacizumab therapy. Volumetric percentage change and early posttreatment volume of enhancing tumor can stratify survival for patients with recurrent glioblastoma receiving bevacizumab therapy.
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURE
Dr. Wen has served on the board of Merck, Novartis, EMD, Genentech, and Serono; has received payment for lectures, including serving on speakers bureau for Merck, Genentech/Roche, and GlaxoSmithKline; and has received research support from Amgen, AstraZeneca, Boehringer Ingelheim, Esai, Exelixis, Genentech/Roche, Geron, MEdimmune, Merck, Novartis, Sanofi-Aventis, and Vascular Biogenics. Dr. Norden has been a consultant for Parexel and Brigham and Women's Hospital TIMI research group. Dr. Reardon has been a consultant for Genentech/Roche and has served on an advisory board for Genentech.