No age-adjusted or histologic-adjusted assessments of the association between extent of resection and risk of either recurrence or death exist for neurosurgical patients who undergo resection of low-grade glioma using intraoperative magnetic resonance image (MRI) guidance.
The current data included 156 patients who underwent surgical resection of a unifocal, supratentorial, low-grade glioma in the MRI suite at Brigham and Women's Hospital between January 1, 1997, and January 31, 2003. Estimates of disease-free and overall survival probabilities were calculated using Kaplan–Meier methodology. The association between extent of resection and these probabilities was measured using a Cox proportional hazards model. Observed death rates were compared with the expected death rate using age-specific and histologic-specific survival rates obtained from the Surveillance, Epidemiology, and End Results Registry.
Patients who underwent subtotal resection were at 1.4 times the risk of disease recurrence (95% confidence interval [95% CI], 0.7–3.1) and at 4.9 times the risk of death (95% CI, 0.61–40.0) relative to patients who underwent gross total resection. The 1-year, 2-year, and 5-year age-adjusted and histologic-adjusted death rates for patients who underwent surgical resection using intraoperative MRI guidance were 1.9% (95% CI, 0.3–4.2%), 3.6% (95% CI, 0.4–6.7%), and 17.6% (95% CI, 5.9–29.3%), respectively: significantly lower than the rates reported using national data bases.
The role of surgical intervention in the treatment of low-grade gliomas is not known with certainty. Despite the many projects1–20 that have examined the association between both the use and the extent of surgical resection with outcomes for neurosurgical patients who are diagnosed with low-grade gliomas,3 the relation remains unclear. In addition, although previous studies have examined the role of intraoperative magnetic resonance imaging (MRI) in tumor volume measurement,18, 20 the reports that exist generally are relatively small in size, present estimates that were not adjusted for age and/or histology, vary by study patient composition, or present study results in nonstandard statistical format. No study to date has utilized measurements of tumor volume and resection obtained using intraoperative (MRI) to calculate age-adjusted and histology-adjusted estimates of recurrence and survival rates (with associated standard errors) for patients with low-grade glioma.
The MRI suite at Brigham and Women's Hospital (BWH) in Boston is one of the first such operating rooms in the United States.21, 22 The creation of this suite represents a collaborative project of BWH and General Electric that began in 1990 and continues to this day. Currently, > 700 patients have undergone craniotomy in this unit. A benefit of this machinery is that tumor resections that are judged complete by either surgical field of view or preoperative imaging may be confirmed with reimaging while the patient still is in the operating room. In addition, associated MRI software allows for a more precise estimation of initial and resected MRI tumor volume, permitting an improved measurement of the exposure variable. Over the past few years, patients who present to BWH who have presumed low-grade glioma based on preoperative imaging studies typically undergo surgical resection within the MRI operating suite. The data for this series of patients were examined here under the hypothesis that a more precise measurement of the extent of surgical resection, like that obtained through use of intraoperative MRI, may help to clarify the role of resection in patients with low-grade glioma. In addition, an effort is made here to present these results in a statistically rigorous fashion that will allow other investigators to utilize these results for comparative purposes.
MATERIALS AND METHODS
Eligible study participants for this analysis included all patients who underwent resection within the intraoperative MRI operating suite at BWH in Boston who were diagnosed with a unifocal, histologically confirmed, low-grade (Grade 1 or 2), supratentorial, nonoptic pathway glioma between January 1, 1997 and January 31, 2003, and who had a minimum of 1 year of follow-up (n = 156 patients). The histologic subtypes included in this analysis were astrocytoma (International Classification of Diseases for Oncology [ICD-O] codes 9400, 9410, and 9420), mixed glioma (ICD-O code 9382), and oligodendroglioma (ICD-O code 9450). Pilocytic astrocytomas, gemistocytic astrocytomas, and gangliogliomas were not included. In addition, patients who were identified with chronic seizures or epilepsy also were excluded from this analysis. The study was approved by the Institutional Review Board at BWH.
There were two primary endpoints measured in this study: 1) time to recurrence and 2) time to death. Data on death were obtained from medical chart review as well as through use of the Social Security Death Index and the National Death Index. Data on recurrence were obtained through medical chart review, review of the BWH Brain Tumor Registry, or contact with treating physicians (oncologist, radiation oncologist). A patient was considered to have recurrent disease if it was revealed either by MRI imaging or by surgical resection or biopsy. Information on the following variables was obtained for all patients: age at diagnosis, gender, tumor volume, surgically resected tumor volume, histology, MIB-1 index, radiation therapy (yes or no), date of radiation therapy, chemotherapy (yes or no), date of chemotherapy, medical history (myocardial infarction, stroke, hypertension, cancer, diabetes), and history of smoking or alcohol use.
The extent of resection was measured using MRI studies that were obtained during surgery. Imaging is performed at time of initial positioning, at dural opening, and again at the completion of surgery. For this group of patients, the objective is complete resection, unless there is restriction due to the presence of tumor in eloquent areas of the brain. For patients who have preoperative imaging studies that indicate the involvement of eloquent areas, the procedure is performed using monitored and local anesthesia along with intraoperative cortical stimulation rather than with general anesthesia. The MRI machinery used in this operating suite is a 0.5-Tesla, intraoperative MRI system that was developed initially in collaboration with engineers from General Electric Medical Systems (Schenectady, NY) and consists of 2 vertically oriented, superconducting magnets with coils in separate but communicating cryostats.21, 22 Tumor volume measurements were conducted using three-dimensional (3D) Slicer software tools that were developed by our surgical planning laboratory in the Department of Radiology at BWH.21–25 These software tools comprise a wide range of operations, including capabilities for coregistration of different imaging data sets, manual or semiautomated segmentation, and generation of various 3D models and reformats. All of the imaging data sets (fast spoiled-gradient [FSPGR], T1, T2, gradient sequences) from each patient are transferred from the 0.5-T MR console to a UNIX network, where the images can be viewed and brought up for interactive segmentation (MRX; GE Medical Systems) on an Ultra 10 Workstation (Sun Microsystems, Mountain View, CA). In our patients, the majority of tumor measurements were performed on 3D spoiled-gradient (SPGR) images (with or without gadolinium contrast), in which the slices are thin (1.5–2.5 mm), allowing for more accurate delineation of the tumor. Interactive manual segmentation was performed by two operators (neurosurgeon and neuroradiologist) on both preoperative and postoperative SPGR sequences for all patients. For the few instances in which SPGR images were deemed inadequate due to motion or technical limitations, T2 sequences were used instead with thicker (5 mm) and fewer slices through the tumor. The tumor is outlined manually (freehand with a mouse) on each slice in which the interactive software connects all points to form a closed contour. The 3D Slicer editing feature also allows the display of multiple imaging sequences in multiple planes on the same patient simultaneously in 2D slices or 3D formats, facilitating a more accurate definition of the tumor margins that otherwise may be detected suboptimally on a single imaging plane or sequence. Volumetric calculations are then performed automatically by summation of all voxels of the enclosed tumor area on each slice that the 3D Slicer program had labeled from the segmentations. Residual tumor volume after resection is obtained in an identical fashion on postoperative SPGR images. An patient was considered to have a complete or macroscopic total resection of their lesion if SPGR or T2 imaging revealed no residual lesion at final imaging.
Comparison of patients by descriptor variables was done using a chi-square test or a Fisher exact test for discrete variables and a Student t test for continuous variables. Estimates of disease-free and overall survival probabilities were calculated using Kaplan–Meier estimates and were compared using a Wilcoxon log-rank test in SAS software.26 Hazard rates were computed using a Cox proportional hazards model27 in SAS.26
The expected numbers of deaths were estimated by determining the cumulative incidence for each patient according to his or her age at diagnosis, year at diagnosis, histology, and duration of follow-up. These data was calculated from the survival rates obtained from the Surveillance, Epidemiology, and End Results (SEER) Program as presented by the Central Brain Tumor Registry of the United States (CBTRUS) (Table 2).28 These rates are available for 1-year, 2-year, 5-year, and 10-year periods and include data from the SEER Program for the years 1973–1999.28 The length of follow-up for which a particular rate applied to an individual was determined for each patient, and the corresponding expectation was calculated by cumulating the product of the length of follow-up and the applicable rate. The relative risk was calculated by determining the ratio of the number of observed deaths to the expected number of deaths. Significance tests and 95% confidence intervals (95% CI) were based on the assumption under the null hypothesis that the observed number of deaths had a Poisson distribution with mean equal to the expected number of deaths.
Descriptive statistics for the sample are presented overall and by histologic subtype in Table 1. The majority of patients were diagnosed with oligodendroglioma (n = 95 patients), 35 patients were diagnosed with astrocytoma, and 26 patients were diagnosed with mixed tumors. The mean age of the sample was 41.6 years, and 45.9% of patients were women. The three histologic subgroups did not differ by mean age, gender, length of follow-up, mean tumor volume, mean residual tumor volume, extent of resection, mean MIB-1 value, use of chemotherapy and/or radiation therapy, or comorbid conditions.
Table 1. Descriptive Statistics of Study Sample by Histologic Subtype
All (n = 156)
Oligodendroglioma (n = 95)
Astrocytoma (n = 35)
Mixed (n = 26)
Mean age (yrs)
Gender (% female)
Tumor volume (mL)
Resected (mean %)
No. receiving radiation
No. receiving chemotherapy
No. of deaths
No. of recurrences
Mean follow-up (yrs)
The median follow-up was 3 years with a minimum follow-up of 1 year and a maximum follow-up of 6.8 years. At the time of last follow-up, 37 patients had developed a recurrence (18 oligodendrogliomas, 11 astrocytomas, and 8 mixed tumors). Among the patients with recurrent lesions, 32 recurrences were confirmed histologically by surgical resection, and all were confirmed by MRI imaging. Thirteen patients had died (8 patients with oligodendrogliomas, 3 patients with astrocytomas, and 2 patients with mixed tumors). The primary cause of death in all instances was attributed to glioma. Using data from the SEER Registry, the expected number of deaths in this group after correcting for age at diagnosis, date of diagnosis, and histologic subtype is 31 deaths (P = 0.05). There were no significant differences with regard to either the overall recurrence rates (log-rank test: P = 0.34) or the death rates (log-rank test: P = 0.63) by histologic subgroup; thus, estimates of these rates are presented for the overall sample (Figs. 1, 2). For all 3 histologic groups combined, the 1-year, 2-year, and 5-year recurrence rates (with 95% CIs) were 3.2% (95% CI, 2.5–6.1%), 9.4% (95% CI, 4.4–14.4%), and 44.6% (95% CI, 31.1–58.1%), respectively. The 1-year, 2-year, and 5-year death rates were 1.9% (95% CI, 0.3–4.2%), 3.6% (95% CI, 0.4–6.7%), and 17.6% (95% CI, 5.9–29.3%), respectively (see Table 2).
Table 2. One-Year, Two-Year, and Five-Year Relative Survival Rates for Astrocytoma, Oligodendroglioma, and Mixed Low-Grade Gliomas by Age Group: Surveillance, Epidemiology, and End Results Data, 1973–1999a
Histology (ICD code)
Age group (yrs)
Relative survival rate (%)
ICD: International Classification of Diseases.
The Surveillance, Epidemiology, and End Results Program of the National Cancer Institute and data from the Central Brain Tumor Registry of the United States (CBTRUS) website.
Astrocytoma (9410, 9420)
Mixed glioma (9382)
Individuals who developed recurrences did not differ by age, gender, resection status (complete vs. partial), mean MIB-1 value, use of chemotherapy, comorbid conditions, or histologic subtype. Individuals who died were significantly more likely to have residual tumor, to have received chemotherapy and/or radiation therapy (as a result of their residual tumor), to be older (46.8 years vs. 41.0 years), and to have a large initial tumor burden. There were no significant differences between the two groups with respect to gender, histologic subtype, comorbid conditions, or mean MIB-1 value.
Complete tumor resection was possible for 56 of 156 patients. No significant difference in time to recurrence was noted between individuals who underwent partial resection versus those who underwent total resection (log-rank test: P = 0.5). After adjustment for age, it was noted that patients who underwent partial resection were at 1.4 times the risk of recurrence (95% CI, 0.7–3.1) as patients who underwent total resection. However, the unadjusted overall death rate among patients who underwent total resection (1 of 56 patients) was significantly lower compared with the death rate (8 of 100 patients) among patients who underwent less than total resection (log-rank test: P = 0.05). After adjustment for age, it was noted that patients who underwent partial resection were at 4.9 times the risk of death (95% CI, 0.61–40.0) as patients who underwent total resection (Figs. 1 and 2).
The optimal management of adult, supratentorial, low-grade gliomas remains unclear.1–20 Current primary treatment options include surgical resection and/or radiotherapy, although definitive evidence for an optimal treatment remains elusive. Advances in molecular epidemiology indicate that certain histologic subtypes of low-grade gliomas, such as oligodendrogliomas, may have varied response to chemotherapeutic agents based on chromosomal changes, including loss of chromosome 1p and/or 19q, and, hence, may have improved survival with use of these agents.8, 29 Currently, however, these drugs are reserved primarily for use in individuals with lesions that include anaplastic components or high proliferative (MIB-1) indices.
Given the uncertainty regarding the association between surgical resection (defined as a binary exposure variable) and survival, investigators have made numerous attempts to examine this association in greater depth by dividing surgical patients according to the extent of surgical resection. The majority of those studies presented survival rates according to three categories–gross total resection, subtotal resection, or biopsy/no resection, while some investigators attempt to utilize a more numeric estimate of extent of resection by creating ordered categories, such as quartiles or deciles. A recent review of the literature that examined whether the extent of surgical resection is a significant prognostic factor in defining survival for patients diagnosed with low-grade glioma concluded that resection indeed may be a positive prognostic factor for survival, although the evidence is limited3; and a formal meta-analysis would be difficult given the limited statistical information available across studies.
The data in the current series of patients were examined under the hypothesis that a more precise measurement of the extent of surgical resection, like that obtained through use of intraoperative MRI, may help to clarify the role of resection in patients with low-grade glioma. Several other groups have utilized MRI readings to examine the association between extent of glioma resection and survival, although most included only high-grade lesions, or included a mix of high-grade and low-grade lesions, or were limited in sample size.18, 20 Although the risk estimates in our study suggest a relation between the extent of surgical resection and survival (i.e., 1 in 56 patients died in the gross total resection group vs. 8 in 100 patients died in the subtotal resection group), our results were not statistically significant. A number of explanations are possible for this finding. First, it is possible that surgical resection does not confer a benefit with respect to the survival of patients with low-grade glioma. However, it is also possible that the benefit may be real and that our data, although relatively large in terms of neurosurgical analyses, lacked the statistical power to detect a statistically significant effect, as evidenced in part by the wide range of the confidence intervals presented here. Because national data indicate that survival times differ significantly by histologic subgroups within the overall classification of “low-grade glioma,” it also is possible that the joint analysis of histologically diverse groups of patients, for whom the association between surgical resection and survival may have differing magnitude and direction, may result in an overall estimate of no effect (although, in these data, no difference in outcome was appreciated by histology). Furthermore, the benefit may be seen only with more extended follow-up—in these data, the median survival was approximately 3 years.
Despite the limited evidence indicating an association between extent of resection and time to recurrence or time to death, we did note significantly improved survivals for this group of patients overall relative to age-adjusted and histology-adjusted rates obtained from national data bases. The results from this analysis indicate that the age-adjusted survival rates for our patients were significantly longer than expected based on data from the CBTRUS/SEER sources. Because SEER data include all individuals with a given diagnosis, regardless of whether they have undergone surgical resection, comparison of our data with SEER data, although crude, may suggest a benefit to surgical resection. Because the current survival data were corrected for age at diagnosis, year of diagnosis, and histologic subtype, to some extent, this should correct for possible variations in survival due to competing causes of death (i.e., an older individual would be more likely to have additional illnesses that may lead to death), improved survival over time due to improvements in surgical technique and in general medical care,30 and differences in survival due to histologic subtype, respectively. Additional explanations for the improved survival noted in our series include possible nonhomogeneity of pathologic subtypes within the nationally reported data (i.e., anaplastic lesions placed within the nonanaplastic category, mixing of pathologic subtypes). Although all diagnoses reported to the SEER registries are confirmed by pathology reports, these reports are not re-reviewed uniformly and, thus, represent the opinions of a wide range of pathologists who may have varied levels of experience in the diagnosis of neurologic disorders. In a recent assessment of patients with brain tumors from the Connecticut Tumor Registry (a SEER site), absolute concordance rates of 49–81% were observed for the pathology review, with mixed tumors having relatively high discordance rates relative to pathology review.31
Numerous other retrospective studies have presented survival rates for patients with low-grade gliomas by surgical resection extent (gross total, subtotal, or biopsy); however, they are difficult to compare with the current data, because most of those studies did not present estimates by specific age or histologic groups, and nor did they provide confidence interval estimation for risk estimates. Furthermore, many of the previous studies included subtypes of patients that were not included in the current study, such as patients with diagnoses of epilepsy. The majority of studies that were similar to our study, however, reported lower survival rates than noted here.3
The data presented in the current study are retrospective and hospital-based, with all of the selection biases inherent to such data. At our institution, the majority of patients with presumed low-grade hemispheric gliomas are treated with surgical resection within the intraoperative MRI suite with the surgical objective being complete resection. This treatment selection is reflected in the relatively small variance estimates for total tumor volume resected (i.e., most patients have the majority of MRI-visualized tumor removed [i.e., > 80%], with the exception of patients who have lesions in eloquent areas of the brain). It is possible that this small variation in resection volume percentage may have affected our ability to detect any treatment effect, if it exists.
The current data suggest a benefit associated with surgery, but this finding should be considered preliminary and will require formal confirmation either within the context of a randomized clinical trial, with patients matched by surgeon or institution according to management preference, or a large, population-based project. To date, results from at least four clinical trials that have enrolled adult patients diagnosed with low-grade glioma have been published.29, 32–36 None of those trials randomized according to surgical resection; rather, they were randomized according to the use of radiotherapy and/or chemotherapy, although it is interesting to note that, within those trials, one of the primary predictors of outcome was the extent of surgical resection. To our knowledge, no population-based effort exists that has focused exclusively on patients with low-grade gliomas; thus, the commencement of such studies in the near future are likely to generate valuable information regarding outcomes for this group of patients.