Impact of Intraoperative Stimulation Brain Mapping on
Glioma Surgery Outcome: A Meta-Analysis
Philip C. De Witt Hamer, Santiago Gil Robles, Aeilko H. Zwinderman, Hugues Duffau, and Mitchel S. Berger See accompanying editorial doi: 10.1200/JCO.2011.40.6959 Philip C. De Witt Hamer, Neurosurgical Center Amsterdam, Vrije Universiteit University Medical Center; Aeilko H. Zwinderman, Academic Medical Center, University of Amsterdam, the Netherlands; Santiago Gil Robles, Hospital Quiron, Madrid, Spain; Hugues Duffau, Ho pital Gui de Chauliac, Le Centre Hospitalier Regional Universita- ire de Montpellier, and Institut National de la Sant et de la Recherche Mdi- cale 1051, Institute for Neurosciences of Montpellier, Montpellier, France; and Mitchel S. Berger, Brain Tumor Research Center, University of Califor- nia, San Francisco, San Francisco, CA. Submitted July 24, 2011; accepted February 29, 2012; published online ahead of print at www.jco.org on April 23, 2012. Authors disclosures of potential con- icts of interest and author contribu- tions are found at the end of this article. Corresponding author: Philip C. De Witt Hamer, MD, PhD, Neurosurgical Center Amsterdam, VU University Medical Center, De Boelelaan 1117, 1081 HV Amsterdam, the Netherlands; e-mail: p.dewitthamer@vumc.nl. 2012 by American Society of Clinical Oncology 0732-183X/12/3099-1/$20.00 DOI: 10.1200/JCO.2011.38.4818 A B S T R A C T Purpose Surgery for inltrative gliomas aims to balance tumor removal with preservation of functional integrity. The usefulness of intraoperative stimulation mapping (ISM) has not been addressed in randomized trials. This study addresses glioma surgery outcome on the basis of a meta-analysis of observational studies. Methods A systematic search retrieved 90 reports published between 1990 and 2010 with 8,091 adult patients who had resective surgery for supratentorial inltrative glioma, with or without ISM. Quality criteria consisted of postoperative neurologic examination details and follow-up timing. New postoperative neurologic decits were categorized on the basis of timing and severity. Meta-analysis with a Bayesian random effects model determined summary event rates of decits as well as gross total resection rate and eloquent locations. Meta-regression analysis explored heterogeneity among studies. Results Late severe neurologic decits were observed in 3.4% (95% CI, 2.3% to 4.8%) of patients after resections with ISM, and in 8.2% (95% CI, 5.7% to 11.4%) of patients after resections without ISM (adjusted odds ratio, 0.39; 95% CI, 0.23 to 0.64). The percentages of radiologically conrmed gross total resections were 75% (95% CI, 66% to 82%) with ISM and 58% (95% CI, 48% to 69%) without ISM. Eloquent locations were involved in 99.9% (95% CI, 99.9% to 100%) of resections with ISM and in 95.8% (95% CI, 73.1% to 99.8%) of resections without ISM. Relevant sources of heterogeneity among studies were ISM, continent, and academic setting. Conclusion Glioma resections using ISM are associated with fewer late severe neurologic decits and more extensive resection, and they involve eloquent locations more frequently. This indicates that ISM should be universally implemented as standard of care for glioma surgery. J Clin Oncol 30. 2012 by American Society of Clinical Oncology INTRODUCTION Resective surgery is a widely used treatment for in- ltrative glioma, along with irradiation and chemo- therapy. The annual incidence for glioma is six per 100,000. Inltrativegliomas invadethebrain, relent- lessly recur, transform into higher-grade gliomas, and are invariably lethal. 1-3 Brain surgery for inltrative gliomas aims to balance extensive removal of glioma tissue inltrat- ing the brain with preservation of functional integ- rity by avoiding critical brain structures. The extent of glioma resection independently correlates with patient survival, 4-6 but when resective surgery ex- tends too far into inltrated critical structures, the patients condition permanently worsens and both quality of life andsurvival are compromised. Several neurosurgical advances aimto improve patient out- come. To localize critical brain structures and tu- mor, techniques such as preoperative functional neuroimaging, neuronavigation, uorescent dyes, magnetic resonance imaging (MRI) in the surgi- cal eld, and intraoperative stimulation mapping (ISM) have been used. 7 ISM is used to map brain function and to monitor neurologic performance, oftenunder local anesthesia. 8 Theusefulness of these techniques andtheir impact onneurologic outcome (as a measure of surgical safety) have not been ad- dressed in randomized trials or by meta-analysis. Here, we present event rates of newneurologic decits after resective brain surgery in adults with supratentorial inltrative glioma by performing a JOURNAL OF CLINICAL ONCOLOGY R E V I E W A R T I C L E 2012 by American Society of Clinical Oncology 1 http://jco.ascopubs.org/cgi/doi/10.1200/JCO.2011.38.4818 The latest version is at Published Ahead of Print on April 23, 2012 as 10.1200/JCO.2011.38.4818 Copyright 2012 by American Society of Clinical Oncology Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. meta-analysis of published observational studies. In particular, we addressed the impact of ISM and other determinants on neuro- logic outcome. METHODS Search Strategy Weaimedtoidentifyall publications reportingnewneurologicdecits in adult patients after resective glioma surgery, irrespective of the study design, setting, or language, inaccordance withMeta-Analysis of Observational Stud- ies in Epidemiology (MOOSE) guidelines. 9 Our data sources included the National Library of Medicine (PubMed/MEDLINE) andthe Excerpta Medica Database (EMBASE). Relevant reviews and editorials were assessed for publi- cations of potential interest, but no further publications were identied. The publication period was restricted to January 1, 1990, to January 1, 2010 (ac- cessed April 12, 2010). MeSHand EMTREE subject headings were combined for patient-related, treatment-related, and outcome-related characteristics (Data Supplement). Abstracts of identied studies were reviewed, and any study reporting neurologic outcome was included for full-text analysis by two independent observers (with 3- and 4-year qualications as neurosurgical attendings, re- spectively). In cases of multiple publications from the same cohort, the most recent report was chosen. Study Selection Criteria Publications included in our analysis had an adult population of 20 or more, provideddetails of newdecits todetermineseverityat patient level, and described the postoperative timing of neurologic assessment. Publications were excluded on the basis of location of gliomas other than supratentorial intraparenchymal, event rates of specic neurologic decits only, or glioma grading other than WHO grade 2 to 4. Disagreement on inclusion between observers was resolved by discussion or by using a third observer for adjudi- cation. The inter-rater agreement was calculated. In accordance with the MOOSE guidelines, several other quality criteria were scored that were not used for inclusion but were used to dene sets of publications of high and intermediate quality: aim of the study, consecutive recruitment, data collec- tion, loss of follow-up and missing data, surgical techniques, denition of eloquence, preoperative neurologic condition, eligibility criteria, histopathol- ogy, and discussion of possible confounders and bias. We searched the refer- ences by hand to identify additional studies. Outcome Measures and Denitions The primary outcome measure was the event rate of new postoperative neurologic decits. Decits were categorized according to severity (severe or less severe) andtimingof assessment (earlyor late). This resultedinfour decit event rates as end points. In addition, we included mortality from any cause within 30 days after resection. Decits were considered severe when they involved muscle strength grade 1 to 3 on the Medical Research Council Scale, aphasia or severe dyspha- sia, hemianopia, or a vegetative state. All other neurologic decits were con- sidered less severe, including grade 4 monoparesis, isolated central facial palsy or other cranial nerve decit, dysnomia, somatosensory syndrome, or parietal syndrome. Decits within 3 months after surgery were considered early and decits after 3 months were considered late, as is customarily considered the cutoff for permanency of postoperative neurologic decits. To assess the relation between ISMand neurologic outcome more specically, decits were extractedat study level andalsowithinstudies for patient populations that had resective surgery with and without ISM. Because considerable heterogeneity was expected across observational studies, several potential sources of heterogeneity were speciedandextracted from the publications. We distinguished sources related to publication, pop- ulation, or management characteristics. Publication-related characteristics were publication quality, year and continent of publication, study setting and design, prospective versus retrospective data collection, and the data source. Patient populationrelated characteristics were mean age; glioma volume and diameter; percentage of patients with preoperative decits, second surgery, and large gliomas, according to the authors denition; patients with gliomas located eloquently according to the authors denition; and patients with low-grade glioma (WHO grade 2) compared with high-grade glioma (WHO grade 3 or 4). 10 Treatment-related characteristics were preoperative diagnos- tics and intraoperative techniques. Preoperative diagnostics included func- tional MRI, white matter tractography, and functional magnetic source imaging. Intraoperative techniques includedpercentage of patients withresec- tions using ISM. For ISM, we distinguished electrostimulation frommotor or somatosensory evoked potentials. Other intraoperative techniques included intraoperativeMRI, neuronavigation, andultrasonography. Thepercentageof patients in whomgross total resections were obtained according to postoper- ative neuroimaging was also extracted. Statistical Analysis The endpoints were expressedas logit-transformedevent rates andwere analyzed in a random effects meta-analysis model. A random effects model was applied because indications for intervention, techniques, patient charac- teristics, and outcome assessments were likely to be dissimilar across studies. The randomeffects model took into account both within-study and between- study variation. The I 2 statistic quanties the percentage of total variation betweenstudies that is attributable toheterogeneity rather thantochance. 11-13 The relation between the potential sources of heterogeneity and outcome measures was explored by using logistic meta-regression modeling of covari- ates. Logistic regression coefcients were calculated. Odds ratios (ORs) and 95% Bayesian CIs were determined from regression coefcients and consid- ered statistically signicant for covariates when the 95% CI excluded one. A multivariate model was built on the basis of the deviance information crite- rium. 14 Bayesianhierarchical meta-analysis andmeta-regressionwere applied (Data Supplement), because event rates were frequently close to or equal to zero, and end point and covariate data were partially missing. 15-18 Given the event rates and the number of patients, the number of events in a study was representedbyabinomial distribution. Thelogit-transformedevent rates were assumed to be normally distributed with unknown means and unknown precisions using mean 0 and variance 1 10 4 as vague priors. Vague priors were chosen to primarily reect inference from the presented data without substantive prior knowledge. For the precision of between-study variations, we selected uniform distributions between 0 and 100. The median values of posterior distributions were used as summary event rate estimates with 95% credibility intervals based on three parallel chains of 50,000 Markov chain Monte Carlo simulation samples after a sufcient number of burn-ins using Gibbs sampling by JAGS software, version 2.2.0 (Jags Software, Newark, DE; http://mcmc-jags.sourceforge.net) called from the R2jags package, version 0.02-11 (http://cran.r-project.org/web/packages/R2jags/index.html) for R, version 2.12.1 (R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org). Sampling traces and distributions and Gelman- Rubin diagnostics were obtained by using the coda package for R, version 0.14-2 (http://cran.r-project.org/web/packages/coda/index.html). 19 No evi- dence against convergence was identied. Sensitivity analysis was performed by using alternative vague priors and by analyzing the high-quality subset. Similar posterior inferences were ob- tained. Publication bias was explored by plotting the event rate versus sample size and by funnel plot regression using the inverse variance as weight and zero-cell correction by adding 0.5. 20,21 RESULTS The search strategy retrieved 5,167 publications (Fig 1). After screen- ing, 366candidate publications were reviewedinfull text, including14 additional publications retrieved from citations. Consensus was reachedon90for inclusion, of which65were highquality and25were intermediate quality. The initial agreement between observers was substantial (Cohens 0.63). The study characteristics of the 90 publications are listed in the Data Supplement. The events mainlyconsistedof motor andlanguage De Witt Hamer et al 2 2012 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. decits (DataSupplement). Themedianfollow-upwas 7days for early outcome and 180 days for late outcome. The cohorts varied between 20 and 648 patients, with a median of 50 patients. Patients were includedconsecutivelyin45studies. Datawere collectedprospectively in14 studies. Nearly all studies reported observational cohorts, except for two randomized studies. These trials were also considered obser- vational with respect to neurologic outcome, because the primary outcome measure consisted of extent of resection. One trial random- ized the type of anesthesia, and the other randomized the use of uorescent dye to detect tumor. 22,23 Neurologic Events The overall meta-analysis of 90 studies included 8,091 patients. Not all studies allowed extraction of all end points. The summarized event rates are plotted in Figure 2. The event rates for each study are provided in the Data Supplement. Early severe decits were observed in 22.9% (95% CI, 16.5% to 30.6%) of 6,066 patients from 73 publi- cations, andearly decits of any severity were observedin30.3%(95% CI, 21.9%to 40.0%). Late severe decits were observed in 4.6%(95% CI, 3.3% to 6.1%) of 6,095 patients from 75 publications, and late decits of any severity were observedin7.1%(95%CI, 5.3%to9.0%). Mortality rates could be extracted from all 90 studies and were ob- served in 0.26%(95%CI, 0.01%to 0.50%). No evidence for publica- tion bias due to lack of small studies with high event rates was identied (P .814; Data Supplement). The heterogeneity of event rates was considerable for early de- cits (I 2 97.4%for severe decits and97.8%for any decits) andwas substantial for late decits (86.1% and 87.0%, respectively). The het- erogeneity for mortality was acceptable (56.7%). Subgroup Analysis of Sources of Heterogeneity and the Multivariate Model Because of this heterogeneity, we deemed an exploration of po- tential sources necessary (Data Supplement). Because avoiding per- manent neurologic decits is of primary importance, we focused on late severe decits for the subgroup analysis. The following factors were indicated as being signicantly related to late severe decits (in decreasing order of explainedvariance): (1) the percentage of patients with resections using ISM, particularly with electrostimulation, was 3.4%(95%CI, 2.1%to 5.0%) of late severe events (OR, 0.42; 95%CI, 0.23 to 0.72) rather than evoked potentials (5.7%; 95% CI, 2.9% to 10.2%; OR, 0.72; 95% CI, 0.36 to 1.40) compared with resections without ISM(7.7%; 95%CI, 5.0%to11.5%); (2) the continent, either NorthAmerica (4.0%; 95%CI, 2.2%to6.6%; OR, 0.34; 95%CI, 0.2to 0.83) or Europe (4.1%; 95%CI, 2.7%to 5.9%; OR, 0.35; 95%CI, 0.14 to 0.76) compared with other continents (11.3%; 95% CI, 5.2% to 21.5%); and (3) an academic study setting (4.3%; 95% CI, 3.0% to 5.7%; OR, 0.29; 95% CI, 0.09 to 0.99) compared with other settings (13.3%; 95% CI, 4.6% to 33.8%). A fourth signicant factor was whether ISMwas used for all patients in the study cohort (3.1%; 95% CI, 1.9% to 4.7%; OR, 0.43; 95% CI, 0.23 to 0.79) or in a subset of patients (5.5%; 95% CI, 2.7% to 10.0%; OR, 0.77; 95% CI, 0.34 to 1.76) compared with cohorts in which ISMwas not used (7.0%; 95% CI, 4.4%to10.6%). Obviously, this is avariant of the rst factor. These Publications excluded on third pass (full text) (n = 276) No details of neurologic deficits (n = 88) ) 4 6 = n ( y g o l o h t a p r e h t O ) 1 2 = n ( s t n e i t a p 0 2 n a h t r e w e F ) 7 1 = n ( e g a u g n a l e l b a t a l s n a r t n U ) 6 1 = n ( n o i t a c o l r e h t O ) 6 = n ( y r e g r u s e v i t c e s e r o N ) 5 = n ( w e i v e R Report on single type of deficit (n = 4) ) 3 = n ( a t a d t l u d a e l b a r a p e s o N ) 3 = n ( t r o h o c e t a c i l p u D ) 3 = n ( y p a r e h t n o i t a n i b m o C ) 2 = n ( t n e m s s e s s a f o g n i m i t o N Publications excluded on second pass (abstract) (n = 809) ) 7 3 3 = n ( e m o c t u o c i g o l o r u e n o N ) 0 9 1 = n ( s t n e i t a p 0 2 n a h t r e w e F ) 2 4 1 = n ( e l b a l i a v a n u t x e t l l u F ) 5 3 = n ( w e i v e R ) 4 3 = n ( y r e g r u s e v i t c e s e r o N ) 0 3 = n ( t n e m t a e r t n o i t a n i b m o C Report on single type of deficit (n = 24) ) 1 2 = n ( n o i t a c o l r e h t O ) 0 2 = n ( y g o l o h t a p r e h t O ) 0 1 = n ( a t a d t l u d a e l b a r a p e s o N Publications excluded on first pass (title) (n = 4,005) ) 8 6 6 = n ( s e t a c i l p u D ) 7 3 3 , 3 = n ( e l t i t f o s i s a b n O Potentially relevant publications identified by search (N = 5,167) MEDLINE (n = 2,984) EMBASE (n = 2,183) Publications included (n = 90) in meta-analysis High quality (n = 65) Intermediate quality (n = 25) Publications retrieved for more detailed evaluation (n = 1,162) Candidate publications (n = 352) Additional publications from reference lists (n = 14) Fig 1. Flow chart of publication selection. Meta-Analysis of Glioma Surgery Outcome www.jco.org 2012 by American Society of Clinical Oncology 3 Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. factors were considered the major sources of heterogeneity, as also substantiated by reduction of I 2 in the subgroups. The multivariate model had anoptimal t by using the study setting as a confounder to ISMon which the adjusted ORs were based. Intraoperative Stimulation Mapping Because ISM is the most important source of heterogeneity, we extracted subpopulations that had resections with and without ISM from the study cohorts (Fig 2). Early decits could be extracted from 42 publications for 3,602 patients after resections with ISM and from 26 publications for 1,537 patients after resections without ISM. Late decits couldbe extractedfrom41publications for 3,230patients after resections with ISM and from 33 publications for 1,731 patients after resections without ISM. Earlysevere decits occurredmore frequently inpatients after resections withISM(36.0%; 95%CI, 24.9%to49.1%) thanwithout ISM(11.3%; 95%CI, 5.9%to20.2%; adjustedOR, 4.25; 95% CI, 1.80 to 10.2). Importantly, event rates subsided to substan- tially fewer late severe decits withISM(3.4%; 95%CI, 2.3%to4.8%) than without ISM (8.3%; 95% CI, 5.6% to 11.3%; adjusted OR, 0.39; 95%CI, 0.23to0.64). Comparable results were obtainedfor decits of any severity. This indicates that reversible temporary loss of function of critical brain structures is more frequent with ISM, but irreversible neurologic damage is more effectively avoided, in comparison with resective surgery, without ISM. Of course, the favorable event rate needs to be in balance with tumor control. Because fewsurvival data were reported, we decidedto take extent of resection into account for tumor control. To illustrate this with an extreme example, neurologic events can be further re- duced by restricting resections to biopsy samples without having any therapeutic impact. The relation between functional and oncologic outcome for resective surgery with or without ISM was plotted in Figure 3Aas a percentage of gross total resections versus percentage of late severe decits. Evidently, the use of ISMdoes not compromise the gross total resection rate. On the contrary, these results indicate that gross total resections are achieved more frequently with ISM (74.9%; 95%CI, 67.1%to81.9%) thanwithout ISM(58.1%; 95%CI, 47.4%to 68.6%; OR, 2.15; 95%CI, 1.18 to 3.87). Moreover, the patient populations with ISM appear to be more complex because eloquent locations were involved in all patients. This is depicted in Figure 3B as percentage of eloquent localizations versus percentage of late severe decits. Gross total resections are generally more likely achieved in patients with less complex gliomas that are not in proximity to critical brain regions. Conversely, the closer the relation between glioma and critical brain region, the more likely a partial resection is. In this respect, the meta-regression lines in Figure 3 support the idea that ISM is generally not useful in patients with gliomas in non-eloquent loca- tions but is all the more important in patients with more complex gliomas to determine where to end a partial resection to avoid permanent neurologic decits. Thus, ISMis valuable in optimizing resective surgery by minimizing the rate of late severe decits and maximizing the extent of (partial) resection, particularly for glio- mas involving eloquent brain regions. DISCUSSION The main nding of this meta-analysis is that resective surgery with ISM is associated with reduction in late severe neurologic decits in adult patients with supratentorial inltrative gliomas two times greater thanthat with surgery without ISM, after anearly increase in temporary severe decits. The benecial neurologic A B I 2 (%) 95% CI (%) 97.8 96.8 to 98.5 97.5 96.4 to 98.4 95.3 92.9 to 97.0 Percentage 95% CI (%) 22.9 16.5 to 30.6 36.0 24.9 to 49.1 11.3 5.9 to 20.2 Early Neurologic Outcome Any Deficits Severe Deficits Late Neurologic Outcome Glioma resections With stimulation mapping Without stimulation mapping Glioma resections With stimulation mapping Without stimulation mapping Percentage 95% CI (%) 30.3 21.9 to 40.0 47.9 35.0 to 61.7 14.2 7.4 to 25.4 Publications 73 42 26 N 6,066 3,602 1,537 Publications 75 41 33 N 6,095 3,230 1,731 I 2 (%) 95% CI (%) 97.4 96.2 to 98.3 97.4 96.1 to 98.3 94.8 92.1 to 96.6 I 2 (%) 95% CI (%) 87.0 80.0 to 92.3 81.5 70.7 to 89.1 82.5 72.2 to 89.5 I 2 (%) 95% CI (%) 86.1 78.0 to 91.8 70.4 53.8 to 82.6 78.5 64.7 to 88.0 Percentage 95% CI (%) 7.1 5.3 to 9.0 6.4 4.4 to 8.8 9.4 6.4 to 13.1 Percentage 95% CI (%) 4.6 3.3 to 6.1 3.4 2.3 to 4.8 8.3 5.6 to 11.3 0 20 40 60 80 0 20 40 60 80 0 5 10 15 20 0 5 10 15 20 Any Deficits Severe Deficits e g a t n e c r e P e g a t n e c r e P e g a t n e c r e P e g a t n e c r e P Fig 2. Neurologic outcome of glioma resections. Meta-analysis results using a Bayesian random effects model have been summarized with 95% Bayesian CIs for all glioma resections (black diamonds), for glioma resections with intraoperative stimulation mapping (large gold squares), and without intraoperative stimulation mapping (small blue squares). Square sizes are drawn proportional to precision. Neurologic outcome is categorized by timing and severity. Timing distinguishes between early and late outcome at least 3 months after resection. Severity distinguishes between severe neurologic decits and any neurologic decits, including severe and less severe decits. The number of publications from which relevant data could be extracted is listed along with corresponding numbers of patients, the summary percentages with 95% CIs, and the measures of heterogeneity between studies (I 2 with 95% CIs). De Witt Hamer et al 4 2012 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. outcome from resective glioma surgery with ISM does not com- promise the extent of resection and involves eloquent locations more frequently. These ndings are in accordance with the concept that critical structures exist in the brain that cannot be resected without loss of functional integrity but that can be reliably mapped by electrostimu- lation during brain surgery. Electrostimulation interferes with neuro- logic performance by temporarily inactivating brainregions withhigh resolution, to judge before tissue removal whether a tumor can be resected without permanent loss of function. The observed transient neurologic decits usually subside within a few weeks to 3 months after resectionandare due tothe nearness of critical brainstructures to the resection cavity. In the margin of the resection cavity, resection- induced contusion, edema, and hypoperfusion are factors that likely contribute to these reversible decits. Another explanation could be the relocation of involved brain networks. 24 ISMenables more exten- sive tumor removal for better tumor control, 25 while reliably identify- ing critical brain structures so that permanent decits can be avoided at the price of transient decits. Several strategies can be used to minimize these transient decits in patients for whoma transient decit is considered unac- ceptable. First, the distance to the critical structures can be varied to create a safety margin by either increasing the stimulator current or by impinging the stimulator in the resection cavity wall for stimulation ahead of the resection margin. This distance has been shown to be an important factor for the time to recovery from transient decits. 26,27 Obviously, an increase of this distance com- petes with maximizing tumor removal, which may be considered more acceptable in poor-prognosis patients with less benet from an aggressive approach. Second, subpial dissection and particular care to avoid vascular damage helps minimize hypoperfusion and ischemia. Late and permanent decits are usually due to inadvertent postresection ischemia which can be evaluated by using diffusion- weighted MRI. 28,29 Third, perioperative corticosteroids minimize re- active edema. Fourth, relocation of functional networks may be facilitated by intense and specic rehabilitationafter surgery or, inthe future, by preoperative enhancement of brain plasticity, for instance by using transcranial magnetic stimulation. 24,30 To the best of our knowledge, this is the rst study to system- atically summarize safety of resective glioma surgery. The compar- ison of absolute risks of events with subgroup analysis of confounding factors is the best evidence available to assess the value of ISMwithout data fromrandomized trials. One strength of our study was the broad search strategy for articles covering mod- ern neurosurgical practice that were published over the last two decades. Thus, we were able to reasonably estimate the rate of postoperative neurologic events. In addition, the impact of several factors, including management techniques such as ISM, was as- sessed by meta-regression analysis. Sensitivity analysis indicates that our ndings were robust to model assumptions. Clearly, these ndings should be interpreted within the limita- tions of a meta-analysis based on observational studies, because data B A 0 With stimulation mapping Without stimulation mapping Publications 27 23 56 N 2,372 1,557 5,451 Percentage 95% CI (%) 74.8 66.5 to 81.7 58.3 48.4 to 68.0 68.6 62.6 to 74.4 Publications 46 24 69 N 3,783 1,232 5,752 Percentage 95% CI (%) 99.9 99.9 to 100 95.8 73.1 to 99.8 99.9 99.8 to 100 P u b l i c a t i o n s 7 5 4 1 3 3 N 6 , 0 9 5 3 , 2 3 0 1 , 7 3 1 P e r c e n t a g e
9 5 %
C I
( % )
4 . 6
3 . 3
t o
6 . 1 3 . 4
2 . 3
t o
4 . 8 8 . 3
5 . 6
t o
1 1 . 3 L a t e
S e v e r e
D e f i c i t s
( % ) Gross Total Resections (%) 25 20 15 10 5 20 40 60 80 0 0 0 1 With stimulation mapping Without stimulation mapping L a t e
S e v e r e
D e f i c i t s
( % ) Eloquent Locations (%) 25 20 15 10 5 20 40 60 80 100 240 120 60 20 Fig 3. Glioma surgery outcome after resections with and without intraoperative stimulation mapping. (A) The relation between oncologic and neurologic outcome is plotted as a bubble chart with percentages of gross total resections (x-axis) and percentages of late severe neurologic decits (y-axis) for patient populations after resection with (gold circles) and without (blue circles) intraoperative stimulation mapping. Sizes of circles are proportional to study cohort size. Regression lines are plotted. Lines are not straight because of inverse logit transformation. Summary estimates and 95% Bayesian CIs for all glioma resections (black diamonds), resections with intraoperative stimulation mapping (large gold squares), and without intraoperative stimulation mapping (small blue squares) are plotted and listed in the axis margins. (B) The relation between eloquent localization and neurologic outcome is plotted with percentage of eloquent localizations (x-axis) and percentages of late severe neurologic decits (y-axis). Color codes, sizes of circles, and logistic regression lines are similar to those in (A). Meta-Analysis of Glioma Surgery Outcome www.jco.org 2012 by American Society of Clinical Oncology 5 Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. are liable to be confounded by many factors, which is reected by the levels of heterogeneity. We therefore restrictedinclusionby methodo- logic quality and, within the restricted set, we distinguished studies of high and intermediate quality. In addition, we analyzed potential confounders by subgroup analysis. Furthermore, several sources of bias are possible. First, publicationbias canresult inour ndings being rather precise, withsmall 95%CIs, but less accurateas anestimationof the ground truth of outcome in general neurosurgical practice, be- cause published neurologic outcome does not necessarily reect gen- eral clinical practice. Our results pertain more to glioma patients treatedinlarge-volume academic hospitals dedicatedtoneuro-oncol- ogy than to patients treated in general hospital settings that infre- quently provide neuro-oncologic care. This is substantiated by the overall low rate of mortality, which is known to correlate with large- volume centers. 31,32 Second, the selectionbias of our ndings couldbe due to patient selection with various indications for surgical interven- tion. In this respect, the course of disease in patients who were denied surgerywouldbe of interest. Astricter indicationpolicylikelyprovides better neurologic outcome in a published cohort, whereas in referral centers, patients with more complex tumors are treated. Another source of bias concerns patient selection criteria for ISM, which is mainly considered in eloquently located gliomas, as is conrmed by the data. As a result, patients selected for ISM are more prone to neurologic decits. It is inconceivable that baseline risk levels are different because of the relative contraindications for ISMunder local anesthesia, such as dense presurgical motor or language decits, ex- tensive tumor mass effect, emotional instability, extreme obesity, or infancy. Third, neurologic assessment is subject to a certain level of suspicion. This observation bias can result in overestimation of neu- rologic events in studies that aim to assess functional outcome and, conversely, in underestimation of events in studies that determine other disease aspects, such as oncologic outcome. The frequency and timing of follow-up can be similarly affected by study perspective. Fourth, in meta-analyses the effect size can be inaccurate when many small studies are included, because these are liable to less quality of design, execution, and analysis, or because these focus on a subgroup of selected patients. Taking together the potential impact of this bias and differential baseline risks of study cohorts, the true difference in event rates between glioma resections with and without ISM is more likely to be larger than that observed rather than smaller. These ndings could have several implications for clinical practice. The estimates of neurologic events will likely serve patient counseling anddecisions for surgical management. Hence, our obser- vations consolidate arguments for universal adoption of ISMas stan- dard of care for glioma surgery because ISMincreases both safety and efcacy of tumor removal, particularly in eloquent locations. ISM shouldbe universally implementedinacademic neurosurgical centers withdedicatedneuro-oncologic care andcalls for more liberal referral patterns. ISMwas usedinonly19%of gliomaresections between1997 and 2000 in the United States. 33 There seems to be two general exceptions to the standard use of ISM. First, patients inwhomaradiologicallycompleteresectioncanbe achieved at a distance from critical brain structures do not benet fromISM. To verify eloquence, a reasonably pragmatic approachis to use empirical maps of residual tumor after resection rather than the intuition of the individual surgeon. 34 The second exception concerns patients in whom the advantage of decreased late neurologic decits does not offset the time to resolution of the transient decits. For instance, an older patient who is not in optimal condition with a presumed glioblastoma in an eloquent region of which an extent of resection of at least 80% 35 cannot be attained, is probably a poor candidate for ISM. As a rule of thumb this applies when the expected increase in survival from an aggressive resection approximates reha- bilitation time. Ultimately, a randomized controlled trial to determine the im- pact of ISM on surgical safety and survival could account for the bias and confounders inherent in our observations. So far, randomized controlled trials for surgical interventions have proven difcult to accomplish. Even more so, because our observations reinforce the assumption that it is notoriously difcult, if not impossible for ethical reasons, torecruit a control groupof patients withgliomas ineloquent brain regions for resective surgery without ISM. In conclusion, glioma resections using ISMare associated with a reductionof late severe neurologic decits more thantwotimes that of surgery without ISM and more extensive resection while involving eloquent locations more frequently. These observations indicate that ISM should be universally implemented as standard of care for gli- oma resections. AUTHORS DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST The author(s) indicated no potential conicts of interest. AUTHOR CONTRIBUTIONS Conception and design: All authors Collection and assembly of data: Philip C. De Witt Hamer, Santiago Gil Robles Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors REFERENCES 1. DeAngelis LM: Brain tumors. N Engl J Med 344:114-123, 2001 2. Wen PY, Kesari S: Malignant gliomas in adults. N Engl J Med 359:492-507, 2008 3. Behin A, Hoang-Xuan K, Carpentier AF, et al: Primary brain tumours in adults. Lancet 361:323-331, 2003 4. Smith JS, Chang EF, Lamborn KR, et al: Role of extent of resection in the long-term outcome of low-grade hemispheric gliomas. J Clin Oncol 26: 1338-1345, 2008 5. Stummer W, Reulen HJ, Meinel T, et al: Extent of resection and survival in glioblastoma multiforme: Identication of and adjustment for bias. Neurosurgery 62:564-576, 2008 6. Sanai N, Berger MS: Glioma extent of resec- tion and its impact on patient outcome. Neurosur- gery 62:753-764, 2008; discussion 264-266 7. Berger MS, Hadjipanayis CG: Surgery of in- trinsic cerebral tumors. Neurosurgery 61:279-304, 2007; discussion 304-305 8. Szele nyi A, Bello L, Duffau H, et al: Intraoper- ative electrical stimulation in awake craniotomy: Methodological aspects of current practice. Neuro- surg Focus 28:E7, 2010 9. Stroup DF, Berlin JA, Morton SC, et al: Meta- analysis of observational studies in epidemiology: A pro- posal for reportingMeta-analysis of Observational Studies in Epidemiology (MOOSE) group. JAMA 283: 2008-2012, 2000 10. Louis DN, Ohgaki H, Wiestler OD, et al: WHO Classication of Tumours of the Central Nervous System (ed 4). Geneva, Switzerland, World Health Organization, 2007 11. Higgins JP, Thompson SG, Deeks JJ, et al: Measuring inconsistency in meta-analyses. BMJ De Witt Hamer et al 6 2012 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved. 327:557-560, 2003 12. Viechtbauer W: Condence intervals for the amount of heterogeneity in meta-analysis. Stat Med 26:37-52, 2007 13. Higgins JP, Thompson SG: Quantifying heter- ogeneity in a meta-analysis. Stat Med 21:1539- 1558, 2002 14. Spiegelhalter DJ, Best NG, Carlin BP, et al: Bayesian measures of model complexity and t. J R Stat Soc B 64:583-639, 2002 15. Smith TC, Spiegelhalter DJ, Thomas A: Bayesian approaches to random-effects meta- analysis: A comparative study. Stat Med 14:2685- 2699, 1995 16. Hemming K, Hutton JL, Maguire MG, et al: Meta-regression with partial information on sum- mary trial or patient characteristics. Stat Med 29: 1312-1324, 2010 17. Higgins JP, Thompson SG, Spiegelhalter DJ: A re-evaluation of random-effects meta-analysis. J R Stat Soc Ser A Stat Soc 172:137-159, 2009 18. Thompson SG, Higgins JP: How should meta- regression analyses be undertaken and interpreted? Stat Med 21:1559-1573, 2002 19. Gelman A, Rubin DB: Inference from iterative simulation using multiple sequences. Stat Sci 7:457- 472, 1992 20. Egger M, Davey Smith G, Schneider M, et al: Bias in meta-analysis detected by a simple, graphical test. BMJ 315:629-634, 1997 21. Macaskill P, Walter SD, Irwig L: A comparison of methods to detect publication bias in meta- analysis. Stat Med 20:641-654, 2001 22. Gupta DK, Chandra PS, Ojha BK, et al: Awake craniotomy versus surgery under general anesthesia for resection of intrinsic lesions of eloquent cortex: A prospective randomised study. Clin Neurol Neu- rosurg 109:335-343, 2007 23. Stummer W, Pichlmeier U, Meinel T, et al: Fluorescence-guided surgery with 5-aminolevulinic acid for resection of malignant glioma: A ran- domised controlled multicentre phase III trial. Lan- cet Oncol 7:392-401, 2006 24. Duffau H: Lessons from brain mapping in surgery for low-grade glioma: Insights into associa- tions between tumour and brain plasticity. Lancet Neurol 4:476-486, 2005 25. Chang EF, Clark A, Smith JS, et al: Functional mapping-guided resection of low-grade gliomas in elo- quent areas of the brain: Improvement of long-term survivalClinical article. J Neurosurg 114:566-573, 2011 26. Gil-Robles S, Duffau H: Surgical management of World Health Organization Grade II gliomas in eloquent areas: The necessity of preserving a margin around functional structures. Neurosurg Focus 28:E8, 2010 27. Haglund MM, Berger MS, Shamseldin M, et al: Cortical localization of temporal lobe language sites in patients with gliomas. Neurosurgery 34:567- 576; discussion 576, 1994 28. Ulmer S, Braga TA, Barker FG 2nd, et al: Clinical and radiographic features of peritumoral infarction following resection of glioblastoma. Neu- rology 67:1668-1670, 2006 29. Smith JS, Cha S, Mayo MC, et al: Serial diffusion-weighted magnetic resonance imaging in cases of glioma: Distinguishing tumor recurrence from postresection injury. J Neurosurg 103:428-438, 2005 30. Desmurget M, Bonnetblanc F, Duffau H: Con- trasting acute and slow-growing lesions: A new door to brain plasticity. Brain 130:898-914, 2007 31. Cowan JA Jr, Dimick JB, Leveque JC, et al: The impact of provider volume on mortality after intracranial tumor resection. Neurosurgery 52:48-53, 2003 32. Barker FG 2nd, Curry WT Jr, Carter BS: Sur- gery for primary supratentorial brain tumors in the United States, 1988 to 2000: The effect of provider caseload and centralization of care. Neuro Oncol 7:49-63, 2005 33. Chang SM, Parney IF, Huang W, et al: Pat- terns of care for adults with newly diagnosed ma- lignant glioma. JAMA 293:557-564, 2005 34. Ius T, Angelini E, Thiebaut de Schotten M, et al: Evidence for potentials and limitations of brain plasticity using an atlas of functional resectability of WHO grade II gliomas: Towards a minimal com- mon brain. Neuroimage 56:992-1000, 2011 35. Sanai N, Polley MY, McDermott MW, et al: An extent of resection threshold for newly diagnosed glioblastomas. J Neurosurg 115:3-8, 2011
Meta-Analysis of Glioma Surgery Outcome www.jco.org 2012 by American Society of Clinical Oncology 7 Downloaded from jco.ascopubs.org on April 24, 2012. For personal use only. No other uses without permission. Copyright 2012 American Society of Clinical Oncology. All rights reserved.