|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 25, No 16 (June 1), 2007: pp. 2288-2294 © 2007 American Society of Clinical Oncology. DOI: 10.1200/JCO.2006.08.0705 Epidermal Growth Factor Receptor Variant III Status Defines Clinically Distinct Subtypes of Glioblastoma
From the Departments of Radiation Oncology, Biostatistics and Applied Mathematics, Pathology, Neuro-Oncology, and Neurosurgery, The University of Texas M.D. Anderson Cancer Center, Houston, TX; Divisions of Biostatistics and Experimental Pathology and Department of Medical Oncology, Mayo Clinic, Rochester, MN; Department of Neurosurgery, University of California San Francisco School of Medicine, San Francisco, CA; and Neurosurgical Service, Massachusetts General Hospital, Boston, MA Address reprint requests to Kenneth Aldape, MD, Department of Pathology, Unit 85, The University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Blvd, Houston, TX 77030; e-mail: kaldape{at}mdanderson.org
Purpose: The clinical significance of epidermal growth factor receptor variant III (EGFRvIII) expression in glioblastoma multiforme (GBM) and its relationship with other key molecular markers are not clear. We sought to evaluate the clinical significance of GBM subtypes as defined by EGFRvIII status. Patients and Methods: The expression of EGFRvIII was assessed by immunohistochemistry in 649 patients with newly diagnosed GBM. These data were then examined in conjunction with the expression of phospho-intermediates (in a subset of these patients) of downstream AKT and Ras pathways and YKL-40 as well as with known clinical risk factors, including the Radiation Therapy Oncology Group's recursive partitioning analysis (RTOG-RPA) class. Results: The RTOG-RPA class was highly predictive of survival in EGFRvIII-negative patients but much less predictive in EGFRvIII-positive patients. These findings were seen in both an initial test set (n = 268) and a larger validation set (n = 381). Similarly, activation of the AKT/MAPK pathways and YKL-40 positivity were predictive of poor outcome in EGFRvIII-negative patients but not in EGFRvIII-positive patients. Pair-wise combinations of markers identified EGFRvIII and YKL-40 as prognostically important. In particular, outcome in patients with EGFRvIII-negative/YKL-40–negative tumors was significantly better than the outcome in patients with the other three combinations of these two markers. Conclusion: Established prognostic factors in GBM were not predictive of outcome in the EGFRvIII-positive subset, although this requires confirmation in independent data sets. GBMs negative for both EGFRvIII and YKL-40 show less aggressive behavior.
The epidermal growth factor receptor variant III (EGFRvIII) is the most common mutation of the epidermal growth factor receptor (EGFR) in glioblastomas (GBMs) and is present in 25% to 33% of all GBMs but only in GBMs showing EGFR amplification and overexpression.1 This variant is the result of the deletion of exons 2 to 7 (the extracellular protein-binding domain) and results in a constitutively active form of the receptor. Because it is not present in normal tissue, it is a potential target for tumor-specific therapy. Currently, considerable effort is being put into the development of anti-EGFRvIII agents for such strategies as immunotherapy/vaccination2-4 and molecular/antibody neutralization.5-8 However, anti–EGFR kinase therapy trials in GBM have been largely unsuccessful, even in patients in whom the gene is overexpressed.9 However, there is clinical evidence that EGFRvIII can predict responses to erlotinib and gefitinib,10,11 but the overall prognostic relevance of EGFRvIII in GBM remains unclear. The few observations that have been made in this regard are that EGFRvIII is prognostic in anaplastic astrocytoma but not GBM1,12; EGFRvIII is prognostic only in GBMs that exhibit EGFR amplification13; and EGFRvIII is a negative prognostic factor.14,15 However, possible reasons for these discrepant findings are that these studies have examined relatively small numbers of patients and have omitted other established molecular markers in the analyses. An additional possible reason includes variability in the methodology used, including different antibodies and scoring systems. Inclusion of additional markers, such as AKT/Ras pathway markers, has not been performed. These pathways are an important consideration because they are thought to play a critical role in the virulence of GBMs through their hyperactivation as opposed to their mutation.16 Indeed, activated downstream effectors of Ras have been shown clinically to correlate with decreased response to treatment and/or shorter survival times in patients with GBM.17-20 Regulators of Ras activation in GBM include, among others, EGFR,21 EGFRvIII,22,23 and YKL-40.24,25 To get a better understanding of the prognostic significance of EGFRvIII, we specifically examined the complex relationships between EGFRvIII and the activated downstream Ras pathway intermediates and YKL-40, as well as traditional clinical prognostic factors, in a large series of GBM patients.
Patients All patients had newly diagnosed GBM and paraffin-embedded tumor tissue available. The initial patient group consisted of 268 patients from The University of Texas M.D. Anderson Cancer Center and has been previously described.26 EGFRvIII expression was also studied in an additional 381 patients to increase the sample size and validate some of the results in the initial cohort, for a total of 649 patients. If enough tissue was available, staining for YKL-40 was also performed. The 381 patients in the validation group consisted of 210 patients from two clinical trials performed at the University of California, San Francisco,27,28 104 patients from clinical trials performed by the North Central Cancer Treatment Group,29-34 and 67 patients from the tissue bank at M.D. Anderson Cancer Center. All patients were treated from 1992 to 2003. All but six patients received radiation therapy in various fractionation schemes. The revised Radiation Therapy Oncology Group recursive partitioning analysis (RTOG-RPA) classification system for GBMs,35-37 which encompasses all of the clinical prognostic variables (ie, older age, lower Karnofsky performance score [KPS], extent of surgery, and so on), was used for comprehensive clinical risk stratification. When appropriate (in 80 of 649 patients, 12%), Eastern Cooperative Oncology Group performance scores were converted to KPS, as suggested in the literature,38,39 to ascertain the RTOG-RPA class for all patients. Although in most patients the RTOG-RPA was retrospectively assigned, the data used to determine the RTOG-RPA class were determined from the requisite clinical data at the time of diagnosis. The RTOG-RPA was assigned while blinded to survival.
Immunohistochemistry
Statistical Analysis
Description of Independent Patient Sets In the initial group of 268 GBM patients, tumors in 84 patients (31%) expressed EGFRvIII, whereas tumors in the remaining 184 patients (69%) did not express EGFRvIII. In these patients, the panel of markers also included panEGFR (57% positive) and p-EGFR (31% positive) as well as YKL-40 and phosphorylated intermediates of AKT and MAPK, as previously performed26 (Appendix Table A1, online only). The median age and survival time in this initial group were 59 years and 50 weeks, respectively. The median age and survival time in the validation group, which consisted of 381 patients, were 55 years and 56 weeks, respectively. There was no significant difference in age (P = .64) or survival (P = .28) between the two groups. YKL-40 expression was analyzed in 241 of these patients (when tissue was available). The institution from which the patients came did not have a significant impact on survival in both univariable and multivariable analyses that accounted for other prognostic variables (all P > .439).
Analysis of Initial Patient Set To determine whether the presence of a molecular marker altered the prognostic impact of another marker, we performed an initial univariable screen (Fig 1). Patients were stratified by the status of one molecular marker, whereas the remaining molecular markers, as well as age and KPS (because these factors were highly significant in predicting survival), were tested across these stratifications using Cox regression. Unique in these analyses was the disappearance or blunting of prognostic impact of all examined markers in the EGFRvIII-positive patients (numerical data showing these relationships are shown in Appendix Fig A1, online only). Because age has historically been the most robust prognostic marker in GBM, the observation that age was not a predictor of outcome in EGFRvIII-positive patients prompted further investigation.
Closer inspection revealed that there were few long-term (2-year) survivors among EGFRvIII-positive patients compared with EGFRvIII-negative patients (5% v 21%, respectively) and that there was little variability in the long-term survivorship regardless of the status of other markers. In particular, as shown in Table 1, the expected associations of YKL-40, p-MAPK, p-mTOR, p-70S6K, age, and KPS with survival were identified in patients with EGFRvIII-negative tumors. However, these prognostic associations tended to be blunted in the EGFRvIII-positive patients because neither significant associations nor trends in association between most of these markers and survival were seen within this group. The effect of EGFRvIII was still apparent after accounting for panEGFR status as well (Appendix Table A2, online only).
Interestingly, EGFRvIII positivity also seemed to diminish the prognostic effect of the RTOG-RPA classification system. That is, whereas the RTOG-RPA class was highly predictive of survival in the EGFRvIII-negative subset (P < .001), it did not seem to be associated with survival in the EGFRvIII-positive subset (P = .810). However, the numbers of class III and class V+VI patients in the EGFRvIII-positive set were small (n = 10 and n = 16, respectively). To increase the sample size, we examined EGFRvIII in a validation cohort.
Analysis of Validation Cohort
Multivariable Survival Analysis To systematically account for all other possible molecular interactions, a multivariable analysis of the clinical prognostic factors, all of the molecular markers, and marker-marker interaction terms was performed in the initial set (n = 268) of patients. The Cox model revealed that age, KPS, EGFRvIII, YKL-40, and the EGFRvIII/YKL-40 interaction term were independent prognostic factors within the initial group (Table 2). The interaction terms of all of the other molecular markers were not significant. The hazard ratios showed that the detrimental effects of EGFRvIII and YKL-40 were not additive or synergistic.
Because YKL-40 seemed to be an independent molecular prognostic factor and showed a statistically significant interaction with EGFRvIII, the YKL-40 status was ascertained in the validation patient set in all patients in whom there was sufficient tissue available (241 of 381 patients, 63%). This subset of the validation set was used in subsequent multivariable analyses. As in the initial group, Cox modeling of the validation set revealed that age, KPS, EGFRvIII, YKL-40, and the EGFRvIII/YKL-40 interaction term were independent prognostic factors, along with extent of resection. Collectively, age, KPS, EGFRvIII, YKL-40, the EGFRvIII/YKL-40 interaction term, and extent of resection were independent prognostic factors for the entire set of patients in which these markers were examined (n = 509). The relevance of the statistical interaction between EGFRvIII and YKL-40 is best illustrated in Figure 3. Patients with an EGFRvIII-negative/YKL-40–negative status (81 of 509 patients, 16%) defined a favorable subgroup with a significantly longer survival time (median survival time, 85 weeks) compared with the survival time in patients with the other three marker combinations (median survival time, 47 to 56 weeks; P < .001). There was no significant difference in survival between the EGFRvIII-positive only group (n = 36), YKL-40–positive only group (n = 282), and EGFRvIII- and YKL-40–positive group (n = 110; all P > .588). Because of the statistical interaction, positivity for both markers did not have an additive or synergistic detrimental effect on patient survival time.
CART Analysis Because the EGFRvIII status seemed to dictate the prognostic accuracy of the clinical and molecular markers, a regression tree was generated for the initial group by performing a CART analysis of the EGFRvIII-negative and -positive patients to determine whether patients could be better stratified according to risk. The CART analysis failed to generate an optimal tree for the EGFRvIII-positive patients from the molecular data and clinical variable data (age, KPS, and extent of resection) in the initial patient set (Appendix Fig A2A, online only). However, a tree was identified in the EGFRvIII-negative patients and showed that age and YKL-40 status defined three risk groups (Appendix Fig A2B). The prognostic importance of age, YKL-40, p-mTOR, and p-MAPK in the EGFRvIII-negative tumors is estimated by their variable importance scores of 100, 67, 12, and 2, respectively (Appendix Fig A2C). The remaining clinical and molecular factors had scores of 0. The variable importance scores reflect the contribution each variable makes in classifying or predicting the target variable. The variable used to split the root node (first division) is ranked as the most important and receives a score of 100. A score of 0 indicates that a variable played no role in the tree generation. We then performed a CART analysis of the patients in the validation set for whom EGFRvIII and YKL-40 data were available (n = 241). Again, although no regression tree was seen for the EGFRvIII-positive patients (Appendix Fig A2D), YKL-40 status and age emerged as the most important prognostic factors in the EGFRvIII-negative patients, whereas KPS and surgery did not affect the tree splitting (Appendix Figs A2E and A2F). Although the relative prognostic importance of YKL-40 and age was different between the initial and validation sets, it is noteworthy that the same two variables (age and YKL-40 status) emerged as the two most important factors.
When the EGFRvIII-positive patients from both the test and validation groups were combined (n = 146), CART analysis still failed to generate an optimal tree (Fig 4A). The composite regression tree generated for all of the EGFRvIII-negative tumors (n = 363) using the validated variables of age and YKL-40 status is shown in Figure 4B. The variable importance scores were 100 for YKL-40 and 15 for age (Fig 4C). The median survival times for the CART-derived groups 1 (YKL-40 negative), 2 (YKL-40 positive; age < 47 years), and 3 (YKL-40 positive; age
This report both extends and clarifies the prognostic relevance of EGFRvIII by elucidating two new findings. First, we found that, although the expression of EGFRvIII itself does not seem to be a strong prognostic factor in GBM, certain characteristics distinguish EGFRvIII-positive tumors from EGFRvIII-negative tumors. Second, we reinforced our prior finding of YKL-40 as a prognostic marker by identifying an interaction of this marker with EGFRvIII. In particular, we found that tumors negative for both markers seemed to be less clinically aggressive than tumors positive for one or both markers. This is illustrated by the 2-year survival rates, which were 43% in the former group and only 12% in the latter group. We further found that clinical risk factors, as defined by the RTOG-RPA classification, that are well known as strong predictive factors in malignant glioma were predictive in the patients with EGFRvIII-negative tumors but not in the patients with EGFRvIII-positive tumors; this was an unexpected finding that was then validated in an independent sample set. We do note that the number of EGFRvIII-positive patients is smaller than the number of EGFRvIII-negative patients, and there was a limited number of patients (n = 134) in one of the RTOG-RPA subclasses (V+VI) in the EGFRvIII-positive subgroup. Nevertheless, the greatest hazard ratio with respect to RTOG-RPA class in survival is seen in the EGFRvIII-negative patients, and it seems to diminish in the EGFRvIII-positive patients, but the patient numbers in the RPA classes of V and VI are too small to make definitive statistical conclusions. A parallel finding in our study was that molecular markers of the downstream Ras pathways, including p-70S6K and p-MAPK, which were prognostic in the EGFRvIII-negative patients, showed no consistent relationship with survival time in the EGFRvIII-positive patients. Our CART analyses similarly identified patient age and YKL-40 negativity as important factors in the EGFRvIII-negative group, but no such factors could be identified in the EGFRvIII-positive group. We note that the results of the CART analysis were not identical in the initial versus validation groups. Specifically, age was most important in the initial set followed by YKL-40, whereas the validation set showed that YKL-40 was the most important factor followed by patient age. However, an overall consistency is observed by the fact that YKL-40 and age emerged as robust risk stratifiers in both sample sets. These data also clarify the prognostic relevance of activated Ras/PI3K pathway intermediates. The status of these markers is known in 268 patients, and their incidences are listed in Appendix Table A1. Although significantly associated with survival in previous reports19,20,26,44 and within the univariable survival analyses of the initial set of patients, these markers are not independent prognostic factors when age, KPS, extent of surgery, and the status of YKL-40 and EGFRvIII are included within the analysis. We have previously shown that YKL-40 correlates with the Ras/PI3K phospho-intermediates in human GBM tumor specimens.26 YKL-40 is thought to stimulate these pathways.24,25 It is possible that YKL-40 status may serve as a surrogate measurement of Ras/PI3K activation, whereas EGFRvIII status, which had no correlation with these markers (data not shown), may represent a different yet equally virulent pathway.
The authors indicated no potential conflicts of interest.
Conception and design: Christopher E. Pelloski, Kenneth Aldape Provision of study materials or patients: Amy B. Heimberger, Dima Suki, Michael D. Prados, Susan M. Chang, Fred G. Barker II, Jan C. Buckner, C. David James, Kenneth Aldape Collection and assembly of data: Shiao Y. Woo, Amy B. Heimberger, Fred G. Barker II, Kenneth Aldape Data analysis and interpretation: Christopher E. Pelloski, Karla V. Ballman, Alfred F. Furth, Li Zhang, E. Lin, Erik P. Sulman, J. Matthew McDonald, W.K. Alfred Yung, Howard Colman, Shiao Y. Woo, Amy B. Heimberger, Dima Suki, Susan M. Chang, Fred G. Barker II, Jan C. Buckner, Kenneth Aldape Manuscript writing: Christopher E. Pelloski, Karla V. Ballman, Erik P. Sulman, Krishna Bhat, J. Matthew McDonald, W.K. Alfred Yung, Howard Colman, Michael D. Prados, Susan M. Chang, Fred G. Barker II, Jan C. Buckner, C. David James, Kenneth Aldape Final approval of manuscript: Christopher E. Pelloski, Karla V. Ballman, Alfred F. Furth, Li Zhang, E. Lin, Krishna Bhat, J. Matthew McDonald, W.K. Alfred Yung, Howard Colman, Shiao Y. Woo, Amy B. Heimberger, Dima Suki, Michael D. Prados, Susan M. Chang, Fred G. Barker II, Jan C. Buckner, C. David James, Kenneth Aldape
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Aldape KD, Ballman K, Furth A, et al: Immunohistochemical detection of EGFRvIII in high malignancy grade astrocytomas and evaluation of prognostic significance. J Neuropathol Exp Neurol 63:700-707, 2004[Medline] 2. Heimberger AB, Crotty LE, Archer GE, et al: Epidermal growth factor receptor VIII peptide vaccination is efficacious against established intracerebral tumors. Clin Cancer Res 9:4247-4254, 2003 3. Beers R, Chowdhury P, Bigner D, et al: Immunotoxins with increased activity against epidermal growth factor receptor vIII-expressing cells produced by antibody phage display. Clin Cancer Res 6:2835-2843, 2000 4. Wu AH, Xiao J, Anker L, et al: Identification of EGFRvIII-derived CTL epitopes restricted by HLA A0201 for dendritic cell based immunotherapy of gliomas. J Neurooncol 76:23-30, 2006[CrossRef][Medline] 5. Perera RM, Narita Y, Furnari FB, et al: Treatment of human tumor xenografts with monoclonal antibody 806 in combination with a prototypical epidermal growth factor receptor-specific antibody generates enhanced antitumor activity. Clin Cancer Res 11:6390-6399, 2005 6. Mishima K, Johns TG, Luwor RB, et al: Growth suppression of intracranial xenografted glioblastomas overexpressing mutant epidermal growth factor receptors by systemic administration of monoclonal antibody (mAb) 806, a novel monoclonal antibody directed to the receptor. Cancer Res 61:5349-5354, 2001 7. Allen LF, Lenehan PF, Eiseman IA, et al: Potential benefits of the irreversible pan-erbB inhibitor, CI-1033, in the treatment of breast cancer. Semin Oncol 29:11-21, 2002[Medline] 8. Modjtahedi H, Moscatello DK, Box G, et al: Targeting of cells expressing wild-type EGFR and type-III mutant EGFR (EGFRvIII) by anti-EGFR MAb ICR62: A two-pronged attack for tumour therapy. Int J Cancer 105:273-280, 2003[CrossRef][Medline] 9. Rich JN, Reardon DA, Peery T, et al: Phase II trial of gefitinib in recurrent glioblastoma. J Clin Oncol 22:133-142, 2004 10. Haas-Kogan DA, Prados MD, Tihan T, et al: Epidermal growth factor receptor, protein kinase B/Akt, and glioma response to erlotinib. J Natl Cancer Inst 97:880-887, 2005 11. Mellinghoff IK, Wang MY, Vivanco I, et al: Molecular determinants of the response of glioblastomas to EGFR kinase inhibitors. N Engl J Med 353:2012-2024, 2005 12. Liu L, Backlund LM, Nilsson BR, et al: Clinical significance of EGFR amplification and the aberrant EGFRvIII transcript in conventionally treated astrocytic gliomas. J Mol Med 83:917-926, 2005[CrossRef][Medline] 13. Shinojima N, Tada K, Shiraishi S, et al: Prognostic value of epidermal growth factor receptor in patients with glioblastoma multiforme. Cancer Res 63:6962-6970, 2003 14. Heimberger AB, Hlatky R, Suki D, et al: Prognostic effect of epidermal growth factor receptor and EGFRvIII in glioblastoma multiforme patients. Clin Cancer Res 11:1462-1466, 2005 15. Feldkamp MM, Lala P, Lau N, et al: Expression of activated epidermal growth factor receptors, Ras-guanosine triphosphate, and mitogen-activated protein kinase in human glioblastoma multiforme specimens. Neurosurgery 45:1442-1453, 1999[CrossRef][Medline] 16. Guha A, Feldkamp MM, Lau N, et al: Proliferation of human malignant astrocytomas is dependent on Ras activation. Oncogene 15:2755-2765, 1997[CrossRef][Medline] 17. Chakravarti A, Delaney MA, Noll E, et al: Prognostic and pathologic significance of quantitative protein expression profiling in human gliomas. Clin Cancer Res 7:2387-2395, 2001 18. Pelloski CE, Mahajan A, Maor M, et al: YKL-40 expression is associated with poorer response to radiation and shorter overall survival in glioblastoma. Clin Cancer Res 11:3326-3334, 2005 19. Mawrin C, Diete S, Treuheit T, et al: Prognostic relevance of MAPK expression in glioblastoma multiforme. Int J Oncol 23:641-648, 2003[Medline] 20. Choe G, Horvath S, Cloughesy TF, et al: Analysis of the phosphatidylinositol 3'-kinase signaling pathway in glioblastoma patients in vivo. Cancer Res 63:2742-2746, 2003 21. Chakravarti A, Chakladar A, Delaney MA, et al: The epidermal growth factor receptor pathway mediates resistance to sequential administration of radiation and chemotherapy in primary human glioblastoma cells in a RAS-dependent manner. Cancer Res 62:4307-4315, 2002 22. Li B, Yuan M, Kim IA, et al: Mutant epidermal growth factor receptor displays increased signaling through the phosphatidylinositol-3 kinase/AKT pathway and promotes radioresistance in cells of astrocytic origin. Oncogene 23:4594-4602, 2004[CrossRef][Medline] 23. Moscatello DK, Holgado-Madruga M, Emlet DR, et al: Constitutive activation of phosphatidylinositol 3-kinase by a naturally occurring mutant epidermal growth factor receptor. J Biol Chem 273:200-206, 1998 24. Ling H, Recklies AD: The chitinase 3-like protein human cartilage glycoprotein 39 inhibits cellular responses to the inflammatory cytokines interleukin-1 and tumour necrosis factor-alpha. Biochem J 380:651-659, 2004[CrossRef][Medline] 25. Recklies AD, White C, Ling H: The chitinase 3-like protein human cartilage glycoprotein 39 (HC-gp39) stimulates proliferation of human connective-tissue cells and activates both extracellular signal-regulated kinase- and protein kinase B-mediated signalling pathways. Biochem J 365:119-126, 2002[CrossRef][Medline] 26. Pelloski CE, Lin E, Zhang L, et al: Prognostic associations of activated mitogen-activated protein kinase and Akt pathways in glioblastoma. Clin Cancer Res 12:3935-3941, 2006 27. Prados MD, Wara WM, Sneed PK, et al: Phase III trial of accelerated hyperfractionation with or without difluromethylornithine (DFMO) versus standard fractionated radiotherapy with or without DFMO for newly diagnosed patients with glioblastoma multiforme. Int J Radiat Oncol Biol Phys 49:71-77, 2001[CrossRef][Medline] 28. Prados MD, Larson DA, Lamborn K, et al: Radiation therapy and hydroxyurea followed by the combination of 6-thioguanine and BCNU for the treatment of primary malignant brain tumors. Int J Radiat Oncol Biol Phys 40:57-63, 1998[CrossRef][Medline] 29. Buckner JC, Schomberg PJ, McGinnis WL, et al: A phase III study of radiation therapy plus carmustine with or without recombinant interferon-alpha in the treatment of patients with newly diagnosed high-grade glioma. Cancer 92:420-433, 2001[CrossRef][Medline] 30. Elliott TE, Dinapoli RP, O'Fallon JR, et al: Randomized trial of radiation therapy (RT) plus dibromodulcitol (DBD) versus RT plus BCNU in high grade astrocytoma. J Neurooncol 33:239-250, 1997[CrossRef][Medline] 31. Rao RD, Thome SD, O'Fallon J, et al: Safety of thrice-daily hyperfractionated radiation and BCNU for high-grade gliomas. Int J Radiat Oncol Biol Phys 53:376-384, 2002[CrossRef][Medline] 32. Rajkumar SV, Buckner JC, Schomberg PJ, et al: Phase I evaluation of radiation combined with recombinant interferon alpha-2a and BCNU for patients with high-grade glioma. Int J Radiat Oncol Biol Phys 40:297-302, 1998[CrossRef][Medline] 33. Rajkumar SV, Buckner JC, Schomberg PJ, et al: Phase I evaluation of preirradiation chemotherapy with carmustine and cisplatin and accelerated radiation therapy in patients with high-grade gliomas. Neurosurgery 44:67-73, 1999[CrossRef][Medline] 34. Rajkumar SV, Buckner JC, Schomberg PJ, et al: Phase I and pharmacokinetic study of preirradiation chemotherapy with BCNU, cisplatin, etoposide, and accelerated radiation therapy in patients with high-grade glioma. Int J Radiat Oncol Biol Phys 42:969-975, 1998[CrossRef][Medline] 35. Curran WJ Jr, Scott CB, Horton J, et al: Recursive partitioning analysis of prognostic factors in three Radiation Therapy Oncology Group malignant glioma trials. J Natl Cancer Inst 85:704-710, 1993 36. Scott CB, Scarantino C, Urtasun R, et al: Validation and predictive power of Radiation Therapy Oncology Group (RTOG) recursive partitioning analysis classes for malignant glioma patients: A report using RTOG 90-06. Int J Radiat Oncol Biol Phys 40:51-55, 1998[CrossRef][Medline] 37. Shaw EG, Seiferheld W, Scott C, et al: Reexamining the radiation therapy oncology group (RTOG) recursive partitioning analysis (RPA) for glioblastoma multiforme (GBM) patients. Int J Radiat Oncol Biol Phys 57:S135-S136, 2003 (suppl) 38. Kobayashi K, Takeda F, Teramukai S, et al: A cross-validation of the European Organization for Research and Treatment of Cancer QLQ-C30 (EORTC QLQ-C30) for Japanese with lung cancer. Eur J Cancer 34:810-815, 1998[CrossRef][Medline] 39. Conill C, Verger E, Salamero M: Performance status assessment in cancer patients. Cancer 65:1864-1866, 1990[CrossRef][Medline] 40. Simmons ML, Lamborn KR, Takahashi M, et al: Analysis of complex relationships between age, p53, epidermal growth factor receptor, and survival in glioblastoma patients. Cancer Res 61:1122-1128, 2001 41. Harrell FE Jr: Hmisc: Harrell Miscellaneous, R package version 3.1-2. http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/Hmisc 42. Harrell FE Jr: Design: Design Package, R package version 2.0-12. http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/Design 43. Harrell FE Jr, Lee KL, Califf RM, et al: Regression modelling strategies for improved prognostic prediction. Stat Med 3:143-152, 1984[Medline] 44. Chakravarti A, Zhai G, Suzuki Y, et al: The prognostic significance of phosphatidylinositol 3-kinase pathway activation in human gliomas. J Clin Oncol 22:1926-1933, 2004 Submitted July 10, 2006; accepted March 12, 2007. This article has been cited by other articles:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
INTRODUCTION
2 analyses were performed, as appropriate, to determine differences in means between groups. Classification and regression tree (CART) analysis was performed using software from Salford Systems (San Diego, CA). Unsupervised regression tree analysis of clinical and molecular factors was performed with nodal limits set to no less than 20% of the sample population to determine risk groups of meaningful size. Concordance statistics (c-statistics) were generated for the Cox regression models by calculating Somers' d using the method of Harrell.
47 years) were 85, 62, and 45 weeks, respectively. The differences in survival times were significant between all three CART groups (all P < .008; 
