Colorectal cancer has an incidence of approximately. Review

Transcription

1 Annals of Internal Medicine Review Systematic Review: Anti Epidermal Growth Factor Receptor Treatment Effect Modification by KRAS Mutations in Advanced Colorectal Cancer Issa J. Dahabreh, MD; Teruhiko Terasawa, MD, PhD; Peter J. Castaldi, MD, MS; and Thomas A. Trikalinos, MD Background: KRAS mutations have been extensively investigated as predictive biomarkers for treatment of advanced colorectal cancer with the anti epidermal growth factor receptor (EGFR) antibodies cetuximab and panitumumab. Purpose: To summarize whether KRAS mutation status modifies effects of anti-egfr based treatments for patients with advanced colorectal cancer and whether KRAS status predicts clinical outcomes among such patients. Data Sources: MEDLINE and 2 curated genetics databases (through 24 March 2010) were searched for observational studies. MEDLINE, the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, and the Database of Abstracts of Reviews of Effects (through 1 September 2010) were searched for randomized, controlled trials. No search was restricted by language. Study Selection: Three reviewers screened titles and abstracts to identify published studies assessing KRAS mutations as predictors of overall and progression-free survival or treatment failure for patients who received anti-egfr based therapy for metastatic colorectal cancer. Data Extraction: Three investigators extracted data on population and study-design characteristics, including quality items, and on outcomes of interest. Random-effects meta-analyses were done on nonoverlapping studies. Data Synthesis: In 4 reanalyses of randomized trials of anti-egfr based therapy versus best supportive care or cytotoxic chemotherapy, no significant benefit was found for overall or progression-free survival from anti-egfr based treatment among KRAS-positive patients (hazard ratio [HR], 1.0). However, evidence favors anti-egfr therapy among KRAS wild-type patients; the relative HR across KRAS-positive and wild-type patients was 1.30 (95% CI, 0.95 to 1.78) for overall survival and 2.22 (CI, 1.74 to 2.84) for progression-free survival by random-effects meta-analysis. In 13 cohorts of patients who received anti-egfr antibodies, the summary HR for overall survival was 1.79 (CI, 1.48 to 2.17), with better survival in wild-type patients. The corresponding HR for progression-free survival was 2.11 (CI, 1.74 to 2.55 [16 cohorts]). In random-effects bivariate meta-analysis of 22 studies, the summary sensitivity of KRAS mutations for predicting lack of response was 0.49 (CI, 0.43 to 0.55), and summary specificity was 0.93 (CI, 0.87 to 0.97). Limitations: Limited evidence from randomized studies exists. Patient-level data are needed to assess modifiers of the mutationby-treatment interaction. Publication bias could be a concern. Conclusion: KRAS mutations are consistently associated with reduced overall and progression-free survival and increased treatment failure rates among patients with advanced colorectal cancer treated with anti-egfr antibodies. Primary Funding Source: Agency for Healthcare Research and Quality. Ann Intern Med. 2011;154: For author affiliations, see end of text. Colorectal cancer has an incidence of approximately cases per year in the United States and is the third leading cause of cancer-related death in both men and women (1). Up to 80% of patients are diagnosed with advanced disease (2) and have an adverse prognosis, with an estimated median survival of approximately 2 years despite treatment with combination cytotoxic chemotherapy (3). Recent efforts to improve outcomes have focused on the combination of standard chemotherapy with agents targeting biological pathways central to colorectal cancer pathogenesis (4, 5). Epidermal growth factor receptor (EGFR) has been considered a good candidate for targeted cancer therapy since its discovery (6, 7). Colorectal cancer is a model disease for EGFR-targeted treatments because the EGFR gene is frequently amplified in colorectal cancer and is overexpressed at the RNA and protein levels in most tumors (8, 9). The development of monoclonal antibodies that target EGFR began in the 1980s and has culminated in the clinical testing and approval of 2 monoclonal antibodies: cetuximab and panitumumab (7). These agents, either alone or in combination with chemotherapy, have been shown to improve overall and progression-free survival and have an adverse effect profile that is distinct from that of conventional cytotoxic agents (10, 11). Because not all patients respond to anti-egfr monoclonal antibodies, and because of the risk for adverse effects associated with their use and their substantial cost, there is particular interest in identifying predictors of treatment benefit or lack thereof. One hypothesis that has been explored in many studies is that genetic aberrations of the genes encoding downstream effectors of EGFR-mediated signaling could be associated with resistance to anti-egfr antibody therapy. Among all See also: Print Editors Notes Web-Only Appendix Figures CME quiz Conversion of graphics into slides 2011 American College of Physicians 37

2 Review KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Context KRAS mutations may mark resistance to anti epidermal growth factor receptor (EGFR) antibody treatments for patients with advanced colorectal cancer. Contribution This systematic review of trials and cohort studies found that KRAS mutations were consistently associated with increased rates of treatment failure and reduced survival in adults with advanced colorectal cancer treated with anti-egfr antibodies. Caution Patient-level data about pathologic features of tumors and other potential prognostic factors were not available. Implication Benefits of anti-egfr monoclonal antibody treatment of colorectal cancer may be limited to patients without KRAS mutations. The Editors EGFR effectors, KRAS, a member of the rat sarcoma virus (ras) gene family of oncogenes, is a central signaling node that integrates signaling cascades controlling gene transcription, including many EGFR-mediated pathways. KRAS is frequently mutated in epithelial cancer, including colorectal cancer, resulting in activation of the downstream RAS-RAF-MAPK or PI3K pathways, regardless of whether the EGFR is activated or pharmacologically blocked (constitutive activation) (12). It has been hypothesized that the presence of KRAS mutations could abrogate the effects of anti-egfr antibodies and preclude any beneficial effects of antibody therapy. Therefore, KRAS mutations may predict resistance (lack of response) to anti-egfr antibodies, and testing for KRAS mutations can inform treatment choices. In fact, many retrospective studies have suggested that both cetuximab and panitumumab have limited efficacy in patients with activating KRAS mutations (13). More recently, several analyses of randomized, controlled trials (RCTs) comparing anti-egfr based therapy with alternative treatments have assessed the ability of KRAS mutations to predict clinical outcomes (14 17). We report a systematic review of the published evidence on the ability of KRAS testing to predict response to treatment with cetuximab or panitumumab in patients with advanced colorectal cancer. METHODS Our 2 main objectives were to summarize whether KRAS mutation status modifies the comparative effect of anti-egfr based therapies versus best supportive care or cytotoxic chemotherapy and to summarize whether KRAS status predicts clinically relevant outcomes among patients treated with anti-egfr antibodies. To address the first objective, a reasonable alternative to de novo RCTs of molecular testing versus no testing is to use reanalyses of already completed RCTs that provide information on the interaction of KRAS mutation status with anti-egfr treatment. Such repurposed RCTs provide higher-level evidence for biomarker validation compared with the retrospective cohorts that are much more common in the biomarker literature (18). The second objective can be informed by noncomparative cohorts of patients treated with anti-egfr and stratified by KRAS mutation status (including treatment groups of repurposed RCTs). This systematic review is based on a technology assessment conducted by the Tufts Evidence-Based Practice Center, using standard systematic review methods and following a prespecified protocol. The technology assessment and its supplements are available at Literature Search and Eligibility Criteria We searched PubMed, MEDLINE, and the Human Genome Epidemiology Literature Finder (19) through 24 March 2010 to identify studies reporting on at least 1 of the following outcomes: overall mortality; recurrence, relapse, or disease progression; and treatment failure, and the corresponding time-to-event outcomes (for example, overall or progression-free survival) for patients with colorectal cancer receiving treatment with cetuximab or panitumumab, stratified by KRAS mutation status. We set no geographic or language restrictions. We also searched an online database of predictive studies of colorectal cancer (dbcpco [ (20). Additional targeted searches for RCTs were done. We searched MEDLINE, the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, and the Database of Abstracts of Reviews of Effects through 1 September To increase yield, we also searched the references of all retrieved manuscripts and relevant review articles. All of the investigators screened the first 300 citations to ensure that screening criteria were well understood and applied uniformly. Thereafter, 3 authors screened nonoverlapping sets of the remaining citations. Potentially eligible studies were retrieved and reviewed in full text. Eligible studies were published studies that reported on at least 10 patients who had received a diagnosis of metastatic colorectal cancer and received treatment with anti-egfr antibodies alone or in combination with cytotoxic chemotherapy, used genotyping methods to identify KRAS mutations, and reported the outcomes of interest stratified by mutational status. Because biomarker studies do not typically define time-to-event outcomes with the exception of overall survival, we accepted the definitions used in the primary studies (21). Treatment failure was defined as lack of response by radiologic criteria, as used in each study. We considered all study designs (prospective and retrospective), treatment settings (first line and second line or higher), and treatment strategies (monotherapy and 38 4 January 2011 Annals of Internal Medicine Volume 154 Number 1

3 KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Review combination with cytotoxic chemotherapy). During fulltext review, we only considered studies published in English. Studies that did not identify any KRAS mutation carriers could not contribute to the analyses and were excluded. We also excluded studies conducted in the adjuvant or neoadjuvant setting. Because evidence suggests that combinations of anti-egfr antibodies with bevacizumab (an antivascular endothelial growth factor antibody) and chemotherapy may be harmful (22), we considered such studies separately. Identification of Publications With Overlapping Populations We took particular care to identify publications with at least partially overlapping populations by comparing authors, centers, recruitment periods, and patient demographic characteristics. We depicted the overlap by using undirected graphs in which each publication is represented by an ellipse, and ellipses of publications with at least partial overlap are linked. This provides a visual representation of potentially overlapping publications: The larger the number of connected publications, the more extensive the overlap in a topic. All publications were included in descriptive analyses, but only nonoverlapping publications were included in meta-analyses (23). This method avoids overcounting in quantitative analyses but may introduce undercounting and is thus a conservative approach. Data Extraction Three reviewers independently extracted data. Each study was reviewed by 2 reviewers, and discrepancies were resolved by consensus including a third reviewer. From each eligible study, we extracted bibliographic information and information on study design; inclusion criteria; patient and treatment characteristics; details of genetic testing and specific mutations; definitions of outcomes; quality-related items; and numerical data for the outcomes of interest, namely overall survival, progression-free survival, and treatment failure. If a study did not explicitly report these data, but cited other publications instead, we consulted the cited publications. For time-to-event outcomes (overall and progressionfree survival), we extracted hazard ratios (HRs) comparing patients with KRAS mutations with wild-type patients, as well as the corresponding variance (24). For comparative RCTs of anti-egfr based therapy versus alternative treatments, we extracted survival statistics for the treatment effect separately for KRAS-positive and wild-type patients. Survival statistics were extracted by using methods described elsewhere (24 26). We also extracted the total number of wild-type and KRAS-positive patients and the number of patients in each of these groups who had treatment failure. Quality Assessment We abstracted information on aspects of the design and conduct of the individual studies that were deemed important for interpreting their results (27). These items include description of patient sampling, administered treatments, design of the parent study (repurposed RCT or prospective or retrospective cohort), whether tissue samples were available for the substantial majority of study participants, whether clinically representative patients were assessed, assay methodology, definitions of clinical end points and methods of measurement, and analysis methods. These items are largely consistent with the REMARK (REcommendations for tumor MARKer prognostic studies) guidelines (28). Evidence Synthesis We considered overall survival the primary outcome and progression-free survival and treatment failure secondary outcomes. For our first objective (to summarize the interaction effects of anti-egfr treatment by KRAS mutation status), we calculated relative HRs for overall and progression-free survival of anti-egfr treatment in KRASpositive over wild-type (KRAS-negative) patients as the ratio of the respective HRs. Similarly, we calculated the relative odds ratio (OR) for treatment failure as the ratio of the respective ORs for KRAS-positive over wild-type patients. Variances of the log-transformed relative HRs and relative ORs were calculated by using standard methods (29). A relative HR (or OR) of 1.0 indicates that the HR (or OR) of the anti-egfr based treatment is the same in both KRAS-positive and KRAS-negative patients. A value greater than 1.0 means that the anti-egfr effect is more favorable in wild-type patients than in KRAS-positive patients, whereas a value less than 1.0 means the opposite. These analyses of the interaction between anti-egfr treatment and KRAS mutation status are possible only in (repurposed) RCTs of anti-egfr based treatments with cytotoxic chemotherapy or best supporting care. We examined whether KRAS predicts clinically relevant outcomes among patients treated with anti-egfr (our second objective) by calculating HRs for overall and progression-free survival in KRAS-positive over wild-type patients. In addition, we quantified the sensitivity and the specificity of KRAS mutations (index test) to predict treatment failure (reference standard) (30 32). For the second objective, we included the anti-egfr groups of RCTs as independent cohorts. To summarize evidence on survival outcomes, we did random-effects meta-analysis of HRs, relative HRs, and ORs by using the DerSimonian Laird model (33). We tested for between-study heterogeneity with the Q statistic (considered significant at the 0.10 level) and quantified its extent with the I 2 statistic (34). Systematic differences in the effect sizes of more and less precise (bigger and smaller) studies were tested with the Egger regression test (35) and only for the primary outcome. This and other similar tests (36) are often misleadingly referred to as publication-bias tests (37, 38). To assess the accuracy of KRAS mutations for predicting treatment failure, we did a bivariate random-effects 4 January 2011 Annals of Internal Medicine Volume 154 Number 1 39

4 Review KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Table 1. Characteristics of Eligible Studies of Nonresectable Metastatic Colorectal Cancer* Study, Year (Reference) First-line setting RCTs Bokemeyer et al, 2009 (17) Hecht et al, 2009 (46) Tol et al, 2009 (47) Van Cutsem et al, 2009 (16) Folprecht et al, 2010 (48) Country Study Design Sampling Treatment Regimen (Sample Size) Multinational Multinational Netherlands Multinational Germany and Austria Retrospective, based on a phase 2 RCT Prospective; analysis of a phase 3 RCT Retrospective, based on a phase 3 RCT Retrospective, based on a phase 3 RCT Retrospective, based on a phase 2 RCT Convenience Prospective patient selection Convenience Convenience Convenience Median Age (Range), y Men, % Median Follow-up (Range), mo Cetuximab chemotherapy 59 (24 82) 50 NR (n 113) Chemotherapy (n 120) 60 (30 82) 55 Panitumumab bevacizumab Approximate: 55 Oxaliplatin-based chemotherapy (n 528) 61 (28 88) chemotherapy: Bevacizumab chemotherapy (n 525) Approximate: 62 (22 89) Cetuximab bevacizumab 62 (33 80) (NR) chemotherapy (n 368) Bevacizumab chemotherapy (n 368) 62 (27 83) 56 Cetuximab chemotherapy 61 (22 79) 62 NR (n 277) Chemotherapy (n 263) 62 (22 79) 53 Cetuximab FOLFOX Median, NR (n 56) (IQR, 57 71) Cetuximab FOLFIRI 62 (56 68) 64 (n 55) (0 26) Irinotecan-based chemotherapy: 9 (0.3 24) Single-group studies Yen et al, 2009 (49) Tabernero et al, 2010 (50)** Yen et al, 2010 (51) Taiwan RC and SC Convenience Cetuximab chemotherapy (n 76) Spain PC and MC Convenience Cetuximab chemotherapy (n 62) Taiwan RC and SC Convenience Cetuximab chemotherapy (n 95) 64 (39 83) (4 34) 65 (39 80) 63 NR 66 (39 86) 58 NR Second-line or higher setting RCTs Amado et al, 2008 (14) Karapetis et al, 2008 (15) Single-group studies Moroni et al, 2005 (52) Lièvre et al, 2006 (53) Benvenuti et al, 2007 (54) Di Fiore et al, 2007 (55) Frattini et al, 2007 (56) Khambata-Ford et al, 2007 (57) Cappuzzo et al, 2008 (58) De Roock et al, 2008 (59) Di Nicolantonio et al, 2008 (60) Freeman et al, 2008 (61) Gonçalves et al, 2008 (62) Lièvre et al, 2008 (63) Belgium and Italy Multinational Prespecified subgroup analysis of a phase 3 RCT Prospective analysis of an RCT Prospective BSC panitumumab Approximate: 63 NR patient selection (n 208) BSC (n 219) 62 (27 82) Approximate: 63 (27 83) 64 Convenience BSC cetuximab (n 198) Total: 63 (29 88) 64 NR BSC cetuximab (n 196) Italy RC and SC Convenience Cetuximab chemotherapy, panitumumab (n 31) 65 (41 85) 71 NR France RC and MC Convenience Cetuximab chemotherapy 62 (41 78) 63 NR (n 30) Italy RC and SC Convenience Cetuximab chemotherapy, 61 (39 84) 63 NR panitumumab (n 48) France RC and MC Convenience Cetuximab chemotherapy NR NR (n 59) Switzerland RC and SC Unclear Cetuximab chemotherapy 67 (29 84) 67 NR (n 27) Multinational PC and MC Unclear Cetuximab (n 110) 61 (25 89) 50 NR Italy RC and SC Convenience Cetuximab chemotherapy (n 85) Belgium RC and MC Convenience Cetuximab chemotherapy (n 113) Italy and RC and MC Convenience Cetuximab chemotherapy, Switzerland panitumumab (n 113) 63 (NR) 64 NR Mean, (25 473) (SD, 11) 63 (NR) 64 NR Multinational RC and MC Convenience Panitumumab (n 62) 62 (29 85) 60 NR France RC and SC Convenience Cetuximab chemotherapy (n 32) France RC and MC Convenience Cetuximab chemotherapy (n 89) 58 (36 78) (NR) NR NR 10 (NR) 40 4 January 2011 Annals of Internal Medicine Volume 154 Number 1

5 KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Review Table 1 Continued Study, Year (Reference) Lurje et al, 2008 (64) Personeni et al, 2008 (65) Bibeau et al, 2009 (66) Garm Spindler et al, 2009 (67) Jacobs et al, 2009 (68) Laurent-Puig et al, 2009 (69) Loupakis et al, 2009 (70) Loupakis et al, 2009 (71) Molinari et al, 2009 (72) Muro et al, 2009 (73) Oden-Gangloff et al, 2009 (74) Perrone et al, 2009 (75) Prenen et al, 2009 (76) Paule et al, 2009 (77) Sartore-Bianchi et al, 2009 (78) Sartore-Bianchi et al, 2009 (79) Sohn et al, 2009 (80) Souglakos et al, 2009 (81) Chung et al, 2010 (82) Graziano et al, 2010 (83) Montagut et al, 2010 (84) Perkins et al, 2010 (85) Scartozzi et al, 2010 (86) Country Study Design Sampling Treatment Regimens (Sample Size) Median Age (Range), y Men, % Median Follow-up (Range), mo Multinational Retrospective analysis of a phase 2 trial Convenience Cetuximab (n 130) NR ( 54 to 65) (NR) Belgium RC and MC Convenience Cetuximab chemotherapy 59 (26 80) 56 NR (n 96) France RC and MC Convenience Cetuximab chemotherapy 60 (23 81) (NR) (n 69) Denmark RC and SC Convenience Cetuximab chemotherapy 61 (38 77) 52 8 (2 30) (n 71) Belgium RC and MC Convenience Cetuximab chemotherapy NR NR NR (n 220) France RC and MC Convenience Cetuximab chemotherapy NR NR NR (n 173) Italy RC and MC Convenience Cetuximab chemotherapy 62 (38 78) (NR) (n 102) Italy RC and MC Convenience Cetuximab chemotherapy 61 (42 77) 55 NR (n 138) Switzerland RC and SC Convenience Cetuximab irinotecan 67 (48 94)*** 63 NR (n 12) Japan Retrospective analysis Convenience Panitumumab (n 52) 59 (23 77) (5 42) of a multicenter, phase 2 trial France RC and SC Convenience Cetuximab chemotherapy Mean, NR (n 64) (20 82) Italy RC and SC Convenience Cetuximab chemotherapy 58 (36 78) 63 NR (n 32) Belgium RC and SC Convenience Cetuximab chemotherapy 61 (26 89) 60 NR (n 200) France RC and SC Convenience Cetuximab chemotherapy 60 (29 80) 65 NR (n 23) Italy and RC and SC Convenience Cetuximab chemotherapy, 64 (26 85) 65 NR Switzerland panitumumab (n 110) Italy RC and SC Convenience Cetuximab chemotherapy, 63.5 (26 85) 65 NR panitumumab (n 132) Korea RC and SC Convenience Cetuximab chemotherapy 58 (28 77) 61 NR (n 66) United States RC and SC Convenience Cetuximab chemotherapy 59 (23 86) 52 NR and (n 92) Greece United States RC and SC Convenience Cetuximab (n 88) 57 (NR) 57 NR Italy RC and SC Convenience Cetuximab chemotherapy (n 138) Spain RC and SC Convenience Cetuximab chemotherapy (consecutive (n 48) patients) France RC and SC Convenience Cetuximab chemotherapy, panitumumab (n 42) Italy RC and SC Convenience Cetuximab chemotherapy (n 112) 65 (41 77) 54 NR Approximately NR 61 (44 78) (NR) 64 (36 81) 63 NR BSC best supportive care; FOLFIRI irinotecan, fluorouracil, and folinic acid; FOLFOX oxaliplatin, fluorouracil, and folinic acid; IQR interquartile range; MC multicenter; NR not reported; PC prospective cohort; RC retrospective cohort; RCT randomized, controlled trial; SC single center. * Numbers have been rounded to the nearest integer. Number of included patients typically assessed for baseline clinical characteristics, which is not necessarily the same as the number assessed for KRAS mutations (in the first-line, second-line, or higher settings) or clinical outcomes (in the first-line setting only). Data were abstracted for the subgroup of patients assessed for KRAS mutations (in the first-line setting only) or had KRAS mutational testing (in the second-line or higher settings). Availability of tissue was an inclusion criterion for the study. It is unclear whether the assessment of KRAS mutations was also prospectively planned. Data were abstracted for the subgroup of patients who had KRAS mutation testing. ** Some patient characteristics were extracted from reference 50. Time since diagnosis of metastatic disease. 114 patients were assessed in a pooled analysis. This study also included a small (n 67) cohort from France that received standard treatment for colorectal cancer. All patients in this cohort were wild-type for KRAS mutations. Data were not extracted for these patients. KRAS mutations 12, 13, 61, and 146 were assessed. Seventy-six patients were assessed for KRAS mutations 61 and 146 only. Applicable to patients who received cetuximab only. The other 26 patients did not receive cetuximab. *** Data extracted from the entire cohort, including those who did not receive cetuximab. Only 16 samples were available for KRAS genotyping. This was augmented with 8 additional samples from another phase 1 panitumumab study in patients with metastatic disease, for a total of 24 KRAS samples. The study enrolled a total of 168 patients, of whom 100 received cetuximab plus chemotherapy. Of those 100 patients, 92 received this treatment in the second-line setting, and the study reports clinical outcomes only for these patients. Data refer to the overall cohort (168 persons). Data refer to the overall cohort (88 persons). 4 January 2011 Annals of Internal Medicine Volume 154 Number 1 41

6 Review KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Figure 1. Forest plot of HRs for overall survival comparing KRAS-mutated and wild-type patients. Study, Year (Reference) Amado et al, 2008 (14) Cappuzzo et al, 2008 (58) Freeman et al, 2008 (61) Gonçalves et al, 2008 (62) Karapetis et al, 2008 (15) Garm Spindler et al, 2009 (67) Jacobs et al, 2009 (68) Laurent-Puig et al, 2009 (69) Sohn et al, 2009 (80) Van Cutsem et al, 2009 (16) Graziano et al, 2010 (83) Sartore-Bianchi et al, 2009 (78) Yen et al, 2010 (51) Overall (I 2 = 56%; P = 0.007) HR (95% CI) 1.69 ( ) 1.14 ( ) 2.00 ( ) 1.32 ( ) 2.08 ( ) 1.28 ( ) 2.00 ( ) 1.71 ( ) 2.42 ( ) 1.42 ( ) 2.30 ( ) 1.72 ( ) 6.76 ( ) 1.79 ( ) HR (95% CI) Better Survival in KRAS Mutant Better Survival in KRAS Wild-Type Each study is shown by the point estimate of the HR (size of square is proportional to the weight of each study) and 95% CI; the summary HR and its 95% CI by random-effects calculations are depicted by the diamond. Values greater than 1 indicate that patients with the KRAS mutation have reduced survival compared with wild-type patients. HR hazard ratio. meta-analysis of sensitivity and specificity by using the exact binomial likelihood (39, 40). We also calculated summary positive and negative likelihood ratios with their corresponding CIs (41). Likelihood ratios quantify the value of mutations for ruling in (positive likelihood ratio) or ruling out (negative likelihood ratio) lack of response to treatment. Subgroup Analyses and Meta-regression We did subgroup analyses to assess whether specific factors may influence the ability of KRAS mutations to predict response to anti-egfr antibodies. Subgroups were defined by percentage of patients who had not been exposed to previous chemotherapy, use of the anti-egfr antibodies in combination with chemotherapy or as monotherapy, and the specific antibody used. The effect of these covariates on the log HR of KRAS mutations for overall and progression-free survival was assessed by using univariate random-effects meta-regressions. The respective analyses for treatment failure used bivariate meta-regressions in which covariates were allowed to have distinct effects on sensitivity and specificity (42 44). Software Statistical analyses were conducted with Stata SE, version 11.1 (StataCorp, College Station, Texas), and Meta- Analyst, version 3.0 beta (Tufts Medical Center, Boston, Massachusetts) (45). P values for all comparisons were 2-tailed, and statistical significance was defined as a P value less than 0.05 for all tests except those for heterogeneity. Role of the Funding Source This project is based on a technology assessment done under contract with the Agency for Healthcare Research and Quality. The funding source had no role in the design, conduct, and analysis of the study or in the decision to submit the manuscript for publication. RESULTS Literature Flow and Identification of Overlapping Publications Appendix Figure 1 (available at summarizes the literature flow. A total of 45 publications were considered eligible for this systematic review. The log of included and excluded studies is in the appendices of the original technology assessment. Table 1 summarizes the clinical settings and therapeutic regimens investigated. The technology assessment contains details about the designs of these studies, the methods and results of mutational testing, and the outcomes assessed. Cetuximab and panitumumab have been investigated in several phase 2 and phase 3 clinical trials. Many of the centers participating in these trials have genotyped tissue samples from their patient cohorts for KRAS mutations and presented several post hoc analyses, often including 42 4 January 2011 Annals of Internal Medicine Volume 154 Number 1

7 KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Review additional patients treated outside the clinical trials. Furthermore, pooled analyses of patients included in previous publications have also been published. Appendix Figure 2 (available at suggests that the 45 publications correspond to 24 nonoverlapping studies because there are 24 unconnected components: that is, single ovals or groups of connected ovals. Only the largest publication from each unconnected component of Appendix Figure 2 was included in quantitative analyses. Patient Populations and Treatment Regimens Most patients had received previous treatment with at least 1 chemotherapy regimen; both the number and types of treatment regimens administered varied across studies. Mean or median participant age was 65 years or older in 6 of 40 studies that reported relevant information, and 60 years or older in 31 of 42 studies that reported relevant information. Table 1 summarizes clinical settings and treatment strategies. Cetuximab was evaluated both as monotherapy and in combination with cytotoxic chemotherapy; panitumumab was evaluated almost exclusively as monotherapy. Anti-EGFR antibody dosing was commonly not reported, particularly in reports of retrospective studies; many of the patients in the 45 publications were enrolled in larger, multicenter prospective clinical trials that used standard drug doses. Study Design In the 45 publications, 12 to 440 patients were analyzed for KRAS status and treated with anti-egfr antibodies. Median follow-up duration ranged from 9 to 26 months and was commonly not reported. Only 2 studies explicitly stated that sample collection and subgroup analysis by KRAS testing was a prespecified aim of the study. In the remaining 43 studies, outcome assessment was frequently described as prospective, whereas KRAS testing was done only on available archival samples (14). Most studies (39 studies) assessed the ability of KRAS testing to predict outcomes only in patients who received anti-egfr antibodies. This design essentially assumes that KRAS testing has no predictive ability for colorectal cancer outcomes in patients who did not receive anti-egfr antibodies. Six studies reported biomarker analyses based on RCTs in which the control groups received cytotoxic chemotherapy (16, 17), best supportive care (14, 15), or chemotherapy in combination with bevacizumab (47, 61). In the 6 RCTs, 1479 patients treated with anti-egfr antibodies and 1476 patients treated in the comparator groups were assessed for KRAS mutations. Addition of bevacizumab to cetuximab plus chemotherapy has been shown to result in reduced survival and increased toxicity; therefore, we considered the 2 RCTs in which bevacizumab was used (47, 61) as a separate subgroup. Mutation Testing Methods and Results The methods and results of KRAS mutation analyses are presented in detail in the technology assessment. Most studies only assessed codons 12 and 13 of the KRAS gene, using direct sequencing or allele-specific methods. The commonly used technologies are commercially available, and their analytic validity is appropriate for clinical use. Modification of the Treatment Effect by KRAS Mutations We used the results of the 4 comparative RCT based biomarker analyses in which the control groups received cytotoxic chemotherapy or best supportive care to quantify the treatment-by-biomarker interaction. For overall survival, 3 RCTs provided data (14 16), and 1 indicated a statistically significant benefit for using cetuximab (compared with best supportive care) among KRAS wild-type patients (15). Among KRAS-positive patients, anti-egfr therapy did not seem to offer clinically significant benefits. As shown in Appendix Figure 3, top (available at the random-effects summary relative HR comparing the treatment effect on overall survival between KRAS-positive and wild-type patients was 1.30 (95% CI, 0.95 to 1.78; P 0.103), indicating that benefits from anti-egfr regimens are probably limited to KRAS wildtype patients. Four studies provided data on progression-free survival; all 4 suggested that anti-egfr antibody treatment is beneficial among KRAS wild-type patients but not among Sensitivity, % Figure 2. Bivariate meta-analysis of sensitivity and specificity of KRAS mutations for predicting lack of response to anti-egfr based treatment Specificity, % Study estimate 95% confidence region Summary point Study results are plotted in the receiver-operating characteristic curve space. Each study is represented by a circle, whose size is proportional to the study size. EGFR epidermal growth factor receptor. 4 January 2011 Annals of Internal Medicine Volume 154 Number 1 43

8 Review KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Table 2. Main and Subgroup Meta-analysis Results for Overall and Progression-Free Survival Subgroup Studies (Mutations/ Patients), n (n/n) Heterogeneity (P Value; I 2 Value) HR (95% CI) P Value Interaction P Value Overall survival All studies 13 (651/1695) 0.007; 56% 1.79 ( ) NA Treatment regimens Monotherapy 4 (216/517) 0.05; 61% 1.61 ( ) Combination chemotherapy 9 (435/1178) 0.02; 55% 1.91 ( ) Setting First-line 2 (146/372) 0.001; 93% 2.98 ( ) Second-line or higher 11 (505/1323) 0.21; 24% 1.68 ( ) Anti-EGFR antibody Cetuximab 11 (558/1456) 0.003; 63% 1.80 ( ) Panitumumab 2 (93/239) 0.63; 0% 1.76 ( ) Progression-free survival All studies 16 (741/1945) 0.003; 57% 2.11 ( ) NA Treatment regimens Monotherapy 4 (204/517) 0.15; 43% 2.52 ( ) Combination chemotherapy 12 (537/1428) 0.008; 57% 1.99 ( ) Setting First-line 4 (216/530) 0.001; 81% 2.24 ( ) Second-line or higher 12 (525/1415) 0.07; 41% 2.08 ( ) Anti-EGFR antibody Cetuximab 14 (648/1706) 0.003; 59% 2.03 ( ) Panitumumab 2 (93/239) 0.81; 0% 2.70 ( ) EGFR epidermal growth factor receptor; HR hazard ratio; NA not applicable. KRAS-positive patients (14 17). One study suggested that treatment with anti-egfr antibody based therapy was harmful compared with standard cytotoxic chemotherapy (17). The summary relative HR was 2.22 (CI, 1.74 to 2.84; P 0.001), indicating that benefit from anti-egfr antibodies is limited to KRAS wild-type patients (Appendix Figure 3, middle). Regarding response to treatment, all 4 RCTs demonstrated a statistically significant benefit of anti-egfr antibody treatment compared with alternative therapies among KRAS wild-type patients but not KRAS-positive patients. There seemed to be no effect among patients with mutated KRAS. The summary relative OR for treatment response was 3.27 (CI, 1.70 to 6.31; P 0.001), indicating that anti-egfr antibody treatment benefits on response to treatment are limited to KRAS wild-type patients, although substantial uncertainty was found around the relative OR estimates of the studies conducted in the second-line setting (Appendix Figure 3, bottom) (14, 15). The results from the comparative RCT based biomarker analyses, in which the control groups received cytotoxic chemotherapy in combination with bevacizumab, are presented in Appendix Figure 4 (available at In general, the combination of cetuximab, bevacizumab, and chemotherapy resulted in worse outcomes compared with chemotherapy plus bevacizumab alone, regardless of mutational status and without significant treatment-by-biomarker interaction. Effect of KRAS Mutations on Clinical Outcomes Among Patients Treated With Anti-EGFR Antibodies The technology assessment details the findings from the eligible studies with the clinical outcomes of mortality, progression, and response to treatment. For these analyses, we did not consider the anti-egfr antibody groups of bevacizumab plus chemotherapy trials. Mortality In all 20 publications that reported relevant information, median overall survival in KRAS-positive patients was shorter compared with wild-type patients and ranged from 4.4 to 17.5 months for patients with KRAS mutations and from 6.6 to 24.9 for wild-type patients. Figure 1 shows the meta-analysis of the 13 nonoverlapping studies (651 mutations in 1695 patients) with data on overall survival. The summary HR was 1.79 (CI, 1.48 to 2.17; P 0.001), indicating that KRAS mutations are associated with shorter survival, and there was evidence of between-study heterogeneity (P 0.007; I 2 56%). Some evidence suggested that smaller studies were systematically different from larger studies (P 0.04). Disease Progression In all 30 of 42 publications that reported disease progression, median progression-free survival or time-toprogression was shorter among patients with KRASpositive tumors than with wild-type patients. Median progression-free survival and time to progression ranged 44 4 January 2011 Annals of Internal Medicine Volume 154 Number 1

9 KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Review from 1.3 to 7.6 months for patients with KRAS mutations and from 1.4 to 12.3 months for wild-type patients. In total, 16 studies (741 mutations in 1945 patients) were considered nonoverlapping and provided data for the meta-analysis of progression-free survival. The summary HR was 2.11 (CI, 1.74 to 2.55; P 0.001), indicating a detrimental effect of KRAS mutations on progression-free survival; there was evidence for substantial between-study heterogeneity (P 0.003; I 2 57%). Treatment Failure, by Imaging Criteria Overall, treatment failure rates were higher in patients with KRAS mutations than in wild-type patients. Particularly in studies of patients who had received previous chemotherapy, the response rates in the presence of KRAS mutations were typically very low and often 0. Figure 2 shows the summary sensitivity and specificity of KRAS mutations for predicting treatment failure from a bivariate meta-analysis of 22 studies with nonoverlapping populations (841 mutations in 2242 patients). The summary specificity was 0.49 (CI, 0.43 to 0.55), and the summary sensitivity was 0.93 (CI, 0.87 to 0.97), corresponding to a summary positive likelihood ratio of 7.35 (CI, 3.72 to 14.50) and a summary negative likelihood ratio of 0.55 (CI, 0.49 to 0.61) for predicting treatment failure. Subgroup Analyses and Meta-regression For both overall and progression-free survival, none of the assessed covariates affected the predictive value of KRAS mutations beyond that expected by chance (Table 2). In contrast, for treatment failure, KRAS testing had high summary specificity (0.90) in all study subgroups, except those of studies conducted in the first-line setting. The specificity of KRAS mutations in the first-line setting was statistically significantly different from that in the second-line setting (P for interaction). Sensitivity was low (0.47 to 0.57) in all evaluated subgroups (Table 3). DISCUSSION Our report is an overview of KRAS mutations as a biomarker of resistance to anti-egfr antibodies for patients with advanced colorectal cancer. We found that KRAS mutations are strong predictors of reduced overall survival and progression-free survival and increased rates of treatment failure. Quantitative evidence that the benefits of anti-egfr targeted treatment in colorectal cancer are largely limited to patients with no KRAS mutations is provided; this treatment-by-biomarker interaction was particularly evident for the outcomes of progression-free survival and response to treatment, but congruent results were obtained for overall survival. Hence, our results provide support for a recent Provisional Clinical Opinion by the American Society of Clinical Oncology and the labeling changes implemented by the U.S. Food and Drug Administration restricting the use of anti-egfr monoclonal antibodies to patients with colorectal cancer with tumors that test negative for KRAS mutations (87). We evaluated activating KRAS mutations for their ability to predict differential response to EGFR-based treatment: that is, as a biomarker that informs a pharmacogenetic interaction. Predictive or pharmacogenetic effects of biomarkers (ability to identify treatment-effect modification) should be distinguished from prognostic effects (ability to predict outcomes regardless of treatment) (88). Ideally, to assess predictive effect of KRAS mutations, patients who did and did not receive EGFR-based therapy need to be tested for treatment-by-kras mutation interactions (18), as was done in 6 repurposed RCTs that we identified. Most of the available studies analyzed only patients who received EGFR-based treatments and, by definition, do not provide information on treatment-by-kras mutation interactions. We argue that they can still be considered supportive of a predictive (pharmacogenetic) effect of KRAS mutations because mutations seem to have little Table 3. Main and Subgroup Meta-analysis Results for the Predictive Accuracy of KRAS Mutations Subgroup Total Studies (KRAS-Positive/ Total Patients), n (n/n) Sensitivity (95% CI) Interaction P Value Specificity (95% CI) Interaction P Value Positive Likelihood Ratio (95% CI) Negative Likelihood Ratio (95% CI) All studies 22 (841/2242) 0.49 ( ) NA 0.93 ( ) NA 7.35 ( ) 0.55 ( ) Treatment regimen Monotherapy 6 (256/621) 0.54 ( ) ( ) ( ) 0.48 ( ) Combination chemotherapy 16 (585/1621) 0.47 ( ) 0.92 ( ) 6.16 ( ) 0.57 ( ) Previous chemotherapy First-line 5 (243/624) 0.57 ( ) ( ) ( ) 0.56 ( ) Second-line or higher 17 (598/1618) 0.47 ( ) 0.96 ( ) ( ) 0.55 ( ) EGFR antibody Cetuximab 19 (738/1979) 0.48 ( ) NA 0.91 ( ) NA 5.50 ( ) 0.57 ( ) Panitumumab 3 (103/263) NA NA NA NA EGFR epidermal growth factor receptor; HR hazard ratio; NA not applicable. 4 January 2011 Annals of Internal Medicine Volume 154 Number 1 45

10 Review KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer or no ability to predict outcomes in patients receiving standard chemotherapy (89, 90). Therefore, the relationship of KRAS mutations on clinical outcomes in the noncomparative studies essentially can be viewed as a predictive or pharmacogenetic effect. Two of 6 repurposed RCTs used a combination of cetuximab, bevacizumab, and chemotherapy in the experimental group, which has been shown to result in reduced efficacy and increased toxicity compared with bevacizumab plus chemotherapy combinations (22). No statistically significant pharmacogenetic (treatment-by-kras status) interaction occurred in these 2 trials. The predictive ability of KRAS mutations seems to be similar in cetuximab and panitumumab therapy, although the bulk of available evidence for this subgroup comparison was related to studies assessing panitumumab as monotherapy and, in all cases, in patients pretreated with cytotoxic chemotherapy. Panitumumab is approved for use as a single agent for patients with disease progression who are receiving or following fluoropyrimidine, oxaliplatin, and irinotecan chemotherapy regimens (91). The predictive effect of KRAS mutations may be modified by previous exposure to chemotherapeutic agents (92). In our analysis, the ability of KRAS mutations to predict treatment failure seemed to be lower in studies conducted in the first-line setting compared with studies in patients who have received cytotoxic chemotherapy. A plausible explanation is that in the salvage setting, many patients are already resistant to cytotoxic chemotherapy and most of their benefit comes from the anti-egfr component of the treatment regimens. In patients who have been previously treated (salvage setting), the modifying effect of KRAS mutations on the anti-egfr therapy is not diluted by the effect of cytotoxic chemotherapy and is more readily observable. These observations need to be viewed with caution, given that they did not seem to apply to survival outcomes and require confirmation by further research focused in the first-line setting. Ideally, this could be studied in future RCTs of anti-egfr agents by prespecifying analyses by KRAS status. Another option is to repurpose already completed RCTs, in which the drugs of interest are tested against a suitable comparator, by genotyping tissue samples from enrollees (18, 93). In the latter case, genetic analyses in archival tissue from RCT enrollees would be done and associated with the prospectively recorded clinical outcomes (18). Properly stored tissue samples can be genotyped with negligible error long past the collection date. Provided that samples are collected from all RCT participants (even if a proportion is missing completely at random), retrospective associations of somatic mutations with treatment outcomes should be unconfounded. However, statistical corrections for repeated testing would still need to be done to control for spurious findings (type I error) (18). Several limitations need be considered when interpreting our results. First, our criteria to identify overlapping publications are conservative, in that we may have excluded publications that do not have extensive overlap with the included ones. This approach ensures that overcounting of patients did not occur. Randomized evidence is limited, and few studies seem to have been specifically designed to address the questions that were considered in this review. Furthermore, we had no access to primary data, such as sex, age, or pathologic features of the tumor, that may also have predictive importance and may be modifiers of the observed relationship of KRAS mutations and treatment effect. Finally, we found some evidence that smaller observational studies provided more extreme estimates of the HR for overall survival compared with larger studies. Although it is customary to ascribe this pattern to publication bias, many reasons, including unexplained betweenstudy heterogeneity and chance, can result in similar patterns (37, 38). Nevertheless, as in all systematic reviews based on the published literature, publication bias may exist. However, it is unlikely that the consistent and relatively large pharmacogenetic interaction is a result of publication bias alone. In conclusion, we provide evidence that KRAS mutations are predictive of survival, disease progression, and treatment failure in patients with advanced colorectal cancer treated with anti-egfr antibodies. We show that the benefits of anti-egfr antibody therapy are largely limited to KRAS wild-type patients, particularly regarding progression-free survival and response to treatment. Evidence indicates that the predictive effect is stronger among patients who have received previous cytotoxic chemotherapy, but more evidence is needed to establish this observation. Further research is warranted to identify subgroups of the disease in which the predictive effect may be stronger and to integrate several molecular biomarkers, such as BRAF, PTEN, or PIK3CA aberrations (13), along with KRAS mutations in predictive instruments, to further improve predictive accuracy. From Center for Clinical Evidence Synthesis, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Boston, Massachusetts, and Nanakuri Sanatorium, Fujita Health University School of Medicine, Tsu, Japan. Grant Support: By the Agency for Healthcare Research and Quality (contract HHSA ). Potential Conflicts of Interest: Drs. Dahabreh, Castaldi, and Trikalinos: Grants received (money to institution): Agency for Healthcare Research and Quality. Disclosures can also be viewed at /authors/icmje/conflictofinterestforms.do?msnum M Requests for Single Reprints: Thomas A. Trikalinos, MD, Center for Clinical Evidence Synthesis, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, 800 Washington Street (Box 63), Boston, MA, 02111; , TTrikalinos@tuftsmedicalcenter.org. Current author addresses and author contributions are available at January 2011 Annals of Internal Medicine Volume 154 Number 1

11 KRAS Mutations and Anti-EGFR Therapy for Colorectal Cancer Review References 1. Jemal A, Siegel R, Ward E, Hao Y, Xu J, Thun MJ. Cancer statistics, CA Cancer J Clin. 2009;59: [PMID: ] 2. Alberts SR, Wagman LD. Chemotherapy for colorectal cancer liver metastases. Oncologist. 2008;13: [PMID: ] 3. Cunningham D, Atkin W, Lenz HJ, Lynch HT, Minsky B, Nordlinger B, et al. Colorectal cancer. Lancet. 2010;375: [PMID: ] 4. Weinstein IB. Cancer. Addiction to oncogenes the Achilles heal of cancer. Science. 2002;297:63-4. [PMID: ] 5. Deschoolmeester V, Baay M, Specenier P, Lardon F, Vermorken JB. A review of the most promising biomarkers in colorectal cancer: one step closer to targeted therapy. Oncologist. 2010;15: [PMID: ] 6. Baselga J. Why the epidermal growth factor receptor? The rationale for cancer therapy. Oncologist. 2002;7 Suppl 4:2-8. [PMID: ] 7. Gschwind A, Fischer OM, Ullrich A. The discovery of receptor tyrosine kinases: targets for cancer therapy. Nat Rev Cancer. 2004;4: [PMID: ] 8. Spano JP, Lagorce C, Atlan D, Milano G, Domont J, Benamouzig R, et al. Impact of EGFR expression on colorectal cancer patient prognosis and survival. Ann Oncol. 2005;16: [PMID: ] 9. Kuramochi H, Hayashi K, Nakajima G, Kamikozuru H, Yamamoto M, Danenberg KD, et al. Epidermal growth factor receptor (EGFR) mrna levels and protein expression levels in primary colorectal cancer and corresponding liver metastases. Cancer Chemother Pharmacol. 2010;65: [PMID: ] 10. Jonker DJ, O Callaghan CJ, Karapetis CS, Zalcberg JR, Tu D, Au HJ, et al. Cetuximab for the treatment of colorectal cancer. N Engl J Med. 2007;357: [PMID: ] 11. Van Cutsem E, Peeters M, Siena S, Humblet Y, Hendlisz A, Neyns B, et al. Open-label phase III trial of panitumumab plus best supportive care compared with best supportive care alone in patients with chemotherapy-refractory metastatic colorectal cancer. J Clin Oncol. 2007;25: [PMID: ] 12. Bos JL. ras oncogenes in human cancer: a review. Cancer Res. 1989;49: [PMID: ] 13. Siena S, Sartore-Bianchi A, Di Nicolantonio F, Balfour J, Bardelli A. Biomarkers predicting clinical outcome of epidermal growth factor receptortargeted therapy in metastatic colorectal cancer. J Natl Cancer Inst. 2009;101: [PMID: ] 14. Amado RG, Wolf M, Peeters M, Van Cutsem E, Siena S, Freeman DJ, et al. Wild-type KRAS is required for panitumumab efficacy in patients with metastatic colorectal cancer. J Clin Oncol. 2008;26: [PMID: ] 15. Karapetis CS, Khambata-Ford S, Jonker DJ, O Callaghan CJ, Tu D, Tebbutt NC, et al. K-ras mutations and benefit from cetuximab in advanced colorectal cancer. N Engl J Med. 2008;359: [PMID: ] 16. Van Cutsem E, Köhne CH, Hitre E, Zaluski J, Chang Chien CR, Makhson A, et al. Cetuximab and chemotherapy as initial treatment for metastatic colorectal cancer. N Engl J Med. 2009;360: [PMID: ] 17. Bokemeyer C, Bondarenko I, Makhson A, Hartmann JT, Aparicio J, de Braud F, et al. Fluorouracil, leucovorin, and oxaliplatin with and without cetuximab in the first-line treatment of metastatic colorectal cancer. J Clin Oncol. 2009;27: [PMID: ] 18. Simon RM, Paik S, Hayes DF. Use of archived specimens in evaluation of prognostic and predictive biomarkers. J Natl Cancer Inst. 2009;101: [PMID: ] 19. Lin BK, Clyne M, Walsh M, Gomez O, Yu W, Gwinn M, et al. Tracking the epidemiology of human genes in the literature: the HuGE Published Literature database. Am J Epidemiol. 2006;164:1-4. [PMID: ] 20. Savas S, Younghusband HB. dbcpco: a database of genetic markers tested for their predictive and prognostic value in colorectal cancer. Hum Mutat. 2010; 31: [PMID: ] 21. Allegra C, Blanke C, Buyse M, Goldberg R, Grothey A, Meropol NJ, et al. End points in advanced colon cancer clinical trials: a review and proposal. J Clin Oncol. 2007;25: [PMID: ] 22. Mayer RJ. Targeted therapy for advanced colorectal cancer more is not always better [Editorial]. N Engl J Med. 2009;360: [PMID: ] 23. Senn SJ. Overstating the evidence: double counting in meta-analysis and related problems. BMC Med Res Methodol. 2009;9:10. [PMID: ] 24. Parmar MK, Torri V, Stewart L. Extracting summary statistics to perform meta-analyses of the published literature for survival endpoints. Stat Med. 1998; 17: [PMID: ] 25. Michiels S, Piedbois P, Burdett S, Syz N, Stewart L, Pignon JP. Metaanalysis when only the median survival times are known: a comparison with individual patient data results. Int J Technol Assess Health Care. 2005;21: [PMID: ] 26. Tierney JF, Stewart LA, Ghersi D, Burdett S, Sydes MR. Practical methods for incorporating summary time-to-event data into meta-analysis. Trials. 2007;8: 16. [PMID: ] 27. Rothman KJ, Greenland S, Lash TJ. Modern Epidemiology. 3rd ed. Baltimore: Lippincott Williams & Wilkins; McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM; Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics. Reporting recommendations for tumor marker prognostic studies (REMARK). J Natl Cancer Inst. 2005;97: [PMID: ] 29. Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ. 2003;326:219. [PMID: ] 30. Irwig L, Tosteson AN, Gatsonis C, Lau J, Colditz G, Chalmers TC, et al. Guidelines for meta-analyses evaluating diagnostic tests. Ann Intern Med. 1994; 120: [PMID: ] 31. Leeflang MM, Deeks JJ, Gatsonis C, Bossuyt PM; Cochrane Diagnostic Test Accuracy Working Group. Systematic reviews of diagnostic test accuracy. Ann Intern Med. 2008;149: [PMID: ] 32. Linardou H, Dahabreh IJ, Kanaloupiti D, Siannis F, Bafaloukos D, Kosmidis P, et al. Assessment of somatic k-ras mutations as a mechanism associated with resistance to EGFR-targeted agents: a systematic review and meta-analysis of studies in advanced non-small-cell lung cancer and metastatic colorectal cancer. Lancet Oncol. 2008;9: [PMID: ] 33. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7: [PMID: ] 34. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327: [PMID: ] 35. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics. 1994;50: [PMID: ] 36. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315: [PMID: ] 37. Terrin N, Schmid CH, Lau J. In an empirical evaluation of the funnel plot, researchers could not visually identify publication bias. J Clin Epidemiol. 2005; 58: [PMID: ] 38. Lau J, Ioannidis JP, Terrin N, Schmid CH, Olkin I. The case of the misleading funnel plot. BMJ. 2006;333: [PMID: ] 39. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol. 2005;58: [PMID: ] 40. Chu H, Cole SR. Bivariate meta-analysis of sensitivity and specificity with sparse data: a generalized linear mixed model approach [Letter]. J Clin Epidemiol. 2006;59:1331-2; author reply [PMID: ] 41. Harbord RM, Whiting P. metandi: Meta-analysis of diagnostic accuracy using hierarchical logistic regression. Stata Journal. 2009;9: Knapp G, Hartung J. Improved tests for a random effects meta-regression with a single covariate. Stat Med. 2003;22: [PMID: ] 43. Thompson SG, Higgins JP. How should meta-regression analyses be undertaken and interpreted? Stat Med. 2002;21: [PMID: ] 44. van Houwelingen HC, Arends LR, Stijnen T. Advanced methods in metaanalysis: multivariate approach and meta-regression. Stat Med. 2002;21: [PMID: ] 45. Wallace BC, Schmid CH, Lau J, Trikalinos TA. Meta-Analyst: software for meta-analysis of binary, continuous and diagnostic data. BMC Med Res Methodol. 2009;9:80. [PMID: ] 46. Hecht JR, Mitchell E, Chidiac T, Scroggin C, Hagenstad C, Spigel D, et al. A randomized phase IIIB trial of chemotherapy, bevacizumab, and panitumumab compared with chemotherapy and bevacizumab alone for metastatic colorectal cancer. J Clin Oncol. 2009;27: [PMID: ] 47. Tol J, Koopman M, Cats A, Rodenburg CJ, Creemers GJ, Schrama JG, et al. Chemotherapy, bevacizumab, and cetuximab in metastatic colorectal cancer. N Engl J Med. 2009;360: [PMID: ] 48. Folprecht G, Gruenberger T, Bechstein WO, Raab HR, Lordick F, Hartmann JT, et al. Tumour response and secondary resectability of colorectal liver metastases following neoadjuvant chemotherapy with cetuximab: the CELIM randomized phase 2 trial. Lancet Oncol. 2010;11: [PMID: ] 49. Yen LC, Yeh YS, Chen CW, Wang HM, Tsai HL, Lu CY, et al. Detection of KRAS oncogene in peripheral blood as a predictor of the response to cetuximab 4 January 2011 Annals of Internal Medicine Volume 154 Number 1 47