Studies were identified by searching electronic databases, conference abstracts, regulatory documents, and trial registries. MEDLINE was searched from January 2000 to November 2010 for English language abstracts. This search was adapted for four additional databases (Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Database of Abstracts of Reviews of Effects, and Health Technology Assessments Database) and limited to publications between 2000 and 2010, with no language restrictions. We also searched Conference Papers Index (via CSA) from 2009 to 2010 and hand searched selected scientific conferences from 2009 to November 2010. Relevant studies were also identified by searching clinicaltrials.gov
, NIH RePORTER, Current Controlled Trials (International Standard Randomized Controlled Trial Number Register), WHO International Clinical Trials Registry Platform, and FDA regulatory documents via Drugs@FDA. The last search for all databases was performed on November 24, 2010. Search details are provided in Search Strategies document ().
Two investigators independently reviewed 3,365 abstracts and 191 articles against a priori specified inclusion criteria (). For all key questions, we considered studies that included persons with metastatic colorectal cancer being treated with cetuximab or panitumumab, either alone or in combination with other chemotherapeutic agents. Studies that only included patients with locally advanced disease were excluded. Testing included assays for mutations in BRAF, NRAS, PIK3CA, and protein expression for PTEN and AKT. We also included studies examining PTEN and AKT mutations and gene copy number. We excluded testing for EGFR protein expression or gene copy number; upstream molecular drivers (i.e., EGFR ligands epiregulin and amphiregulin); and molecular targets not directly part of the EGFR signaling cascade, but mediators in adjacent pathways. We considered any study reporting one or more of the following outcomes: overall survival, clinical response to treatment (e.g., progression free survival, time to progression), health-related quality of life, radiologic evidence of tumor progression, or potential adverse effects (e.g., incorrect genotype assignment leading to incorrect treatment assignment, delayed treatment, negative psychological effects, and ethical, legal, and social issues/concerns). We did not exclude studies based on study design or study quality. Excluded studies are listed in .
Two investigators independently assessed the quality of each study using the quality criteria proposed by the EGAPP working group [9
], supplemented by the Newcastle Ottawa Scale developed for observational studies [10
], and reporting standards checklist (REMARK) developed for prognostic and predictive studies [11
]. Articles were rated good-, fair-, or marginal-quality. Good-quality studies were those that met the following criteria: prospective design; large, well-defined, and representative study population; genetic testing was described well; blinded assessment of genetic testing in relation to outcome; homogeneous treatment; low rate of missing data; sufficiently long follow-up; and well-described and well-conducted analysis of outcomes. Fair-quality studies did not meet all the criteria, but did not have any fatal flaws in study design. Marginal- or poor-quality studies had significant flaws or lack of reporting that implied bias affecting interpretation of study results. Disagreements about inclusion and quality were resolved by consensus with a third reviewer.
One investigator extracted all relevant data from the studies into evidence tables that included the following study details: critical features of study design and quality, funding source, patient characteristics (e.g., age, sex, race/ethnicity, Eastern Cooperative Oncology Group [ECOG] performance status, metastatic disease), treatment regimen and setting, genetic testing details (e.g., gene mutation(s)/protein expression, tumor sample, assay technique, scoring method), frequency of gene mutation or protein expression, a priori specified study outcomes (stratified by KRAS wild-type if available), and any outcome representing potential adverse effects. A second reviewer verified all extracted data.
We identified 27 studies (reported in 34 articles); however there was a significant overlap of populations studied. Our evidence synthesis focuses on studies with independent patient populations and, when possible, results in persons with metastatic colorectal cancer that have KRAS
wild-type tumors. We focus on three primary outcomes— tumor response (or disease control if tumor response is not reported) based on radiographic findings, progression-free survival (PFS), and overall survival (OS). Studies either used RECIST or World Health Organization (WHO) criteria (based on radiographic findings) to assess tumor response or disease control. Most individual studies reported tumor response rates with or without an odds ratio (OR). For tumor response, we also calculated the true positive fraction (TPF or clinical sensitivity) and false positive fraction (FPF or 1-specificity) if sufficient data were reported in persons with KRAS
wild-type tumors [12
]. For continuous outcomes (survival), in addition to reporting hazard ratios (HR), we also report absolute differences between groups in weeks or months of median progression-free or overall survival.
We summarize results qualitatively and provide these results in tables for easy comparison across studies representing unique populations. For tumor response based on imaging, we attempted quantitative synthesis (meta-analyses) for sensitivity, specificity, and odds ratios to evaluate the predictive value for each genetic test with sufficient data. Due to overlapping populations and lack of outcome reporting for individuals with KRAS
wild-type tumors, only 3 studies could be included in the meta-analyses. We attempted bivariate analyses for sensitivity and specificity (of BRAF
and PTEN testing) simultaneously [13
], as well as univariate meta-analyses for sensitivity, specificity, and diagnostic odds ratios, separately using random effects models [14
]. However, the small number of studies and clinical heterogeneity among studies prohibited us from producing meaningful combined estimates. We instead focused on the best available evidence (e.g., single large, well-reported study) to provide the best estimate of clinical validity. We also considered how additional studies with independent and overlapping populations confirmed, disagreed, and/or contributed additional information to the best evidence detailed.
In addition to a summary of evidence table, we also provide a summary table focusing on the strength of the body of evidence, based on the GRADE (Grading of Recommendations Assessment, Development and Evaluations) approach [16
]. The following four domains were assessed: risk of bias, consistency, directness, and precision. The overall strength of evidence was graded as high, moderate, low, or very low (insufficient).