|Home | About | Journals | Submit | Contact Us | Français|
These studies were conducted to determine the relationship between quantitative tumor gene expression and risk of cancer recurrence in patients with stage II or III colon cancer treated with surgery alone or surgery plus fluorouracil (FU) and leucovorin (LV) to develop multigene algorithms to quantify the risk of recurrence as well as the likelihood of differential treatment benefit of FU/LV adjuvant chemotherapy for individual patients.
We performed quantitative reverse transcription polymerase chain reaction (RT-qPCR) on RNA extracted from fixed, paraffin-embedded (FPE) tumor blocks from patients with stage II or III colon cancer who were treated with surgery alone (n = 270 from National Surgical Adjuvant Breast and Bowel Project [NSABP] C-01/C-02 and n = 765 from Cleveland Clinic [CC]) or surgery plus FU/LV (n = 308 from NSABP C-04 and n = 508 from NSABP C-06). Overall, 761 candidate genes were studied in C-01/C-02 and C-04, and a subset of 375 genes was studied in CC/C-06.
A combined analysis of the four studies identified 48 genes significantly associated with risk of recurrence and 66 genes significantly associated with FU/LV benefit (with four genes in common). Seven recurrence-risk genes, six FU/LV-benefit genes, and five reference genes were selected, and algorithms were developed to identify groups of patients with low, intermediate, and high likelihood of recurrence and benefit from FU/LV.
RT-qPCR of FPE colon cancer tissue applied to four large independent populations has been used to develop multigene algorithms for estimating recurrence risk and benefit from FU/LV. These algorithms are being independently validated, and their clinical utility is being evaluated in the Quick and Simple and Reliable (QUASAR) study.
Although adjuvant chemotherapy is the standard of care in stage III colon cancer, its routine use in patients with stage II colon cancer is controversial.1–10 The Quick and Simple and Reliable (QUASAR) study11 showed that adjuvant chemotherapy with fluorouracil (FU) plus leucovorin (LV) produces a small (approximately 3%) survival benefit in stage II colon cancer, which must be balanced with its toxicity, including toxic deaths (approximately 0.5%). This narrow therapeutic index underscores the importance of selecting the appropriate patients for adjuvant treatment.
In current practice, clinical and pathologic markers (ie, intestinal perforation/obstruction, pathologic stage T4, presence of lymphatic/vascular invasion, high tumor grade, < 12 nodes examined) can identify a minority of patients with stage II disease who have higher recurrence risk, but they do not adequately assess recurrence risk for individual patients. To address this issue, the use of molecular markers, such as microsatellite instability (MSI)/mismatch repair (MMR), LOH 18q, and levels of expression of individual genes or groups of genes12–21 has been investigated. Some recent studies suggest MMR deficiency (ie, MSI high) may identify a small percentage (approximately 15%) of patients with stage II disease who receive little benefit from FU/LV.22 However, the clinical utility of these markers remains under study.23
Here, we report the application of the quantitative reverse transcription polymerase chain reaction (RT-qPCR) platform developed for the Oncotype DX Breast Cancer Assay (Genomic Health, Inc, Redwood City, CA)24–26 in four independent colon cancer studies to generate the 12-gene recurrence score and 11-gene treatment score algorithms that, if validated, will quantify the risk of recurrence as well as the likelihood of differential treatment benefit of FU/LV adjuvant chemotherapy for individual patients with stage II colon cancer.
Samples from four independent cohorts of patients with stage II or stage III colon cancer treated with surgery alone (National Surgical Adjuvant Breast and Bowel Project [NSABP] C-01/C-02 or Cleveland Clinic [CC] study) or surgery plus FU/LV (NSABP C-04, NSABP C-06) were studied (Appendix Table A1, online only; Appendix Fig A1, online only).27–30 Prespecified criteria for being evaluable were as follows: eligibility for the parent clinical studies; availability of the fixed, paraffin-embedded (FPE) tumor block from initial diagnosis; presence of sufficient tumor (ie, ≥ 5% of tissue area occupied by invasive cancer cells in the guide hematoxylin and eosin slide); pathology diagnosis of colon adenocarcinoma (excluding signet ring carcinoma); adequate RNA to perform quantitative RT-qPCR analysis (≥ 1,069 ng for C-01/C-02 and C-04 and ≥ 587 ng for CC and C-06); and sufficient RNA quality by predefined metrics.
For each patient, RNA was extracted from three pooled 10-μm sections obtained from archived FPE colon tumor tissue. Nontumor elements were commonly identified on the guide hematoxylin and eosin slide reviewed for each patient and were removed by manual microdissection before transfer to the extraction tube.
Assessment of tumor grade was performed according to WHO criteria31 by an academic surgical pathologist with sub-specialty expertise in gastrointestinal pathology. The extracted RNA was quantified and then analyzed by RT-qPCR.32 For the C-01/C-02 and C-04 cohorts, two 384 well plates, which contained a total of 761 unique assays (ie, 761-gene panel), were used for each sample. With the exception of four assays (three K-ras mutations and one BRAF mutation), all assays were designed to detect the expression levels of wild-type genes. The panel of 761 candidate genes (Appendix Table A2, online only) was constructed from published gene expression profiling data and from biologic pathways identified as functionally important in colon cancer.17–21 For the CC and C-06 cohorts, one 384-well plate, which contained 375 unique assays (ie, 375-gene panel), was used for each sample. The genes for the 375-gene panel were chosen from the 761-gene panel on the basis of the strength of the association of their level of expression with recurrence risk and chemotherapy benefit in the C-01/C-02 and C-04 studies. Gene expression measurements were normalized relative to five reference genes. Among the available samples across the four cohorts, only eight were excluded because of inadequate RT-qPCR expression.
MMR status was assessed by immunohistochemistry for MLH1 and MSH2 (which identify > 90% of the MMR-deficient tumors) on fixed, primary colon tumor tissue in the CC study.33
FPE tissue sections were prepared by either NSABP or CC personnel and were shipped to Genomic Health, Inc (Redwood City, CA), where the expression profiling was performed, blinded to the clinical data. The expression data and the clinical/pathology data were independently locked and then merged to construct the analysis data set for each study.
The primary objective of all four studies was to identify genes associated with recurrence-free interval (RFI), defined as the time from surgery to first colon cancer recurrence. Deaths before recurrence were considered censoring events. Second primary cancers were considered neither events nor censoring events. Secondary end points were disease-free survival and overall survival.
Prespecified univariate (primary analysis) and multivariate relationships between clinical outcomes and categorical or continuous variables (eg, gene expression) were modeled using Cox proportional hazards regression.34 All baseline patient characteristics related to RFI (P < .20) were included in the multivariate analysis for a given study. Hazard ratios (HRs) were tested for significance using the likelihood ratio test.35 For univariate models of gene expression and RFI, an unadjusted P value less than .05 was considered significant. A test of interaction was used to identify genes that predict treatment benefit; because such tests have lower power compared with the main effect tests, an unadjusted P value of less than .10 was considered significant. No adjustment for multiplicity was applied. To estimate the false discovery rate (FDR), the Benjamini-Hochberg method was used within each study,36 and a permutation-based method37 was used across studies.
For each of the 375 genes assessed in the four studies, univariate t tests were performed to identify mean differences in gene expression between patients with stage II and stage III disease in each study. Additionally, Cox proportional hazards regression models of gene expression, stage, and the interaction of gene expression and stage, stratified by study, were examined, and a P value of less than .10 for interaction was considered significant. In the absence of strong evidence of stage differences, data across stages were combined for gene discovery and algorithm development.
To identify clusters of coexpressed genes and to facilitate the understanding of important biologic pathways, unsupervised hierarchical clustering of genes was performed using Pearson r as the distance measure for gene expression and the unweighted pair-group average as the amalgamation method.35 Similar results were obtained using other methods, such as principal component analysis.
A smaller subset of genes significantly and consistently related to risk of recurrence was identified by examining the results across studies. Multiple factors were considered for gene inclusion in algorithm development, including, but not limited to, the known role of the genes in important biologic pathways, analytic performance, and range of expression. The final gene panels and algorithms for prediction of recurrence risk (ie, recurrence score; Table 1) and chemotherapy benefit (ie, treatment score; Table 2) were derived as described in the text of the Appendix (online only).
Bootstrap methods38,39 were used to evaluate the extent to which recurrence risk differed among the recurrence risk groups defined by recurrence score for patients with stage II disease. A total of 1,000 bootstrap samples were drawn randomly with replacement from the pooled data set, taking variability between studies into account. Kaplan-Meier estimates of recurrence risk at 3 years were obtained for each recurrence risk group. Median recurrence risk estimates across all bootstrap samples and percentile CIs were reported. A similar approach was used to assess the results of the final multigene algorithm to predict FU/LV benefit, in which patients were divided in chemotherapy benefit groups on the basis of both their recurrence scores and treatment scores (Appendix Table A3, online only). Data were analyzed independently by the NSABP Biostatistical Center, Cleveland Clinic, and Genomic Health, Inc for individual studies. Analyses across four studies were conducted by Genomic Health.
The final numbers of evaluable patients were 270 in the C-01/C-02, 765 in the CC, 308 in the C-04, and 508 in the C-06 cohort. The outcomes and clinical/demographic characteristics of evaluable patients with tumor blocks were similar to those observed in the parent NSABP studies.
The baseline characteristics of the three NSABP cohorts were generally similar; patients from CC differed in age, percentage of right-sided tumors, number of lymph nodes examined, and percentage of stage II versus stage III disease (Appendix Table A1). Univariate Cox proportional hazards regression identified nodal status (0 positive nodes and ≥ 12 nodes examined, 0 positive nodes and < 12 nodes examined, 1 to 3 or ≥ 4 positive nodes) as the most significant clinical/pathologic predictor of RFI (P < .001) in all studies (Appendix Tables A4, A5, A6, and A7, online only; Appendix Figs A2A, A2B, A2C, and A2D, online only). T stage, available in adequate numbers of patients in CC, was associated with RFI (T4 v other; P = .003; Appendix Table A5). MMR was not associated with RFI in CC (P = .27; Appendix Table A5).
Univariate analysis identified 143 genes as significantly related to RFI in the C-01/C-02 cohort, 119 in the CC cohort, 143 in the C-04 cohort, and 169 in the C-06 cohort; 27%, 16%, 27%, and 11% of these genes, respectively, were expected to be false discoveries. When studies were pooled, the FDR was markedly lower. In the multivariate analysis, 43%, 74%, 50%, and 84% of the genes identified in the univariate analysis retained significance in C-01/C-02, CC, C-04, and C-06, respectively, and had similar HRs in both analyses (Appendix Figs A3 and A4, online only, for surgery-alone studies; similar results for studies of surgery + FU/LV not shown), which suggests that gene expression contributes information about recurrence beyond standard clinical and pathologic covariates. In these multivariate analyses, the contribution of nodal status was consistently statistically significant.
The relationship between gene expression, tumor stage, and RFI was investigated across studies. In univariate analyses, six of the 375 genes had significant (P < .05) mean differences in expression between patients with stage II and stage III disease in all four studies. Thirty-three genes had a significant interaction of gene expression and stage (P < .1) in Cox proportional hazards models stratified by study; 32 of these 33 genes were potential false positives (FDR = 97%), which suggests that interaction between gene expression and stage was weak. The coexpression of genes examined using cluster analysis was virtually identical in patients with stage II and stage III disease. Agreement between univariate HRs for patients with stage II and stage III disease is illustrated in Figure 1 for genes significantly associated with RFI in both surgery-alone studies and at least one study of surgery plus FU/LV. HRs were generally similar with overlapping CIs, with a few exceptions that could be chance findings due to multiplicity of testing across multiple genes and studies. These results support pooling data across stages for gene discovery and algorithm development.
Recurrence risk genes were expected to have a similar relationship with RFI when measured in patients treated with surgery alone or surgery followed by FU/LV. A total of 48 (13%) of the 375 genes studied in all four development studies were significantly (P < .05) associated with RFI in both surgery alone studies and at least one study of surgery plus FU/LV. Fewer than one of these 48 genes is expected to be a false discovery. Cluster analysis identified two relatively distinct gene groups: a stromal gene group (containing several subgroups such as early response) and a cell cycle gene group (Fig 2). Higher expression of stromal genes (eg, BGN, FAP, GADD45B, and PAI) was associated with higher risk of recurrence, whereas higher expression of cell cycle genes (eg, Ki–67, MYBL2, and MCM2) was associated with lower risk of recurrence.
In contrast to recurrence risk genes, the genes predictive of differential FU/LV benefit are required to exhibit a different relationship with outcome (ie, different HRs) in patients treated with surgery alone compared with patients treated with surgery plus FU/LV. A total of 66 (18%) of 375 genes studied in all four development studies had interactions of gene expression and treatment that were significant at the less than .10 level if the data across the four studies were pooled (13 genes with P < .01; 45 genes with P < .05); four of these genes were also associated with risk of recurrence at the P < .05 level. Approximately 37 of these 66 genes are expected to be false discoveries; this was expected, given the lower statistical power associated with the analysis of interaction. Among 66 potentially predictive genes, there were a large number of genes involved in multiple stages of the cell cycle and apoptosis (ie, MAD2L1, AURKB, BIK, BUB1, CDC2), and higher expression was associated with greater differential benefit from FU/LV (Appendix Fig A5, online only). There were also a prominent stress response/hypoxia signature (ie, HSPE1, NR4A1, RhoB, HIF1A); multifunctional transcription factors (RUNX1, CREBBP, KLF5); and genes associated with wnt signaling (AXIN2 and LEF), MMR (MSH2 and MSH3), and angiogenesis (EFNB2). Higher expression of some of these genes (eg, RUNX1, CREBBP, KLF5, and EFNB2) was associated with lower benefit from FU/LV.
Seven of the 48 recurrence risk genes and six of the 66 chemotherapy benefit genes were selected to create the final recurrence score and treatment score algorithms (Appendix Figs A1 and A5; described in the Appendix). The results of bootstrap analyses to assess the predictive ability of the recurrence score are listed in Table 1 for patients with stage II disease treated with surgery alone (C-01/C-02 and CC cohorts). Patients were divided into three recurrence risk groups on the basis of the calculated recurrence score (ie, < 30, 30-40, and ≥ 41). The recurrence score separated the 632 patients with stage II disease into groups that had a sizable difference in estimates of risk of recurrence between the high- and the low-risk groups.
The results of bootstrap analyses to assess the predictive ability of the treatment score are listed in Table 2 for the 870 patients with stage II disease treated with surgery alone or surgery plus FU/LV. Patients were divided into three benefit groups (Appendix). For comparison, the overall 3-year risk of recurrence of patients with stage II disease was 14% for the surgery-alone group and was 10% for patients treated with surgery plus FU/LV. The correlation between the recurrence score and treatment score was relatively low (r = −0.4) in these studies, which suggests that the determinants of recurrence risk and differential FU/LV benefit may be distinct.
The performance of these algorithms was evaluated on the data set used for algorithm development; hence, the results in Tables 1 and 2 are likely to be optimistic. These algorithms will be validated on an independent data set of patients with stage II colon cancer from the QUASAR study.11
Our strategy for discovering genes related to recurrence risk and differential benefit with adjuvant FU/LV chemotherapy has been to perform multiple, large, independent studies to identify those genes most consistently and strongly related to clinical outcome (Appendix Fig A1).40,41 The results reported here are based on data from 1,851 patients, using standardized assay technology in a single laboratory. This approach is in contrast to algorithms developed from much larger numbers of genes and significantly smaller sample sizes.17–21
We have identified 48 genes that have significant and similar relationships with RFI and 66 genes that have different relationships with RFI in patients treated with surgery alone compared with patients treated with surgery plus FU/LV: the former genes are likely to predict recurrence, whereas the latter genes are likely to predict differential benefit with adjuvant FU/LV therapy. A large proportion of these genes remain significantly associated with RFI after analysis is controlled for the effects of numerous clinical/pathologic covariates, including nodal status, which also contributed significantly to prediction of recurrence risk.
This report highlights several challenges to biomarker development in colon cancer. First, our ability to identify genes predictive of differential treatment benefit was limited by the lack of large, randomized clinical trials with tumor specimens (beyond the QUASAR validation trial). Second, the lower power of the test of interaction leads to a high FDR among the candidate predictive genes, of greater than 50% in our studies. Finally, the question of whether stage II and III disease are biologically similar or dissimilar is unresolved. However, for the vast majority of genes, we saw no strong difference between stages in the relationship of gene expression and RFI, and an additional analysis that compared patients who had stage II disease and ≥ 12 nodes examined with patients who had stage III disease confirmed these findings (data not shown).
Our strategy has led to the discovery of recurrence risk genes that can be confidently associated with clinical outcome and are generally different from the genes identified in our breast cancer studies (with the exceptions of Ki-67 and MYBL2)42 or the genes previously reported in colon cancer.17–21 Higher expression of cell cycle genes was associated with an increased RFI; this observation is similar to that reported for Ki-67 in colon cancer43,44 but is opposite to the relationship observed in breast cancer studies.42 The association of stromal gene expression with colon cancer recurrence provides an elegant molecular explanation for Dukes' original observation that invasion is the critical characteristic that should be used in staging colorectal cancer.45,46
Some of the genes that were identified as predictive of chemotherapy benefit are not unexpected. Sensitivity to FU should be affected by factors related to the level of proliferation (ie, cell-cycle related genes), the induction of apoptosis and hypoxia,47 FU metabolism, and MMR.48 However, it is not clear why the expression of other genes, such as GJB2 or HES6, which are associated with gap junction communication and notch signaling respectively, would predict FU benefit. The quantitative expression of the genes related to FU activation/metabolism (TS, DPD, TP)15,49–54 or to the markers (hMLH1, hMSH2) associated with MMR55–57 were not associated with differential FU/LV benefit in our studies; current guidelines by the American Society of Clinical Oncology conclude that there is insufficient evidence to recommend the use of these markers as predictors of response to therapy.23
This report describes our process for identifying genes that can be used to estimate recurrence risk and differential FU/LV chemotherapy benefit on the basis of the relationship of quantitative gene expression at the time of diagnosis and clinical outcome in patients with stage II/III colon cancer treated with surgery or surgery plus FU/LV. The results of these four studies have been used to develop a multigene assay (Figs 3 and and4)4) for prediction of recurrence risk and differential benefit of adjuvant FU/LV chemotherapy that can be used to divide patients with stage II colon cancer into groups with different likelihoods of recurrence and treatment benefit. The requirement for external validation is being addressed in QUASAR,11 a large, independent study of patients with stage II colon cancer randomly assigned to surgery alone or to surgery followed by adjuvant FU/LV chemotherapy.
We thank Barbara C. Good, PhD, Director of Scientific Publications for the National Surgical Adjuvant Breast and Bowel Project, for editorial assistance; and Meike Labusch, Angela Chen, Anhthu Nguyen, Bhavin Padhiar, Debjani Dutta, Jayadevi Krishnakumar, Jennie Jeong, Jenny Wu, Hyun Soo Son, Mei-Lan Liu, Mylan Pho, Ranjana Ambannavar, Lauren Intagliata, Freda Lane, James Hackett, and Jeanne Yue from Genomic Health, Inc, for their contributions to sample and data processing and data analysis.
To develop a tumor gene expression assay for use with tumor blocks that are routinely prepared following surgery, we used a multistep approach.
First, a real-time reverse transcriptase polymerase chain reaction method to quantify the expression of hundreds of genes in RNA isolated from three 10-μm sections of fixed, paraffin-embedded tumor tissue was developed.24
Second, we selected 761 candidate genes from the published literature, genomic databases, pathway analysis, and from microarray-based gene expression profiling experiments performed with fresh frozen tissue.17–21 The 761 candidate genes are listed in Appendix Table A2 (online only).
Third, we performed four independent clinical studies in a total of 1,851 patients with colon cancer to test the relationship between the expression of the candidate genes and time to recurrence. We reasoned that in any single gene expression study a considerable number of genes may correlate with outcome as a result of chance alone. To identify true positives, we performed multiple independent studies to identify whether expression of any of the candidate genes correlated with recurrence across the studies. We hypothesized that the genes most highly correlated with recurrence would survive evaluation across diverse patients and treatments, and we selected heterogeneous populations for development of the gene list. After National Surgical Adjuvant Breast and Bowel Project (NSABP) C-01/C-02 and NSABP C-04 were conducted, the gene list was reduced to 375 genes on the basis of the strength of the relationship between gene expression and recurrence risk as well as the association of gene expression with chemotherapy benefit; and the NSABP C-06 patients and Cleveland Clinic cohorts were studied with 375 genes (Appendix Table A2). There were 48 genes for which expression was associated with recurrence in three of four studies at an unadjusted P value less than .05, and 66 genes had an interaction of gene expression and chemotherapy treatment significant at an unadjusted P < .1.
Fourth, we used the results from the four development studies to select the final gene panels and to design algorithms to compute a recurrence score (RS) and a treatment score (TS) for each tumor sample. The list of gene candidates was narrowed down using a number of considerations that included, but were not limited to, the strength of the associations with recurrence, the consistency of performance across studies, consistency of performance in patients with stage II and stage III disease, and analytic performance. The correlation of the expression of the genes with respect to each other was analyzed by unsupervised cluster analysis and by principal component analysis. Two major gene groups were identified among the prognostic genes: a stromal group (containing extracellular matrix genes, such as BGN, FAP, INHBA, and SPARC; early response genes, such as GADD45B; and invasion genes, such as PAI), and a cell cycle group (genes such as Ki-67, MYBL2, MCM2). Increased expression of stromal genes was associated with increased risk of recurrence, whereas increased expression of cell cycle genes was associated with decreased risk of recurrence. Among the 66 predictive genes, there were several multifunctional transcription factors (RUNX1, TCF1, CREBBP, KLF5) and genes involved in cell cycle and apoptosis (eg, MAD2L1, AURKB, BIK, TOP2A, BUB1, CDC2), hypoxia/stress response (HSPE1, NR4A1, RhoB, HIF1A, CREBBP, EPAS), wnt signaling (AXIN 2 and LEF), mismatch repair (MSH2 and MSH3), and angiogenesis (EFNB2). Analyses were then performed to determine the appropriate number of terms to include in the model and the functional forms of the variables. For this purpose, we used correlation analysis, dimension reduction (including stepwise variable selection and classification and regression trees), Martingale residual analysis, and bootstrap resampling. Multiple analyses across the four studies were conducted to determine whether models with a larger number of genes were better able to predict risk of recurrence and/or chemotherapy benefit and whether having fewer genes resulted in loss of robustness. We observed that relatively parsimonious models with anywhere from six to 10 recurrence-risk genes and five to six chemotherapy-benefit genes were adequate. The selection of the final seven recurrence-risk and six chemotherapy-benefit genes was based primarily on the strength of their performance across all studies and the consistency of primer/probe performance in the assay.
|Characteristic||No. of Patients||Patients by Study Cohort|
|NSABPC-01/C-02||NSABP C-04||Cleveland Clinic||NSABP C-06|
|Multiple or unknown||25||10||3.7||8||2.6||0||0.0||7||1.4|
|No. of nodes examined|
|No. of positive nodes|
|Nodal status of examined nodes|
|0 of < 12 examined||365||64||24.5||68||22.4||117||15.3||116||22.8|
|0 of ≥ 12 examined||634||62||23.8||66||21.7||387||50.6||119||23.4|
|≥ 4 overall||241||41||15.7||56||18.4||60||7.8||84||16.5|
Abbreviations: NSABP, National Surgical Adjuvant Breast and Bowel Project.
|No.||Gene Name||Sequence ID||375 Gene Panel||No.||Gene Name||Sequence ID||375 Gene Panel|
|22||ANXA1||NM_000700.1||Yes||405||KCNH2 iso a/b||NM_000238.2|
|23||ANXA2||NM_004039.1||Yes||406||KCNH2 iso a/c||NM_172057.1||Yes|
|39||axin1||NM_003502.2||Yes||422||K-ras mutant 1||GHI_k-ras_mut1||Yes|
|40||BAD||NM_032989.1||Yes||423||K-ras mutant 2||GHI_k-ras_mut2||Yes|
|66||Braf Mutant 1||GHI_BRAF_mut4||Yes||449||MADH4||NM_005359.3||Yes|
|273||FGFR2 isoform 1||NM_000141.2||656||SOS1||NM_005633.2||Yes|
|Chemotherapy Benefit Group||X = 0.859exp[1.839×RSu+3.526+1.781xTSu]−0.859exp[1.839×RSu]|
|Low||X less than 2%|
|Intermediate||X greater than or equal to 2% and less than 6%|
|High||X greater than or equal to 6%|
NOTE. The unscaled recurrence score and unscaled treatment score were used in combination to determine a chemotherapy benefit group for each patient, as the absolute chemotherapy benefit was also a function of baseline recurrence risk.
|Variable||No. of Patients||HR||95% CI||P|
|Female v male||270||0.82||0.57 to 1.19||.299|
|Age, per 1 year increase||270||1.00||0.99 to 1.02||.702|
|Age ≥ 60 v < 60 years||270||1.02||0.70 to 1.48||.929|
|Location of tumor||270||.006|
|Right v left||2.46||1.45 to 4.19|
|Rectosigmoid v left||1.76||1.02 to 3.04|
|Multiple/unknown v left||2.21||0.82 to 5.93|
|Protocol (C-02 v C-01)||270||0.65||0.42 to 1.02||.052|
|Surgery + BCG v surgery only||270||1.37||0.94 to 2.00||.107|
|No. of nodes examined||270||.335|
|< 12 v ≥ 12||1.18||0.81 to 1.73|
|Unknown v≥ 12||1.81||0.82 to 4.00|
|No. of positive nodes||266||< .001|
|1-3 v 0||1.73||1.13 to 2.65|
|≥ 4 v 0||2.94||1.80 to 4.78|
|Nodal involvement status||261||< .001|
|0 of < 12 examined v 0 of ≥ 12 examined||2.45||1.27 to 4.72|
|1-3 v 0 of ≥ 12 examined||2.79||1.50 to 5.19|
|≥ 4 v 0 of ≥ 12 examined||4.74||2.44 to 9.20|
|Stage III v II||270||2.10||1.43 to 3.08||< .001|
|Tumor grade at GHI|
|First reading high v low||270||1.35||0.88 to 2.09||.180|
|Second reading high v low||269||1.51||1.01 to 2.26||.054|
|Segmental/anterior resection v colectomy/hemicolectomy||1.27||0.82 to 1.96|
|Other/unknown v colectomy/hemicolectomy||0.78||0.42 to 1.48|
|Mucinous tumor||270||1.64||0.92 to 2.93||.115|
|Surgery year||270||0.95||0.88 to 1.02||.151|
Abbreviations: RFI, recurrence-free interval; NSABP, National Surgical Adjuvant Breast and Bowel Project; HR, hazard ratio; BCG, Bacillus Calmette-Guérin; GHI, Genomic Health, Inc.
|Variable||No. of Patients||HR||95% CI||P|
|Female v male||765||0.93||0.67 to 1.28||.653|
|Age, per 1 year increase||765||1.02||1.00 to 1.03||.010|
|Age ≥ 60 v < 60 years||765||1.49||0.98 to 2.26||.054|
|Location of tumor||765||.609|
|Right v left||1.25||0.80 to 1.95|
|Rectosigmoid v left||1.15||0.71 to 1.85|
|No. of nodes examined|
|< 12 v ≥ 12||765||1.66||1.17 to 2.36||.006|
|No. of positive nodes||765||< .001|
|1-3 v 0||2.03||1.42 to 2.91|
|≥ 4 v 0||4.22||2.67 to 6.67|
|Nodal involvement status||765||< .001|
|0 of < 12 examined v 0 of ≥ 12 examined||1.50||0.90 to 2.50|
|1-3 v 0 of ≥ 12 examined||2.26||1.53 to 3.33|
|≥ 4 v0 of ≥ 12 examined||4.69||2.89 to 7.60)|
|Stage III v II||765||2.43||1.76 to 3.36||< .001|
|GHI tumor grade high v Low||765||0.87||0.61 to 1.26||.460|
|CC tumor grade high vLow||761||1.28||0.89 to 1.86||.195|
|Segmental/anterior resection v colectomy/hemicolectomy||0.94||0.65 to 1.36|
|Other/unknown v colectomy/hemicolectomy||0.94||0.41 to 2.15|
|Mucinous tumor||765||0.64||0.40 to 1.02||.049|
|1986-1990 v 1981-1985||0.90||0.59 to 1.38|
|1991-1995 v 1981-1985||0.56||0.34 to 0.94|
|1996-2000 v 1981-1985||0.88||0.57 to 1.35|
|T stage T4 v T1, T2, and T3||765||1.74||1.22 to 2.48||.003|
|Hollandes v formalin||1.53||0.91 to 2.59|
|Zenkers or Bouins v formalin||2.06||1.06 to 4.02|
|MMR deficient v proficient||712||0.75||0.45 to 1.27||.266|
Abbreviations: RFI, recurrence-free interval; CC, Cleveland Clinic; HR, hazard ratio; GHI, Genomic Health, Inc; MMR, mismatch repair.
|Variable||No. of Patients||HR||95% CI||P|
|Age ≥ 60 v < 60 years||308||0.76||0.50 to 1.15||.191|
|Age, per 1 year increase||308||0.99||0.97 to 1.01||.240|
|Female v male||308||0.81||0.53 to 1.23||.322|
|Black v white||0.65||0.28 to 1.49|
|Other v white||1.26||0.51 to 3.12|
|Location of tumor||308||.972|
|Right v left||1.13||0.65 to 1.97|
|Rectosigmoid v left||1.11||0.63 to 1.95|
|Multiple/unknown v left||0.95||0.22 to 4.09|
|Segmental/anterior resection v colectomy/hemicolectomy||1.13||0.73 to 1.75|
|Other/unknown v colectomy/hemicolectomy||0.96||0.38 to 2.40|
|Tumor grade by GHI|
|First reading||308||1.37||0.81 to 2.32||.260|
|Second reading||308||1.04||0.64 to 1.68||.884|
|No. of nodes examined||308||.245|
|< 12 v ≥ 12||1.37||0.90 to 2.10|
|Unknown v ≥ 12||1.97||0.61 to 6.39|
|Tumor stage III v II||308||1.44||0.94 to 2.21||.089|
|No. of positive nodes||307||.002|
|1-3 v 0||1.06||0.64 to 1.73|
|≥ 4 v 0||2.43||1.47 to 4.04|
|Nodes involvement||304||< .001|
|0 of < 12 examined v 0 of ≥ 12 examined||2.27||1.11 to 4.66|
|1-3 v 0 of ≥ 12 examined||1.66||0.83 to 3.32|
|≥ 4 v 0 of ≥ 12 examined||3.83||1.90 to 7.72|
|Mucinous tumor||308||1.53||0.77 to 3.04||.255|
|Surgery date, quarters||308||1.00||0.88 to 1.13||.975|
|Surgery year||308||0.98||0.63 to 1.53||.924|
Abbreviations: RFI, recurrence-free interval; NSABP, National Surgical Adjuvant Breast and Bowel Project; HR, hazard ratio; GHI, Genomic Health, Inc.
|Variable||No. of Patients||HR||95% CI||P|
|Female v male||508||1.14||0.80 to 1.61||.463|
|Age, per 1 year increase||508||1.00||0.99 to 1.02||.632|
|Age ≥ 60 v < 60 years||508||1.13||0.79 to 1.62||.485|
|Black v white||0.98||0.53 to 1.82|
|Other/unknown v white||0.53||0.17 to 1.68|
|Location of tumor||508||.805|
|Right v left||1.10||0.70 to 1.73|
|Rectosigmoid v left||0.95||0.59 to 1.53|
|Multiple/unknown v left||0.55||0.08 to 4.07|
|No. of nodes examined|
|< 12 v ≥ 12||507||1.09||0.77 to 1.30||.632|
|Tumor stage III v II||508||2.84||1.91 to 4.22||< .001|
|T1 v T3||1.51||0.37 to 6.11|
|T2 v T3||0.83||0.41 to 1.70|
|No. of positive nodes||508||< .001|
|1-3 v 0||2.19||1.41 to 3.38|
|≥ 4 v 0||4.49||2.84 to 7.09|
|0 of < 12 examined v 0 of ≥ 12 examined||1.70||0.84 to 3.41|
|1-3 v 0 of ≥ 12 examined||2.91||1.59 to 5.35|
|≥ 4 v 0 of ≥ 12 examined||5.98||3.21 to 11.1|
|GHI tumor grade|
|First reading||508||0.95||0.63 to 1.42||.794|
|Second reading||508||1.08||0.75 to 1.56||.668|
|Segmental/anterior resection v resection v colectomy/hemicolectomy||0.89||0.61 to 1.28|
|Other/unknown v colectomy/hemicolectomy||1.80||0.66 to 4.92|
|Mucinous tumor||508||0.71||0.44 to 1.14||.144|
|Surgery date, quarters||508||0.97||0.90 to 1.05||.427|
|Surgery year||508||0.88||0.66 to 1.17||.367|
Abbreviations: RFI, recurrence-free interval; NSABP, National Surgical Adjuvant Breast and Bowel Project; HR, hazard ratio; GHI, Genomic Health, Inc.
See accompanying editorial on page 3904
Supported by Public Health Service Grants No. U10-CA-37377, U10-CA-69974, U10-CA-12027, and U10-CA-69651 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services, and by Genomic Health, Inc.
Presented in part at the 42nd Annual Meeting of the American Society of Clinical Oncology, June 2-6, 2006, Atlanta, GA, and at the Annual Gastrointestinal Cancers Symposium of the American Society of Clinical Oncology, January 25-27, 2008, Orlando, FL.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
Clinical trial information can be found for the following: NCT00427570.
Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
Employment or Leadership Position: Kim M. Clark-Langone, Genomic Health, Inc (C); Margarita Lopatin, Genomic Health, Inc (C); Drew Watson, Genomic Health, Inc (C); Frederick L. Baehner, Genomic Health, Inc (C); Steven Shak, Genomic Health, Inc (C); Joffre Baker, Genomic Health, Inc (C); J. Wayne Cowens, Genomic Health, Inc (C) Consultant or Advisory Role: Michael J. O'Connell, Genomic Health, Inc (U) Stock Ownership: Kim M. Clark-Langone, Genomic Health, Inc; Margarita Lopatin, Genomic Health, Inc; Drew Watson, Genomic Health, Inc; Steven Shak, Genomic Health, Inc; Joffre Baker, Genomic Health, Inc; J. Wayne Cowens, Genomic Health, Inc Honoraria: Greg Yothers, Genomic Health, Inc Research Funding: None Expert Testimony: None Other Remuneration: None
Conception and design: Michael J. O'Connell, Ian Lavery, Soonmyung Paik, Margarita Lopatin, Frederick L. Baehner, Steven Shak, Joffre Baker, J. Wayne Cowens, Norman Wolmark
Financial support: Steven Shak
Administrative support: Soonmyung Paik, J. Wayne Cowens, Norman Wolmark
Provision of study materials or patients: Ian Lavery, Soonmyung Paik
Collection and assembly of data: Ian Lavery, Greg Yothers, Soonmyung Paik, Kim M. Clark-Langone, Drew Watson, Frederick L. Baehner, Steven Shak, J. Wayne Cowens
Data analysis and interpretation: Michael J. O'Connell, Ian Lavery, Greg Yothers, Soonmyung Paik, Kim M. Clark-Langone, Margarita Lopatin, Drew Watson, Steven Shak, Joffre Baker, J. Wayne Cowens
Manuscript writing: Michael J. O'Connell, Greg Yothers, Soonmyung Paik, Kim M. Clark-Langone, Margarita Lopatin, Frederick L. Baehner, Steven Shak, Joffre Baker, J. Wayne Cowens
Final approval of manuscript: Michael J. O'Connell, Ian Lavery, Greg Yothers, Soonmyung Paik, Kim M. Clark-Langone, Margarita Lopatin, Drew Watson, Frederick L. Baehner, Steven Shak, Joffre Baker, J. Wayne Cowens, Norman Wolmark