Reply to L. Casadaban et al.

We presented univariable results according to the REMARK guidelines for associations between ColDx score and prognostic factors for recurrence-free interval (RFI; Appendix Table A1 [online only] in our article). As Casadaban et al point out, ColDx is associated with T-stage and lymphovascular invasion but not the number of nodes examined, perineural invasion, or tumor grade. It is not clear why such a relationship would be expected. The assay was designed to be independent from other known prognostic clinical factors and to add new prognostic information. As Casadaban et al suggest, we considered the subgroup of high-risk patients who we defined as exhibiting any one of the following clinical characteristics: obstruction or perforation (six patients), lymphovascular invasion (42 patients), fewer than 12 nodes sampled (176 patients), or microsatellite instability low or stable (283 patients; n 5 317; 80 RFI events). RFI was then compared between highrisk patients and low-risk patients as determined by ColDx score. Results were significant at P5 .05, with a hazard ratio of 1.62 (95%CI, 0.99 to 2.68). Thus, ColDx provides further discrimination in this higher-risk subgroup. The number of events was too small to make this comparison in the low-risk subgroup. In the parent trial, Alliance C9581, investigators sought to determine whether the use of edrecolomab—a relatively nontoxic adjuvant therapy—would demonstrate an overall survival benefit in a cohort of patients with resected, stage II colon cancer that excluded patients with high-risk factors. Patients were considered disease-free postsurgery. Thus, tumor response was not a study end point. Patient samples were obtained before treatment with edrecolomab. Overall survival and disease-free survival between treated and untreated patients were essentially equivalent (Fig 2A in our article). Nonetheless, under the case-cohort design in our validation study, we randomly selected patients stratified by assigned treatment and accounted for stratification in the analysis. Overall, toxicity was low. A maximum of grade 3 toxicity was reported for 242 (29.4%) of 823 participants who reported adverse events with edrecolomab treatment, and 48 patients (5.8%) experienced a maximum grade 4 toxicity. No individual adverse event was reported in . 5% of patients, the most prevalent of which was diarrhea. One death occurred within 30 days of completing edrecolomab therapy and was not attributed to treatment. This validation study used the same primary end point on which the gene signature was developed. Among patients who were studied in the Alliance C9581 trial, we found that it is important to distinguish between disease-related death and other causes of death in this low-risk, older patient population with stage II disease. We found large differences in outcome by sex and age for all-cause mortality that were primarily caused by association of these factors with death as a result of other causes. Including deaths as a result of other causes as an event may unduly bias results. Regarding sample insufficiency, in clinical testing, the quality control fail rate that was observed for the study is not unusual considering the average age of the formalin-fixed, paraffinembedded tissue used in the validation study (average age, 13.2 years). This limitation is acknowledged in the manuscript. In addition, average quality control fail rate within fresh formalinfixed, paraffin-embedded tissue is 5%. We stated the reason for the different prognostic score cut points in our article, which was “migration of the ColDx assay from the Affymetrix GeneChip System 3000 7G scanner to the Affymetrix microarray platform GeneChip System 3000Dx v.2.” With respect to the association with lymphocyte proliferation and activation of biologic functions with recurrence-free survival in colorectal liver metastases, the validation study was performed within primary tissuematerial. Biologic signaling within metastatic tissue is inherently different from that found within primary tissue material. That said, the most significant molecular pathways measured by the ColDx assay are detailed by Kennedy et al, among which are TGF-b and chemokine signaling, and both are associated with lymphocyte proliferation and recruitment. It is not unusual that two assays, such as the 12-gene recurrence score and ColDx, have good discrimination and calibration but do not agree with one another in individual probability predictions. It is more relevant to determine which assay is better calibrated and has better discrimination—that is, which assay is better at generating estimates that are closer to observed values. We agree with Casadaban et al that further studies are needed to demonstrate the ability of the gene expression signature to predict treatment benefit. Despite its limitations, our study was prospectively planned and used specimens and clinical data from a cohesive, well-conducted clinical trial. The results demonstrate the additive prognostic value of the measure.