Predictive Value of Defective Mismatch Repair in Adjuvant Colon Cancer

TO THE EDITOR: Sargent et al add to our knowledge about the predictive value of mismatch repair (MMR) status in patients with stage II and III colon cancer. Ng and Schrag and Kerr and Midgley offer insightful commentaries on their article. Importantly, as Kerr and Midgley imply, the study performed by Sargent et al must be viewed foremost as a confirmatory study seeking to determine the relevance of the findings of Ribic et al. As such, the primary outcome reported by Sargent et al regarding the value of MMR status as a predictive marker is the interaction test between MMR status and treatment efficacy from the multivariable Cox model for disease-free survival (DFS) in the 457 patients from their international collaboration. This test was not significant (P .18). This means that the data are insufficient to show that MMR status is predictive of response to chemotherapy. This should not be confused with their analyses showing that patients with proficient MMR showed significant DFS benefit from treatment (hazard ratio, 0.67; 95% CI, 0.37 to 0.83) whereas those with deficient MMR did not (hazard ratio, 1.39; 95% CI, 0.46 to 4.15). The paradox is resolved by noting the overlapping CI. The number of patients with deficient MMR was small (n 70), estimates of survival were imprecise, and CIs were wide: it could not be proven that treatment was beneficial in patients with deficient MMR, but the data were insufficient to exclude a major benefit (a possible 54% improvement in DFS). Although the pooled analysis offered by Sargent et al may be interesting, it cannot serve the same function as a confirmatory study. Including the initial study patients in the expanded pool results in ascertainment bias.