Question special

Among the prognostic and predictive CDX2 questions addressed by this paper, the most interesting is certainly the potential predictive effect of CDX2 for treatment determinations in the stage II setting. As others have already mentioned, one strong limitation of this analysis is the lack of randomized data sources for evaluating the interaction. Given the data at hand however, another limitation is the small sample size and corresponding very low number of events that appears to be driving the marginally significant (p = 0.02) treatment-by-CDX2 interaction in stage II (shown in the middle and right-hand panels of Figure 5A). Specifically, from the Kaplan-Meier plots, it appears that only a small number of DFS events occurred in either treatment group (2 on chemotherapy and approximately 9 without chemotherapy) among the 48 CDX2-negative stage II patients. Although 2 vs. 9 produces a very large separation in DFS curves in the CDX2- subgroup, these are very small raw numbers, and the p-value of most interest could have easily shifted with different observations for just 1 or 2 patients. External validation will be very important here.