We thank Sclafani et al for their comments on our study and would like to address some of the questions raised. First, we agree that the value of multimodality strategies has indisputably been proven in the treatment of resectable gastric cancer, and one of the challenges in the coming years will be to determine which (neo)adjuvant approach should be used in gastric cancer. In fact, our currently accruing CRITICS (ChemoRadiotherapy after Induction ChemoTherapy in Cancer of the Stomach) trial is addressing this question by randomly assigning patients to receive either perioperative chemotherapy or preoperative chemotherapy combined with postoperative chemoradiotherapy (CRT). However, pursuing a solution to this important problem does not justify neglecting other relevant questions. Obviously, it was never our goal to use a retrospective analysis to question the main conclusion of a randomized phase III trial (in this case, the beneficial effect of postoperative CRT in gastric cancer). However, since publication of the Intergroup 0116 results, there has been debate on combining extended lymphadenectomy with postoperative CRT. In a separate report, the investigators of the Intergroup 0116 trial themselves concluded that D-level designation failed to significantly correlate with survival, but that the power to detect such interaction was low, which emphasizes the relevance of our study. Performing retrospective analyses always introduces selection bias, and study design should take this into account. Because of baseline differences between the two groups in our study, no univariate survival analyses were performed. Rather, patients were included in a Cox regression to correct for these baseline differences. However, differences in time of accrual were not overcome by this regression method, and we agree that this should be taken into account when interpreting the results. The more intensive follow-up in the CRT group underscores the difference in local recurrence, with fewer recurrences in the more intensively monitored group. Furthermore, Sclafani et al draw additional conclusions on the basis of two different graphs. However, in Cox models, comparing curves between different graphs is highly unreliable and should be avoided, because each Cox regression only internally corrects for covariates. A recent update of our retrospective analysis (data unpublished)—with a median follow-up of 27 months as compared with the original 19 months— confirms the results regarding patterns of recurrence, whereas in the D1 subgroup, there is now a significant difference in overall survival favoring the CRT group (hazard ratio, 1.68; P .04). No significant difference was found in the D2 subgroup. When performing clinical research, every investigator would prefer a phase III randomized trial over a retrospective analysis or perfect matching conditions when performing a retrospective study. However, in hypothesis-generating clinical research, comparison of historical phase II and III data can play an important role, particularly because the final results of our CRITICS study will not be available for at least 5 more years.