In Reply.—Dr Bodner correctly points out that we have not fully explained how wide lymphadenectomy favorably influenced survival in our study of potentially curative resection for gastric carcinoma. We must, however, recognize that any retrospective comparative analysis may be biased by nonrandom clustering of patient characteristics due to the way patients were selected for treatment or how they were selected for comparison by the investigator. If the bias is small, and other strong evidence supports our hypothesis, we may perform the analysis and choose to accept the limitations of its conclusions.
In this regard, the possible bias of having the stage N0, N1 groups "erroneously contain[ing] some patients with N2 tumors," as referred to by Dr Bodner, was quite small. As Table 2 of our article indicated, only 15% (30/202) of the listed patients with NON1-N2 tumors had N2 tumors. To examine the possible effect of such a bias,