To the Editor.—We would like to comment on the recent article by Shiu and coworkers.1
Multivariate analysis is a powerful tool in the evaluation of retrospective data in a field where prospective studies have not been undertaken, and this article should therefore be welcomed. We have, however, doubts whether total gastrectomy per se has an adverse influence on survival, as the authors contend. We suspect that significant variables not considered in this series are masked by the risk factor "total gastrectomy" in the Cox model. The inclusion of "tumor size" in the multivariate analysis is not sufficient proof for the independence of "total gastrectomy."
In the final Cox model (Table 41), no variables show up that represent the extent of the tumor as "tumor size" or TNM stage. These variables correlate highly with "total gastrectomy," rendering the question unsolved whether total gastrectomy or rather the more advanced