To the Editor: Potti et al. (Aug. 10 issue)1 apply a metagene model to the profiling of non–small-cell lung cancer (NSCLC) and demonstrate superior performance in predicting tumor recurrence and survival, as compared with a clinical model. We believe that the impressively contrasting results could be partially due to the incompleteness of the clinical model the authors used. Classifying NSCLC into squamous-cell carcinoma and adenocarcinoma has not been predictive for prognosis in general. However, subtypes of adenocarcinoma — bronchioloalveolar carcinoma and mixed adenocarcinoma with a bronchioloalveolar component, which account for approximately 20% of cases of early-stage NSCLC — have a much better prognosis than do other subtypes.2 Potti et al. did not consider these adenocarcinoma subtypes. In addition, the literature3 and our recent work demonstrate that the histologic grade is a significant predictor of both tumor recurrence and survival,4 and there is a high correlation between histologic features and gene-expression profiles.5 Our work also shows that incorporating the adenocarcinoma subtype and histologic grade into clinical models would provide a prediction very similar to that of a well-validated, 50-gene panel for survival.