TO THEEDITOR: We write in response to the articles of Hirsch et al 1 and Takano et al 2 and the editorial by Johnson and Janne 3 concerning predictive epidermal growth factor receptor (EGFR) markers for sensitivity to tyrosine kinase inhibition in patients with non–small-cell lung cancer. These and other authors have concluded from their analyses of gene copy number and mutational status that the presence of classical exon 19 deletions or exon 21 L858R point mutations and/or increased gene copy number are predictive of a higher response rate in patients with non–small-cell lung cancer. We agree that this is indeed true, although the exceptionally high response rates for mutation-positive patients (sometimes approaching 100%) reported from retrospective analyses of highly selected patients and, frequently, from a single institution are unlikely to be confirmed in larger, prospectively designed, multicenter trials in which patients are less homogeneous and strict response and validation criteria are applied. For example, the response rate for patients with mutations in the trial reported by Hirsch et al 1 was only 50%. 4 Similarly, in the BR.21 trial in which erlotinib was compared with placebo, 5 the response rate was 25% in patients with classical mutations compared with 9% both in the entire study and in patients with novel mutations. 6 Most retrospective analyses of phase II trials of EGFR inhibitors, including those published in the Journal of Clinical Oncology, 1,2 have also reported superior survival for patients with classical mutations and/or increased gene copy number. Furthermore, most authors have either directly or by inference suggested that the superior survival was a direct result of treatment with the EGFR inhibitor, thereby indicating a differential effect of therapy in the subgroup of patients with mutations or high copy number. This would mean that these factors were predictive markers of a greater survival benefit. We maintain that it is inappropriate to draw such conclusions from phase II studies that did not have untreated control arms. There is increasing evidence to suggest that the presence ofEGFR mutations is prognostic. Significantly longer survival has been reported recently for untreated mutation-positive patients from surgical series. 7 In the placebo arm of BR.21, patients with classical mutations demonstrated longer median survival time than patients with wild-type or novelEGFR variants (9.1 v 3.5 and 3.5 months, respectively). This suggests that classical EGFR mutations confer a favorable prognosis and that the superior survival reported for mutation-positive patients from single-arm studies may not have been a result of a greater benefit from the EGFR inhibitor in these patients. In fact, no significantly greater survival benefit was seen in classical or novel mutation-positive patients in BR.21 compared with the survival benefit seen in patients with wild-type EGFR (hazard ratio 0.65, 0.67, and 0.73, respectively). 6 Similar results were reported from a trial of chemotherapy with or without erlotinib in which the survival of patients with mutations was significantly longer than the survival of patients with wild-type tumors (P .001). However, there was no further survival benefit for mutation-positive patients when erlotinib was added to chemotherapy. 8 In contrast, the results of BR.21 gene copy analyses hint at a possible differential effect on survival from erlotinib in patients with high gene copy number compared with patients with low copy number (hazard ratio 0.44 and 0.85, respectively), although theP value for interaction was negative (P .1), and thus, these analyses must be considered exploratory. 6 Prognostic factors are patient and tumor factors that predict patient outcome (usually survival) and are independent of treatment administered. Predictive factors are clinical, cellular, and molecular markers that predict response of the tumor to treatment (either in terms of tumor shrinkage or a survival benefit from treatment). In essence, therefore, prognostic factors define the effects of tumor characteristics on the patient, whereas predictive factors define the effect of treatment on the tumor. Evidence is growing to suggest that EGFR mutations are favorable prognostic markers of survival and that they are predictive markers of response in terms of tumor shrinkage. There is no evidence yet to suggest that they are predictive of a differential effect of EGFR inhibitor therapy on survival. The subgroup of patients with mutations clearly is biologically distinct, but an interaction of this subset with treatment and survival has yet to be demonstrated.