In their review of Pulmonary Resection in Metastatic Colorectal Cancer, Pfannschmidt et al.1 consider carcinoma embryonic antigen (CEA) as an adverse feature in the prognosis of patients who have had pulmonary metastasectomy for colorectal cancer, a consistent finding in multivariable analyses of clinical case series.2,3 Kanemitsu et al.3 published a predictive model on the basis of 313 patients operated on between 1989 and 1998. Their five prognostic factors were primary histology, hilar or mediastinal lymph node involvement, number of metastases, preoperative CEA level, and extrathoracic disease. However, CEA was not statistically significant in the most recent Memorial Sloane Kettering report.4 They found that in patients younger than 65 years, female sex, apparent disease free interval (DFI) less than 1 year, and number of metastases more than three to be predictive of recurrence and went on to make an explicit recommendation: “medical management alone should be considered standard for patients who have both three or more pulmonary metastases and less than 1 year DFI.” This contradiction prompts a consideration of the role of CEA in the surveillance of patients with colorectal cancer and in the selection of patients for surgery and of the limitations of multivariate analysis as a basis for clinical decision making. First, to deal briefly with the statistical issue, the failure to find a “p value” does not mean that CEA would have had no bearing on outcome. If at the case selection stage, there was a clinical policy (explicit or unstated) to avoid operating on patients with high CEA, the range of CEA data will have been curtailed at the input stage; there is a point at which it cannot then emerge as a prognostic factor. Other factors will then come to the fore. It should also be remembered that there is a limit to the number of factors that will emerge as significant and an inevitable degree of variation in their apparent ranking. The contradiction merits attention for surgeons who have to make decisions about how to use CEA in their selection of patients.5,6 In the systematic review by Fiorentino et al.,7 approximately a third of 51 included publications reported the preoperative CEA in a way that could be used in the analysis (Figure 1). Approximately half of these patients in the operated series had an increased CEA in the earlier era falling to approximately a third more recently. This illustrates the point alluded to above; there may have been a conscious avoidance of metastasectomy for patients with high CEA. The knowledge provided in earlier reports might quite rightly have altered the practice, and hence the case mix, for later reports. When CEA and survival data were available, these showed (Figure 2) an evident and consistent relationship: patients with increased CEA have consistently worse survival. This suggests that the importance of a raised CEA as an adverse prognostic feature for survival should not be forgotten just because it now features less prominently in the input data set. In the course of developing the PulMiCC trial,8,9 we discovered that gastroenterologists and thoracic surgeons were at cross purposes in the use of CEA in patients with recurrence of colorectal cancer. While as thoracic surgeons we were interpreting CEA as an adverse feature,3 some gastroenterologists were using the CEA measurement as a surveillance tool: when CEA became increased, this prompted a computed tomography scan and possible referral to thoracic surgery if this revealed pulmonary metastases. The practice may have stemmed in part from a philosophy of care of colorectal cancer in which routine CEA testing was used to prompt “second look” surgery to deal with local recurrence10 and was at least temporally related to the growth in hepatic metastasectomy for colorectal cancer.11 CEA measurement was then shown to be not cost effective, because although it picked up asymptomatic patients earlier, it made no impact on survival.10,12,13 From the thoracic surgeons’ viewpoint, it would seem reasonable to obtain a CEA measurement, and if it is high, it should go in the balance pan adding weight against metastasectomy. It is critical that the statistical findings are used appropriately in surgical decision making. To rather stretch the point, but to illustrate the point, here is a wry Clinical Operational Research Unit, University College London, London, UK. Disclosure: The author declares no conflicts of interest. Address for correspondence: Tom Treasure, MD, Clinical Operational Research Unit, University College London, 4 Taviton Street, London WC1H 0BT, United Kingdom. E-mail: tom.treasure@gmail.com Copyright © 2010 by the International Association for the Study of Lung Cancer ISSN: 1556-0864/10/0506-0179