Preoperative carcinoembryonic antigen as an outcome predictor in colon cancer.

OBJECTIVE Several reports have shown that certain pre-operative CEA intervals can be predictive of long-term outcomes and have subsequently implied that preoperative CEA may be useful to assess the risk of recurrence or death as a continuous number for individual cases. This analysis assesses if this hypothesis is valid after correction for confounders. METHODS All colon cancer patients operated on at Massachusetts General Hospital from 2004 through 2011 were considered for retrospective review. Association between outcomes and preoperative CEA was measured in intervals and as a linear relationship. RESULTS Of the 1,071 patients operated for colon adenocarcinoma, 621 (57.9%) had a preoperative CEA drawn and were included in the analysis. In models using intervals, preoperative CEA did show association with (disease-free) survival, but this was shown to be chiefly a surrogate for metastatic presentation. In linear approaches adjusted for metastatic presentation, CEA loses all correlations with metastatic disease (P = 0.84), survival (P = 0.11), survival duration (P = 0.42) and disease-free interval (P = 0.94). CONCLUSIONS Extrapolating the predictive value of certain preoperative CEA intervals to a continuous approach for use in a case-for-case basis is unjustified. Preoperative CEA may be a useful risk estimator but has limited significance for predictions of long-term outcomes in individual cases.