Diagnostic accuracy of serum-carcinoembryonic antigen in recurrent colorectal cancer: a receiver operating...

Serial measurements of carcinoembryonic antigen (CEA) are frequently used in the follow-up after colorectal cancer (CRC), but its usefulness remains debatable. Choosing the appropriate cut-off point is crucial to the diagnostic accuracy (DA) of continuous test variables. Receiver operating characteristic curve (ROC) analysis is the appropriate statistical method for this purpose, but has not been applied in previous studies.

One hundred ninety-four consecutive patients surgically treated with curative intent for CRC between July 1996 and June 1999 had systematic follow-up for five years. Follow-up included imaging, coloscopy and serial CEA measurements. Complete data including CEA measurements were available from 153 patients. ROC analysis of CEA was done with regard to detection of recurrent disease.

Depending on the chosen cut-off value of CEA, DA varied widely within the normal range (CEA 4 U/ml provided the highest sensitivity (0.78) and specificity (0.91), compared to a sensitivity and specificity at the upper normal range (CEA = 10 U/ml) of 0.51 and 0.99, respectively. Thirty-three patients (24%) developed recurrence. Among 11 (5%) asymptomatic patients diagnosed by elevated CEA levels, only two patients (1.5%) were amenable to secondary curative surgery. A threefold increase of CEA in an individual patient had the same DA as the best cut-off value (>4 U/ml).

Diagnostic accuracy of CEA in follow-up after curative surgery for CRC is influenced by the chosen cut-off value. A threefold increase of CEA may indicate recurrent disease. The value of serial measurement of CEA was limited.