What carcinoembryonic antigen level should trigger further investigation during colorectal cancer follow-up? A systematic review and secondary analysis of a randomised controlled trial

Background Following primary surgical and adjuvant treatment for colorectal cancer, many patients are routinely followed up with blood carcinoembryonic antigen (CEA) testing. Objective To determine how the CEA test result should be interpreted to inform the decision to undertake further investigation to detect treatable recurrences. Design Two studies were conducted: (1) a Cochrane review of existing studies describing the diagnostic accuracy of blood CEA testing for detecting colorectal recurrence; and (2) a secondary analysis of data from the two arms of the FACS (Follow-up After Colorectal Surgery) trial in which CEA testing was carried out. Setting and participants The secondary analysis was based on data from 582 patients recruited into the FACS trial between 2003 and 2009 from 39 NHS hospitals in England with access to high-volume services offering surgical treatment of metastatic recurrence and followed up for 5 years. CEA testing was undertaken in general practice. Results In the systematic review we identified 52 studies for meta-analysis, including in aggregate 9717 participants (median study sample size 139, interquartile range 72–247). Pooled sensitivity at the most commonly recommended threshold in national guidelines of 5 µg/l was 71% [95% confidence interval (CI) 64% to 76%] and specificity was 88% (95% CI 84% to 92%). In the secondary analysis of FACS data, the diagnostic accuracy of a single CEA test was less than was suggested by the review [area under the receiver operating characteristic curve (AUC) 0.74, 95% CI 0.68 to 0.80]. At the commonly recommended threshold of 5 µg/l, sensitivity was estimated as 50.0% (95% CI 40.1% to 59.9%) and lead time as about 3 months. About four in 10 patients without a recurrence will have at least one false alarm and six out of 10 tests will be false alarms (some patients will have multiple false alarms, particularly smokers). Making decisions to further investigate based on the trend in serial CEA measurements is better (AUC for positive trend 0.85, 95% CI 0.78 to 0.91), but to maintain approximately 70% sensitivity with 90% specificity it is necessary to increase the frequency of testing in year 1 and to apply a reducing threshold for investigation as measurements accrue. Limitations The reference standards were imperfect and the main analysis was subject to work-up bias and had limited statistical precision and no external validation. Conclusions The results suggest that (1) CEA testing should not be used alone as a triage test; (2) in year 1, testing frequency should be increased (to monthly for 3 months and then every 2 months); (3) the threshold for investigating a single test result should be raised to 10 µg/l; (4) after the second CEA test, decisions to investigate further should be made on the basis of the trend in CEA levels; (5) the optimal threshold for investigating the CEA trend falls over time; and (6) continuing smokers should not be monitored with CEA testing. Further research is needed to explore the operational feasibility of monitoring the trend in CEA levels and to externally validate the proposed thresholds for further investigation. Study registration This study is registered as PROSPERO CRD42015019327 and Current Controlled Trials ISRCTN93652154. Funding The main FACS trial and this substudy were funded by the National Institute for Health Research Health Technology Assessment programme.