The rate of false-positive diagnosis of colorectal liver metastases in patients undergoing resection with the development of a novel, externally validated risk score

Background: Diagnostic error in patients undergoing resection of colorectal liver metastases (CRLM) is unusual but exposes patients to unnecessary risks associated with treatment. The primary aim of this study was to determine the rate of and risk factors for a false-positive diagnosis of colorectal liver metastases in patients undergoing hepatic resection. The secondary aim was to develop and validate a risk score to predict a false-positive diagnosis. Methods: Patients were identified from prospectively maintained databases. Patients who underwent a first liver resection for presumed colorectal liver metastases were divided into 2 groups: CRLMPOS (colorectal liver metastases present on histology or appearance of complete pathologic response to preoperative chemotherapy) and CRLMNEG (all others). Univariable analysis and multivariable binary logistic regression were used to identify risk factors for CRLMNEG. Risk scores were developed for CRLMNEG both with and without the use of preoperative carcinoembryonic antigen and were validated on an external cohort. Results: 3.1% of patients in both test and validation cohorts were CRLMNEG (39/1,252 and 59/1,900, respectively). CRLMNEG patients had fewer (P =.006) and smaller lesions (P <.001) with lower serum levels of carcinoembryonic antigen (P <.001), T (P =.031) and N (P <.001) and a lower Dukes’ stage of the primary (P <.001). The risk score performed well (area under the receiver operating characteristic curve 0.869; standard error = 0.030; P <.001) with reasonable performance on validation (area under receiver operating characteristic curve 0.743; standard error = 0.058; P <.001]). Conclusion: A false-positive diagnosis of colorectal liver metastases affected the same proportion of patients in 2 unrelated cohorts. This study identified risk factors for false-positive diagnosis with development of a novel risk score supported by external validation.