Abstract
OBJECTIVE: To improve the diagnostic capacity of serum biomarkers for colorectal cancer (CRC), we introduced three novel indicators, namely, the C-X-C motif chemokine ligand 5 (CXCL5), stanniocalcin 2 (STC2), and chitinase 3 like 1 (CHI3L1) and assessed their performances in the detection of CRC. METHODS: A total of 887 serum samples (153 health, 342 polyps, and 392 CRCs) were collected. Concentrations of CXCL5, STC2, and CHI3L1 were measured by the ELISA. CEA and CA199 were determined by electrochemiluminescence. Binary logistic regression was used to build the combination model. ROC analysis was used to evaluate the performance of biomarkers alone or in combination. RESULTS: Model_2 that based on CXCL5, STC2, and CHI3L1 was the best approach in discriminating CRC from non-CRC controls (AUC, 0.943 (0.922-0.960); sensitivity, 0.848; specificity, 0.917; and accuracy, 0.887 in the training cohort and 0.959 (95% CI 0.927-0.980), 0.878, 0.917, and 0.900 in the testing cohort, respectively). In the detection of early CRC, Model_2 revealed AUC, sensitivity, specificity, and accuracy of 0.925 (0.897-0.947), 0.793, 0.917, and 0.886 in the training cohort and those of 0.926 (0.979-0.959), 0.786, 0.931, and 0.898 in the testing cohort. Furthermore, Model_2 exhibited an excellent diagnostic performance in CEA-negative cases (0.938 (0.913-0.957), 0.826, 0.917, and 0.888 in the training cohort and 0.961 (0.925-0.983), 0.887, 0.931, and 0.918 in the testing cohort). As used alone, STC2 achieved the capacities that is second only to that of Model_2 (0.866 (0.837-0.892), 0.859, 0.842, and 0.853 in the training cohort and 0.887 (0.842-0.923), 0.922, 0.799, and 0.853 in the testing cohort). STC2 alone also yielded acceptable results for early CRC detection (0.815 (0.776-0.849), 0.767, 0.849, and 0.829 in the training cohort and 0.870 (0.812-0.914), 0.952, 0.799, and 0.833 in the testing cohort). Moreover, STC2 maintained diagnostic accuracy for CRC patients with negative CEA (0.874 (0.842-0.901), 0.862, 0.849, and 0.853 in the training cohort and 0.898 (0.848-0.936), 0.930, 0.801, and 0.842 in the testing cohort). In comparison, the performances of the CEA and CA199 based Model_1 were far from satisfactory, especially in early cases (0.767 (0.726-0.805), 0.491, 0.863, and 0.771 in the training cohort and 0.817 (0.754-0.870), 0.476, 0.889, and 0.796 in the testing cohort). CONCLUSIONS: STC2 was a promising serum biomarker for CRC diagnosis either used alone or in combination with CXCL5 and CHI3L1.