[Serum expression and combined diagnostic value of LBP, ITIH3, and CXCL7 in colorectal cancer].

Objective: To screen for novel candidate serum biomarkers of colorectal cancer (CRC) using serum proteomics, and to evaluate the value of combining them with conventional tumor markers to construct a multivariable diagnostic model for the non-invasive auxiliary diagnosis of CRC. Methods: This study adopted a retrospective case-control design and consecutively enrolled patients with colorectal diseases and healthy individuals undergoing routine health examination at the Department of Clinical Laboratory, Sun Yat-sen University Cancer Center, between August 2023 and August 2024. Candidate protein markers were screened by four-dimensional data-independent acquisition (4D-DIA) mass spectrometry and subsequently validated by enzyme-linked immunosorbent assay (ELISA) in 623 participants (267 with CRC, 256 healthy controls, and 100 with colorectal polyps). On this basis, a machine learning diagnostic model incorporating multiple variables was constructed, and a nomogram was generated. Between-group differences were compared using non-parametric methods such as the Kruskal-Wallis test, and diagnostic performance was evaluated and compared using the area under the ROC curve (AUC) in combination with the DeLong test. Results: A total of 140 differentially expressed proteins were identified by 4D-DIA proteomic analysis, and stepwise screening selected three candidate markers-lipopolysaccharide-binding protein (LBP), inter-alpha-trypsin inhibitor heavy chain 3 (ITIH3), and C-X-C motif chemokine ligand 7 (CXCL7)-for the large-sample validation cohort. In the validation cohort of 623 participants, the serum concentrations of LBP, ITIH3, and CXCL7 in the CRC group were all significantly higher than those in the healthy control and colorectal polyp groups (Kruskal-Wallis H values of 161.77, 232.62, and 126.89, respectively; all P<0.001), and their AUCs for discriminating CRC from all non-tumor participants were 0.797, 0.857, and 0.760, respectively, all superior to CEA (AUC=0.711) and CA19-9 (AUC=0.606). The three-protein combination of ITIH3, CXCL7, and LBP further improved performance, achieving an AUC of 0.905 for discriminating CRC from all non-tumor participants; in stage I CRC, the AUCs of LBP, ITIH3, and CXCL7 for discriminating CRC from all non-tumor participants were 0.701, 0.792, and 0.789, respectively, all superior to CEA (AUC=0.594) and CA19-9 (AUC=0.537), and the ITIH3+CXCL7 combination further reached an AUC of 0.853. In the independent test set, the seven-variable logistic regression model achieved an AUC of 0.919 (95%CI: 0.872-0.955) for discriminating all-stage CRC from all non-tumor participants, significantly outperforming the CEA+CA19-9 reference model (AUC=0.704; DeLong P<0.001); for stage I CRC, the model yielded AUCs of 0.879, 0.882, and 0.870 for distinguishing stage I CRC from all non-tumor participants, healthy controls, and benign diseases, respectively, all significantly superior to the CEA+CA19-9 reference model (corresponding AUCs of 0.651, 0.664, and 0.645; DeLong P=0.003, 0.003, and 0.012, respectively). The model was well calibrated (Hosmer-Lemeshow test χ²=6.31, P=0.613); decision curve analysis showed that its net benefit exceeded that of the CEA+CA19-9 model across threshold probabilities of 0.05-0.50, with this advantage being more pronounced in early-stage CRC. A nomogram derived from the model showed that the total points contributed by ITIH3, LBP, and CXCL7 (90.8) exceeded those of CEA and CA19-9 (1.3). Conclusion: The combined detection of LBP, ITIH3, and CXCL7 together with conventional markers can improve auxiliary diagnostic performance for CRC, particularly early-stage CRC.