A nomogram based on autoantibodies for noninvasive detection of AFP-negative hepatocellular carcinoma: a multicenter study.

Background: Diagnosing AFP-negative hepatocellular carcinoma (HCC) is challenging. Autoantibodies to tumor-associated antigens have been extensively investigated as serum biomarkers.

Methods: We employed serological proteome analysis and protein microarray to identify potential autoantibodies for HCC, followed by a two-center and two-independent-phase validation and evaluation using ELISA in patients with AFP-negative HCC (ANHCC). LASSO regression addressed multicollinearity among biomarkers. Four machine-learning methods developed diagnostic models for ANHCC. ROC analysis and various evaluation indicators were applied to assess the performance.

Results: Eight autoantibodies out of sixteen candidates, including Survivin, NPM1, GNAS, SRSF2, GNA11, PTCH1, GAPDH, and HSP90, were validated as superior biomarkers. The Logistic regression model was optimal for ANHCC, achieving an area under the ROC (AUC) of 0.883 in the training dataset and an AUC of 0.840 in the validation dataset. When tested on the entire HCC patient cohort, which included both ANHCC and AFP-positive patients (APHCC), with ANHCC accounting for 37.5%, the AUC reached 0.825, with a sensitivity of 66.4%, and a specificity of 84.2%. Combining this model with AFP improved efficacy, yielding an AUC of 0.945, an IDI of 23.1%, and an NRI of 21.1% compared to using AFP alone.

Conclusion: The Logistic regression model demonstrates superior diagnostic performance for ANHCC. Integrating this model with AFP enhances the entire HCC diagnosis.