Development and validation of reassigned CEA, CYFRA21-1 and NSE-based models for lung cancer diagnosis and prognosis prediction

Background: The majority of lung cancer(LC) patients are diagnosed at advanced stage with a poor prognosis. However, there is still no ideal diagnostic and prognostic prediction model for lung cancer. Methods: Data of CEA, CYFRA21-1 and NSE test of patients with LC and benign lung diseases (BLDs) or healthy people from Physical Examination Center was collected. Samples were divided into three data sets as needed. Reassign three kinds of tumor markers (TMs) according to their distribution characteristics in different populations. Diagnostic and prognostic models were thus established, and independent validation was conducted with other data sets. Results: The diagnostic prediction model showed good discrimination ability: the area under the receiver operating characteristic curve (AUC) differentiated LC from healthy people and BLDs (diagnosed within 2 months), being 0.88 and 0.84 respectively. Meanwhile, the prognostic prediction model did great in prediction: AUC in training data set and test data set were 0.85 and 0.8 respectively. Conclusion: Reassigned CEA, CYFRA21-1 and NSE can effectively predict the diagnosis and prognosis of LC. Compared with the same TMs that were considered individually, this diagnostic prediction model can identify high-risk population for LC screening more accurately. The prognostic prediction model could be helpful in making more scientific treatment and follow-up plans for patients.