Development and validation of a tumor marker-based model for the prediction of lung cancer: an analysis of a multicenter retrospective study in Shanghai, China

Background: The incidence and mortality rates of cancer are the highest globally. Developing novel methodologies that precisely, safely, and economically differentiate between benign and malignant lung conditions holds immense clinical importance. This research seeks to construct a predictive model utilizing a combination of diverse biomarkers to effectively discriminate between benign and malignant lung diseases.

Methods: This retrospective study included patients admitted to the two general hospitals in Shanghai from 2014 to 2015. This study was developed using five tumor markers: carcinoembryonic antigen (CEA), carbohydrate antigen 199 (CA199), cytokeratin fragment 21-1 (CA211), squamous cell carcinoma antigen (SCC), and neuron specific enolase (NSE). The entire sample was divided into two groups according to the hospital: 1033 cases were included in the development cohort and 300 cases in the validation cohort. Logistic regression analysis was used for univariate analysis to explore individual correlations between each selected clinical variable and lung cancer diagnostic outcome. Diagnostic prediction models were constructed and validated based on independent prognostic factors identified using multifactorial analysis. A nomogram was created using these tumor markers (age and sex were additionally included) and validated using the concordance index and calibration curves. Clinical prediction models were evaluated using decision curve analysis.

Results: Fully adjusted multivariate analysis showed that the risk of lung cancer was 2.38 times higher in men than in women. CEA positivity was associated with an 13.41-fold increased risk in lung cancer. The area under the curve (AUC) values for the development cohort and validation cohort models were 0.907 and 0.954, respectively. In the established nomogram, the AUC for the receiver operating characteristic curve was 0.907 (95% CI, 0.889-0.925). The validation model confirmed the strong discriminative power of the nomogram (AUC = 0.954). The described calibration curves demonstrated good fit predictions and observation probabilities. In addition, decision curve analysis concluded that the newly established nomogram has important implications for clinical decision making.

Conclusions: Combined prediction models based on CEA, CA199, CA211, SCC, and NSE biomarkers could significantly the differentiation between benign and malignant lung diseases, thus facilitating better clinical decision making.