The predictive value of serum tumor markers for EGFR mutation in non-small cell lung cancer patients with non-stage IA

Objective: The predictive value of serum tumor markers (STMs) in assessing epidermal growth factor receptor (EGFR) mutations among patients with non-small cell lung cancer (NSCLC), particularly those with non-stage IA, remains poorly understood. The objective of this study is to construct a predictive model comprising STMs and additional clinical characteristics, aiming to achieve precise prediction of EGFR mutations through noninvasive means. Materials and methods: We retrospectively collected 6711 NSCLC patients who underwent EGFR gene testing. Ultimately, 3221 stage IA patients and 1442 non-stage IA patients were analyzed to evaluate the potential predictive value of several clinical characteristics and STMs for EGFR mutations. Results: EGFR mutations were detected in 3866 patients (57.9 %) of all NSCLC patients. None of the STMs emerged as significant predictor for predicting EGFR mutations in stage IA patients. Patients with non-stage IA were divided into the study group (n = 1043) and validation group (n = 399). In the study group, univariate analysis revealed significant associations between EGFR mutations and the STMs (carcinoembryonic antigen (CEA), squamous cell carcinoma antigen (SCC), and cytokeratin-19 fragment (CYFRA21-1)). The nomogram incorporating CEA, CYFRA 21-1, pathology, gender, and smoking history for predicting EGFR mutations with non-stage IA was constructed using the results of multivariate analysis. The area under the curve (AUC = 0.780) and decision curve analysis demonstrated favorable predictive performance and clinical utility of nomogram. Additionally, the Random Forest model also demonstrated the highest average C-index of 0.793 among the eight machine learning algorithms, showcasing superior predictive efficiency. Conclusion: CYFRA21-1 and CEA have been identified as crucial factors for predicting EGFR mutations in non-stage IA NSCLC patients. The nomogram and 8 machine learning models that combined STMs with other clinical factors could effectively predict the probability of EGFR mutations.