Association between preoperative carcinoembryonic antigen and occult lymph node metastasis in clinical stage I non-small cell lung cancer: a meta-analysis

Preoperative imaging has limited sensitivity for nodal involvement, and the strength and robustness of the association between serum carcinoembryonic antigen (CEA) and occult lymph node metastasis (OLM) remain incompletely quantified. We conducted a meta-analysis to evaluate this association in clinical stage I non-small cell lung cancer (NSCLC). The PubMed, Embase, Web of Science and Cochrane Library databases were searched from inception to 18 March 2026. Studies reporting odds ratios (ORs) for OLM with preoperative CEA as a dichotomous or continuous variable were included. Random-effects meta-analyses using restricted maximum likelihood (REML) estimation were performed, with univariate ORs as the primary analysis and multivariate ORs pooled separately as a sensitivity analysis. Subgroup analysis (by clinical stage, tumor size, histological type, CEA cut-off value, and Country), meta-regression (by Country, publication year and histological type), leave-one-out and publication-bias analyses were also performed. The protocol was registered in PROSPERO (CRD420261363192). Thirty studies were included. For dichotomous CEA, the univariate pooled OR for OLM was 4.82 (95% CI: 3.66–6.36, P < 0.001, I² = 60.81%, 95% PI: 1.80–12.91) and the multivariate OR 2.74 (95% CI: 2.03–3.69, P < 0.001, I² = 49.34%, 95% PI: 1.10–6.84). In a separate analysis of continuous CEA, each 1-ng/mL increase was associated with OLM (univariate OR 1.16, 95% CI 1.05–1.28, I² = 92.64%, 95% PI 0.81–1.67; multivariate OR 1.14, 95% CI 1.10–1.18, I² = 10.33%, 95% PI 1.07–1.22). For N1 metastasis, the univariate and multivariate ORs for dichotomous CEA were 2.93 (95% CI 1.65–5.19, I² = 79.92%, 95% PI 0.38–22.68) and 2.09 (95% CI 1.29–3.39, I² = 59.90%), respectively, the corresponding ORs for N2 metastasis were 5.14 (95% CI 3.95–6.70, I² = 20.57%, 95% PI 2.94–8.99) and 2.86 (95% CI 1.99–4.12, I² = 41.65%, 95% PI 1.02–8.01). Subgroup analyses stratified by clinical stage, tumor size, histological type, CEA cut-off value, and Country revealed significant positive associations in most strata with a sufficient number of studies. Meta-regression identified Country ( P = 0.032, R² = 36.5%) and histological type ( P = 0.007, R² = 51.3%) as significant moderators, whereas publication year was not significant ( P = 0.879, R² = 0%). Residual I² was 49.1%, 42.4%, and 62.9% for the three models, respectively. Leave-one-out analyses (range 4.54–5.17) and Egger’s regression test ( P = 0.819) indicated the primary result was robust, and no publication bias was detected. Elevated preoperative CEA is associated with an increased risk of OLM in clinical stage I NSCLC, particularly in lung adenocarcinoma. Because all included cohorts were East Asian, these findings require external validation in other populations. Combined predictive models integrating CEA with other parameters warrant development to improve the preoperative prediction of nodal involvement.