Definitions for central lung cancer (CLC) have been ambiguous in guidelines, causing difficulty in selecting candidates for invasive mediastinal staging among patients with radiologically node-negative, early-stage lung cancer. What is the optimal definition for CLC that is robust to interreader and institutional variation to select candidates for invasive mediastinal staging among those with clinical T1N0M0 lung cancer? Two retrospective cohorts were evaluated for the associations of central lung cancer according to 13 definitions based on chest CT scan with occult nodal metastasis. Univariate and multivariate ordinal logistic regression analyses were performed with the pathologic N category as an ordinal outcome. Robust definitions, which retained statistical significance across multireader, dual-institutional datasets, were identified. For these definitions, binary diagnostic performance and interreader agreement were investigated. In the two cohorts, 807 patients (median age, 63 years; interquartile range [IQR], 56-71 years; 410 women; 33 pN1, 48 pN2, and 1 pN3) and 510 patients (median age, 65 years; IQR, 58-71 years; 267 women; 33 pN1, 20 pN2, and no pN3) were included, respectively. Three definitions robust to interreader variation and dataset heterogeneity were identified: definition 7 (concentric lines arising from the midline, inner one-third, medial margin; adjusted OR, 2.01; 95%CI, 1.13-3.51; P= .02), definition 10 (location index-based inner one-third, center; adjusted OR, 3.60; 95%CI, 1.49-8.25; P= .003), and definition 12 (location index-based inner one-third, medial margin; adjusted OR, 3.57; 95%CI, 1.91-6.52; P< .001). Definition 12 showed higher interreader agreement than definition 7 (Cohen κ, 0.80 vs0.66; P= .005). Nevertheless, the sensitivity and positive predictive value of the three definitions were< 50%. Three definitions exhibited robust associations with occult nodal metastasis. However, selecting candidates for invasive mediastinal staging solely based on a central tumor location would be suboptimal.