Blinded ¹⁸ FDG PET-CT re-read versus EBUS-TBNA in stage III _A-C NSCLC: A mediastinal staging concordance study.

Abstract

280 Background: 18 FDG PET-CT is standard for mediastinal staging of NSCLC, but its specificity is low in tuberculosis-endemic regions. Because EBUS-TBNA is an imperfect reference standard, we quantified agreement and directional discordance between PET and EBUS cytology. Methods: This retrospective, blinded diagnostic concordance study included 52 patients with biopsy-confirmed stage III A-C NSCLC who underwent both PET-CT and EBUS-TBNA for initial mediastinal staging. A nuclear medicine specialist, blinded to original reports and EBUS results, re-interpreted de-identified PET-CT DICOM images. EBUS-TBNA served as the imperfect reference standard. Primary endpoint was station-level agreement quantified by Cohen’s κ with patient-cluster bootstrap 95% CIs. Secondary analyses included overall agreement, Byrt/Bishop/Carlin prevalence and bias indices, station-group strata (N2 vs N1) and echelon strata. Station-level diagnostic metrics were estimated with cluster-bootstrap CIs. Patient-level N2/N3 positivity concordance used standard κ and exact McNemar’s test. Discordance characterization compared SUV max and short-axis diameter between PET+/EBUS+ vs PET+/EBUS− stations (Wilcoxon). SUV max discrimination for EBUS positivity was assessed by ROC AUC (DeLong and cluster-bootstrap CIs) and Youden thresholds. Sensitivity analyses tested varying EBUS sensitivity assumptions. Results: Station-level agreement was slight (κ = 0.177; 95% CI: −0.008 to 0.371), with 42.3% discordance. The discordance pattern was anatomically asymmetric: at N2 mediastinal stations (n = 88), κ was 0.255 (fair agreement), driven by PET+/EBUS− discordance (33% FP vs 7% FN) and at N1 hilar stations (n = 16), κ was −0.125, driven by PET−/EBUS+ discordance (44% FN vs 13% FP). Echelon-stratified analysis showed best agreement at echelon 2 (subcarinal/ipsilateral station 4; κ = 0.301) with FP-dominant error, and symmetric poor agreement at echelon 3 (contralateral mediastinum; κ = 0.111). At the patient level, PET showed 91.7% concordance sensitivity but only 35.7% specificity for N2/N3 disease, with strongly directional discordance (McNemar’s p = 0.0004). PET+/EBUS− stations had significantly lower SUV max (median 5.4 vs 8.8, p = 0.002) and smaller nodes (median 12 vs 18 mm, p < 0.001) than concordant positives. Of 13 PET−/EBUS+ stations, 84.6% had no visible node on PET. ROC analysis yielded an AUC of 0.760 (95% CI: 0.627–0.890); the Youden-optimal threshold (SUV max > 7.2) achieved 80.0% specificity versus 11.1% at the traditional 2.5 cutoff. Imperfect-reference sensitivity analysis (EBUS sensitivity 86–97%) minimally changed κ (0.165–0.173). Conclusions: Station-level agreement between PET-CT and EBUS-TBNA is slight. PET has high patient-level sensitivity but poor specificity for N2/N3 disease, causing directional over-calling. EBUS-TBNA is necessary for PET-positive mediastinal nodes.