Mediastinal over-staging with PET-CT in tuberculosis-endemic locally advanced NSCLC: SUV threshold recalibration improves PET/EBUS concordance.

Abstract

To quantify concordance between blinded PET-CT reinterpretation and EBUS-guided transbronchial needle aspiration (EBUS-TBNA) cytology for mediastinal lymph node staging in locally advanced NSCLC in a tuberculosis-endemic lower-middle income country (LMIC). Retrospective, single-institution blinded concordance study of 52 consecutive Stage III_A-C NSCLC patients. A nuclear medicine physician, blinded to all clinical data, reinterpreted deidentified PET-CT images. Patient-level agreement (any N₂/N₃ involvement) was assessed using Cohen's κ and McNemar's test. Among 75 PET-visible EBUS-sampled stations, SUV_max discriminative performance was evaluated by ROC. Candidate thresholds were applied to the patient-level classification to assess performance. Inclusion of nodal stations into radiotherapy volumes was also assessed. Patient-level agreement was fair (κ = 0.26; 95% CI: 0.04-0.48) with strongly directional discordance: 18 patients were PET+/EBUS- versus 2 PET-/EBUS+ (p < 0.001). PET-CT re-read specificity was 36%. Among sampled stations, confirmed malignant nodes had higher SUV_max than false-positives (median 8.8 vs. 5.4; p = 0.002; AUC 0.76). Applying the N₂-derived threshold of ≥ 5.8 to patient-level classification reclassified 9 patients, improving κ from 0.26 (fair) to 0.43 (moderate) and specificity from 36% to 61%. Despite negative or absent cytology, 71% of false-positive and 84% of unsampled PET-positive stations were included in the radiotherapy clinical target volume. In this TB-endemic cohort, PET systematically over-staged mediastinal disease, and PET positivity rather than tissue confirmation drove radiation volume design. An exploratory SUV_max threshold of ≥ 5.8 substantially improved patient-level concordance and needs further validation.