A multiparametric nomogram to reduce unnecessary EBUS-TBNA in suspected lung cancer: development and internal validation in an anthracosis-prevalent cohort

Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) is the first-line invasive modality for evaluating mediastinal lymphadenopathy in suspected lung cancer, yet it entails sedation, cost, and non-diagnostic results. In regions with a high prevalence of anthracosis, 18 F-fluorodeoxyglucose positron emission tomography/computed tomography (18 F-FDG PET/CT) specificity is reduced because pigmented and granulomatous nodes produce 18 F-FDG-avid benign lymphadenopathy mimicking malignancy. We developed and internally validated a multiparametric nomogram to support pre-procedural triage and reduce unnecessary EBUS-TBNA referrals. This single-center retrospective study analyzed 337 consecutive adults who underwent EBUS-TBNA between January 2019 and November 2025 (184 malignant, 153 benign by cytopathology). For each patient, an index node was defined as the station-level node with the highest SUVmax and largest short-axis diameter. Mean Hounsfield unit (HU) and SUVmax were measured by two blinded readers (intraclass correlation coefficients 0.95 and 0.93). Multivariable logistic regression identified independent predictors. Discrimination, calibration, and clinical utility were assessed by area under the curve (AUC), Hosmer-Lemeshow testing with calibration slope, and decision curve analysis. Internal validation used 1,000-iteration bootstrap optimism correction. Sex-specific decision thresholds were derived under a pre-specified false-negative-rate constraint of ≤ 5%. Five independent predictors were retained: female sex (adjusted odds ratio 0.29, 95% confidence interval 0.17–0.49), age ≥ 65 years (0.42, 0.25–0.70), short-axis diameter (1.05/mm, 1.02–1.10), mean HU (0.988/unit, 0.978–0.999), and SUVmax (1.16/unit, 1.09–1.23). Bootstrap bias-corrected AUC was 0.803; calibration was satisfactory (Hosmer-Lemeshow p = 0.420; bias-corrected slope 0.913). Against a SUVmax-only model, the nomogram improved integrated discrimination by 0.132 ( p < 0.001) with a categorical net reclassification improvement of 0.269, mainly through appropriate downgrading of benign nodes (net 20.9%). Sex-specific thresholds ( P < 0.15 women, P < 0.35 men) deferred 14 of 108 women (one missed malignancy of 36) and 26 of 229 men (five missed of 148), preserving negative predictive values above 80%. In a pre-specified anthracosis-versus-malignancy subgroup ( n = 293), bias-corrected AUC improved to 0.817. This multiparametric nomogram stratifies pre-procedural malignancy risk in mediastinal lymphadenopathy referred for EBUS-TBNA in suspected lung cancer. It offers sex-specific thresholds and a universal fail-safe option as a decision-support adjunct for reducing unnecessary invasive sampling in anthracosis-prevalent cohorts. Prospective external multicenter validation is required before broader implementation. Retrospectively registered.