Nondiagnostic Lung Biopsies: Radiologist Performance in Prediction of Malignancy.

RATIONALE AND

OBJECTIVES

To determine outcomes of nondiagnostic lung biopsies and assess thoracic radiologist performance for predicting the likelihood of malignancy.

MATERIALS AND

METHODS

We retrospectively identified all patients who underwent computed tomography (CT)-guided transthoracic or cone-beam CT-guided bronchoscopic lung biopsy between 2020 and 2022 with nondiagnostic results. The final diagnostic outcome (benign vs malignant) was determined through repeat biopsy, surgical excision, or follow-up imaging. Two thoracic radiologists and two senior residents independently assigned a 5-point score after reviewing relevant clinical history, imaging studies, and histopathological data. Diagnostic performance was assessed using sensitivity, specificity, and area under the curve (AUC). Interobserver agreement was measured with the quadratic-weighted Cohen's κ test. Logistic regression was used to model predicted malignancy probability across the 5-point scale.

RESULTS

We included 142 nondiagnostic biopsies in 131 patients (mean age, 66 ± 14.8 years); 36 lesions (25.4%) were malignant. Malignancy rates were highest among lesions with atypical cells (16/18), normal lung tissue (11/20), and necrosis (3/6). Organizing pneumonia was associated with malignancy in 6.7% of cases. Using a score threshold ≥4, both thoracic radiologists achieved an identical sensitivity of 91.7% (95% confidence interval 77.5-98.2%), and specificities of 94.3% (88.1-97.9%) and 89.6% (82.2-94.7%) with no significant difference in AUC (p = 0.32). Interobserver agreement between the two thoracic radiologists was excellent (κ = 0.82). Logistic regression demonstrated <1% predicted malignancy at score 1 ("definitely benign") and >95% malignancy at score 5 ("definitely malignant").

CONCLUSION

Thoracic radiologists accurately predicted malignancy likelihood after nondiagnostic lung biopsy by integrating clinical, imaging, and histopathologic data.