Abstract Rationale With the implementation of lung cancer screening programs, accurate malignancy risk assessment of pulmonary nodules has become increasingly important. The LIONS PREY (LP) model was developed to enhance predictive accuracy by incorporating longitudinal nodule growth dynamics and, in its updated version, optional PET-CT parameters (SUVmax). Depending on data availability, three LP variants can be applied: LP baseline (LP-bl) without follow-up data, LP follow-up (LP-fu) including temporal size change when at least two CT scans are available, and LP integrating SUVmax (LP-S) including PET-CT information as an optional variable. This study aimed to validate the LP in patients undergoing navigational bronchoscopy and to compare its performance with the established Brock and Herder models. Methods Patients who underwent index navigational bronchoscopy for peripheral pulmonary lesions between December 2019 and March 2024 were retrospectively analyzed. Malignancy risk was calculated using the three LP variants, as well as the Brock and Herder scores. Model performance was assessed using receiver operating characteristic (ROC) curve analysis, area under the curve (AUC) comparison (DeLong test), calibration (O/E-ratio), and classification accuracy. Results A total of 193 lesions were analyzed, including 134 (69.4%) malignant and 59 (30.6%) benign. Malignant lesions were associated with increased growth dynamics (2.7 ± 4.1 mm vs. 1.0 ± 4.3 mm, p 0.001), and spiculation (56.0% vs. 23.7%, p 0.001). All LP variants significantly outperformed Brock and Herder in ROC analysis (Fig. 1). The LP-bl achieved an AUC of 0.91 95%-CI: 0.84-0.97, LP-fu 0.96 0.92-1.0, and LP-S 0.96 0.92-1.0 (p 0.001 for all comparisons vs. Brock and Herder). The Herder and Brock models showed AUCs of 0.73 0.60-0.86 and 0.59 0.45-0.74, respectively. The LP variants also demonstrated the highest correct classification rate (LP-bl: 89.4%; LP-fu: 91.1%; LP-S: 92.9%) compared with 72.5% for Herder and 60.8% for Brock (p 0.001). Calibration analysis for Brock and Herder showed a significant underestimation of malignancy compared to all three LP models (LP-bl: 1.06; LP-fu: 1.03; LP-S: 1.03; Brock: 1.38; Herder: 1.19). Conclusions The updated LP variants demonstrated significantly superior predictive accuracy and calibration compared with the Brock and Herder models in patients undergoing navigational bronchoscopy. All three LP variants showed excellent malignancy prediction in high-risk cohorts, maintaining high performance both when follow-up or PET-CT information was available and when such data were absent, underscoring their flexibility and clinical utility for malignancy risk stratification. This abstract is funded by: None
Büscher et al. (Fri,) studied this question.