External Validation of the LIONS-PREY score with the NLST data
Principal Investigator
Name
Luca Salhoefer
Degrees
M.D.
Institution
University Hospital Essen
Position Title
Senior Researcher
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1484
Initial CDAS Request Approval
Dec 2, 2025
Title
External Validation of the LIONS-PREY score with the NLST data
Summary
With the ongoing surge in chest CT requests, whether in clinical routine or systematic screening programs, accurate malignancy prediction of lung nodules is becoming increasingly paramount in a patient’s journey. Additionally, the success of lung cancer screening programs is tightly bound to accurate malignancy prediction of lung nodules. Most programs use the Brock or Mayo models for newly detected nodules, while the Herder model refines predictions after PET/CT. Recently, Doerr et al. developed a new scoring system for screening-detected nodules: LIONS PREY (Lung lesION Score PREdicts malignancY). In the inaugural study of 382 biopsies or resected nodules, it outperformed the Mayo score significantly with by overall correct classification of 96.0 % (Doerr F, Giese A, Höpker K, et al. LIONS PREY: A New Logistic Scoring System for the Prediction of Malignant Pulmonary Nodules. Cancers (Basel). 2024;16(4):729). In two validation studies, one with 523 patients who underwent lung resection and a second one with 193 lesion biopsies from navigational bronchoscopy, LIONS PREY outperformed the Brock, Mayo, and Herder (including PET/CT data) models (manuscripts in preparation, Abstracts submitted for the 2026 ATS congress). To foster the integration into clinical practice, validation on large multicenter datasets is mandatory. We hypothesize that the LIONS PREY score is more accurate than established scoring systems to estimate the malignancy risk of lung nodules from the National Lung Screening Trial. Those results could disrupt clinical practice and might lead to changes in the management of lung nodules. Among those could be a reduction of medical interventions (e.g., surgery, biopsy) and less follow-up imaging, leading to more personalized care.
Aims
- External Validation of the LIONS PREY score on the NLST dataset
Objective: Apply the LIONS PREY malignancy prediction model to all screen-detected pulmonary nodules in the NLST dataset and determine its diagnostic accuracy.
Hypothesis: LIONS PREY will demonstrate superior discrimination, calibration, and overall accuracy compared to established models (Brock, Mayo, and Herder).
Outcomes: AUC, sensitivity, specificity, PPV/NPV, calibration curves, decision curve analysis.
- Compare LIONS PREY with established risk-prediction models for clinical decision-making impact.
Objective: Quantitatively compare LIONS PREY to Brock, Mayo, and Herder models using identical NLST nodules to determine improvements in classification accuracy and risk-stratification performance.
Hypothesis: LIONS PREY will reclassify a meaningful proportion of nodules into more appropriate management categories (e.g., avoiding unnecessary biopsies or surgeries).
Outcomes: Net reclassification improvement (NRI), integrated discrimination improvement (IDI), proportion of nodules moved to different guideline-based management pathways.
- Estimate the potential clinical and health-system impact of replacing current models with LIONS PREY.
Objective: Model the downstream effect of LIONS PREY–based decision making on patient management, including the number of invasive procedures, follow-up imaging studies, and potential over- or under-diagnosis.
Hypothesis: Use of the LIONS PREY score will reduce unnecessary interventions and follow-up imaging while maintaining or improving early cancer detection.
Outcomes: Reduction in surgeries/biopsies for benign lesions, changes in follow-up intervals, cost-effectiveness, and potential reductions in patient morbidity.
Collaborators
Luca Salhoefer University Hospital Essen
Fabian Doerr University Hospital Essen
Marcel Opitz University Hospital Essen
Johannes Haubold University Hospital Essen