Post-imaging pulmonary nodule mathematical prediction models: are they clinically relevant?

Area of Science:

• Pulmonary Medicine
• Radiology
• Oncology

Background:

• Post-imaging mathematical prediction models (MPMs) assess lung cancer risk using demographic and imaging data for solid pulmonary nodules.
• Existing MPMs often show suboptimal performance when applied to local patient cohorts without recalibration.
• Longitudinal imaging data's impact on MPM accuracy for solid nodules remains an area for investigation.

Purpose of the Study:

• To evaluate the hypothesis that calibrating MPM risk score thresholds to a local study cohort improves performance.
• To compare the pre- and post-calibration performance of four established MPMs.
• To determine if longitudinal imaging of solid nodules enhances MPM prediction accuracy.

Main Methods:

• A cohort of 317 individuals with computed tomography-detected solid nodules (80 malignant, 237 benign) was analyzed.
• Four MPMs (Mayo Clinic, Veteran's Affairs, Brock University, Peking University) were evaluated.
• A web-based application was developed to facilitate local cohort calibration and MPM performance analysis; 30 patients with repeated imaging were assessed longitudinally.

Main Results:

• Calibrated thresholds improved MPM accuracy, with Mayo Clinic and Brock University models showing the best performance (AUC=0.63, 0.61).
• Veteran's Affairs (AUC=0.51) and Peking University (AUC=0.55) models showed less improvement after calibration.
• No significant accuracy improvements were observed with repeated imaging sessions over time.

Conclusions:

• Calibration of MPMs to institutional cohorts is crucial for selecting the best-performing model and enhancing accuracy.
• The Brock University model demonstrated stable performance for solid nodules ≥8 mm but offered moderate refinement potential.
• MPM application is recommended primarily for initial evaluations, as longitudinal imaging did not increase predictive accuracy.