Exploratory Treatment-Selection Model of Intraoperative Cone-Beam Computed Tomography During Percutaneous...

Background/Objectives: Intraoperative cone-beam computed tomography (CBCT) can detect residual fragments (RFs) during percutaneous nephrolithotomy (PCNL), enabling immediate removal and improving stone-free status. However, CBCT requires a hybrid operating room (OR), which is often limited in availability. This study explores patient and stone characteristics associated with CBCT eligibility and develops an exploratory treatment-selection model estimating stone-free probabilities conditional on CBCT use. Methods: We performed a retrospective study of a previously conducted randomized controlled trial evaluating intraoperative CBCT during PCNL in a tertiary care center. We compared CBCT-eligible cases versus ineligible cases, and cases achieving grade C (≤4 mm) stone-free status versus those with RFs. A multivariate exploratory treatment-selection model was developed using the strongest potential predictors of stone-free status. Internal validation was performed using bootstrapping. The model was also assessed for predicting grade A (0 mm) stone-free status. Results: The only significant difference between CBCT-eligible (n = 160) and ineligible (n = 60) cases was stone composition (p = 0.022). The final model included intraoperative CBCT (p = 0.003), stone size (p = 0.024), and composition (p = 0.044). Model-based estimates suggested smaller differences in predicted stone-free probabilities with CBCT in solitary stones. The AUC was 0.81 (95% CI: 0.73-0.88) for grade C and 0.75 for grade A (95% CI: 0.67-0.82) outcomes. Internal validation demonstrated moderate optimism, indicating potential overfitting. Conclusions: This exploratory treatment-selection model estimates conditional stone-free probabilities with and without CBCT. The findings suggest variation in expected benefit across stone characteristics but should be considered hypothesis-generating. The model is not intended for clinical decision-making and requires external validation before implementation.