Marieke van Son

147 PREDICTION MODELS FOR BF AFTER FOCAL SALVAGE HDR-BT Cox proportional hazards models Table 2 presents the results from multivariable Cox regressions for model 1 and 2. At baseline (model 1), four variables were identified as significant predictors of BF: age (HR 0.94), pre-salvage PSA (HR 2.19), GTV (HR 1.05), and pre-salvage PSADT (HR 0.87 and 1.18 for PSADT and PSADT’, respectively). For model 2, six predictors were identified: age (HR 0.92), pre-salvage PSADT (HR 0.89 and 1.16), pre-salvage PSA (HR 4.47), seminal vesicle involvement (HR 1.49), post-salvage time to PSA nadir (HR 0.82), and PSA reduction (HR 0.98). Although seminal vesicle involvement was not statistically significant in model 2 (p=0.14), its exclusion affected AIC notably and therefore it remained in the model. The ranges of the continuous variables in our dataset are displayed in Supplementary File C. Calibration and internal validation Calibration curves at 12, 24, and 36 months for both models are depicted in Figure 1. Cal- ibration was reasonable up to 24 months. Internal validation showed an optimism of 0.15 and 0.19 for model 1 and 2, respectively. The β-coefficients were therefore adjusted with a factor of 0.85 (model 1) and 0.81 (model 2). The C-statistic was adjusted from 0.75 to 0.73 (95% CI: 0.66-0.81) for model 1 and from 0.85 to 0.84 (95% CI: 0.78-0.90) for model 2. The full regression equation for both models can be found in Supplementary file D. Figure 1 – Calibration plots for model 1 (upper row) and model 2 (lower row) depicting the ob- served (y-axis) versus the predicted probability (x-axis) of biochemical disease-free survival (bDFS) at 12, 24, and 36 months, respectively. Vertical bars indicate the 95% confidence interval. The grey diagonal line depicts the ideal line for complete concordance between observed and predicted probabilities. The blue crosses indicate the optimism-corrected probabilities. 8