Marieke van Son

142 CHAPTER 8 months, and tumor stage ≤T3b. All study patients provided written informed consent. A waiver from the IRB was obtained for patients treated off-protocol. Study and treatment details have been described previously[9,17]. Pre-treatment procedures Patients underwent pre-treatment 3T multiparametric (mp) MRI (including T2-weighted, diffusion-weighted, and dynamic contrast enhanced sequences) and 68Ga-PSMA- PET-CT or 18F-Choline-PET-CT scans. Initially, PSMA-PET/MRI-targeted biopsies were performed in all patients. However, since the accuracy of Gleason score assessment is debated in irradiated prostate tissue and because biopsies were predominantly positive, biopsies were no longer performed from the end of 2017 onward[18–20]. A dose of 19 Gray (Gy) was prescribed to the clinical target volume (CTV), which con- sisted of the MRI- and PET-CT-visible lesion (gross tumor volume [GTV]) plus a 5 mm margin. The planning target volume (PTV) was equal to the CTV. Dose constraints to organs at risk were according to protocol and included rectum D1cc and bladder D1cc <12 Gy, and urethra D10% <17.7 Gy [9]. Follow-up and outcome assessment Follow-up consisted of outpatient clinical visits combined with PSA measurements at 1 and 3 months, every three months the first year, biannually the second year, and an- nually thereafter up to 10 years. The outcome, BF, was defined according to the Phoenix definition (PSA nadir + 2 ng/ml). In case of BF, follow-up imaging was performed with Ga68-PSMA-PET-CT to assess loco-regional recurrence and/or metastatic disease. Candidate variables for model building To minimize the risk of overfitting, a sample size calculation was performed up front to calculate the number of candidate predictors allowed for multivariable testing. As- suming a 0.05 acceptable difference in apparent and adjusted R-squared, an expected R-squared of 0.15, an overall event rate of 0.2 (200 events per 1000 person-years fol- low-up), and a shrinkage factor of 0.8, would allow for seven candidate variables with 150 patients and 61 events[21]. For model 1, six candidate variables were selected for multivariable testing based on clinical knowledge and literature[10,15,16]. For model 2, three additional variables were tested, thereby accepting a small increase in chance of overfitting. For model 1, the variables assessed pre-salvage included: age at FS- HDR-BT, seminal vesicle involvement, GTV (cm3), PSADT (months), PSA (ng/ml), and MRI-based T-stage (T1, T2, and T3 based on NCCN criteria). PSADT was obtained using the Memorial Sloan Kettering Cancer tool (available via: https://www.mskcc.org/ nomograms/prostate/psa_doubling_time). For model 2, CTV D95% (dose to 95% of the CTV, in Gy), time to PSA nadir (months) and PSA reduction (ratio between pre-salvage PSA and PSA nadir, in %) were added.