Session Item

Magnetic resonance (MR)-guided focal salvage high-dose-rate brachytherapy (FS-HDR-BT) for localized radiorecurrent prostate cancer (PCa) is associated with low toxicity rates. However, biochemical failure (BF) remains common. We developed two prediction models for BF to (1) enhance patient selection for FS-HDR-BT at baseline and (2) to identify patients at high-risk of BF during follow-up.

The models were based on a prospective cohort of 150 patients with localized radiorecurrent PCa, treated with FS-HDR-BT between 2013 and 2020 at the University Medical Center Utrecht. For model 1, the following pre-salvage characteristics were selected for multivariable analysis: age, gross tumour volume (GTV), PSA-level, PSA-doubling time (PSADT), and seminal vesicle involvement. For model 2, clinical target volume (CTV) D95% (dose to 95% of the CTV), post-salvage time to PSA nadir, and PSA reduction (ratio between pre-salvage PSA and post-salvage PSA nadir, in %) were added. Restricted cubic splines with 3 knots were used for non-linear continuous variables. Pre-salvage PSA-level was log-transformed. Predictors of BF (Phoenix-definition) were identified by multivariable Cox proportional hazards regression using backward elimination (based on lowest AIC). Internal validation was performed with 2000 bootstrap resamples and the C-statistics and hazard ratios (HR) were adjusted accordingly. Nomograms were constructed, and three risk groups were identified based on the 25th and 75th percentile of the linear predictor (total score).

Median follow-up was 25.1 months (IQR 13.5-38.1). Sixty-one patients (41%) experienced BF after a median of 29.7 months (IQR 23.5-43.6). At baseline (model 1), age, GTV, pre-salvage PSA, and pre-salvage PSADT were predictive of BF (Table 1). For model 2, age, pre-salvage PSA, seminal vesicle involvement, time to PSA nadir, and percentage PSA reduction were predictive of BF (Table 1). The adjusted C-statistics were 0.73 and 0.84 and calibration was acceptable. The nomogram belonging to model 1 is shown in Figure 1A. Estimated 2-year biochemical disease-free survival was 84%, 70%, and 31% for model 1 (Figure 1B), and 100%, 71%, and 5% (model 2) for the low-, intermediate-, and high-risk group, respectively.

Figure 1 Nomogram (A) belonging to Model 1 and corresponding Kaplan-Meier survival curves (B) for the risk groups based on the total score.

This study provides two models for prediction of BF in patients with localized radiorecurrent PCa treated with FS-HDR-BT. Our findings support that both pre- and post-salvage PSA characteristics are important predictors of BF in patients treated with focal therapy for radiorecurrent PCa. Model 1 could be used for patient selection for FS-HDR-BT. During follow-up, an updated risk prediction can be obtained with model 2, which can support patient guidance. Potentially, these models can also be used for other salvage techniques, for which external validation remains necessary.