Development and Internal Validation of a Novel Nomogram Predicting the Outcome of Salvage Radiation Therapy for Biochemical Recurrence after Radical Prostatectomy in Patients without Metastases on Restaging Prostate-specific Membrane Antigen Positron Emission Tomography/Computed Tomography

Take Home Message We developed a novel nomogram for predicting outcomes of salvage radiation therapy (sRT) after prostate-specific membrane antigen positron emission tomography/computed tomography for patients with biochemical recurrence of prostate cancer after radical prostatectomy. This nomogram may facilitate better counseling of patients regarding early oncological outcomes after sRT. Patients with high risk of developing biochemical progression may be candidates for more extensive treatment.


Introduction
Radical prostatectomy (RP) is one of the main curative treatment options for patients with localized prostate cancer (PCa).Despite good long-term oncological outcomes, approximately 20-40% of patients experience biochemical recurrence (BCR) after RP [1][2][3].Results from the RAVES, RADICALS, and GETUG AFU-17 randomized clinical trials indicated that (observation followed by) early salvage radiation therapy (sRT) is oncologically noninferior to (immediate) adjuvant RT, but has lower toxicity rates and should therefore be considered a standard of care [4][5][6].As a result, the majority of contemporary patients undergo prostatespecific antigen (PSA) surveillance after RP.
Prostate-specific membrane antigen (PSMA) positron emission tomography/computed tomography (PET/CT) imaging is advised for patients with PSA >0.2 ng/ml and rising after RP and if the results will influence subsequent treatment decisions [7].Negative PSMA PET/CT findings should not delay sRT if otherwise indicated.Currently, owing to the enhanced detection of metastases with PSMA PET/CT [8,9], the number of patients with (oligo)metastatic disease has dramatically increased.In addition, local recurrence of disease is visualized more frequently, guiding potential local salvage treatment strategies such as sRT.However, a substantial number of patients with early BCR of PCa after RP undergo restaging PSMA PET/CT that shows no evidence of disease [10][11][12].In these patients, sRT is also recommended, assuming that they have local recurrent disease.Nomograms that include several biochemical and pathological parameters have been developed to predict sRT outcomes for patients with PCa BCR [13,14].However, these nomograms were based on outcomes for patients who did not undergo imaging with modern techniques for restaging, such as PSMA PET/CT.Our group recently demonstrated that a cohort of patients who under-went PSMA PET/CT for restaging before sRT had better shortterm oncological outcomes after sRT in comparison to a cohort of patients without PSMA PET/CT before sRT [15].Consequently, nomograms predicting sRT outcomes that were developed before the introduction of PSMA PET/CT imaging may no longer be as accurate.Our aim was to develop a novel nomogram to predict short-term oncological outcomes for patients who underwent PSMA PET/CT imaging that revealed no metastases before sRT for PCa BCR.

Study design and inclusion and exclusion criteria
We evaluated all patients who underwent PSMA PET/CT imaging before sRT to the prostatic fossa for PCa BCR (PSA 0.  [20].For all patients, the equivalent dose in 2-Gy fractions (EQD2) was calculated.

Outcome variables and statistical analysis
The study aim was to develop a novel nomogram to predict biochemical progression of disease within 1 yr after sRT without concomitant hormonal treatment in patients who underwent PSMA PET/CT before sRT.
Biochemical progression of disease was defined as PSA 0.2 ng/ml above the post-RT nadir, or the start of additional treatment after sRT (at the discretion of the treating physician).
Multivariable logistic regression analysis was performed using a backward elimination procedure with a significance level of p = 0.157 [21].PSA at sRT initiation (continuous), pathological grade group (GG; categorical: 1-2 vs 3 vs 4-5), pathological N stage (categorical: pNx vs pN0), pathological T stage (categorical: pT2 vs pT3a vs pT3b), surgical margin status (categorical: negative vs positive), time between RARP and sRT (continuous), PSA doubling time (continuous), PSMA PET/CT findings (categorical: negative vs local recurrence of disease), and biochemical persistence (BCP) after RARP (categorical: no vs yes) were included as potential predictors.The discriminative ability of the model was quantified using the area under the receiver operating characteristic curve (AUC).A calibration plot was generated to assess the agreement between predicted risk and observed risk of biochemical progression within 1 yr after sRT.To this end, patients were grouped by deciles for predicted risk.The average observed risk was then plotted against the average predicted risk for each of the ten groups.Decision curve analysis (DCA) was conducted to visualize the net benefit as a function of the threshold probability.The threshold probability reflects the variation in predicted risk that patients or doctors consider the minimum requirement for undergoing a specific intervention.A higher net benefit for the nomogram means that prediction of biochemical progression within 1 yr after sRT was correct for more patients.
To predict the risk for patients with missing variables, a multiple imputation procedure was used.All model and performance analyses were conducted across the imputed data sets and Rubin's rules were applied to obtain pooled results [22].Predicted risk values were subse-quently calculated for the complete data sets.Internal validation was performed via bootstrapping with 250 samples to correct for overfitting of the model.For all performance measures, the following steps were applied.First, Rubin's rules were applied to obtain the pooled coefficients.Second, internal validation was performed to determine the shrinkage factor, which was subsequently applied to the pooled coefficients.These coefficients were then used to calculate the performance metrics (eg, AUC, DCA).Analyses were performed in RStudio v4.1.2(R Foundation for Statistical Computing, Vienna, Austria) with the mice and psfmi packages [22].

Baseline characteristics
We included 302 patients who underwent PSMA PET/CT before sRT at a median age of 68 yr (IQR 64-72).The median time between RARP and restaging PSMA PET/CT was 26 mo (IQR 14-56; Table 1).According to the European Association of Urology (EAU) BCR risk categories, 62/302 patients (21%) were considered low risk and 183 (60%) were considered high risk; the risk category could not be calculated for 57 patients (19%) because of missing data for pathological GG or PSA doubling time.

Novel prediction model
PSA at sRT initiation (continuous), pathological GG (categorical: 1-2 vs 3 vs 4-5), surgical margin status (categorical: negative vs positive), PSA doubling time (continuous), local recurrence on PSMA PET/CT (categorical: negative vs miTr), and BCP after RARP (categorical: no vs yes) were included in the final model (Table 2).The AUC for this model was 0.72 (95% confidence interval [CI] 0.64-0.79;Supplementary Table 1).The novel nomogram is freely available online (https://www.evidencio.com/models/show/3019).The calibration plot in Figure 1A demonstrates that the nomogram is well calibrated, with good correlation between predicted and observed outcome.The range for the predicted values is 0-50%, with explained variance (R 2 ) of 0.151.DCA revealed that the nomogram yielded an increase in net benefit at a threshold probability of 4% (Fig. 1B).

3.5.
Clinical applicability Among patients with a predicted risk of <10% of developing biochemical progression after sRT, only three of 70 (4.3%)actually experienced biochemical progression 1 yr after sRT (Table 3).Conversely, among patients with a predicted risk of >20% according to the novel nomogram, 36/117 (30.8%) actually experienced biochemical progression 1 yr after sRT.

Internal validation
The regression coefficients for the final model were multiplied by a shrinkage factor of 0.817, as derived from bootstrapping.The optimism-corrected AUC and R 2 variance for the model are 0.681 and 0.095, respectively.

Discussion
Traditional imaging modalities are unreliable in excluding concurrent systemic disease in patients with rising PSA.
Given the side effects of sRT and the reported reduction in health-related quality of life [23][24][25], there is reluctance to offer local salvage treatment that may ultimately prove futile in the presence of coexisting metastatic disease.
Besides, in patients with measurable but nonprogressing PSA after RP, sRT can be considered as overtreatment.Nevertheless, sRT is recommended even in the absence of abnormalities on imaging.As PSMA PET/CT has greater diagnostic accuracy for detection of metastatic disease in comparison to conventional imaging techniques (92% vs 65% in the proPSMA-study [8]), it may have a substantial influence on management decisions for patients with BCR [26,27].We developed a novel nomogram for prediction of biochemical progression of disease within 1 yr after sRT in patients with BCR PCa who underwent PSMA PET/CT imaging for restaging.The final predictive model included PSA at sRT initiation, pathological GG, surgical margin status, PSA doubling time, PSMA PET/CT findings, and BCP after RARP.The AUC for this model was 0.72 (95% CI 0.64-0.79).
As PSMA PET/CT imaging has only been routinely performed in patients with BCR after RARP for a few years, our selection of the outcome variable for the predictive nomogram was limited to biochemical progression within 1 yr after sRT.In the ongoing PSMA-SRT randomized trial [28], PSMA PET/CT was able to detect disease outside the prostatic fossa in 25% (26/103) of patients.Historically, without PSMA PET/CT imaging for restaging, these patients would probably experience biochemical progression after  local salvage treatment.In the present we only included patients without metastatic disease on PSMA PET/CT.Hence, we believe that our biochemical progression rate within 1 yr after sRT (19%) might not increase substan-tially after a longer follow-up period and is therefore an accurate surrogate for long-term biochemical progression.All current nomograms for predicting oncological outcomes after sRT were developed before the introduction of PSMA PET/CT.Stephenson et al [14] developed a nomogram predicting the probability of 6-yr biochemical progressionfree survival after sRT in a multi-institutional cohort of 1540 patients.Their nomogram, which includes 11 parameters such as surgical margin status and PSA doubling time, had an AUC of 0.69.Campbell et al [13] developed a model that includes 12 parameters in a cohort of 1005 patients and obtained an AUC of 0.74 for predicting BCR at 5 yr after sRT.In the present study, the AUC for prediction of biochemical progression after sRT was 0.72.As we used a standardized method for selecting variables, pathological T stage and the time between RARP and sRT, for instance, were not included in the final model.Although these poten-  prognostic variables were not associated with our outcome variable and were therefore not included in the final model consisting of six parameters, it can be hypothesized that they would slightly increase the predictive ability of the nomogram.However, we chose to increase the clinical applicability of the model and therefore excluded variables that were not (significantly) associated with the outcome variable.
Zamboglou et al [29] recently evaluated metastasis-free survival and patterns of metastatic disease in patients undergoing PSMA PET/CT-guided sRT.They found that patients with PET-positive lymph nodes on PSMA PET/CT had significantly worse oncological outcomes than patients without metastatic disease on PSMA PET/CT.In the present study, we excluded patients with metastatic disease on PSMA PET/CT from the analysis.Patients with metastatic disease on restaging PSMA PET/CT are probably more likely to experience recurrence after local salvage treatment.To date, however, it is not known whether local sRT in patients with metastatic disease on PSMA PET/CT improves oncological outcomes in comparison to no local salvage treatment.Further prospective trials are warranted to investigate the role of local salvage treatment in patients with (oligo-) metastatic disease.
We feel that use of a predictive nomogram can add to counseling for patients with BCR after RARP.For the group of patients predicted to have <10% risk of developing biochemical progression according to our novel nomogram, only 4.3% had experienced biochemical progression 1 yr after sRT.On the basis of these findings, administration of sRT without concomitant hormonal therapy should be encouraged.As we did not include patients who did not receive any sRT, we were unable to compare these sRT outcomes with the natural history of PSA progression in patients undergoing surveillance instead of immediate sRT after RARP.Among patients predicted to have >20% risk of developing biochemical progression according to our nomogram, 30.8% actually experienced biochemical progression within 1 yr after sRT.It may be worth investigating whether a more extensive radiation treatment strategy or addition of concomitant hormonal treatment, as proposed by Pollack et al [30] and Spratt et al [31], would reduce the risk of disease progression in these patients.
Our study is not free of limitations.First, the retrospective nature may have potentially introduced patient selection biases.Second, it should be noted that PSMA PET/CT scans were reported as part of routine clinical practice and were not part of a prospective clinical trial or centrally reviewed.Therefore, different scan protocols, PSMA PET tracers, and PET scanners were used.Third, there were minor differences in the total sRT dose and the number of fractions between the participating centers.Since the target volume was defined according to the EORTC guidelines [20] and the EQD2 is similar, the study is not hampered by the inclusion of patients with extended radiation fields.Fourth, we included a relatively small number of patients and did not externally validate our novel nomogram, which will be of great importance before applying the nomogram in clinical practice.Finally, as BCR of PCa after sRT is not directly correlated with metastases-free or overall survival, it may not be a reliable surrogate for long-term oncological outcomes.

Conclusions
We developed a novel nomogram for prediction of biochemical progression of disease within 1 yr after sRT in patients with BCR of PCa who underwent PSMA PET/CT imaging for restaging purposes, without evidence of metastatic disease.To the best of our knowledge, this is the first nomogram to include PSMA PET/CT findings for patients undergoing sRT.The novel nomogram shows good performance and could result in better counseling of patients regarding oncological outcomes after sRT and thus more informed decision-making.

Table 2 -
Multivariable logistic regression analysis for prediction of biochemical progression at 1 yr after sRT in 302 patients who underwent PSMA PET/CT before sRT

Table 3 -
Proportion of patients experiencing biochemical progression according to different hypothetical cutoffs for the novel nomogram on when to perform salvage radiation therapy to the prostatic fossa without concomitant hormonal therapy