Prognostic risk classification for biochemical relapse-free survival in patients with oligorecurrent prostate cancer after [68Ga]PSMA-PET-guided metastasis-directed therapy

Purpose Since the success of prostate-specific membrane antigen-positron emission tomography (PSMA-PET) imaging for patients with oligorecurrent prostate cancer (ORPC), it is increasingly used for radiotherapy as metastasis-directed therapy (MDT). Therefore, we developed a prognostic risk classification for biochemical relapse-free survival (bRFS) for patients after PSMA-PET-guided MDT after radical prostatectomy. Methods We analyzed 292 patients with local recurrence (LR) and/or pelvic lymph node (LN) lesions and/or up to five distant LN, bone (BM), or visceral metastases (VM) detected with [68Ga]PSMA-PET imaging. Median follow-up was 16 months (range 0–57). The primary endpoint was bRFS after MDT. Cox regression analysis for risk factors was incorporated into a recursive partitioning analysis (RPA) with classification and regression tree method. Results PSA at recurrence ≥ 0.8 ng/mL, BM, and VM was significantly associated with biochemical relapse. RPA showed five groups with tenfold cross-validation of 0.294 (SE 0.032). After building risk classes I to IV (p < 0.0001), mean bRFS was 36.3 months (95% CI 32.4–40.1) in class I (PSA < 0.8 ng/mL, no BM) and 25.8 months (95% CI 22.5–29.1) in class II (PSA ≥ 0.8 ng/mL, no BM, no VM). LR and/or pelvic LNs caused relapse in classes I and II. Mean bRFS was 16.0 months (95% CI 12.4–19.6) in class III (PSA irrelevant, present BM) and 5.7 months (95% CI 2.7–8.7) in class IV (PSA ≥ 0.8 ng/mL, no BM, present VM). Conclusion We developed and internally validated a risk classification for bRFS after PSMA-PET-guided MDT. Patients with PSA < 0.8 ng/mL and local relapse only (LR and/or pelvic LNs) had the most promising bRFS. PSA ≥ 0.8 ng/mL and local relapse only (LR and/or pelvic LNs) indicated intermediate risk for failure. Patients with BM were at higher risk regardless of the PSA. However, those patients still show satisfactory bRFS. In patients with VM, bRFS is heavily decreased. MDT in such cases should be discussed individually.


Introduction
In 1995, Weichselbaum et al. hypothesized that an oligometastatic state in cancer progression exists [1]. On that basis, the oligorecurrent paradigm was also introduced into prostate cancer (PC) management. However, several descriptions of oligometastatic PC exist, and definitions vary between less than three to five distant lesions [2]. Data show a better outcome for patients with oligometastatic PC than for patients with advanced metastatic disease [3]. Recently, Ost et al. demonstrated that patients with oligorecurrent PC treated with metastasis-directed therapy (MDT) had improved freedom from androgen deprivation therapy (ADT) (13 vs. 21 months) [4]. Furthermore, Decaestecker et al. reported a prolonged time to ADT and no grade III toxicity after repeated stereotactic body radiotherapy (SBRT) [5]. Interestingly, even by using less sensitive choline positron emission tomography (PET) imaging for staging in the trials mentioned above [4,5], a good response to MDT was observed.
Nevertheless, standard-of-care treatment for oligorecurrent PC remains empiric salvage radiotherapy (RT) for suspect local recurrence in the prostatic bed or ADT for patients with distant metastases. Due to limited treatment and diagnostic options, local therapy of the metastatic sites has been rarely used in the past. With the increasing application of prostatespecific membrane antigen (PSMA)-PET imaging, the treatment of patients with few metastases has changed. PSMA-PET imaging has become an effective tool for staging and defining targets for precise treatment of patients with biochemical relapse after radical prostatectomy (RP) [6,7]. Whereas in the past, radiation oncologists had to administer empiric treatment to the prostate bed mostly without evidence in imaging; today, PSMA-PET helps to visualize recurrent tumor sites and distant lesions and provides a highly sensitive means for estimating the accurate tumor load.
PSMA-PET imaging enriches the possibilities of MDT with conventional fractionated RT, stereotactic body RT (SBRT), or surgery. These can prolong time to initiation of ADT [2] and bear the possibility of cure.
Up to now, the optimal selection of patients for MDT is unknown. The majority of patients will develop progression, despite PSMA-PET staging and MDT. Therefore, clinical risk classifications have become an accepted instrument for weighing the risk-benefit ratio. To our knowledge, no such tools have been published for patients with oligorecurrent PC treated with MDT using highly sensitive PSMA-targeted imaging. In the present analysis, we aimed to develop a prognostic risk classification predicting biochemical relapse-free survival (bRFS) after PSMA-PET-guided MDT for oligorecurrent prostate cancer after prior RP.

Patients
We established a multi-institutional, retrospective database, and collected data of 379 patients from six German and Swiss centers. The institutional review boards of all participating centers approved the study (BASEC-Nr. 2017-01499). The subjects were free from distant metastases (M0) at initial diagnosis. We defined the relapse as biochemical failure and recurrence diagnosed by PSMA-PET imaging. The patients exhibited oligorecurrent disease with local recurrence and/or pelvic lymph node lesions (N1) and/or distant metastases (M1a/1b/1c) in PSMA-PET imaging. We included patients with up to five distant lymph node, bone, or visceral lesions and any PSA level at PSMA-PET-based diagnosis. There was no restriction on the total number of pelvic lymph node lesions. The database included patients treated with MDT for PC metastases independently of the primary therapy. In the present analysis, we included 292 patients with initial RP and a subsequent diagnosis of oligorecurrent PC with positive findings in PSMA-PET imaging. Patients with other primary therapy for PC (e.g., primary RT) or with previous postoperative non-PSMA-PET-based RT were exclude from this analysis (n = 87). The flow diagram is shown in Fig. 1.

PSMA-PET imaging and RT
Each patient received PET imaging with a [ 68 Ga]-labeled PSMA targeting ligand [8]. Imaging was performed according to the joint EANM and SNMMI guidelines [9]. PSMA-PET imaging was acquired in conjunction with either computed tomography (CT) or magnetic resonance imaging (MRI). The CT was performed as a diagnostic CT or as a low-dose CT. Intravenous and if necessary oral iodinated contrast agents were used if the patients had no contraindications. CT scans were acquired in the portal venous phase. MRI scans where performed with contrast agents, when possible. PET scans were conducted approximately 60 min after bolus injection of the ligand (mean activity 162.5 MBq; range 87.0-291.0 MBq). When possible, furosemide 20 mg was given to reduce tracer collection in the urinary tract system. One specialist in nuclear medicine and one radiologist or a dual boarded nuclear medicine physician and radiologist interpreted the acquired imaging. Focal tracer uptake higher than the surrounding background and not associated with physiologic uptake was considered as malignant.
MDT as conventional fractionated RT or SBRT was performed at the discretion of the study centers. Furthermore, ADT was administered according to the guidelines of the participating departments.

Endpoints
The primary endpoint was bRFS, which was measured from MDT to biochemical relapse. We defined biochemical relapse as PSA levels ≥ 0.2 ng/mL above PSA nadir after RT. When PSA levels did not respond to RT, pre-RT PSA level with a rise of ≥ 0.2 ng/mL was defined as a relapse. Follow-up was performed according to institutional protocol.

Statistical analysis
We applied univariate and multiple Cox regression to assess significance for risk factors for biochemical relapse. We evaluated initial tumor stage [10] (T2c ≤ vs. ≥ T3a), initial nodal status [11] (N0 vs. N1), initial Gleason score [10,11] (7a ≤ vs. ≥ 7b), and initial resection status [11] (R0 vs. R1). Further, we evaluated the risk factors PSA persistence > 0.1 ng/mL after RP [12], PSA level at PMSA-PET imaging/recurrence [10,11], age at relapse as well as the PSMA-PET-related factors local recurrence (no vs. yes), pelvic lymph node lesions (no vs. yes), distant metastases (no vs. yes), and the number of lesions at PSMA-PET imaging. We used receiver operating characteristic (ROC) analysis to determine cutoff values for non-dichotomous variables. To account for the possible confounder of additive ADT administration and to show robustness of the model, multiple Cox regression was calculated unadjusted and adjusted for additive ADT. Risk factors which showed significance in both models were included into the risk classification. We used recursive partitioning analysis (RPA) with the classification and regression tree (CRT) method to analyze the significant factors. bRFS at 12 months was the primary endpoint in the decision tree calculation. Decision tree analysis is a nonparametric automatic statistical learning algorithm that combines variables. The results are shown in a decision tree with splits based on the variables. The Gini index was applied to minimize node impurity after splitting. End nodes of the decision tree were grouped from A to E. For internal validation, we used tenfold cross-validation. In the tenfold cross-validation scheme, patients were randomly assigned into ten equal subgroups. Nine subsets were utilized as a training cohort. The tenth group was used as a validation set to test the performance. Subsequently, risk classes I to IV (low to very high risk) were built by combining end node groups with a similar outcome in Kaplan-Meier estimator for the overall group. We used the Kaplan-Meier method to calculate bRFS of the risk classes. All statistical analyses were performed with SPSS version 21 (IBM, Armonk, USA). A p value of less than 0.05 was defined as statistically significant.

Results
All patients were treated for oligorecurrent PC with conventional fractionated RT and/or SBRT between April 2013 and January 2018. The median age of all patients was 70 years (range 46-95 years). Median follow-up was 16 months (range 0-57 months). Table 1 shows the patient characteristics, and Table 2 shows the numbers of biochemical relapse and local control. Table 3 shows the results of the univariate Cox regression of prognostic factors affecting bRFS. ROC analysis determined 0.8 ng/mL as the significant cutoff value for PSA before RT. A PSA persistence after RP ≥ 0.1 ng/mL, PSA levels prior to RT ≥0.8 ng/mL, the presence of bone metastases, and the presence of visceral lesions were significantly associated with relapse in the univariate analysis. Multiple Cox regression unadjusted and adjusted for additive ADT is shown in Table 4. PSA levels ≥ 0.8 ng/mL, the presence of bone metastases, and the presence of visceral metastases remained significant factors in both models.
We included the three significant factors of both models into the RPA using the CRT method. Figure 2 shows the decision tree with the end nodes A to E. Tenfold crossvalidation demonstrated a risk for miscalculation of 0.294 (standard error 0.032), which results in 70.6% accuracy.
The Kaplan-Meier estimator showed a mean bRFS of 36.3 months (95% CI 32.4-40.1 months) for group A (PSA < 0.8 ng/mL, no bone lesions), of 25.8 months (95% CI 22.5-29.1 months) for group B (PSA ≥ 0.8 ng/mL, no bone or visceral lesions), of 15.6 months (95% CI 11.3-20.0 months) for group C (PSA < 0.8 ng/mL, present bone lesions), of 16.6 months (95% CI 11.2-22.0 months) for group D (PSA ≥ 0.8 ng/mL, present bone lesions), and of 5.7 months (95% CI 2.7-8.7 months) for group E (PSA ≥ 0.8 ng/mL, no bone lesions, present visceral lesions). Subsequently, we built risk classes with a similar outcome: risk class I with low risk (group A), class II with intermediate risk (group B), class III with high risk (groups C and D), and class IV with very high risk (group E). We used the Kaplan-Meier estimator to compare survival among risk classes for the overall group (see Fig. 3). The log-rank test showed significant differences among groups with p < 0.0001.
The risk classification showed that patients with PSA levels < 0.8 ng/mL and no bone metastases according to PSMA-PET experience low risk (class I). Patients diagnosed with PSA levels ≥ 0.8 ng/mL and no bone metastases nor visceral metastases prior to MDT were at intermediate risk (class II). PSMA-PET-positive local recurrence and/or pelvic lymph node lesions caused the PSA rise in those two classes. Patients with bone metastases before MDT independently of  PSMA-PET, prostate-specific membrane antigen-positron emission tomography; PSA, prostate-specific antigen; RT, radiotherapy; SBRT, stereotactic body radiotherapy; EQD2/1.5Gy, equivalent dose in 2 Gy for alpha/beta of 1.5 Gy; ADT, androgen deprivation therapy the PSA value showed a faster relapse (class III). Patients with a PSA level ≥ 0.8 ng/mL, no bone metastases, and the presence of visceral metastases were at very high risk for decreased bRFS (class IV). Table 5 shows the risk classes with the respective bRFS and a description of the prognostic factors. Distribution, duration, and medication of additive ADT use in the risk classes are shown in Table 6.

Discussion
The present analysis aimed to develop a prognostic risk classification for bRFS after PSMA-PET-guided MDT for patients with oligorecurrent PC. Three variables emerged as prognostic factors for decreased bRFS in multiple Cox regression: PSA level ≥ 0.8 ng/mL, presence of bone metastases and visceral lesions.   [4]. However, both prospective trials did not use modern PSMA-PET imaging for staging. To our knowledge, the present study is the largest retrospective analysis with patients suffering from oligorecurrent PC treated with PSMA-PET-guided MDT.
Our classification shows that patients with low PSA levels (< 0.8 ng/mL) and pelvic lymph node lesions and/or local recurrence only benefit the most from PSMA-PET-guided MDT (class I). Patients with PSA level ≥ 0.8 ng/mL and pelvic lymph node lesions and/or local recurrence only are at intermediate risk for treatment failure (class II). Hence, our data show a favorable outcome for patients treated with PSMA-PET-guided MDT who are diagnosed with local relapse only. Recently, Schmidt-Hegemann et al. reported a good response to PSMA-PET-guided salvage RT. The authors showed a bRFS of 78% after a median follow-up of 23 months for patients with local relapse only [20]. Furthermore, early treatment of oligorecurrent PC at low PSA levels is desirable. It remains not surprising that the PSA level before MDT is a prognostic factor for biochemical relapse. For local salvage RT, the dictum shifted to "the earlier, the better." Several studies have shown that a high pre-treatment PSA level before local salvage RT is associated with a decreased biochemical and oncological outcome [21][22][23][24]. Early salvage treatment should be initiated at PSA levels of less than 0.5 ng/mL [11]. However, data on the prognostic value of pre-PSA level before MDT in patients with oligorecurrent PC remains rare. Our data showed a pre-treatment PSA level of 0.8 ng/mL as the cutoff value. Although this value seems high, treatment of oligorecurrent PC appears to be beneficial when administered in patients with PSA levels < 0.8 ng/mL or preferably earlier.
Patients with present bone metastases were at higher risk for biochemical relapse (class III). However, those patients still show satisfactory bRFS. Bone metastases are common in patients with metastatic PC [25]. Muacevic et al. reported a local control rate of 95.5% at 24 months in patients treated with robotic SBRT for bone metastases [26]. In a previous study, Habl et al. showed a local progression-free survival (PFS) of 100% at 24 months for patients with bone lesions treated with SBRT. However, the median PSA-PFS was 6.9 months, and the median distant PFS was only 7.4 months [6]. In our study, in-field local control of bone lesions was 100%; bRFS of patients with bone lesions was 16.0 months. The limited effectiveness in systemic control in the series of Habl et al. might be owed to the inability to detect all present metastatic lesions [6] since the authors staged the patients with PET imaging using either the [ 68 Ga]PSMA or the [ 11 C]choline tracer. Therefore, the less sensitive choline-PET imaging might not have detected all lesions. Although PSMA-PET imaging shows a massive improvement in sensitivity at low PSA levels as compared with choline-PET, the detection rate of recurrences is still approximately 50% at PSA levels below 0.2 ng/mL [27]. Perera et al. reported rates of 58% and 76% for PSA levels of 0.2-1.0 and 1.0-2.0 ng/mL for PET scans with [ 68 Ga]PSMA tracers [28]. The optimal threshold value of PSA indicating a high chance of detection of tumor relapse in PSMA-PET imaging remains a topic of discussion.
Patients with visceral metastases experience heavily reduced bRFS (class IV). Although visceral metastases are not common in patients with prostate cancer [25], survival is decreased in general. Gandaglia et al. reported a reduced median overall survival (OS) and PC-specific survival (PSS) in patients with visceral metastases (median OS 16 months, median PSS 26 months) [25]. Only patients with synchronous bone  [29]. The reduced outcome was a reason for defining visceral metastases as highvolume or high-risk disease in the CHAARTED [30] and LATITUDE [31] trial. The data on MDT for patients with visceral metastases is still limited. Ost et al. showed no difference in outcome after MDT between patients with nodal versus non-nodal metastases as visceral lesions [4]. However, the study recruited only one patient with visceral lesions. Although we evaluated a good in-field local control of 100% in patients with visceral metastases, the systemic effect seemed suboptimal. Patients developed biochemical relapse rapidly after local treatment. Therefore, patients with visceral lesions might exhibit a different, more aggressive tumor biology than patients with bone or lymph node metastases. Such patients might benefit from additional ADT to MDT [32].
However, in such cases, MDT should be discussed individually. Bearing in mind the reduced bRFS, palliative ADT alone might be a valid alternative.
Our study has certain limitations. Because of the retrospective nature, the management in terms of treatment was not prospectively defined and differed between institutions. Administration of additive ADT was not standardized and might be a confounder. To account for this problem, we calculated a multiple Cox regression unadjusted and adjusted for additive ADT to show robustness of the model. Furthermore, RT technique and treatment differed between institutions with conventional RT on the one hand to extreme hypofractionated schedules for distant metastasis on the other hand. This must be considered when interpreting the results. However, all visible lesions and thus all the tumor burden was treated with good in- Fig. 2 Recursive partitioning analysis (RPA) with the classification and regression tree (CRT) method (n = 197, data of 95 patients was censored at 12 months). End nodes are marked with groups A to E. (PSMA-PET, prostate-specific membrane antigen-positron emission tomography; PSA, prostate-specific antigen; ng/mL, nanogram/milliliter) field local control (see Table 2). Therefore, influence on bRFS should be low. The group of patients with visceral metastases was small. Therefore, the results must be interpreted with caution. However, the findings are plausible as several studies show the poor outcome of patients with visceral lesions. The results will be validated externally for further evidence.

Compliance with ethical standards
Conflict of interest Arnd-Christian Müller declares scientific cooperation with Electa®, Philips® and Siemens®. Matthias Eiber is consultant for Blue Earth Diagnostics® and ABX® and has patent rights on rhPSMA. All other authors have no conflict of interest to declare.
Ethical approval The institutional review boards of all participating centers approved the study (BASEC-Nr. 2017-01499).
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.