Clinical Outcomes in Locally Advanced Prostate Cancer Between Radical Prostatectomy and Radiotherapy With Androgen Deprivation Therapy: A Propensity Score-matched Analysis

Yu-Cheng Lu National Taiwan University Hospital, National Taiwan University Chao-Yuan Huang National Taiwan University Hospital, National Taiwan University Chia-Hsien Cheng National Taiwan University Hospital, National Taiwan University Kuo-How Huang National Taiwan University Hospital, National Taiwan University Yu-Chuan Lu National Taiwan University Po-Ming Chow National Taiwan University Hospital, National Taiwan University Yi-Kai Chang National Taiwan University Hospital, National Taiwan University Yeong-Shiau Pu National Taiwan University Hospital, National Taiwan University Chung-Hsin Chen National Taiwan University Hospital, National Taiwan University Shao-Lun Lu National Taiwan University Hospital, National Taiwan University Keng-Hsueh Lan National Taiwan University Hospital, National Taiwan University Pei-Ling Chen National Taiwan University Hospital, National Taiwan University Jian-Hua Hong (  d07528012@ntu.edu.tw ) National Taiwan University


Introduction
Prostate cancer (PC) is the second most frequent malignancy diagnosis made in men and the fth leading cause of death worldwide. 1 There were over 1.4 million new cases of PC and 375,304 related deaths around the world in 2020. 2 Diagnostic estimates of loco-regional PC are over 90% in the United States. 3 In Taiwan, 58% of newly diagnosed cases of PC had loco-regional disease and, 11% had locally advanced disease between 2004 and 2012. 4 The management of locally advanced PC remains a challenge for urologists Traditionally, the risk of PC is strati ed by serum prostate-speci c antigen (PSA), Gleason score (GS) of the prostate biopsy and digital rectal exam (DRE). However, DRE is a subjective test with potential interobserver variability and GS has shown discrepancy between prostate biopsy and radical prostatectomy specimens. 5,6 PSA values are also in uenced by patient age and prostate volume. In the last decade, magnetic resonance imaging (MRI) of the prostate has become an essential diagnostic tool for local staging. The Prostate Imaging Reporting and Data System (PI-RADS) version 2 was updated in 2015 and developed to promote global standardization in the interpretation and reporting of the prostate MRI examination. Many studies have investigated the accuracy of MRI in local staging. 7 Robotic-assisted laparoscopic radical prostatectomy (RP) with pelvic lymph node dissection and radiotherapy (RT) combined with long-term androgen deprivation therapy (LTADT) are currently standard treatment options for locally advanced PC. 8 Both treatments incorporate a multimodality approach in this high-risk population to improve oncologic outcomes at the expense of distinct potential complications. RP has the advantage of more accurate disease staging with fewer bowel/rectal problems compared to RT, while anesthesia risk and associated higher risk of impotence and incontinence are still of concern. 9 On the other hand, RT with LTADT does not require hospitalization and has a lower risk of urinary incontinence but does convey a higher rate of radiation cystitis, bowel/rectal problems and possible side effects of ADT. 9,10 Standard treatment options for locally advanced PC include RP with pelvic lymph node dissection and RT alone or combined with hormone therapy, but the optimal therapy is still controversial. Several retrospective series found an advantage for RP but only on the basis of a low level of evidence, while others favored RT because of superior outcomes when adding LTADT. 10 However, RT dose and duration of ADT were heterogeneous. One ongoing prospective randomized study, SPCG-15 trial, comparing primary RP and RT plus ADT in locally advanced PC might provide valuable information in this speci c population in the future. 11 To date, optimal management remains uncertain in locally advanced PC. In this study, we aimed to compare clinical outcomes in locally advanced PC between the use of RP and RT combined with LTADT.

Inclusion and exclusion criteria
Between January 1, 2008 and November 31, 2018, 533 PC patients with clinically T stage 3/4 de ned by MRI were recruited. There were 160 patients in the RP group and 373 patients in the RT group. Thirty-eight patients being operated on at other hospitals, 2 patients undergoing neoadjuvant ADT, and 6 patients with adjuvant RT after operation in the RP group were excluded. We restricted patients to those with RT and 1.5~3 years ADT on the basis of NCCN guideline recommendations. A total of 166 patients were thus excluded. Three patients who experienced biochemical recurrence (BCR) during ADT treatment were also excluded. Nine patients lost to follow-up, with 3 in the RP group and 6 in the RT group, were also excluded. Of these, 309 patients were included in our study: 111 in the RP group and 198 in the RT group.

Statistical methods for clinical variables and de nition of outcomes
The study was approved by the Institutional Review Board and Ethics Committee of National Taiwan University Hospital (IRB 201911084RINC). A review was conducted of retrospectively obtained clinical data taken from the electronic medical records. Patient information was anonymized and de-identi ed prior to analysis. Informed consent from all subjects could be eliminated in this retrospective study according to Institutional Review Board and Ethics Committee of National Taiwan University Hospital (IRB 201911084RINC) regulation. For each group, descriptive statistics were used to summarize the clinical presentation (age at diagnosis, biopsy GS grade group, PSA at diagnosis (iPSA) and clinical T stage by MRI). Continuous variables were shown as median (range) and categorical variables as number (percentage). The Mann-Whitney U-test was performed to determine statistical signi cance for continuous variables between the groups while chi-square test or Fisher's exact test was used for categorical variables. To reduce selection bias, we performed propensity score (PS) matching at a 1:1 ratio of the following variables: age at diagnosis, biopsy GS grade group, iPSA and clinical T stage. Of these, 136 patients were included in our study. BCR was de ned as two consecutive times of PSA ≥0.2 ng/ml in the RP group and rising PSA of 2 ng/ml above the nadir (Phoenix criteria) in the RT group. Local recurrence was de ned as lymphadenopathy or tumor recurrence in the pelvis by computed tomography (CT) or MRI. Metastasis was de ned as distant metastasis in imaging (CT, MRI or bone scan). Kaplan-Meier analysis was performed to analyze BCR-free survival, local recurrence-free survival, metastasis-free survival and overall survival. To avoid immortal-time bias, the elapsed time for BCR-free survival analysis was calculated from the end of ADT treatment in the RT group and the operation date in the RP group to the date of BCR. In local recurrence-free survival, we calculated the follow-up time from the end of RT treatment. Univariate and multivariate Cox proportional hazards models were used to analyze the relationships between clinical variables and oncologic outcomes including BCR, local recurrence, metastasis and overall survival. All statistical analyses were performed using SPSS version 22.0 (IBM, Armonk, USA). Two-sided P values were calculated and a level of <0.05 was considered statistically signi cant.

Results
Patient characteristics are summarized in Table 1. A total of 309 patients were included. There were 111 patients (35.9%) in the RP group and 198 patients (64.1%) in the RT group. The median follow-up period was 62.8 months (interquartile range, IQR: 33.8-89.5) in the RP group and 56.1 months (IQR: 41.4-85.9) in the RT group (p=0.542). RT patients were older than RP patients (p<0.001). A total of 171 patients (86.4%) in the RT group were older than 65 years, compared to 51 (45.9%) in the RP group. RT patients had higher iPSA levels (p<0.001) and GS grade group (p<0.001). There were 91 patients (46.0%) in the RT group with iPSA levels ≥20 ng/ml and 21 patients (18.9%) in the RP group. A total of 138 patients (69.6%) in the RT group had GS grade group ≥3, compared to 56 (48.6%) in the RP group. RT patients also had more advanced clinical T stage (p<0.001). In the RT group, 90 patients (45.4%) had T stage ≥T3b, and 7 patients (3.5%) had T stage 4. In the RP group, 19 patients (17.1%) had T stage ≥T3b and no patient had T stage 4. All cohort patients and PS-matched patients in the RP group had poor BCR-free survival (log-rank test, p< 0.001, Figure 1). The univariate and multivariate analyses of predictors of BCR are demonstrated in Table  2. PS-matched patients with RT treatment were associated with reduced risk of BCR (hazard ratio (HR): 0.162, 95% con dence interval (95%CI): 0.072-0.366, p< 0.001). There was a higher risk of BCR in patients with more advanced MRI T stage (p= 0.014). GS grade group could predict BCR only in univariate analysis (p= 0.019) but failed to predict BCR in multivariate analysis (p= 0.065). In Kaplan-Meier analysis, there was no signi cant difference in local recurrence-free survival (log-rank test, p= 0.155, Figure 2), metastasis-free survival (log-rank test, p= 0.250, Figure 3) and overall survival (log-rank test, p= 0.502, Figure 4) between the two groups. In Table 3, there were no independent variables, including treatment methods, to predict local recurrence-free survival, metastasis-free survival and overall survival in the univariate analysis.

Discussion
To date, no published randomized trial has so far compared RP to RT plus ADT in locally advanced PC. 11 The most optimal strategy in this high-risk subpopulation remains controversial. Most retrospective studies have inevitable selection bias, heterogeneous treatment protocols and unclear outcome de nitions. 10 In this retrospective study, used the PS-matched method, strictly limited patient enrollment, and clear-cut outcome de nition, trying to provide direction for decision-making. We found that treatment with RP had a higher risk of BCR compared to the RT group among locally advanced PC patients.  13 MRI showed a higher sensitivity than did DRE for detection of non-organ-con ned PC (59 vs 41%, p< 0.01) in terms of corresponding pathologic T stage, and furthermore, with incorporation of MRI instead of DRE staging alone, the surgical treatment strategy would be altered in 27% of patients. In other words, DRE or transrectal ultrasonography (TRUS) alone is not accurate enough for local staging (T stage), and MRI is still one of the best imaging tools for assessing ECE in clinical practice. 14 Historically, men with locally advanced PC have been managed mostly with RT with ADT, while RP has been discouraged due to concerns about positive surgical margin, inadequate local control and side effects. 15  showed that RP has better survival than RT. Younger men and those with intermediate-or high-risk localized PC bene t more from surgery during 15 years of follow-up. 18 Consistently, a higher PSA and older age were noted in the RT group in the current cohort. In summary, the pathologic reports of prostatectomy, the dose of RT and duration of ADT treatment were not described in the majority of studies, and the discrepancies might in uence the CSM. 17,18 One ongoing prospective randomized SPCG-15 trial with a similar trial setting as the current study might deliver valuable information regarding this speci c population in the future. 11 Hackman et al. noted that adjuvant RT following prostatectomy prolonged biochemical recurrence-free survival compared with RP alone. 19  Comparing to RT plus LTADT, one of the irreplaceable bene ts of RP for patients in locally advanced PC is the ability to acquire accurate pathologic staging. Pathologic staging provided more reliable information to guide adjunctive therapies based on more precise data than bio-clinical variables including clinical T stage, biopsy GS or PSA. Indeed, 22 to 63% of PC initially de ned as high risk have been found to have organ-con ned disease following RP. 21 Stephen et al. also showed that 57% of patients initially classi ed as D'Amigo high-risk PC have organ-con ned disease at RP. 22 In addition, discrepancies in GS have frequently been found between biopsy and RP, such as up to 52.2% of GS over 8 tumors at biopsy had score downgrading at RP. 23 In the Mayo Clinic, 26% of PC patients with clinical T3 were downgraded to pathologic T2. (14) Similarly, the cT3-4 stage has shown to be inaccurate in up to 33% of cases at RP. 24 MRI images demonstrated moderate sensitivity for clinical T3 following prostatectomy (area under the curve of ROC: 0.61, 95%CI 0.54-0.67 ). 7 In our cohort, 58 (49.6%) clinical T3 diseases de ned by MRI were downgraded to pathologic T2 at RP. There was a comparable rate of T stage discrepancy in our cohort compared to the previous trials. In addition, 50 out of 60 (83%) and 42 out of 51 (82%) patients before and after 2015 respectively were found to have clinical T3 disease on MRI, and there was no statistically different prevalence (p=0.891). Among these patients, 36 patients (60%) diagnosed before 2015 were downgraded to pathologic T2, compared to 22 patients (43.1%) diagnosed after 2015, and despite a trend toward decreased discrepancy, there was no statistical difference (p=0.076).
Local disease control using different doses of RT in patients with PC is a critical issue. The total dose of 6000~7020 centigray (cGy) is currently recommended for localized advanced PC on the basis of the most updated guideline. 8 In our study, nearly all patients received RT with 7800 cGY in 39 fractions. Nevertheless, the most ideal total dose of RT is still under investigation and could in uence oncologic outcomes. Local failure after RT is an independent factor of overall survival, CSM and metastasis-free survival in high-grade localized PC. 25 Few randomized control trials (RCTs) have revealed that dose escalation (range: 7400-8000 cGy) has a signi cant impact on BCR, metastasis and CSM. And there are still inconsistent data on the effect of oncologic outcomes. However, the MRC RT01 RCT demonstrated that dose escalation (7400 vs 6400 cGy) showed an advantage in BCR, but the advantage did not translate into the improvement of overall survival. 26  Dose escalation might be related to more toxic effects. Michalski et al. noted that dose escalation (7920 vs 7020 cGy) showed higher rates of toxic effects. 28 The 5-year rates of 2 or greater rectum and genitourinary tract toxic effects were 21 and 12% in high-dose arm and 15 and 7% with 7020 cGy. When dose escalation was applied, the rates of severe late side effects (>grade 3) were 2~3% for rectum and 2~5% for the genitourinary tract. 29 Consistently, 7 patients (3.5%) suffered from severe radiation cystitis and received blood clot evacuation in our RT group. In terms of treatment-related complications, 25 cases of complication were recorded in the RP group with the majority no more than Clavien-Dindo Grade III, while 8 major complications were recorded in the RT group. Among the 25 complications in the RP group, 12 were Clavien-Dindo Grade I, 8 were Grade II and 5 were Grade III (3 patients, lymphatic leakage; 1 patient, pleural effusion; 1 patient, need for laparoscopic foreign body removal for incarcerated drainage tip). In the RT group, 196 patients (98.99%) received high-dose RT with 7800 cGY in 39 fractions. Among the 8 major complications, 1 patient died because of refractory radiation proctitis bleeding, while 7 patients suffered from severe radiation cystitis and needed blood clot evacuation. In summary, although high-dose RT with LTADT showed better biochemical control, the mid-term survival outcomes were similar to those in treatment with RP in our cohort, and the relatively higher complications rate should not be underestimated. The optimal treatment strategies still need a large cohort to determine the risk-bene t.
To the best of our knowledge, this is the rst study to compare RP alone, without adjuvant RT, with RT plus LTADT and take the immortal-time bias into consideration. RP alone allowed us the opportunity to observe the natural course of disease after prostatectomy. The potential for immortal time bias, also known as guarantee-time bias, exists whenever an analysis that is timed from enrollment is compared across groups de ned by a classifying event occurring sometime during follow-up. 30 It could be challenging for investigators to recognize when immortal time bias in uences the outcome of analyses. When treating BCR as the outcome, the PSA level would be in uenced by ADT treatment and interfered with the evaluation of outcome if immortal time bias was not well considered. However, there were scarce data exploring this bias in previous studies, and investigations retrieved from a database showed heterogeneous ADT duration. 31 In our cohort, patients in the RT group received pre-de ned and consistent duration of 1.5-3 years of ADT. If we were to extend the duration of ADT treatment, the RT group would have longer BCR-free survival, which would in uence the outcome. As a result, the distinct feature of the current study is that we calculated BCR time from the end of ADT treatment to avoid immortal-time bias.
Nevertheless, there were some limitations in our study. First, the retrospective, short follow-up period and small sample size limited extensive analysis and we could only analyze mid-term overall survival within 5 years. There were only 10 deaths (3.2%) in our cohort and no cancer-related death event for analysis with enough statistical power. Second, patient distribution was unbalanced in the two groups. Patients in the RT group were older and had higher iPSA, higher GS and more advanced T stage compared to patients in the RP group. However, the effect could be minimal after adjusting by PS-matched and multivariate analysis. Third, there were no records of Charlson Comorbidity Index and ECOG in our study. These factors might have led to unavoidable selection bias, in uencing the choice of treatment methods, and somehow impacting survival. Finally, we used MRI T stage because clinical T stage by DRE had inherent bias due to its subjective nature and potentially caused variability.
However, there was a lack of central review of the MRI images. After analyzing the discrepancy rate of MRI reports between different time periods, it remained stable with no statistical difference. We therefore believed that the inter-observer variation in the current study was limited because all the radiologists in this high-volume tertiary referral medical center were well experienced.

Conclusion
Among patients with locally advanced PC, treatment with RP had a higher risk of BCR but no signi cant difference in the risk of local recurrence, metastasis, and overall survival compared to RT plus LTADT therapy. This information may provide insight for clinicians and patients for decision-making regarding their preference for either treatment strategy. A further prospective study for patients with advanced PC would be necessary. For each treatment group, we generated Kaplan-Meier survival curves for biochemical recurrence (BCR) according to management method, and calculated a log-rank test to evaluate the association of BCR with different management method. We also compared the BCR free survival between all cohort patients and propensity-score matched patients.

Figure 2
Kaplan-Meier analysis of local recurrence free survival in propensity-score matched patients For each treatment group, we generated Kaplan-Meier survival curves for local recurrence according to management method, and we also calculated a log-rank test to evaluate the association of local recurrence with different management method among propensity-score matched patients.

Figure 3
Kaplan-Meier analysis of metastasis free survival in propensity-score matched patients For each treatment group, we generated Kaplan-Meier survival curves for metastasis according to management method, and we also calculated a log-rank test to evaluate the association of metastasis with different management method among propensity-score matched patients.

Figure 4
Kaplan-Meier analysis of overall survival in propensity-score matched patients For each treatment group, we generated Kaplan-Meier survival curves for overall survival according to management method, and we also calculated a log-rank test to evaluate the association of overall survival with different management method among propensity-score matched patients.