Advertisement

Genomic Prostate Cancer Classifier Predicts Biochemical Failure and Metastases in Patients After Postoperative Radiation Therapy

      Purpose

      To test the hypothesis that a genomic classifier (GC) would predict biochemical failure (BF) and distant metastasis (DM) in men receiving radiation therapy (RT) after radical prostatectomy (RP).

      Methods and Materials

      Among patients who underwent post-RP RT, 139 were identified for pT3 or positive margin, who did not receive neoadjuvant hormones and had paraffin-embedded specimens. Ribonucleic acid was extracted from the highest Gleason grade focus and applied to a high-density-oligonucleotide microarray. Receiver operating characteristic, calibration, cumulative incidence, and Cox regression analyses were performed to assess GC performance for predicting BF and DM after post-RP RT in comparison with clinical nomograms.

      Results

      The area under the receiver operating characteristic curve of the Stephenson model was 0.70 for both BF and DM, with addition of GC significantly improving area under the receiver operating characteristic curve to 0.78 and 0.80, respectively. Stratified by GC risk groups, 8-year cumulative incidence was 21%, 48%, and 81% for BF (P<.0001) and for DM was 0, 12%, and 17% (P=.032) for low, intermediate, and high GC, respectively. In multivariable analysis, patients with high GC had a hazard ratio of 8.1 and 14.3 for BF and DM. In patients with intermediate or high GC, those irradiated with undetectable prostate-specific antigen (PSA ≤0.2 ng/mL) had median BF survival of >8 years, compared with <4 years for patients with detectable PSA (>0.2 ng/mL) before initiation of RT. At 8 years, the DM cumulative incidence for patients with high GC and RT with undetectable PSA was 3%, compared with 23% with detectable PSA (P=.03). No outcome differences were observed for low GC between the treatment groups.

      Conclusion

      The GC predicted BF and metastasis after post-RP irradiation. Patients with lower GC risk may benefit from delayed RT, as opposed to those with higher GC; however, this needs prospective validation. Genomic-based models may be useful for improved decision-making for treatment of high-risk prostate cancer.
      Summary
      One of the main concerns with postprostatectomy RT is that outcomes can vary, resulting in inappropriate treatment. We assessed a validated genomic classifier (GC) to distinguish outcomes in men with high-risk prostate cancer. The GC was able to predict biochemical failure and distant metastasis after RT; significant differences between adjuvant and salvage RT were noted for those with intermediate or high GC scores but not for those with low GC scores.

      Introduction

      Every year, approximately 233,000 men are diagnosed with prostate cancer (PCa) in the United States, and more than 29,000 men die from this disease annually (
      • Siegel R.
      • Naishadham D.
      • Jemal A.
      Cancer statistics, 2013.
      ). The vast majority present with clinically localized disease, and many undergo radical prostatectomy (RP) as their primary treatment. Although RP is performed with curative intent, a proportion of these patients will develop PCa recurrence, particularly those with adverse pathologic features, defined as positive surgical margin, extraprostatic extension, seminal vesicle involvement (SVI), or detectable postprostatectomy prostate-specific antigen (PSA) (
      • Swanson G.P.
      • Hussey M.A.
      • Tangen C.M.
      • et al.
      Predominant treatment failure in postprostatectomy patients is local: Analysis of patterns of treatment failure in SWOG 8794.
      ). Recently, joint consensus guidelines from the American Urologic Association and the American Society for Radiation Oncology (
      • Valicenti R.K.
      • Thompson I.
      • Albertsen P.
      • et al.
      Adjuvant and salvage radiation therapy after prostatectomy: American Society for Radiation Oncology/American Urological Association guidelines.
      ) advocated for postprostatectomy adjuvant radiation in patients at risk for recurrence or with evidence of progression based on detectable PSA or clinical progression. However, the guidelines stressed the importance of a “thoughtful discussion” and a multidisciplinary approach to balance risks and benefits, reflecting the underlying controversy within the field.
      One of the main concerns with postprostatectomy radiation therapy (RT) is that outcomes can vary, and many men will be subjected to unnecessary adjuvant therapy. In the 3 prospective, randomized clinical trials examining adjuvant RT (defined as treatment at the time of undetectable PSA) (
      • Thompson I.M.
      • Tangen C.M.
      • Paradelo J.
      • et al.
      Adjuvant radiotherapy for pathological T3N0M0 prostate cancer significantly reduces risk of metastases and improves survival: Long-term followup of a randomized clinical trial.
      ,
      • Thompson I.M.
      • Tangen C.M.
      • Paradelo J.
      • et al.
      Adjuvant radiotherapy for pathologically advanced prostate cancer: A randomized clinical trial.
      ,
      • Wiegel T.
      • Bottke D.
      • Steiner U.
      • et al.
      Phase III postoperative adjuvant radiotherapy after radical prostatectomy compared with radical prostatectomy alone in pT3 prostate cancer with postoperative undetectable prostate-specific antigen: ARO 96-02/AUO AP 09/95.
      ,
      • Bolla M.
      • van Poppel H.
      • Collette L.
      • et al.
      Postoperative radiotherapy after radical prostatectomy: A randomised controlled trial (EORTC trial 22911).
      ), approximately 50% of patients randomized to observation never developed biochemical failure (BF), even though some patients had detectable PSA after RP. Thus, some clinicians would have advocated for salvage RT as a more selective approach rather than adjuvant RT. At present there is not yet level 1 evidence to support the hypothesis that adjuvant and early salvage RT are equivalent. Further, with salvage RT, 6-year BF-free survival is approximately 40%, and patients with BF have a 60% probability of developing distant metastasis (DM) and 20% probability of PCa-specific death within 10 years (
      • Trock B.J.
      • Han M.
      • Freedland S.J.
      • et al.
      Prostate cancer-specific survival following salvage radiotherapy vs observation in men with biochemical recurrence after radical prostatectomy.
      ). Thus, it is clear that even among a high-risk patient population based on standard clinical features, there remain a significant proportion of patients who may benefit from additional local therapy, whereas others may require systemic therapy. On the other hand, studies have shown that there also exist a significant proportion of patients for whom the disease progression is indolent and who derive little benefit from post-RP RT (
      • Trock B.J.
      • Han M.
      • Freedland S.J.
      • et al.
      Prostate cancer-specific survival following salvage radiotherapy vs observation in men with biochemical recurrence after radical prostatectomy.
      ,
      • D'Amico A.V.
      • Chen M.H.
      • Sun L.
      • et al.
      Adjuvant versus salvage radiation therapy for prostate cancer and the risk of death.
      ).
      Novel biomarkers that can improve our prognostic tools and inform decision making are acutely needed. One such potential platform is the Decipher® genomic classifier (GC) (GenomeDx Biosciences, San Diego, CA) that uses a whole-transcriptome microarray assay from formalin-fixed, paraffin-embedded PCa specimens. This signature was developed and validated as a predictor for clinical metastases after RP in a cohort of men with adverse features (
      • Erho N.
      • Crisan A.
      • Vergara I.A.
      • et al.
      Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy.
      ). Further, it was shown to more accurately predict metastases than individual clinical variables or nomograms (
      • Karnes R.J.
      • Bergstralh E.J.
      • Davicioni E.
      • et al.
      Validation of a genomic classifier that predicts metastasis following radical prostatectomy in an at risk patient population.
      ). Thus, we hypothesized that incorporation of the GC in clinical models would more accurately predict BF and DM in men receiving post-RP RT.

      Methods and Materials

      The Thomas Jefferson University institutional review board reviewed and approved the research protocol under which this validation study was conducted. The study met the PRoBE (
      • Pepe M.S.
      • Feng Z.
      • Janes H.
      • Bossuyt P.M.
      • Potter J.D.
      Pivotal evaluation of the accuracy of a biomarker used for classification or prediction: Standards for study design.
      ) and REMARK (
      • McShane L.M.
      • Altman D.G.
      • Sauerbrei W.
      • et al.
      Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics. REporting reccommendations for tumour MARKer prognostic studies (REMARK).
      ) criteria for prospective-blinded evaluation and analysis of a prognostic biomarker.

       Patient cohort

      The cohort comprised 143 patients with pT3 or margin-positive disease, who may have elevated post-RP PSA, and who underwent post-RP RT at Kimmel Cancer Center, Thomas Jefferson University between 1999 and 2009. Four patients who received neoadjuvant hormone therapy were excluded from the analysis. The GC scores were available for 139 patients only (Fig. e1, available online). Accounting for sample loss, there were no significant differences in the cohort with available GC data in comparison with the original selected cohort (data not shown). The radiation techniques including clinical target division definition, planning target volume generation, RT techniques, doses, and volumes have been described previously (
      • Mishra M.V.
      • Scher E.D.
      • Andrel J.
      • et al.
      Adjuvant versus salvage radiation therapy for prostate cancer patients with adverse pathologic features: Comparative analysis of long-term outcomes.
      ). Further, the use of hormone therapy and irradiation of pelvic lymph nodes was performed at the discretion of the treating radiation oncologist.

       Specimen selection and processing

      After histopathologic review of formalin-fixed, paraffin-embedded (FFPE) tumor blocks from each case by 2 expert genitourinary pathologists, the tumor block with the index lesion was selected for specimen processing. The index lesion was identified as the prostatectomy FFPE block with the highest pathologic Gleason grade, regardless of its volume. Two [x] 0.6-mm diameter tissue biopsy punch tool cores were sampled to enrich for tumor cells from the highest Gleason grade in the index lesion and placed in a microfuge tube for RNA extraction. The RNA extraction and microarray expression data generation were as previously described (
      • Erho N.
      • Crisan A.
      • Vergara I.A.
      • et al.
      Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy.
      ). After microarray quality control using the Affymetrix Power Tools packages (Santa Clara, CA) (
      • Lockstone H.E.
      Exon array data analysis using Affymetrix power tools and R statistical software.
      ), probeset summarization and normalization were performed by the SCAN algorithm (
      • Piccolo S.R.
      • Sun Y.
      • Campbell J.D.
      • et al.
      A single-sample microarray normalization method to facilitate personalized-medicine workflows.
      ), which normalizes each batch individually by modeling and removing probe- and array-specific background noise using only data from within each array (
      • Lockstone H.E.
      Exon array data analysis using Affymetrix power tools and R statistical software.
      ).

       Calculation of GC scores and nomogram scores

      The 22-marker GC was applied to the microarray expression data for each patient sample as previously described (
      • Erho N.
      • Crisan A.
      • Vergara I.A.
      • et al.
      Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy.
      ). Cut points for the GC were estimated using receiver operating characteristic curve–based methods described previously (
      • Karnes R.J.
      • Bergstralh E.J.
      • Davicioni E.
      • et al.
      Validation of a genomic classifier that predicts metastasis following radical prostatectomy in an at risk patient population.
      ). Cancer of the prostate risk assessment post-surgical (CAPRA-S) scores were calculated as described by Cooperberg et al (
      • Cooperberg M.R.
      • Hilton J.F.
      • Carroll P.R.
      The CAPRA-S score: A straightforward tool for improved prediction of outcomes after radical prostatectomy.
      ), and Stephenson 5-year survival probabilities were calculated using the online prediction tool (
      • Stephenson A.J.
      • Scardino P.T.
      • Eastham J.A.
      • et al.
      Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy.
      ).

       Statistical analyses

      Statistical analyses were performed in R, version 3.0, and all statistical tests were 2-sided using a 5% significance level. As previously described, for men who achieved an undetectable PSA level after surgery, BF was defined as a PSA ≥0.4 ng/mL with a subsequent confirmation. For all other patients, including those with a detectable pre-RT PSA, BF was defined as 3 increases in PSA measured at least 6 weeks apart, considering the first PSA rise dated as the BF; or androgen deprivation therapy due to PSA rise (
      • Mishra M.V.
      • Scher E.D.
      • Andrel J.
      • et al.
      Adjuvant versus salvage radiation therapy for prostate cancer patients with adverse pathologic features: Comparative analysis of long-term outcomes.
      ). Undetectable PSA and detectable PSA were defined as PSA levels of ≤0.2 and >0.2 ng/mL immediately before initiation of RT, respectively. Metastatic failure was defined as DM documented in clinical notes and imaging reports. In time-to-event analyses, event times were defined as the time from RT completion date to BF or DM date.
      The χ2, Wilcoxon, or Fisher exact test were used to test for association between categoric variables. Area under receiver operating characteristic curves (AUC), calibration, and Hosmer-Lemeshow goodness of fit, strip plots, univariable, and multivariable (MVA) Cox proportional hazard models were used to compare the GC-based and clinical-only models for predicting BF or DM after RT. The LASSO method for the Cox model was used for identification of the most predictive variables as described previously (
      • Tibshirani R.
      The LASSO method for variable selection in the Cox model.
      ). Cumulative incidence curves were constructed using Fine-Gray competing risks analysis (
      • Fine J.P.
      • Gray R.J.
      A proportional hazards model for the subdistribution of a competing risk.
      ) to estimate the risk of BF or DM while accounting for censoring and death due to other causes.

      Results

      A total of 143 evaluable patients had available archival FFPE blocks for genome-wide expression analysis. Genomic classifier scores were generated for 139 patients (97%) (Fig. e1). Eighty-four percent had Gleason score ≥7 or above; 27% with Gleason 8-10. Eighty-two percent of men had extraprostatic extension, 38% had seminal vesicle invasion, and 75% had positive margins. Fifty-three percent of patients had radiation initiated when PSA was ≤0.2 ng/mL, and 21% received radiation with hormonal therapy. The median follow-up times after RP and after RT were 11.8 and 7.4 years, respectively. After RT, 54 patients (39%) experienced BF, and 10 (7%) developed DM on follow-up (Table 1).
      Table 1Patient characteristics
      Clinicopathologic variableAllNo biochemical failure (column %)Biochemical failure
      Only 10 patients developed distant metastasis among patients with biochemical failure.
      (column %)
      P
      Total N1398554
      Age (y).93
      Using Pearson χ2 test.
       Median606061
       Range(40-76)(40-76)(47-76)
      Race.13
      Using Fisher exact test.
       Caucasian118 (84.9)68 (80)50 (93)
       Black18 (13.0)14 (17)4 (7)
       Hispanic3 (2.2)3 (3)0
      Pre-RP PSA (ng/mL).03
      Using Wilcoxon rank sum test.
       <1090 (64.7)61 (72)29 (53)
       10-2027 (19.4)14 (16)13 (24)
       >2015 (10.8)6 (7)9 (17)
       Unknown7 (5.0)4 (5)3 (6)
      EPE.18
      Using Pearson χ2 test.
       Positive114 (82.0)66 (78)48 (89)
       Unknown1 (0.7)1 (1)0
      SVI.16
      Using Pearson χ2 test.
       Positive53 (38.1)28 (33)25 (46)
      Margin1
      Using Pearson χ2 test.
       Positive105 (75.5)64 (75)41 (76)
      Gleason score.03
      Using Fisher exact test.
       ≤621 (15.1)19 (22)2 (4)
       779 (56.9)49 (58)30 (56)
       8-1038 (27.3)16 (19)22 (40)
       Unknown1 (0.7)1 (1)0
      Pre-RT PSA (ng/mL).008
      Using Fisher exact test.
       ≤0.274 (53.0)51 (60)23 (43)
       >0.2-136 (26.0)22 (26)14 (26)
       >1-518 (13.0)7 (8)11 (20)
       > 58 (5.8)2 (2)6 (11)
       Unknown3 (2.2)3 (4)0
      Time from RP to RT (mo).51
      Using Pearson χ2 test.
       Median4.574.534.64
       Range(1.08-159.67)(1.77-159.67)(1.08-60.91)
      Dose1
      Using Pearson χ2 test.
       Median66.666.666.6
       Range(45-72)(45-72)(60-70.2)
      Field.60
      Using Pearson χ2 test.
       Fossa only110 (79.1)69 (81)41 (76)
       Whole pelvic20 (20.9)16 (19)13 (24)
      ADT.025
      Using Pearson χ2 test.
       Positive29 (20.9)12 (14)17 (32)
      Abbreviations: ADT = androgen deprivation therapy; EPE = extraprostatic extension; PSA = prostate-specific antigen; RP = radical prostatectomy; RT = radiation therapy; SM = surgical margin; SVI = seminal vesicle involvement.
      Only 10 patients developed distant metastasis among patients with biochemical failure.
      Using Pearson χ2 test.
      Using Fisher exact test.
      § Using Wilcoxon rank sum test.
      The distribution of men among low (<0.4), intermediate (0.4-0.6), and high (>0.6) GC risk categories was 41%, 38%, and 22%, respectively (Fig. e2). Nearly all patients who developed DM on follow up (n=10) had intermediate or high GC scores (ie, ≥0.4), except for 1 patient with a borderline GC score (0.395). The GC scores increased with higher Gleason score and tumor stage (Fig. e3). Genomic classifier has a modest correlation to Gleason scores (r2=0.29, P=.0004) and tumor stage (r2=0.15, P=.07). The agreement between observed BF and GC scores demonstrates virtually perfect calibration of GC for predicting BF, with a slope of 1.1 and a Hosmer-Lemeshow P value of .77 (Fig. e4A). We did not observe as good calibration for the post-RT BF endpoint with the Stephenson model (Fig. e4B). Too few events were available to evaluate calibration for the DM endpoint.
      Receive operator characteristic curve analysis was used to determine whether GC could improve prediction of outcome as compared with commonly used clinical risk prediction models for discrimination of BF and DM events (Fig. 2A, B ). The AUC for the post-RP Stephenson nomogram was 0.70 (95% confidence interval [CI] 0.61-0.79) and 0.70 (95% CI 0.49-0.90) for BF and DM endpoints, respectively. For CAPRA-S, the AUC was 0.67 (95% CI 0.58-0.77) and 0.65 (0.44-0.86) for BF and DM endpoints, respectively. Note that neither clinical nomogram was significantly superior to chance in predicting DM because the 95% CI included the AUC of 0.5 for a random model. The AUC for the GC score was 0.75 (95% CI 0.67-0.84) and 0.78 (95% CI 0.64-0.91) for BF and DM endpoints, respectively. Combining the GC with the Stephenson nomogram improved the AUC to 0.78 (95% CI 0.69-0.86) and 0.80 (95% CI 0.68-0.93) for BF and DM, respectively (Fig. 1). A similar improvement in AUC was noted for combining GC with CAPRA-S.
      Figure thumbnail gr1
      Fig. 1Area under the receiver operating characteristic curve (AUC): comparison of genomic classifier (GC)-based and clinical-only risk models for predicting biochemical failure (A) and distant metastasis (B) after postoperative radiation therapy. CAPRA-S = cancer of the prostate risk assessment post-surgical score; CI = confidence interval.
      Figure thumbnail gr2
      Fig. 2Cumulative incidence plots of biochemical failure (A) and distant metastasis (B) for low-, intermediate-, and high-risk genomic classifier (GC) score groups. Cut points were reported previously
      (
      • Parker C.
      • Nilsson S.
      • Heinrich D.
      • et al.
      Alpha emitter radium-223 and survival in metastatic prostate cancer.
      )
      .
      Decision curve analysis was used to determine the clinical utility of the gain in AUC for the GC-based models (Fig. e5). Compared with scenarios in which no prediction model would be used for a postoperative RT treatment decision (ie, “treat all” or “treat none”), the GC-based models had a higher net benefit than clinical models across a wide range of decision threshold probabilities (approximately 20%-75% risk of BF).
      Cumulative incidence plots for the probability of BF and DM show significance for 3 previously reported GC score risk groups (Fig. 2). The 4-year cumulative incidence of BF in patients with low, intermediate, and high GC scores was 13%, 31%, and 49%, respectively (Fig. 2A). By 8 years after RT the difference in BF incidence rates became more pronounced, with cumulative incidence rates of 21%, 48%, and 81% for low, intermediate, and high GC score, respectively (P<.0001). The 8-year cumulative incidence rates of DM were 0, 12%, and 17% for the GC score groups. The incidence rates were signficant (P=.032) despite the small number of DM events on follow-up in this cohort (Fig. 2B).
      Univariable analysis demonstrated that GC and a number of clinical factors such as pre-RP PSA level, seminal vesicle involvement, Gleason score, timing of RT (ie, undetectable vs detectable PSA), radiation dose, and concomitant hormone therapy were all significant predictors of BF (Tables e1 and e2). Only GC and pre-RP PSA level were also significant for DM. In MVA analysis GC, pre-RP PSA level, pathologic Gleason score, and PSA level prior to RT remained significant predictors of BF (Table 2). Again, only GC and pre-RP PSA level were significant for DM. The hazard ratio (HR) for intermediate and high GC was 2.9 and 8.1 in comparison with the low GC risk group (Table 2). The HR estimates for the DM endpoint were only significant for high GC (HR 14.3, P=.005, although because of a small number of events it has a wide confidence interval) (
      • Peduzzi P.
      • Concato J.
      • Feinstein A.R.
      • et al.
      Importance of events per independent variable in proportional hazards regression analysis. II. Accuracy and precision of regression estimates.
      ). Further, we validated the findings from the multiple regression model using penalized regression to ensure that the significance of GC was not an artifact of few metastasis events in the MVA analysis. For both BF and DM, GC was the top variable with a non-zero coefficient, confirming that GC is the most significant variable and that the MVA analysis was robust (Fig. e6).
      Table 2Multivariable Cox proportional hazards analysis of risk factors for postoperative radiation treatment biochemical failure and distant metastasis
      Risk factorBiochemical failureDistant metastasis
      Firth's penalized likelihood method was used due to low number of metastatic events.
      HR (95% CI)PHR (95% CI)P
      GC Intermediate (Ref: GC < 0.4)2.88 (1.21-6.85).022.15 (0.18-39.48).55
      GC High (Ref: GC < 0.4)8.13 (3.40-19.46)<.000114.28 (2.13-210.38).005
      Age at RP1.02 (0.96-1.07).571.01 (0.90-1.15).9
      Caucasian Race (Ref: AAM/Hispanic)2.31 (0.73-7.31).160.39 (0.05-5.34).42
      (Pre-RP PSA) (log2)1.49 (1.06-2.10).022.69 (1.33-5.65).007
      EPE (Ref: Negative)2.04 (0.73-5.66).172.72 (0.21-505).53
      SVI (Ref: Negative)1.08 (0.50-2.32).850.54 (0.07-2.96).49
      SM (Ref: Negative)0.68 (0.31-1.46).322 (0.28-18.45).49
      Pathologic Gleason score (Ref: ≤7)2.21 (1.07-4.56).032.13 (0.30-14.99).4
      Detectable PSA (Ref: Undetectable PSA)3.23 (1.49-6.98).0030.92 (0.08-10.42).91
      Time between RP and RT (mo)0.98 (0.96-1.01).171.03 (0.97-1.09).23
      Radiation dose1.05 (0.91-1.22).491.24 (0.83-1.92).28
      Pelvis radiation (Ref: Fossa only)0.61 (0.25-1.48).282.1 (0.31-15.93).44
      Concomitant hormone therapy (Ref: None)1.14 (0.52-2.49).741.44 (0.26-7.24).66
      Abbreviations:AAM = African American men; CI = confidence interval; GC = genomic classifier; HR = hazard ratio; Ref = referent. Other abbreviations as in Table 1.
      Firth's penalized likelihood method was used due to low number of metastatic events.
      Exploratory analyses were performed to determine whether GC could predict benefit between those treated with RT with either undetectable or detectable PSA. We did not observe the 2 RT groups to differ significantly for any clinical or treatment variable aside from more concomitant hormone therapy and a higher proportion of African Americans in the detectable PSA group (Table e4). The median time from RP to initiation of RT was 4.21 months (interquartile range, 3.4-5.9) and 6.8 months (interquartile range, 3.9-25.8) for undetectable and detectable PSA groups, respectively. Within each GC score group, for patients with BF GC was higher than for patients who did not experience BF, regardless of when RT was initiated (Fig. e7). Cumulative incidence plots for BF and DM, comparing undetectable with detectable PSA RT groups, were stratified by GC risk (Fig. 3). The intermediate and high GC score groups were collapsed into 1 group owing to limitations in sample size for this subset analysis. Within the low GC score group (GC <0.4) there was no difference in cumulative incidence of BF (Fig. 3A) or DM (Fig. 3B) for patients who had RT with undetectable or detectable PSA. However, for the group with GC ≥ 0.4, a 27% absolute difference in BF was observed at 8 years, with a median 4-year PSA-free survival advantage for patients who received RT with undetectable PSA compared with those with detectable PSA (Fig. 3A). Patients with GC ≥ 0.4 who had detectable PSA when RT was initiated had a DM cumulative incidence of 23% by 8 years, compared with just 3% for patients with undetectable PSA (Fig. 3B; P=.03). Similar results were obtained for both BF and DM endpoints in a sensitivity analysis when considering different PSA level thresholds (eg, ≤0.2, 0.21-1, >1.0 ng/mL) at RT initiation (Fig. e8). Finally, in MVA analysis we were able to estimate the hazards at any time point for BF after completion of RT and found that patients with GC ≥ 0.4 who had RT with detectable PSA had an HR of 2.2 (P=.01) in comparison with patients with GC ≥ 0.4 who had RT with undetectable PSA (Table 3). The same trend was observed for the DM endpoint, with an HR of 7.1 (P=.07). For GC low risk patients, no significant differences in hazards were observed for either endpoint when comparing RT initiated with undetectable and detectable PSA.
      Figure thumbnail gr3
      Fig. 3Cumulative incidence plots of biochemical failure and distant metastasis comparing patients treated with radiation therapy (RT) when prostate-specific antigen (PSA) was undetectable (A, C) and detectable (B, D), as stratified by genomic classifier (GC) score risk groups.
      Table 3Multivariable Cox proportional hazards analysis of risk prediction models for postoperative radiation treatment biochemical failure and distant metastasis
      Risk factorMultivariable analysis (BF)Multivariable analysis (DM)
      HR (95% CI)PHR (95% CI)P
      Detectable PSA (Ref: Undetectable PSA)
       GC low-risk subset (<0.4)2 (0.56-7.12).29NA
      Cox regression model does not converge, owing to low number of events (events = 1 out of 10).
      NA
       GC high-risk subset (≥0.4)2.24 (1.19-4.22).017.12 (0.89-57.09).07
      Abbreviation: NA = not applicable. Other abbreviations as in Table 1, Table 2.
      Cox regression model does not converge, owing to low number of events (events = 1 out of 10).

      Discussion

      Despite the publication of the American Urologic Association–American Society for Radiation Oncology consensus guidelines and 3 phase 3 prospective trials demonstrating the benefit of adjuvant therapy as opposed to observation, both in terms of biochemical progression–free survival as well as overall survival (Southwest Oncology Group), there remains controversy regarding the administration of postprostatectomy radiation. Two critical factors are that only 50% of patients with adverse pathologic features treated with surgery alone will develop BF, and even among those patients with a persistently elevated PSA after prostatectomy, approximately 10% may never develop clinical metastases (
      • Shinghal R.
      • Yemoto C.
      • McNeal J.E.
      • et al.
      Biochemical recurrence without PSA progression characterizes a subset of patients after radical prostatectomy. Prostate-specific antigen.
      ). Given the possibility of overtreatment, many clinicians are hesitant to initiate additional therapy (
      • Thompson I.M.
      • Tangen C.M.
      • Klein E.A.
      Is there a standard of care for pathologic stage T3 prostate cancer?.
      ). Currently, 4 prospective randomized clinical trials—Radiotherapy and Androgen Deprivation in Combination After Local Surgery (RADICALS), Radiotherapy Adjuvant versus Early Salvage (RAVES), GETUG-17, and European Organization for Research and Treatment of Cancer 22,043-30,031—are investigating the therapeutic benefit of early salvage RT with or without androgen deprivation therapy compared with adjuvant RT.
      This study provides critical insight into disease aggressiveness, which can have major ramifications for the interpretation of these forthcoming trials. Herein we provide the first validation of the GC score in the postprostatectomy RT setting. We demonstrated that within a group of patients with high-risk features for BF and development of DM, the GC score was able to differentiate outcomes. The GC score improved risk stratification above known clinical classifiers, specifically in terms of development of DM. Moreover, our data suggest that for patients with low GC scores outcomes were not different for men receiving adjuvant versus salvage RT; for those men with intermediate- or high-risk GC the timing of RT initiation in terms of PSA levels significantly altered BF and DM survival outcomes.
      Given the high cumulative incidence of BF and DM in particular, for the high GC score patient group, this is a patient population in whom exploration of intensification of therapy would be warranted. In an unselected patient population, postprostatectomy radiation combined with antiandrogen therapy has been demonstrated to improve freedom from BF as well as reduce the incidence of metastatic PCa (
      • Shipley W.U.
      • Hunt D.
      • Lukka H.
      • et al.
      Initial report of RTOG 9601: A phase III trial in prostate cancer: Anti-androgen therapy (AAT) with bicalutamide during and after radiation therapy (RT) improves freedom from progression and reduces the incidence of metastatic disease in patients following radical prostatectomy (RP) with pT2-3, N0 disease, and elevated PSA levels (Abstr.).
      ) in comparison with radiation monotherapy. A subsequent phase 2 trial examining adjuvant docetaxel after androgen deprivation and RT for high-risk postprostatectomy patients is closed, with results expected within the next several years. With the plethora of new agents approved for metastatic castrate-resistant PCa (
      • Fizazi K.
      • Scher H.I.
      • Molina A.
      • et al.
      Abiraterone acetate for treatment of metastatic castration-resistant prostate cancer: Final overall survival analysis of the COU-AA-301 randomised, double-blind, placebo-controlled phase 3 study.
      ,
      • Scher H.I.
      • Fizazi K.
      • Saad F.
      • et al.
      Increased survival with enzalutamide in prostate cancer after chemotherapy.
      ,
      • De Bono J.S.
      • Logothetis C.J.
      • Molina A.
      • et al.
      Abiraterone and increased survival in metastatic prostate cancer.
      ,
      • Stock P.G.
      • Gores P.F.
      • Kaufmann D.B.
      • et al.
      Immunogenic potential of passenger leucocyte depleted murine islets in-vitro and in-vivo.
      ,
      • Parker C.
      • Nilsson S.
      • Heinrich D.
      • et al.
      Alpha emitter radium-223 and survival in metastatic prostate cancer.
      ,
      • Kantoff P.W.
      • Higano C.S.
      • Shore N.D.
      • et al.
      Sipuleucel-T immunotherapy for castration-resistant prostate cancer.
      ) on the basis of improvements in overall survival, there are a variety of combinations that can be examined clinically. As such, defining those patients most at risk for local treatment failure is critical because many of these therapies have potential morbidity associated with them.
      There are several limitations in the present study. First, there were few DM noted in this patient population. As such, full analysis of DM as an endpoint was limited. Further, this is a retrospective analysis from a single institution, and this data set only included patients receiving RT, and as such, examining whether outcomes from patients treated with observation alone would mimic those treated with RT, particularly for those patients with low GC score, is beyond the scope of this analysis. In addition, some patients received early adjuvant RT, whereas others received later salvage therapy, according to both physician as well as patient preference. The use of androgen deprivation therapy was not universal and reflected inherent biases among the treating physicians. However, this does reflect current treatment practices.
      The data presented in this study confirm and extend the prior publications demonstrating the predictive capability and utility of the GC score (
      • Erho N.
      • Crisan A.
      • Vergara I.A.
      • et al.
      Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy.
      ,
      • Karnes R.J.
      • Bergstralh E.J.
      • Davicioni E.
      • et al.
      Validation of a genomic classifier that predicts metastasis following radical prostatectomy in an at risk patient population.
      ) beyond clinicopathologic parameters. The use of a genomic signature has been demonstrated to alter clinical decision making in approximately 50% of cases (
      • Badani K.
      • Thompson D.J.S.
      • et al.
      Impact of a genomic classifier of metastatic risk on postoperative treatment recommendations for prostate cancer patients: A report from the DECIDE study group.
      ), indicating the potential broad clinical impact. Further, both GC score and clinical factors independently associated with BF and DM in MVA analysis, implying that genomic risk classifiers can be incorporated with added value into clinical care. Furthermore, when translating genomic risk classifiers into clinical practice, it is important that the added information be supplemented with traditional clinical parameters to guide decision making. It is expected that low-risk GC scores in patients with adverse pathology could be used to guide treatment decisions for possibly delayed or deferred irradiation when risk of failure is low. This may benefit patients incontinent of urine, who require additional time to heal after RP or those in whom daily radiation may be burdensome. Given the increasing incidence and costs of PCa treatments (
      • Nguyen P.L.
      • Gu X.
      • Lipsitz S.R.
      • et al.
      Cost implications of the rapid adoption of newer technologies for treating prostate cancer.
      ), avoidance of unnecessary or inadequate treatment would have major implications for the healthcare system and limit the potential overtreatment of patients who will not recur or progress. These observations are particularly salient given the ongoing randomized controlled trials of adjuvant versus early salvage radiation therapy because our results demonstrate 2 distinct groups of patients, who are currently not stratified in the trial designs.

      Acknowledgments

      The authors thank Dr Darby J. S. Thompson (EMMES Canada) and Dr Jason Alter for useful comments and advice relating to study design and analysis; and Dr Christine Buerki, Marguerite du Plessis, Hugh Wellman (GenomeDx), and Dr Firas Abdollah (Vattikuti Institute) for assistance in preparation and revision of the manuscript.

      Supplementary Data

      References

        • Siegel R.
        • Naishadham D.
        • Jemal A.
        Cancer statistics, 2013.
        CA Cancer J Clin. 2013; 63: 11-30
        • Swanson G.P.
        • Hussey M.A.
        • Tangen C.M.
        • et al.
        Predominant treatment failure in postprostatectomy patients is local: Analysis of patterns of treatment failure in SWOG 8794.
        J Clin Oncol. 2007; 25: 2225-2229
        • Valicenti R.K.
        • Thompson I.
        • Albertsen P.
        • et al.
        Adjuvant and salvage radiation therapy after prostatectomy: American Society for Radiation Oncology/American Urological Association guidelines.
        Int J Radiat Oncol Biol Phys. 2013; 86: 822-828
        • Thompson I.M.
        • Tangen C.M.
        • Paradelo J.
        • et al.
        Adjuvant radiotherapy for pathological T3N0M0 prostate cancer significantly reduces risk of metastases and improves survival: Long-term followup of a randomized clinical trial.
        J Urol. 2009; 181: 956-962
        • Thompson I.M.
        • Tangen C.M.
        • Paradelo J.
        • et al.
        Adjuvant radiotherapy for pathologically advanced prostate cancer: A randomized clinical trial.
        JAMA. 2006; 296: 2329-2335
        • Wiegel T.
        • Bottke D.
        • Steiner U.
        • et al.
        Phase III postoperative adjuvant radiotherapy after radical prostatectomy compared with radical prostatectomy alone in pT3 prostate cancer with postoperative undetectable prostate-specific antigen: ARO 96-02/AUO AP 09/95.
        J Clin Oncol. 2009; 27: 2924-2930
        • Bolla M.
        • van Poppel H.
        • Collette L.
        • et al.
        Postoperative radiotherapy after radical prostatectomy: A randomised controlled trial (EORTC trial 22911).
        Lancet. 2012; 366: 572-578
        • Trock B.J.
        • Han M.
        • Freedland S.J.
        • et al.
        Prostate cancer-specific survival following salvage radiotherapy vs observation in men with biochemical recurrence after radical prostatectomy.
        JAMA. 2008; 299: 2760-2769
        • D'Amico A.V.
        • Chen M.H.
        • Sun L.
        • et al.
        Adjuvant versus salvage radiation therapy for prostate cancer and the risk of death.
        BJU Int. 2010; 106: 1618-1622
        • Erho N.
        • Crisan A.
        • Vergara I.A.
        • et al.
        Discovery and validation of a prostate cancer genomic classifier that predicts early metastasis following radical prostatectomy.
        PLoS One. 2013; 8: e66855
        • Karnes R.J.
        • Bergstralh E.J.
        • Davicioni E.
        • et al.
        Validation of a genomic classifier that predicts metastasis following radical prostatectomy in an at risk patient population.
        J Urol. 2013; 190: 2047-2053
        • Pepe M.S.
        • Feng Z.
        • Janes H.
        • Bossuyt P.M.
        • Potter J.D.
        Pivotal evaluation of the accuracy of a biomarker used for classification or prediction: Standards for study design.
        J Natl Cancer Inst. 2008; 100: 1432-1438
        • McShane L.M.
        • Altman D.G.
        • Sauerbrei W.
        • et al.
        Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics. REporting reccommendations for tumour MARKer prognostic studies (REMARK).
        Br J Cancer. 2005; 93: 387-391
        • Mishra M.V.
        • Scher E.D.
        • Andrel J.
        • et al.
        Adjuvant versus salvage radiation therapy for prostate cancer patients with adverse pathologic features: Comparative analysis of long-term outcomes.
        Am J Clin Oncol. 2013 Mar 26; ([Epub ahead of print])
        • Lockstone H.E.
        Exon array data analysis using Affymetrix power tools and R statistical software.
        Brief Bioinform. 2011; 12: 634-644
        • Piccolo S.R.
        • Sun Y.
        • Campbell J.D.
        • et al.
        A single-sample microarray normalization method to facilitate personalized-medicine workflows.
        Genomics. 2012; 100: 337-344
        • Cooperberg M.R.
        • Hilton J.F.
        • Carroll P.R.
        The CAPRA-S score: A straightforward tool for improved prediction of outcomes after radical prostatectomy.
        Cancer. 2011; 117: 5039-5046
        • Stephenson A.J.
        • Scardino P.T.
        • Eastham J.A.
        • et al.
        Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy.
        J Clin Oncol. 2008; 23: 7005-7012
        • Tibshirani R.
        The LASSO method for variable selection in the Cox model.
        Stat Med. 1997; 16: 385-395
        • Fine J.P.
        • Gray R.J.
        A proportional hazards model for the subdistribution of a competing risk.
        J Am Stat Assoc. 1999; 94: 496-509
        • Peduzzi P.
        • Concato J.
        • Feinstein A.R.
        • et al.
        Importance of events per independent variable in proportional hazards regression analysis. II. Accuracy and precision of regression estimates.
        J Clin Epidemiol. 1995; 48: 1503-1510
        • Shinghal R.
        • Yemoto C.
        • McNeal J.E.
        • et al.
        Biochemical recurrence without PSA progression characterizes a subset of patients after radical prostatectomy. Prostate-specific antigen.
        Urology. 2003; 61: 380-385
        • Thompson I.M.
        • Tangen C.M.
        • Klein E.A.
        Is there a standard of care for pathologic stage T3 prostate cancer?.
        J Clin Oncol. 2009; 27: 2898-2899
        • Shipley W.U.
        • Hunt D.
        • Lukka H.
        • et al.
        Initial report of RTOG 9601: A phase III trial in prostate cancer: Anti-androgen therapy (AAT) with bicalutamide during and after radiation therapy (RT) improves freedom from progression and reduces the incidence of metastatic disease in patients following radical prostatectomy (RP) with pT2-3, N0 disease, and elevated PSA levels (Abstr.).
        Int J Radiat Oncol. 2010; 78: S27
        • Fizazi K.
        • Scher H.I.
        • Molina A.
        • et al.
        Abiraterone acetate for treatment of metastatic castration-resistant prostate cancer: Final overall survival analysis of the COU-AA-301 randomised, double-blind, placebo-controlled phase 3 study.
        Lancet Oncol. 2012; 13: 983-992
        • Scher H.I.
        • Fizazi K.
        • Saad F.
        • et al.
        Increased survival with enzalutamide in prostate cancer after chemotherapy.
        N Engl J Med. 2012; 367: 1187-1197
        • De Bono J.S.
        • Logothetis C.J.
        • Molina A.
        • et al.
        Abiraterone and increased survival in metastatic prostate cancer.
        N Engl J Med. 2011; 364: 1995-2005
        • Stock P.G.
        • Gores P.F.
        • Kaufmann D.B.
        • et al.
        Immunogenic potential of passenger leucocyte depleted murine islets in-vitro and in-vivo.
        Horm Metab Res Suppl. 1990; 25: 88-89
        • Parker C.
        • Nilsson S.
        • Heinrich D.
        • et al.
        Alpha emitter radium-223 and survival in metastatic prostate cancer.
        N Engl J Med. 2013; 369: 213-223
        • Kantoff P.W.
        • Higano C.S.
        • Shore N.D.
        • et al.
        Sipuleucel-T immunotherapy for castration-resistant prostate cancer.
        N Engl J Med. 2010; 363: 411-422
        • Badani K.
        • Thompson D.J.S.
        • et al.
        Impact of a genomic classifier of metastatic risk on postoperative treatment recommendations for prostate cancer patients: A report from the DECIDE study group.
        Oncotarget. 2013; 4: 600-609
        • Nguyen P.L.
        • Gu X.
        • Lipsitz S.R.
        • et al.
        Cost implications of the rapid adoption of newer technologies for treating prostate cancer.
        J Clin Oncol. 2011; 29: 1517-1524

      Comments

      Commenting Guidelines

      To submit a comment for a journal article, please use the space above and note the following:

      • We will review submitted comments as soon as possible, striving for within two business days.
      • This forum is intended for constructive dialogue. Comments that are commercial or promotional in nature, pertain to specific medical cases, are not relevant to the article for which they have been submitted, or are otherwise inappropriate will not be posted.
      • We require that commenters identify themselves with names and affiliations.
      • Comments must be in compliance with our Terms & Conditions.
      • Comments are not peer-reviewed.