Performance and Utility of Prognostic Genomic Biomarkers After Prostatectomy: Decipher-ing the Data
Since the initial appreciation of Gleason grading patterns more than 50 years ago,1 there has been a paucity of validated clinical grade biomarkers that independently improve prognostication over standard clinicopathologic variables for men with localized prostate cancer. Recently, multiple commercial prognostic genomic biomarkers have been developed and validated (eg, Decipher [GenomeDx Biosciences Laboratory, San Diego, CA], Oncotype DX [Genomic Health, Redwood City, CA] and Prolaris [Myriad Genetics, Salt Lake City, UT) for prostate cancer that are now tabulated within National Comprehensive Cancer Network guidelines, supported through the Molecular Diagnostic Services Program (MoIDX; Palmetto GBA, Columbia, SC), and covered through Medicare.2 Ultimately, given the high potential for recurrence in a poorly defined subset of men after radical prostatectomy, personalization of treatment is an unmet need in this population.
We thank D’Amico3 for his expert overview regarding the potential benefit of having improved prognostic biomarkers in localized prostate cancer. I could not agree more that the critical issue whenever a new prognostic biomarker is to be adopted is to understand the performance of the test and in what patient population it will provide benefit. Our meta-analysis demonstrated that Decipher was independently associated with a 30% increase in distant metastasis per 0.1 unit on the scale from 0 to 1.0.4 This was independent of Gleason score, margin status, extracapsular extension, seminal vesicle invasion, and lymph node invasion (P < .001).
I wanted to clarify some of the assumptions put forth by D’Amico3 to aid the reader in gaining a more comprehensive understanding of the prognostic ability and utility of genomic biomarkers in prostate cancer.
First, the assumption that most patients with aggressive clinicopathologic features have a high genomic metastatic risk is incorrect. Figure 2 in the article by Spratt et al4 demonstrates that only 26% of patients with clinicopathologic high-risk features, such as lymph node invasion, seminal vesicle invasion, or Gleason score of 8 to 10, would be classified genomically as Decipher high risk. Conversely, 74% of patients with clinicopathologic high-risk disease would be reclassified as low or intermediate genomic risk, of whom the majority would be classified as low genomic risk.
Second, D’Amico3 stated that future studies should be conducted on more clinically favorable-risk patients (eg, Gleason score of 3 + 4, pT2, or margin-negative disease). Fortunately, these data are already available, and 35% of the cohort (n = 300) included such patients without any clinicopathologic high-risk features. In these men, Decipher still demonstrated the independent ability to predict development of distant metastasis (hazard ratio [HR], 1.42; 95% CI, 1.02 to 1.97).
Third, D’Amico3 concluded that only the 16.5% of Decipher high-risk patients would derive benefit from using the biomarker in the postoperative setting. This logic is analogous to concluding that Gleason score should not be used because only 14% of patients had a Gleason score of 9 or 10. A prognostic biomarker must be analyzed on all patients to identify not only the high-risk patients, but also the patients who we thought had aggressive disease but, in fact, do not. I view at least three scenarios where improved prognostication could benefit our patients, as follows: classifying clinically high-risk patients as genomically high risk to warrant further treatment intensification; reclassifying clinically high-risk patients as lower genomic risk to avoid overtreatment; and reclassifying low- or intermediate-risk patients as having genomic high-risk disease to warrant treatment intensification. In total, genomic testing would provide prognostic benefit for at least 575 (67.3%) of 855 patients in our meta-analysis, not 16.5%.
Fourth, D’Amico3 state that “adjusting for use of adjuvant and/or salvage RT [radiotherapy] and ADT [androgen-deprivation therapy] reduced the magnitude of the point estimate of the AHR [adjusted HR] of the Decipher score in the overall study cohort as well as in the intermediate-risk and high-risk subgroups.” What is not mentioned is that the point estimate was reduced for the majority of variables and even to a greater degree for Gleason score. Again, does this mean Gleason score should not be used? Furthermore, D’Amico’s point ignores the fact that for the majority of the men (61%) who received radiotherapy or androgen-deprivation therapy it was salvage treatment, so adjusting for it was essentially adjusting for biochemical recurrence. Therefore, salvage treatment was an intermediate variable, and it is well known that adjusting for an intermediate variable biases the HR toward the null and reduces power.5 Regardless, we demonstrate significant performance of Decipher in nearly every salvage and adjuvant therapy context in Figure 4 on univariable analysis.4 Furthermore, and more importantly, the multivariable analysis was statistically significant and nearly identical to the original model (HR of 1.24 instead of 1.30; both P < .001) when adjusting for adjuvant or salvage therapy in a time-dependent model.
Fifth, D’Amico3 stated that the “primary end point of metastasis may have been subject to ascertainment bias.” This is an important point, and I want to clarify the likely extent and impact of this bias within our study. The strength of a meta-analysis is the ability to capture the extent of heterogeneity across cohorts given potential differences in practice patterns. Our meta-analysis4 demonstrated an I2 of 0% heterogeneity across the five cohorts that spanned academia to community hospitals. Furthermore, it is unlikely that differences in how metastases were diagnosed would be correlated with Decipher, so any differences would likely lead to random misclassification, again biasing the results toward the null. So the effect, if any, of ascertainment differences would likely mean the observed HR is an underestimate.
Therefore, we respectfully disagree with the conclusion of D’Amico3 to not even further test the genomic classifier in prospective adjuvant trials. Prospective testing is necessary to further reduce potential forms of bias. It also is unclear what further data are needed to begin prospective biomarker investigation other than having > 800 patients with whole transcriptomic data with long-term follow-up that consistently demonstrated an independent prognostic ability for a clinically meaningful end point of distant metastasis across five cohorts.
Although it is easy to reject change in prostate cancer,3 rejection of new advancements will ultimately impact our patients by preventing necessary progress. Genomics contains added information over currently used prognostic variables in localized prostate cancer. This is what our meta-analysis has demonstrated.4 Fortunately, the NRG Oncology group, among other groups, is supporting the continued research of genomic biomarkers in the adjuvant therapy trial GU-002 (ClinicalTrials.gov identifier: NCT03070886), which is prospectively using Decipher as a stratification variable.
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.
Consulting or Advisory Role: Dendreon
|1.||Gleason DF, Mellinger GT: Prediction of prognosis for prostatic adenocarcinoma by combined histological grading and clinical staging. J Urol 111:58-64, 1974 Crossref, Medline, Google Scholar|
|2.||Mohler JL, Armstrong AJ, Bahnson RR, et al: Prostate cancer, version 1.2016. J Natl Compr Canc Netw 14:19-30, 2016 Crossref, Medline, Google Scholar|
|3.||D’Amico AV: Decipher postprostatectomy: Is it ready for clinical use? J Clin Oncol 35:1976-1977, 2017 Link, Google Scholar|
|4.||Spratt DE, Yousefi K, Deheshi S, et al: Individual patient-level meta-analysis of the performance of the Decipher genomic classifier in high-risk men after prostatectomy to predict development of metastatic disease. J Clin Oncol 35:1991-1998, 2017 Link, Google Scholar|
|5.||Schisterman EF, Cole SR, Platt RW: Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology 20:488-495, 2009 Crossref, Medline, Google Scholar|