The Decipher® Genomic Classifier for Recurrent Prostate Cancer in the NRG/RTOG 9601 Randomized Clinical Trial, Journal Club - Christopher Wallis & Zachary Klaassen

May 4, 2021

In this UroToday Journal Club, Christopher Wallis and Zachary Klaassen provide insights into the JAMA Oncology publication titled Validation of a 22-Gene Genomic Classifier in Patients With Recurrent Prostate Cancer: An Ancillary Study of the NRG/RTOG 9601 Randomized Clinical Trial. The management of patients with localized and recurrent prostate cancer is driven predominantly at this point by clinicopathologic risk stratification but these factors have a relatively limited ability to accurately identify patients with aggressive disease. Drs. Klaassen and Wallis review the Decipher 22-gene genomic classifier within the context of RTOG 9601 study. This ancillary study used radical prostatectomy specimens from the phase 3 placebo-controlled NRG/RTOG 9601 randomized clinical trial conducted from March 1998 to March 2003. The authors sought to validate the Decipher classifier among patients with recurrent prostate cancer following radical prostatectomy in the context of this trial. Dr. Klaassen reviews evidence in support of the use of the genomic classifier to guide shared decision-making after radical prostatectomy. The evidence also suggests not all men with biochemical recurrence after surgery derive equal and absolute benefits from the addition of hormone therapy to salvage radiotherapy.

Biographies:

Christopher J.D. Wallis, MD, Ph.D., Instructor in Urology, Vanderbilt University Medical Center, Nashville, Tennessee

Zachary Klaassen, MD, MSc, Urologic Oncologist, Assistant Professor Surgery/Urology at the Medical College of Georgia at Augusta University, Georgia Cancer Center


Read the Full Video Transcript

Christopher Wallis: Hello, and thank you for joining us for this UroToday Journal Club. Today, we are looking at a recently published paper, Assessing the Validation of a 22-Gene Genomic Classifier in Patients With Recurrent Prostate Cancer. This is An Ancillary Study of the NRG/RTOG 9601 Randomized Clinical Trial. I'm Chris Wallis, a Fellow in Urologic Oncology at Vanderbilt. And with me, is Zack Klaassen, an Assistant Professor in Urology at the Medical College of Georgia. Here is the citation for this paper recently published in JAMA Oncology and led by Dr. Felix Feng. By way of background, the management of patients with localized and recurrent prostate cancer is driven predominantly at this point by clinicopathologic risk stratification. This relies on PSA, both absolute values and PSA dynamics, as well as clinical staging and Gleason score. However, these factors have a relatively limited ability to accurately identify patients with aggressive disease.

In contrast, Decipher is a 22-gene genomic classifier, which is based on whole transcriptome profiling. This was developed and originally validated in a patient population who had undergone radical prostatectomy. In this cohort, Decipher outperforms risk-stratification based on clinicopathologic features when examined in retrospective analysis. However, for the purposes of this study, the authors sought to validate the Decipher classifier among patients with recurrent prostate cancer following radical prostatectomy in the context of the NRG 9601 trial. This will be the first time that this test is invalidated in a prospective cohort. And so, the authors assessed three predominant hypotheses. First, that the genomic classifier would independently predict the development of metastases, and second, that the genomic classifier could help to personalize the use of ADT with salvage radiotherapy by providing prognostic value.

Finally, the genomic classifier would independently stratify the risk of prostate cancer-specific mortality and overall survival. So the authors utilized the study population from the original NRG 9601 cohort. This trial included patients who had recurrent prostate cancer following radical prostatectomy. At the time of inclusion, they had to have an absolute PSA of 0.2 to 0.4 nanograms per milliliter. They had to have had a high-risk feature at the time of radical prostatectomy, including either extra prosthetic disease manifest as T3 disease or T2 disease and positive surgical margins. Further, there must have been no evidence of nodal and metastatic disease.

Within the context of RTOG 9601, patients were randomized in a one-to-one fashion to salvage radiotherapy with or without the additional bicalutamide. Salvage radiation was given at 64.8 gray in 1.8 gray fractions to the prostate bed. And bicalutamide among those randomized to receive it was given as 150mg daily for two years. In the context of this study and looking at the genomic classifier, the authors undertook RNA processing and analysis based on the specimens collected within the context of the trial. So using the radical prostatectomy specimen, the highest grade tumor focus was identified. Macrodissection was performed, and RNA was extracted.

Among the 760 patients included in the trial, 522 had a radical prostatectomy specimen with sufficient sample to attempt this approach. Following a macrodissection and extraction, 486 patients had sufficient RNA quality to allow for analysis, and further, 352 patients passed quality control and allowed for true derivation of the genomic classifier score. The genomic classifier was scored on a continuous scale from zero to five. And based on previous work, commercial thresholds have been defined as 0.45 to distinguish low from intermediate, and 0.6 to distinguish intermediate from high-risk disease. However, based on a protocol developed for the use of older tissue samples, the lower threshold of 0.40 was also considered.

So the primary outcome in this study was to assess the independent association between the Decipher genomic classifier and the development of metastasis while accounting for the patient, treatment, and tumor characteristics. Secondarily, the authors assessed prostate cancer-specific mortality and overall mortality. And in an exploratory fashion, they looked at the ability of the genomic classifier to prognosticate the second biochemical recurrence metastasis-free survival and progression-free survival. In terms of statistical analysis, the authors assumed a 30% sample loss rate, and a 12-year metastasis rate of 23%. Doing so allowed a 90% power to detect a hazard ratio of 1.13 associated with a 10% change in the Decipher genomic classifier score with a two-sided alpha 0.05.

As a result, this study is adequately powered to assess the prognostic ability of the genomic classifier. However, assessing the predictive question, they relied on assumptions assuming equal sample loss in both arms, 12-year metastasis rate of 14.5% in the experimental group, 23% in the control arm, and 50% of samples with a genomic classifier score of 0.4 or greater and in doing so provided only 35% power to detect an interaction hazard ratio 0.49. Thus, the study is underpowered to assess the predictive value of Decipher. Further, in terms of their analysis, the authors assessed both distant metastases and prostate cancer-specific mortality using cumulative incidence curves and assessed overall survival using the Kaplan-Meier method.

To assess the prognostic performance of Decipher, the authors used Cox proportional-hazards models, with the first method for small event sizes and they performed this both in the overall cohort, as well as subgroups based on the timing of radiotherapy. They adjusted their model to the effect of age, race, Gleason score, tumor stage, PSA, margin status, PSA nadir, and treatment randomization arm. To assess the predictive ability of the Decipher classifier, they performed interaction models using this Decipher score and treatment effect. Primarily they did this without any covariate adjustment, but then they also performed adjustments using the same factors as included in the prognostic evaluation. In order to derive confidence intervals for this approach, they used bootstrapping with 200 resamplings.

At this point, I will now pass it over to Dr. Klaassen to walk us through the results of this analysis.

Zachary Klaassen: Thanks, Chris. So if you look at this as a table, the baseline clinical and genomic characteristics of the cohort, and you can see that overall, these were well-balanced groups with a median age of 64.5 years. The majority of which patients were white at 89%. The majority of these patients also had a phenomenal performance status with a Karnofsky Performance status of 173.9%. The majority of patients had a Gleason 7 prostate cancer, with about 6.5% of patients having a new adjunct hormone use. 75% of patients had positive surgical margins and a median PSA nadir after surgery of 0.1. The median PSA at the trial entry was 0.70, and as Chris previously mentioned, with a very long follow-up of a median of 13 years. The genomic classifier score between the two groups was very similar, with an overall median score of 0.435.

This is the graph looking at the cumulative incidence estimates of distant metastasis by genomic classifier risk group. You can see that the high-risk is in the dark color, intermediate is in orange, and low is in blue. And there is a significant difference between these genomic risk classifier groups with a P-value of 0.03 and an event rate of 15.3% in the high genomic classifier group, 8.7% in the intermediate, and 6.2% in the low genomic classifier group. This is a similar curve looking at prostate cancer-specific mortality stratified by the high, intermediate, and low groups. Once again, an early splitting of these curves with a P-value of 0.01 and an event rate in the high group of 9.8%, in the intermediate group of 2.4%, and in the low group of 0.7%.

This graph looks at overall survival stratified by genomic classifier risk group, and you can see here that once again, a statistically significant P-value of 0.01 within a one-year event rate of 83.2% in the genomic classifier high-risk group, 90.6% in the intermediate, and 94.5% survival in the low-risk group. So, looking at these three tables, in summary, the genomic classifier risk group stratification high, intermediate, and low was able to assess metastasis, prostate cancer-specific mortality, and overall survival. So this is the difference in the effect of treatment based on low, which is a genomic classifier less than 0.45 versus intermediate, to the high group, which is genomic classifier 0.45 to 1 for predicted rates of distant metastasis, prostate cancer-specific mortality, and overall survival at 12 years. And so you can see on these tables that the difference between the treatment effect in the treatment arm minus the placebo group is what these bar graphs are showing.

Furthermore, if the bar graph is above zero, this is a benefit for bicalutamide in each of these groups. And so you can see that there was a benefit for bicalutamide in each of these cohorts for distant metastasis, prostate cancer-specific mortality, and overall survival, but generally, the intermediate and high genomic classifiers had more of a benefit. This is a similar group, but in a subset analysis, looking at patients that were early salvage, so a PSA of less than 0.70. And we can really see here in this set of figures that the intermediate and high genomic classifiers significantly were a better predicting benefit of bicalutamide compared to those with low genomic classifier scores.

This table looks at the multi-variable analysis of genomic classifiers for distant metastasis, death from prostate cancer, and overall survival. And so, I've highlighted the genomic classifier variable in this model, and this model was adjusted for all of these other listed variables. And you can see that for a distant metastasis hazard ratio of 1.17 with a 95% confidence interval of 1.05 to 1.32, this is statistically significant, similarly, in prostate cancer-specific mortality, a significant hazard ratio of 1.39 and a 95% confidence interval of 1.20 to 1.63, and also in overall survival with a hazard ratio of 1.17 and a 95% confidence interval of 1.06 to 1.29. So to summarize this table, after adjusting for all of these factors, both clinical and demographic, the increasing genomic score was associated with the worst distant metastasis, prostate mortality, and overall survival.

This is a subgroup analysis looking at the prognostic performance of the genomic classifiers for distant metastasis in the subgroup. And to summarize, you can see that across all of these subgroups, the genomic classifier had a statistically significant hazard ratio, and this includes treatment, age, Gleason score, PSA nadir, PSA nadir at trial entry, T stage, and surgical margin status.

So there are several important discussion points from this analysis of the RTOG study. And we see that in the current study, this is the first prospective validation of a genomic classifier conducted from a large RCT in men receiving salvage radiotherapy with either placebo or bicalutamide for two years.

This genomic classifier was prognostic for all of the distant metastasis, prostate cancer-specific mortality, and overall survival when evaluated as a continuous score or as a three-tiered risk group system.  Patients with intermediate and high genomic classifier scores had an 88% increased risk of distant metastasis versus those with low genomic classifier scores. Of note, as Chris mentioned earlier, the final cohort was only 352 patients, and there was insufficient power to assess the statistical interaction between the genomic classifier and hormone therapy for any endpoint in the study.

So, in conclusion, these findings represent the first validation in prostate cancer of high throughput, clinical grade, whole transcriptome-based genomic classifier from a prospective randomized controlled trial. This adds evidence to support the use of the genomic classifier to guide shared decision-making after radical prostatectomy. And it shows that not all men with biochemical recurrence after surgery derive equal and absolute benefits from the addition of hormone therapy to salvage radiotherapy. We thank you for your attention, and we hope you enjoyed this UroToday Journal Club.
email news signup