The validated genomic-based classifiers outperform clinical models for predicting postcystectomy bladder cancer recurrence. This [novel genomic signature] may be used to better identify patients who need more aggressive management.
—Anirban P. Mitra, MD, PhD, and colleagues
As reported in the Journal of the National Cancer Institute, Anirban P. Mitra, MD, PhD, of the University of Southern California, and colleagues identified a novel genome-based signature that improves prediction of postcystectomy recurrence in patients with high-risk bladder cancer.1 Use of the signature could help to guide selection of patients for adjuvant therapy in this setting.
Dr. Mitra is the corresponding author for the Journal of the National Cancer Institute article. Siamak Daneshmand, MD, of the University of Southern California, and Peter C. Black, MD, of the University of British Columbia, contributed equally to the work.
This study involved a generation of transcriptome-wide expression profiles using 1.4 million feature arrays on archival tumors from 225 patients with muscle-invasive or node-positive bladder cancer who had undergone radical cystectomy. Genomic classifiers and clinicopathologic classifiers for predicting tumor recurrence were developed in a discovery set (n = 133). The predictive performances of genomic classifiers, clinicopathologic classifiers (individual features and an optimized prognostic model developed in the discovery set including age, gender, pathologic stage, and lymphovascular invasion status), the independent clinical nomogram of the International Bladder Cancer Nomogram Consortium, and combined genomic-clinicopathologic classifiers were assessed in the discovery and independent validation (n = 66) sets.
In the discovery set, patients had a median age of 68.5 years (interquartile range = 63.6–75.6 years), median follow-up was 9.3 years, 51% of patients had tumor recurrence, and 58% of patients had died at last follow-up. In the validation set, median age was 69.8 years (interquartile range = 63.1–74.3 years), median follow-up was 10.8 years, 50% of patients had tumor recurrence, and 68% of patients had died at last follow-up.
In the discovery set, 15 genomic markers were identified, corresponding to RNAs from genome regions that were differentially expressed in tumor recurrence. Associated genes included those involved in cell proliferation/cell-cycle regulation/apoptosis, cell differentiation, transcription regulation, and signaling pathways and transduction (FOXO6, HSD17B7, ARID4B, ENAH, MAP4K3, MARCH7, MECOM, LRBA, MUT, CRCP, SYPL1, ARFGEF1, EHF, METTL7A, and PPP1R12A).
Findings in the Discovery Set
In the discovery set, area under the standard receiver-operating characteristic curve values were 0.88 for the genomic classifiers compared with 0.70 for nodal status, 0.69 for lymphovascular invasion, 0.58 for tumor stage, 0.53 for age, 0.47 for gender, 0.73 for International Bladder Cancer Nomogram Consortium, 0.89 for genomic–International Bladder Cancer Nomogram Consortium, 0.81 for the clinicopathologic model, and 0.93 for genomic-clinical classifiers.
Predictive Ability in the Validation Set
In the validation set, the genomic signature had a survival receiver-operating characteristic curve value of 0.77, higher than the values for the individual clinical parameters of nodal status (0.71), lymphovascular invasion (0.68), tumor stage (0.57), age (0.50), and gender (0.44) and similar to those for the International Bladder Cancer Nomogram Consortium (0.73) and the clinicopathologic model (0.78). The survival receiver-operating characteristic curve value was increased with the addition of the genomic signature to both the International Bladder Cancer Nomogram Consortium (0.82) and the clinicopathologic model (0.86).
On multivariate analysis adjusting for demographic, clinicopathologic, and treatment factors, the genomic signature was the best predictor of tumor recurrence (hazard ratio [HR] = 1.42, P = .005); other significant predictors were nodal status (P = .017), lymphovascular invasion (P = .045), and interaction of the genomic signature and nodal status (P = .03), whereas age, gender, race, tumor stage, and receipt of adjuvant therapy were not independent predictors. Multivariate analysis of the predictive models showed an independent predictive effect when the genomic signature was added to the International Bladder Cancer Nomogram Consortium model (HR = 1.18, P = .016) compared with the International Bladder Cancer Nomogram Consortium model alone (HR = 1.04, P = .62) and when the genomic signature was added to the clinicopathologic model (HR = 1.18, P = .008) compared with the clinicopathologic model alone (HR = 1.10, P = .30).
Reclassification and Decision-Curve Analyses
An analysis in the validation set categorizing patients as high-risk or low-risk based on genomic-clinicopathologic scoring and adjusting for competing risks showed a 4-year probability of tumor recurrence of 81.5% vs 20.6% (P < .001). Further, addition of the genomic signature to the International Bladder Cancer Nomogram Consortium model reclassified 18 patients into different risk categories, with 12 (67%) correctly reclassified based on outcome. The genomic-clinicopathologic model reclassified 12 patients from initial International Bladder Cancer Nomogram Consortium risk categories, with 10 (83%) correctly reclassified.
Decision-curve analysis for decision to treat based on the specificity and sensitivity of predictive models at different thresholds of probability for postcystectomy recurrence showed a greater overall benefit for genome-based classifiers. Thus, for example, for a 50% threshold probability of postcystectomy recurrence, compared with “treat none” and “treat all” options, the genomic–International Bladder Cancer Nomogram Consortium model resulted in a reduction in unnecessary treatment (ie, in low-risk patients) by 27%, compared with 17% for the International Bladder Cancer Nomogram Consortium model alone, and the genomic-clinicopathologic model resulted in a reduction in overtreatment in 28%, compared with 20% for the clinicopathologic model alone.
In the validation cohort, the genomic signature showed better survival predictive ability (survival receiver-operating characteristic curve value = 0.77) than seven other previously reported genomic signatures for muscle-invasive or node-positive disease (receiver-operating characteristic curve values = 0.51–0.72).
Finally, the genomic signature was validated in four additional external bladder urothelial carcinoma data sets (n = 341). For overall survival (all four data sets), hazard ratios ranged from 2.26 to 7.57, with P values of .016 to < .001 and concordance indices of 0.65 to 0.80. For cancer-specific survival (one dataset), the hazard ratio was 2.23, P = .001, and the concordance index was 0.65. For recurrence-free survival (one dataset), the hazard ratio was 3.33, P < .001, and the concordance index was 0.67.
The investigators concluded: “The validated genomic-based classifiers outperform clinical models for predicting postcystectomy bladder cancer recurrence. This [novel genomic signature] may be used to better identify patients who need more aggressive management.” ■
Disclosure: This study was supported by Genome British Columbia. For full disclosures of the study authors, visit jnci.oxfordjournals.org.
1. Mitra AP, Lam LL, Ghadessi M, et al: Discovery and validation of novel expression signature for postcystectomy recurrence in high-risk bladder cancer. J Natl Cancer Inst 106(11):dju290, 2014.
Radical cystectomy is the standard therapeutic option for patients with muscle-invasive bladder cancer. However, 5-year overall survival for high-risk patients with pT3, pT4, pN-negative, and pN-positive M0 bladder cancer after radical cystectomy is only about 50% and ranges from 32% in patients...