Purpose: Pretreatment estimates of seminal vesicle invasion (SVI) are challenging and significantly influence the management of prostate cancer. We sought to improve current models to predict SVI through the development of an SVI prediction genomic signature.
Patients and methods: A total of 15,889 patients who underwent radical prostatectomy (RP) with available baseline clinical, pathology, and transcriptome data were retrieved from the GRID registry (ClinicalTrials.gov identifier: NCT02609269) and other retrospective cohorts. These data were divided into a training (n = 6,766), test (n = 3,363), and two validation (n = 5,062 and 698) cohorts. Multivariable logistic regression was performed to assess the predictive effect of the genomic SVI (gSVI) classifier in the presence of established nomograms (Partin Tables and Memorial Sloan Kettering Cancer Center [MSKCC]).
Results: In the training cohort, univariable filtering identified 2,132 genes that were differentially expressed between RP tumors with and without SVI. Model parameters were tuned to maximize the area under the curve (AUC) in the testing cohort, resulting in a logistic generalized linear model with 581 genes. The gSVI model scores range from 0 to 1. In the first validation set, gSVI showed superior discrimination of patients with and without SVI at RP compared with other prognostic signatures trained to predict distant metastasis or clinical recurrence. Of the 698 patients in the second validation set, gSVI combined with the MSKCC nomogram had a superior AUC (0.86) compared with either nomogram individually (0.81).
Conclusion: The gSVI represents a novel and validated expression signature to predict the presence of SVI before treatment with surgery. This genomic tool adds discriminatory power to existing clinical predictive nomograms and may help with pretreatment counseling and decision making.