Presentation Authors: Alberto Martini*, Dmitry Rykunov, Shivaram Cumarasamy, John P Sfakianos, Ardeshir R Rastinehad, Kenneth G Haines III, Sujit S Nair, Boris Reva, Ashutosh K Tewari, New York, NY
Introduction: The ability to predict non-organ-confined prostate cancer (PCa), as either in the form of extracapsular extension (ECE) or seminal vesicle invasion (SVI) is of pivotal importance for patient counseling and surgical planning. In an era of precision medicine, we aimed to identify genes associated with ECE and SVI and to build a predictive model.
Methods: data from 494 patients with complete transcriptomic profile and SCAN-normalized expression of coding and non-coding genes was identified from our single-center prospectively maintained database. All the patients underwent multiparametric MRI before receiving robot assisted radical prostatectomy. The tissue used for transcriptomic profiling was obtained from the neoplastic area with the highest Gleason Group on the pathology specimen. First, in order to develop the genomic signature, the data was split into a training (50%), test (25%) and validation (25%) cohort.Second, a multivariable binary logistic regression was built to predict alternatively ECE and SVI. Variable included were: PSA, biopsy Gleason grade, highest percentage core involvement, clinical stage from mpMRI (organ-confined vs. capsule abutment vs. ECE vs. SVI) and index lesion&[prime]s greatest diameter.The tumor transcriptional profiles were used to evaluate for biomarkers and to construct genomic signatures for better a classification of disease subtypes, measured by the increase in AUC, with associated 95%CI.
Results: Median (IQR) age at surgery was 53 (57-68) yrs, median (IQR) PSA was 6.5 (5.1-10.2) ng/ml. Overall, 166 (34%) and 62 (13%) patients had ECE and SVI on final pathology, respectively. A signature based on the gene expression values of six coding genes was constructed for the prediction of ECE and SVI. In the validation cohort, the AUC was 0.72 and 0.76 for predicting ECE and SVI, respectively. All variables included in the multivariable model emerged as predictors of both ECE and SVI. The AUC of the models based on clinical parameters, biopsy and mpMRI data was 0.81 (95%CI: 0.77-0.85) and 0.84 (95%CI: 0.79-0.89) for ECE and SVI, respectively. Adding the signatures to the multivariable models improved the discrimination in predicting ECE to 0.83 (0.79-0.87, p=0.01) and in predicting SVI to 0.87 (95%CI: 0.83-0.92; p=0.04).
Conclusions: By integrating clinical, biopsy, mpMRI and genomic data we developed two models that accurately allow for the prediction of non-organ-confined PCa. Our models might serve in patient counseling, in the selection of candidates for potential novel neoadjuvant treatments and for surgical planning.