A multivariate logistic regression model based on 3T mpMRI, clinical, and biopsy parameters for the prediction of prostate cancer extracapsular extension.

Abstract

27 Background: Extracapsular extension (ECE) of prostate cancer (PCa) is a poor prognostic factor associated with progression, recurrence after treatment and increased PCa- related mortality. Accurate staging prior to radical prostatectomy is crucial in avoidance of positive margins and when planning nerve-sparing procedures. This study investigated the predictive value of clinical, biopsy & 3TmpMRI parameters using a multivariate logistic model for per-lesion detection of PCa ECE with wholemount histopathology (WMHP) as reference. Methods: This IRB approved, HIPAA compliant study included 575 patients with 774 true positive PCa lesions, who underwent radical prostatectomy between 7/2010-2/2019. The relationship between pathologic ECE & parameters including clinical; age, prostate specific antigen (PSA) & PSA density (PSAD), biopsy; percentage of positive systematic cores & Gleason score (GS) & 3TmpMRI; prostate volume, number of lesions per patient, size, location, level, PIRADSv2 score, laterality, apparent diffusion coefficient (ADC) value & risk of ECE on MRI was evaluated using bivariate and multivariate analysis. The accuracy of the final model was evaluated using ROC analysis. Results: 27.8% (215/774), 42.9% (332/774) & 29.3% (227/774) of the lesions were PIRADSv2 score 3, 4 & 5 & 59.9% (464/774), 24.7% (191/774) & 17.7% (137/774) were low, intermediate & high risk for ECE, respectively. 23.6% (183/774) of the lesions had ECE on WMHP. On bivariate analysis higher PSA, PSAD, percentage of positive biopsy cores, biopsy GS, size, PIRADSv2 score, ADC value, risk of ECE on MRI, location (posterior), level(midgland & base), bilaterality & lower number of lesions per patient were significant for ECE prediction. The multivariate logistic model included age, PSAD, number of lesions per patient, size, location, level, PIRADSv2 score & risk of ECE on MRI. The AUC for the prediction of ECE for this model was 0.85. Conclusions: This multivariate regression model based on clinical, biopsy & 3TmpMRI parameters have a high predictive value for pathology ECE detection.