Development of a Prognostic Prediction Model for Clinically Significant Prostate Cancer Based on Lesion Zone and...

To guide biopsy decision making via a predictive model for clinically significant prostate cancer (csPCa) in men with positive imaging (PI-RADS 3-5). We aim to elucidate the influence of zone on the predictive accuracy of PI-RADS v2.1, establish a threshold of quantified apparent diffusion coefficient (ADC) values suspicious for csPCa, and create a logistic-regression model, unique in its inclusion of lesion zone and quantified ADC.

Data were retrospectively collected from a single cancer center, with institutional review board review and exemption waiver, on a lesion-level retrospective analysis of prostate multiparametric magnetic resonance imaging (mpMRI) of N = 429 men, harboring N = 546 lesions. MRIs were read and reviewed by a fellowship-trained genitourinary/abdominal imaging radiologist. Pathologic analysis was conducted by a fellowship-trained genitourinary pathologist. All statistical analysis including predictive modeling were conducted by an expert biostatistician.

On multivariate analysis, zone (PZ as compared to TZ) was a predictor of csPCa (OR 2.58, 95% CI 1.50-4.44, P<.0005). A 700 µm²/s ADC value cutoff in isolation had an AUC of 0.56, but our combined model, bolstered by clinical and imaging data, including zone and prostate-specific antigen density, yielded an AUC of 0.78, at an ADC value cutoff of 947 µm²/s. In application, simulated scenarios yielded probabilities between 16% and 82%, therefore above a biopsy omission threshold of 10% or less.

Prostate zone and ADC value quantification each independently contribute to csPCa predictability, but no simulated use of the model yielded biopsy-omission thresholds for patients with MRI visible lesions.