NEAUA Main Site  |  Past and Future Meetings
New England Section of the American Urological Association

Back to 2020 Abstracts


Predicting Cancer Detection Rates from Multiparametric Prostate MRI: Refining the PI-RADS Categorization System
Francisco Ramos, BS1, Aaron Fleishman, MPH2, Ruslan Korets, MD2, Michael Johnson, MD2, Aria Olumi, MD2, Leo Tsai, MD, PhD2, Boris Gershman, MD2.
1Harvard Medical School, Boston, MA, USA, 2Beth Israel Deaconess Medical Center, Boston, MA, USA.

BACKGROUND: The PI-RADS categorization represents the standardized method for assessing risk of prostate cancer in men undergoing multiparametric prostate MRI. However, there exists a substantial discrepancy in widely accepted cancer detection rates for each PI-RADS category as reported in seminal prospective studies (e.g., PROMIS, PRECISION) and cancer-detection rates (CDRs) observed in real-world clinical practice. We hypothesized that CDRs vary according to patient and MRI features beyond what is captured in the PI-RADS categorization. Herein, we examine the associations of clinical and radiographic features with CDRs and develop a predictive model to improve patient counseling and clinical management.
METHODS: We identified men aged 18-89 with a diagnosis of elevated PSA or Gleason 6 prostate cancer on active surveillance and ≥1 PI-RADS 3-5 lesion on prostate MRI who underwent MRI-U/S fusion prostate biopsy in the office or in-bore MRI-targeted biopsy in Interventional Radiology. Only targeted biopsy cores were considered for MRI-U/S fusion biopsy. The associations of clinical and radiographic features with the per-lesion cancer-detection rate (CDR; Gleason 6-10) and clinically-significant cancer detection rate (csCDR; Gleason 7-10) were examined using multivariable logistic regression following a univariable screen at the p<0.10 level. Patients with Gleason 6 prostate cancer on active surveillance were excluded from CDR analyses.
RESULTS: Targeted biopsy was performed for 216 lesions in 169 patients, including MRI-U/S fusion biopsy in 130 lesions and in-bore MRI-targeted biopsy in 86 lesions. Median pre-biopsy PSA was 7.2 ng/mL (IQR 5.2-10.2). Thirty percent of lesions were in biopsy-na´ve patients, while 29% were in patients on active surveillance for Gleason 6 prostate cancer. PI-RADS category was 3 in 17.1%, 4 in 51.9%, and 5 in 31.0% of lesions. Lesions were located in the anterior/transition zone in 58.8% of cases, and median lesion diameter was 13 mm (IQR 9-17 mm). Overall, the CDR was 48% and the csCDR was 26%. In multivariable regression models predicting CDR, number of prior biopsies (OR 0.18, 95%CI 0.06-0.47 for 1; OR 0.16, 95%CI 0.05-0.47 for 2+ versus none), PSA density (OR 1.75, 95% CI 1.15-2.88) and PI-RADS category (OR 5.39, 95%CI 1.43-29.1 for 4; OR 15.1, 95% CI 3.54-88.8 for 5 versus 3) were independently associated with CDR (Table 1). In multivariable models predicting csCDR, abnormal DRE (OR 2.75, 95%CI 1.22-6.19), solitary lesion on MRI (OR 2.42, 95%CI 1.13-5.48 versus 2+), PSA density (OR 1.04, 95%CI 1.00-1.09), and PI-RADS category (OR 3.30, 95%CI 1.03-13.7 for 4; OR 3.44, 95% CI 1.03-14.4 for 5 versus 3) were independently associated with csCDR (Table 1). The bootstrap-adjusted c-index was 0.80 for the CDR model and 0.71 for the csCDR model.
CONCLUSIONS:
In this study, several clinical and radiographic features were independently associated with risk of malignancy in men undergoing MRI-targeted biopsy for elevated PSA. Such models can be operationalized to provide personalized risk-stratification beyond the PI-RADS categorization.


Back to 2020 Abstracts