Clash of the calculators: External validation of prostate cancer risk calculators in men undergoing mpMRI and transperineal biopsy

Abstract Objective To compare the accuracy of the European Randomized Study of Screening for Prostate Cancer (ERSPC) RC, MRI‐ERSPC‐RC, and Prostate Biopsy Collaborative Group (PBCG) RC in patients undergoing transperineal prostate biopsy. Patients and methods We identified 392 patients who underwent mpMRI before transperineal prostate biopsy across multiple public and private institutions between January 2017 and August 2019. The estimated probabilities of detecting PCa and significant PCa were calculated using the MRI‐ERSPC‐RC, ERSPC‐RC, and PBCG‐RC. Receiver operating characteristic (ROC) curves for each calculator were generated and the area underneath the curve (AUC) was compared. Calibration and clinical utility were assessed with calibration plots and decision curve analysis, respectively. Results PCa was detected in 285 patients (72.7%) with significant PCa found in 200 patients (51.1%). ROC curve analysis found the MRI‐ERSPC‐RC outperformed the ERSPC‐RC and PBCG‐RC. For the prediction of PCa, the AUC was 0.756, 0.696, and 0.675 for the MRI‐ERSPC‐RC, ERSPC‐RC, and PBCG‐RC, respectively. The AUC for the prediction of significant PCa was 0.803, 0.745, and 0.746 for the MRI‐ERSPC‐RC, ERSPC‐RC, and PBCG‐RC, respectively. Conclusions Our study validated the ERSPC‐RC, MRI‐ERSPC‐RC, and PBCG‐RC in a cohort undergoing transperineal prostate biopsy with the MRI‐ERSPC‐RC performing the best. These RCs may enable improved shared decision making and help to guide patient selection for biopsy.


| INTRODUC TI ON
Prostate cancer (PCa) is the second most commonly diagnosed cancer in men worldwide with almost 1.3 million new cases in 2018. 1 Traditional approaches to patient selection for prostate biopsy utilizing prostate-specific antigen (PSA) and digital rectal examination (DRE) have contributed to many men having negative biopsies as well as an increased detection of low-risk PCa. 2 Subsequently, many men undergo prolonged follow-up with significant costs and burdens to the patient as well as the healthcare system.
In recent years, multi-parametric Magnetic Resonance Imaging (mpMRI) has become increasingly accessible. There is expanding evidence that pre-diagnostic mpMRI helps to optimize patient selection for biopsy by reducing unnecessary prostate biopsies and improving diagnostic accuracy. [3][4][5] Consequently, standard clinical practice has changed worldwide with many healthcare systems now offering men an mpMRI in the initial workup of PCa prior to their biopsy.
In an attempt to mitigate unnecessary biopsies, several prediagnostic risk calculators have been developed to optimize patient selection prior to biopsy. The European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Biopsy Collaborative Group (PBCG) risk calculators (RC) aim to estimate the likelihood of any PCa and clinically significant PCa. 6,7 With its rise in popularity, mpMRI has been incorporated into the latest ERSPC-RC together with PSA, history of negative prostate biopsy, DRE, prostate volume, and age. 6 The aim of our study was to assess the validity of three RCs; the MRI-ERSPC-RC, the previous ERSPC-RC, and the PBCG-RC. We aimed to compare their respective performances in order to identify which RC provides the superior model in optimizing patient selection for biopsy.

| PATIENTS AND ME THODS
Patients with a clinical suspicion of PCa who underwent mpMRI followed by a transperineal prostate biopsy between January 2017 and August 2019 were identified from a prospectively maintained database. This database comprised patients from multiple institutions encompassing both the public and private healthcare system.
Patients with previously diagnosed PCa or without a pre-biopsy mpMRI were excluded. Furthermore, patients whose age, PSA level, or prostate volume precluded the use of the ERSPC-RC or PBCG-RC were excluded. Approval for this project was granted by our institution's Human Research Ethics Committee.
Transperineal prostate biopsy was performed by a urologist under general anesthesia or sedation with the patient in lithotomy position using a bi-planar ultrasound transducer probe (BK Medical, Peabody, USA) in the rectum and an 18g x 22cm biopsy needle (Bard Max Core Needle, Bard, USA). Prostate mapping was performed in 5-10 mm increments utilizing a brachytherapy template grid (Accucare Template grid, Civco Medical Solutions, UK). All patients underwent systematic biopsy in addition to targeted biopsies from areas of concern identified by mpMRI. The template used and total number of cores taken were in accordance with the Ginsburg protocol. 8 Specimens were assessed by genitourinary pathologists using the International Society of Urological Pathology (ISUP) Grade Group system. 9 In the current study, clinically significant cancer was defined as ISUP Grade Group ≥ 2.
All mpMRIs were performed using a 3 Tesla MRI scanner including diffusion weighted, dynamic contrast enhanced T1 and T2 weighted imaging and reported as per the Prostate Imaging-Reporting and Data System (PI-RADS) by a group of radiologists specializing in uro-oncology. 10 All mpMRIs and prostate biopsies were reviewed at a urology multidisciplinary team meeting and reviewed by dedicated uro-pathologists and uro-radiologists.
Medical records were reviewed and data obtained as guided by the PBCG and ERSPC calculators, including race, age, pre-biopsy PSA, family history, history of previous biopsy, DRE findings, prostate volume, PI-RADS score, and histopathology results. To quantify the discriminative ability of the MRI-ERSPC-RC, ERSPC-RC, and PBCG-RC, receiver operating characteristic (ROC) curves for each RC were generated and plotted as the false negative rate (1-specificity) vs sensitivity. 13 An AUC of 0.5 was considered to demonstrate no discrimination, 0.5-0.7 was considered poor discrimination, 0.7-0.8 was considered acceptable discrimination, 0.8-0.9 was considered excellent discrimination, and greater than 0.9 was considered outstanding. 14 Areas underneath the ROC curve (AUC) were calculated for the respective calculators and compared DeLong's test. 15 Calibration plots were computed by comparing observed proportions of cancer to mean calculated risks by the respective risk calculator deciles observed in the cohort for overall cancer detection and significant cancer risk. 16 The Hosmer-Lemeshow Chi square test was used to compare the observed rates to predicted risks across the deciles for each calculator. For this test, a P < .05 indicates a poor agreement between predicted risks and actual observed risk. Decision curve analysis was performed to assess for the gain derived from the respective risk calculator over the corresponding net benefit curves of referring no patients or all patients to biopsy. 17

| RE SULTS
Eight hundred and sixty-four patients who underwent transperineal prostate biopsy were identified. One hundred and nineteen patients had previously diagnosed PCa and 283 patients did not undergo pre-biopsy mpMRI. A further 70 patients were unable to have their risk estimated using the ERSPC-RC or the PBCG-RC. A total of 392 patients met our final inclusion criteria. The median age, PSA, and prostate volume were 64 years, 6 ng/ml, and 43 ml, respectively. Group ≥ 2 disease. These results are described in more detail in Table 2.  were performed and the gray line represents the decision curve for performing a biopsy on all patients. The green, blue, and red lines represent the benefit of using each respective RCs to determine which patient to biopsy. The MRI-ERSPC-RC was superior to the ERSPC-RC and the PBCG-RC for the prediction of both overall PCa and significant disease as it had the highest net benefit at the majority of threshold probabilities along the x-axis (Figure 3a,b). For the majority of threshold probabilities, the risk calculators performed better than a strategy of performing a biopsy for all patients or never performing a biopsy.
Using a risk threshold of 12.5% on the MRI-ERSPC-RC to identify which patients to biopsy, there would be a reduction of 106 transperineal biopsies. This 27% reduction in biopsies would have Note: Data are presented as median (IQR) or n (%).  Overall, we observed slightly poorer performance in our cohort com-