EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 available at www.sciencedirect.com journal homepage: www.eu-openscience.europeanurology.com Prostatic Disease Comparison of Rotterdam and Barcelona Magnetic Resonance Imaging Risk Calculators for Predicting Clinically Significant Prostate Cancer Juan Morote a,b,y,*, Ángel Borque-Fernando c,y, Marina Triquell a,b, Miriam Campistol a,b, Pol Servian d, José M. Abascal e,f, Jacques Planas a,b, Olga Méndez g, Luis M. Esteban h,à, Enrique Trilla a,b,à a Department of Urology, Vall d́Hebron Hospital, Barcelona, Spain; b Department of Surgery, Universitat Autònoma de Barcelona, Bellaterra, Spain; c Department of Urology, Hospital Miguel Servet, IIS-Aragon, Zaragoza, Spain; d Department of Urology, Hospital Germans Trias I Pujol, Badalona, Spain; e Department of Urology, Parc de Salut Mar, Barcelona, Spain; f Department of Surgery, Universitat Pompeu Fabra, Badalona, Spain; g Biomedical Research in Urology Unit, Vall d́Hebron Research Institute, Barcelona, Spain; h Department of Applied Mathematics, Escuela Universitaria Politécnica La Almunia, Universidad de Zaragoza, Zaragoza, Spain Article info Abstract Article history: Background: Magnetic resonance imaging (MRI)-based risk calculators (MRI-RCs) Accepted March 19, 2023 individualise the likelihood of clinically significant prostate cancer (csPCa) and improve candidate selection for prostate biopsy beyond the Prostate Imaging Associate Editor: Reporting and Data System (PI-RADS). Guillaume Ploussard Objective: To compare the Barcelona (BCN) and Rotterdam (ROT) MRI-RCs in an entire population and according to the PI-RADS categories. Keywords: Design, setting, and participants: A prospective comparison of BCN- and ROT-RC in Prostate cancer 946 men with suspected prostate cancer in whom systematic biopsy was per- Clinically significant formed, as well as target biopsies of PI-RADS 3 lesions. Early detection Outcome measurements and statistical analysis: Saved biopsies and undetected csPCa Magnetic resonance imaging (grade group 2) were determined. Predictive model Results and limitations: The csPCa detection was 40.8%. The median risks of csPCa Risk calculator from BCN- and ROT-RC were, respectively, 67.1% and 25% in men with csPCa, whereas 10.5% and 3% in those without csPCa (p < 0.001). The areas under the curve were 0.856 and 0.844, respectively (p = 0.116). BCN-RC showed a higher net benefit and clinical utility over ROT-RC. Using appropriate thresholds, respectively, 75% and 80% of biopsies were needed to identify 50% of csPCa detected in men with PI-RADS <3, whereas 35% and 21% of biopsies were saved, missing 10% of csPCa detected in men with PI-RADS 3. BCN-RC saved 15% of biopsies, missing 2% of csPCa in men with PI-RADS 4, whereas ROT-RC saved 10%, missing 6%. No RC saved biopsies without missing csPCa in men with PI-RADS 5. y These authors contributed equally as co-first authors. à These authors contributed equally as co-last authors. * Corresponding author. Department of Urology, Vall d́Hebron Hospital, Po Vall d́Hebron 119-129, 08035 Barcelona, Spain. Tel. +34 2746100; Fax: +34 2746028. E-mail address: juan.morote@vallhebron.cat (J. Morote). https://doi.org/10.1016/j.euros.2023.03.013 2666-1683/Ó 2023 The Author(s). Published by Elsevier B.V. on behalf of European Association of Urology. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 47 Conclusions: ROT-RC provided a lower and narrower range of csPCa probabilities than BCN-RC. BCN-RC showed a net benefit over ROT-RC in the entire population. However, BCN-RC was useful in men with PI-RADS 3 and 4, whereas ROT-RC was useful only in those with PI-RADS 3. No RC seemed to be helpful in men with neg- ative MRI and PI-RADS 5. Patient summary: Barcelona risk calculator was more helpful than Rotterdam risk calculator to select candidates for prostate biopsy. Ó 2023 The Author(s). Published by Elsevier B.V. on behalf of European Association of Urology. This is an open access article under the CC BY-NC-ND license (http://creative- commons.org/licenses/by-nc-nd/4.0/). 1. Introduction tional Society of Urologic Pathology (ISUP) grade group 2. BCN MRI-RC has been the first predictive model anal- The early detection of prostate cancer (PCa) has evolved ysed according to the PI-RADS categories, showing that towards clinically significant PCa (csPCa), avoiding unneces- the efficacy in the entire population does not represents sary prostate biopsies and overdetection of insignificant that in each PI-RADS category. In addition, the BCN MRI- tumours. However, PCa suspicion remains based on serum RC has the possibility to select the appropriate threshold prostate-specific antigen (PSA) elevation and abnormal dig- for adjustments in validation studies and according to the ital rectal examination (DRE) [1–3]. This paradigm shift has PI-RADS categories [17]. resulted from the spread of prebiopsy multiparametric Owing to differences between ROT and BCN MRI-RCs, we magnetic resonance imaging (mpMRI), which grades the hypothesise different behaviour for improving candidate likelihood of csPCa through Prostate Imaging Reporting selection for prostate biopsy. We aim to compare the clini- and Data System (PI-RADS) [4]. The negative predictive cal usefulness of ROT and BCN MRI-RCs in a whole popula- value of mpMRI reaches 95%, which makes it possible to tion of suspected PCa men and according to each PI-RADS avoid prostate biopsies for PI-RADS <3 [5,6]. Target biopsies category. of suspicious lesions (PI-RADS 3) increase the csPCa sensi- tivity of systematic-biopsies. However, csPCa detection 2. Patients and methods does not exceed 20% in men with PI-RADS 3, is approxi- mately 50% in men with PI-RADS 4, and reaches 90% in 2.1. Design, setting, and participants men with PI-RADS 5 [7]. A prospective head-to-head comparison of ROT- and BCN-RC in 946 men MRI-predictive models individualise the likelihood of with serum PSA 3.0 ng/ml and/or abnormal DRE, recruited in two aca- csPCa and improve candidate selection for prostate biopsy, demic centres from the Barcelona metropolitan area (PSM and GTiP), although available risk calculators (RCs) and external vali- between 2018 and 2021 was performed. Prebiopsy 3-Tesla mpMRI and dations are essential [8]. Among the 18 MRI-predictive 12-core transrectal ultrasound (TRUS) systematic biopsy were always models developed in the last 5 yr, seven have been vali- performed, and two- to four-core TRUS visual target biopsies were added dated externally and only two RCs are available [9]. Rotter- to PI-RADSv.2 3 lesions. Men undergoing 5-ARIs who had previous PCa, dam (ROT) RC was initially designed from the Rotterdam atypical small acinar proliferation, or high-grade prostatic intraepithelial population of the European Randomised Screening Prostate neoplasia with atypia were excluded. The project was approved by the Cancer (ERSPC) trial [10,11]. ROT MRI-RC has resulted from ethical committee of VHH (PR/AG-317/2017). an adjustment of the previous RCs 3 and 4 to predict PCa and high-grade PCa likelihoods in biopsy-naïve men and 2.2. Intervention those with previous negative prostate biopsy, in 961 men with serum PSA 3.0 ng/ml and/or abnormal DRE, in whom The likelihoods of csPCa from ROT-RC (www.prostatecancer-riskcalcula- systematic biopsy was always performed as a target biopsy tor.com) [12] and BCN-RC (https://mripcaprediction.shinyapps.io/ to PI-RADSv.1 3 lesions. PI-RADSv.1 score (1–5), age (50– MRIPCaPrediction/, second tab ‘BCN2RC: MRI-based model’) [17] were 75 yr), biopsy status (initial vs repeat), serum PSA (0.4–50 assessed. PI-RADSv.2 was introduced in both RCs as the closest value ng/ml), DRE (normal vs abnormal), and prostate volume of serum PSA, prostate volume, and age when out of the accepted range (10–110 ml) were included as predictors [12]. ROT MRI- in ROT-RC. RC has been validated in some populations, although recal- ibrations and risk threshold adjustments have been needed 2.3. Clinically significant PCa definition to assure accurate predictions [13–16]. Barcelona (BCN) MRI-RC was recently designed among 1486 men with PSA ROT-RC predicts high-grade PCa likelihood, defined as a Gleason score of 3.0 ng/ml and/or abnormal DRE, in whom systematic 3 + 4, although the current RC shows this as csPCa [13]. BCN-RC pre- biopsies were always performed as target biopsies to PI- dicts csPCa risk defined as an ISUP grade group of 2 [18]. RADSv.2 3 lesions. BCN MRI-RC has been validated exter- nally and includes the same predictors as ROT MRI-RC with- 2.4. Endpoint variables out range limitation, PI-RADSv.2, and PCa family history (first degree vs no). The csPCa was defined as the Interna- Saved prostate biopsies and missed csPCa were determined. 48 EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 2.5. Statistical analysis Table 1 – Population characteristics Characteristic Measurement Quantitative variables were expressed as median and 25–75 percentiles (interquartile range [IQR]), and qualitative variables in rates. Mann- Number of men 946 Age (yr), median (IQR) 67 (61–72) Witney U test and chi-square test were used to compare medians and Total PSA (ng/ml), median (IQR) 7.4 (5.5–10.9) proportions [19,20]. The discrimination ability of csPCa was analysed Abnormal DRE, n (%) 307 (32.5) with receiver operating characteristic (ROC) curves [21]; areas under Prostate volume (ml), median (IQR) 55 (40–79) Prior negative prostate biopsy, n (%) 293 (31.0) the curve (AUCs) were compared with the test of DeLong et al [22]. Cal- Family history of PCa, n (%) 34 (3.6%) ibration curves analysed the correspondence between predictive proba- PI-RADS, n (%) bilities and observed occurrence of csPCa. The net benefit of RCs over 1–2 235 (24.8) biopsying all men or none was analysed using a decision curve analysis 3 201 (21.2) 4 391 (41.3) (DCA) [23]. Specificities corresponding to 85%, 90%, and 95% sensitivities 5 119 (12.6) were assessed as their 95% confidence intervals (CIs). Clinical utility PCa detection, n (%) 521 (55.1) curves (CUCs) exploring potential rates of avoided biopsies and missed csPCa detection, n (%) 386 (40.8) csPCa detection by PI-RADS category, n (%) csPCa according to continuous probability of csPCa were generated <3 42 (17.9) [24]. Tests with two-sided p < 0.05 were considered statistically signifi- 3 41 (20.4) cant. Statistical analyses were computed using R programming language 4 203 (51.9) v.4.0.3 (The R Foundation for Statistical Computing, Vienna, Austria) and 5 100 (84.0) SPSS v.25 (Statistical Package for Social Sciences; IBM, San Francisco, CA, csPCa = clinically significant PCa; DRE = digital rectal examination; IQR = interquartile range; n = number; PCa = prostate cancer; PI- USA). RADS = Prostate Imaging Reporting and Data System; PSA = prostate- specific antigen. 3. Results 3.1. Population characteristics ROT-RC and 31.8% (28.0–35.8) for BCN-RC (p < 0.001) at 95% sensitivity. The characteristics of participants are summarised in Table 1. In 386 men (40.8%), csPCa was detected—17.9% in 3.3. Behaviour of ROT- and BCN-RC according to the PI-RADS men with PI-RADS <3, 20.4% for PI-RADS 3, 51.9% for PI- categories RADS 4, and 84% for PI-RADS 5. A subset of 209 men Violin plots of csPCa likelihoods in men with and without (22.1%) had age (129), serum PSA (18), or prostate volume csPCa, assessed with both RCs according to PI-RADSv.2 cat- (68) out of the accepted range of ROT-RC. egories, show lower values and a narrow range of ROT-RC predictions (Fig. 4A) than those from BCN-RC (Fig. 4B), 3.2. Behaviour of ROT- and BCN-RC in the whole population although significant differences existed between the medi- The median csPCa likelihood of ROT-RC was 3% (IQR: 2–10) ans of both subsets in all PI-RADS categories. in men without csPCa and 25% (9–46) in men with csPCa (p The AUC of ROT-RC in men with PI-RADS <3 was 0.776 < 0.001), whereas those of BCN-RC were 10.5% (3.6–27.2) (95% CI: 0.661–0.832), whereas the AUC of BCN-RC was and 67.1% (39.3–85.6), respectively (p < 0.001; Fig. 1). 0.774 (0.697–0.850, p = 0.529; Fig. 5A). The AUCs were, The calibration curves of both RCs showed certain under- respectively, 0.836 (0.756–0.916) and 0.838 (0.761–0.914) estimation of csPCa (Fig. 2A and 2B). BCN-RC showed a in men with PI-RADS 3 (p = 0.954; Fig. 5D), 0.829 (0.789– slight underestimation with a calibration in the large of 0.869) and 0.737 (0.677–0.787) in men with PI-RADS 4 (p 0.25 showing a minimum difference between the mean < 0.001; Fig. 5G), and 0.866 (0.785–0.948) and 0.822 observed and the mean predicted values, with a difference (0.723–0.920) in men with PI-RADS 5 (p = 0.323; Fig. 5J). of 1.96 for ROT-RC. The slopes were 0.81 and 1.04 for A small benefit over no-biopsy men with PI-RADS <3 was ROT- and BCN-RC, respectively. BCN-RC showed values near observed in both RCs (Fig. 5B). A clear net benefit of BCN-RC the ideal value of 1. was observed over biopsying all men with PI-RADS 3 from a The discrimination ability of ROT- and BCN-RC for csPCa 10% csPCa probability, with the benefit of ROT-RC being is presented by ROC curves in Figure 3A. The AUCs were lower (Fig. 5E). In men with PI-RADS 4, BCN-RC showed a 0.856 (95% CI: 0.831–0.881) and 0.844 (0.819–0.869) net benefit over biopsying all men from a 17% csPCa proba- respectively (p = 0.116). BCN-RC showed a net benefit over bility; however, ROT-RC showed a small benefit between biopsying all men from a 15% probability threshold; ROT 50% and 75% csPCa probability (Fig. 5H). In men with PI- MRI-RC showed the benefit from a 32% probability thresh- RADS 5, BCN-RC showed a benefit over biopsy-all men from old (Fig. 3B). CUCs showed a larger clinical utility area for 38% probability of csPCa; ROT-RC exhibited a minimal ben- BCN-RC (Fig. 3C); additionally, the morphology and posi- efit from the 83% csPCa probability (Fig. 5K). tioning of CUCs were different between the two RCs, shifted CUCs showed a striking behaviour of both RCs. ROT-RC up and to the left of ROT-RC with respect to BCN-RC. The started in PI-RADS <3 with curves displaced up and to the number of missed csPCa cases and avoided biopsies in each left, evolving with the increase of the PI-RADS category threshold from 1% to 100% are presented in Supplementary towards the graph diagonal line; BCN-RC run through the Table 1. The specificities corresponding to 85%, 90%, and entire area of the graph, evolving from the top left to the 95% sensitivities of csPCa provided by both RCs are pre- bottom right with an increase in the PI-RADS category. CUCs sented in Table 2. There were similar specificities at 85% of BCN-RC showed clear clinical utility in PI-RADS 3 and 4. and 90% sensitivities, with 42.1% (95% CI: 38.0–46.3) for The number of missed csPCa cases and saved biopsies in EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 49 Fig. 1 – Violin plots of csPCa likelihoods of men without and with csPCa estimated with ROT MRI-RC and BCN MRI-RC. BCN = Barcelona; csPCa = clinically significant prostate cancer; MRI-RC = magnetic resonance imaging–based risk calculator; ROT = Rotterdam. Fig. 2 – Calibration curves of (A) ROT MRI-RC and (B) BCN MRI-RC. BCN = Barcelona; MRI-RC = magnetic resonance imaging–based risk calculator; ROT = Rotterdam. each PI-RADS category of ROT- and BCN-RC are presented in missed csPCa according to each PI-RADS category [17]. CUCs Supplementary Tables 2–5. The specificities of both RCs show how no RC appeared clinically helpful for PI-RADS <3 from 85%, 90%, and 95% sensitivities in each PI-RADS cate- because, to identify 50% of 42 csPCa detected (5% of all gory are presented in Table 3. csPCa detected), 75% of men needed biopsy (18.6% of all As the decision of prostate biopsy is made after mpMRI, biopsies) with ROT-RC and 80% (20% of all biopsies) with it is appropriate to define the acceptable percentages of BCN-RC. The rate of csPCa detection in men with PI-RADS 50 EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 Fig. 3 – (A) Discrimination ability of csPCa of BCN MRI-RC and ROT MRI-RC presented with ROC curves, (B) net benefit of BCN MRI-RC and ROT MRI-RC over biopsying all men presented by DCAs, and (C) clinical utility of BCN MRI-RC and ROT MRI-RC showing the percentage of avoided biopsies and missed csPCa according to the threshold probability of csPCa by CUCs. BCN = Barcelona; csPCa = clinically significant prostate cancer; CUC = clinical utility curve; DCA = decision curve analysis; MRI-RC = magnetic resonance imaging–based risk calculator; ROT = Rotterdam. Table 2 – Specificities of BCN MRI-RC and ROT MRI-RC corresponding to 85%, 90%, and 95% sensitivities for csPCa Risk calculator Specificity (95% CI) for sensitivities of 85% p value 90% p value 95% p value BCN MRI-RC (%) 67.3 (63.2–71.1) 0.055 52.0 (47.7–56.2) 0.392 31.8 (28.0–35.8) <0.001 ROT MRI-RC (%) 61.7 (57.5–65.7) 54.7 (50.5–58.9) 42.1 (38.0–46.3) BCN = Barcelona; CI = confidence interval; MRI-RC = magnetic resonance imaging–based risk calculator; ROT = Rotterdam. 3 was 20.4%, and it was acceptable to miss up to 10% of all men from a threshold of >20%, and misclassification of these csPCa (5.2% of all csPCa) cases. ROT-RC saved 21% of csPCa > 60% remained within these thresholds that are not biopsies, whereas BCN-RC saved 35%. In men with PI- in the utility range [23]. In contrast, BCN MRI-RC showed a RADS 4, BCN-RC saved 15% of biopsies (6% of all biopsies), net benefit over biopsying all men above the 15% threshold. missing 2% of csPCa (1% of all csPCa), whereas ROT-RC saved This threshold avoided 40% of prostate biopsies, missing 10% 10% of biopsies, with 6% of csPCa remaining undetected, of csPCa. To know the true clinical value of MRI-RCs is essen- which seems clinically unacceptable. Finally, in PI-RADS 5 tial to assess their behaviour according to the PI-RADS cate- where only 16% of biopsies were unnecessary, no RC gories, because the discrimination ability in the whole assured to avoid any biopsy without missing csPCa. population does not represent that of each PI-RADS [25]. ROT MRI-RC showed no benefit in men with PI-RADS <3, 4, and 5, as did BCN MRI-RC in PI-RADS <3 and 5. BCN MRI- 4. Discussion RC showed a net benefit in men with PI-RADS 4, saving between 6% and 23% of prostate biopsies and missing We were surprised at how far from the ideal was the cali- between 1% and 5% of csPCa detected for risk thresholds bration curve of ROT MRI-RC compared with that of BCN between 13% and 28%. In men with PI-RADS 3, ROT MRI- MRI-RC. The most probable cause was the low and narrow RC with thresholds between 1% and 3% missed between 0% csPCa risk prediction range generated by ROT-RC mainly and 12% of csPCa, saving 0.5–45% of biopsies. BCN MRI-RC located at low predicted probabilities. We also noted that at thresholds between 1% and 11% missed between 0% and 22.1% of analysed suspected PCa men showed age, serum 12% csPCa, saving between 5% and 47% of biopsies. PSA, or prostate volume out of the accepted range of ROT The possible drawback of comparing both MRI-RCs in a MRI-RC. This drawback, not reported until now, can be a population from the same metropolitan area where the consequence of the strict inclusion criteria of the ERSPC trial BCN MRI-RC was developed, we note that csPCa detection in which RCs 3 and 4 were initially developed [10,11]. These rate in the cohort of suspected PCa men in whom ROT limited ranges were not observed in the population of sus- MRI-RC was adjusted was 35.8%, very close to that of pected PCa men in whom the ROT MRI-RC was adjusted 36.9% observed in the BCN MRI-RC development cohort [12], which had similar characteristics as our population [13,17]. The characteristics of participants of this head-to- study [17]. head comparison were different from those of both the The discrimination ability of csPCa of ROT and BCN MRI- development and the adjustment cohorts of BCN and ROT RCs was similar in the entire population; however, DCAs MRI-RCs in terms of age, serum PSA, DRE, and csPCa detec- showed a net benefit of BCN MRI-RC over ROT MRI-RC. Since tion rate of 40.8%. The close origin of this population to that three-quarters of csPCa likelihoods predicted by ROT MRI- of BCN MRI-RC development did not influence the results. RC were below 24%, there was no benefit over biopsying Rather, we believe that the differences between BCN and EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 51 Fig. 4 – Violin plots of csPCa likelihoods, estimated from (A) ROT MRI-RC and (B) BCN MRI-RC, in men without and with csPCa according to the PI-RADS categories. BCN = Barcelona; csPCa = clinically significant prostate cancer; MRI-RC = magnetic resonance imaging–based risk calculator; PI-RADS = Prostate Imaging Reporting and Data System; ROT = Rotterdam. ROT MRI-RCs justify their different usefulness. BCN MRI-RC ROT MRI-RC adjustment and BCN MRI-RC development predicted the likelihood of csPCa defined as grade group 2, comparative cohorts [1], which appears to improve the while ROT MRI-RC defined as Gleason 3 + 4; the age, overall detection of csPCa [26]. serum PSA, and prostate volume ranges were limited in There are specific limitations of predictive models. A ROT MRI-RC and PI-RADSv.1 was used in its adjustment developed predictive model reflects the probability of a cohort, whereas PI-RADSv.2 was used in the BCN MRI-RC condition based on the characteristics at that time. How- development cohort. The expression of csPCa likelihoods ever, changes arising in the same population and those from without and with decimals in ROT and BCN MRI-RCs, the validation cohorts justify the need of recalibrations of respectively, may represent any bias. A limitation of both the models and adjustment of risk thresholds to ensure MRI-RCs and the present study may be that a transperineal accurate predictions [27,28]. BCN MRI-RC reports the nov- approach is currently recommended for prostate biopsy, elty of selecting the risk threshold that can be useful in whereas a transrectal approach was used in the present external validations and selecting appropriate thresholds 52 EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 Fig. 5 – Discrimination ability presented by ROC curves, net benefit presented by DCAs, and clinical utility presented by CUCs, of BCN MRI-RC and ROT MRI-RC according to the PI-RADS categories: (A–C) PI-RADS <3, (D–F) PI-RADS 3, (G–I) PI-RADS 4, and (J–L) PI-RADS 5. BCN = Barcelona; csPCa = clinically significant prostate cancer; CUC = clinical utility curve; MRI-RC = magnetic resonance imaging–based risk calculator; PI-RADS = Prostate Imaging Reporting and Data System; ROT = Rotterdam. for each PI-RADS category [17]. Real-time updating is a 5. Conclusions great challenge for future RCs [29]. Continuous feedback of new cases, big data integration, appropriate machine BCN and ROT MRI-RCs showed different behaviour in this learning algorithms, and federated networking can lead to head-to-head comparative analysis. ROT MRI-RC reported future RCs validated in each site and ensuring accurate a lower and narrower range of csPCa likelihoods than BCN and long-lasting predictions in many places [30]. MRI-RC. BCN MRI-RC presented a net benefit over ROT EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 53 Table 3 – Specificities of BCN MRI-RC and ROT MRI-RC corresponding to 85%, 90%, and 95% sensitivities for csPCa according to the PI-RADSv.2 category Risk calculator Specificity (95% CI) for sensitivities of 85% p value 90% p value 95% p value PI-RADS <3 BCN MRI-RC (%) 49.7 (45.5–54.0) 0.188 44.0 (39.9–48.3) 0.456 9.8 (7.6–12.7) <0.001 ROT MRI-RC (%) 53.8 (49.6–58.0) 41.7 (37.6–45.9) 24.5 (21.1–28.3) PI-RADS 3 BCN MRI-RC (%) 78.1 (74.4–81.4) <0.001 50.6 (46.4–54.8) 0.102 15.6 (12.8–19–0) <0.001 ROT MRI-RC (%) 60.0 (55.8–64.0) 45.6 (41.4–49.8) 25.5 (22.0–29.4) PI-RADS 4 BCN MRI-RC (%) 62.2 (58.1–66.2) <0.001 49.5 (45.3–53.7) <0.001 39.9 (32.9–41.0) <0.001 ROT MRI-RC (%) 29.7 (26.0–33.7) 20.2 (17.0–23.8) 4.6 (3.1–6.8 PI-RADS 5 BCN MRI-RC (%) 63.2 (59.0–67.1) 1.000 52.6 (48.4–56.8) 0.456 36.8 (32.9–41.0) <0.001 ROT MRI-RC (%) 63.2 (59.0–67.1) 60.5 (56.3–64.6) 52.6 (48.4–56.8) BCN = Barcelona; CI =confidence interval; MRI-RC = magnetic resonance imaging–based risk calculator; PI-RADS = Prostate Imaging Reporting and Data System; ROT = Rotterdam. MRI-RC and grater clinical utility in the entire population. screening, diagnosis, and local treatment with curative intent. Eur Urol 2021;79:243–62. According to the PI-RADS category, BCN MRI-RC was helpful [2] Van Poppel H, Roobol MJ, Chapple CR, et al. Prostate-specific in men with PI-RADS 3 and 4, whereas ROT MRI-RC was antigen testing as part of a risk-adapted early detection strategy for helpful only in men with PI-RADS 3. No MRI-RC was helpful prostate cancer: European Association of Urology position and in men with PI-RADS <3 and 5. recommendations for 2021. Eur Urol 2021;80:703–11. [3] Van Poppel H, Hogenhout R, Albers P, van den Bergh RCN, Barentsz JO, Roobol MJ. European model for an organised risk-stratified early Author contributions: Juan Morote had full access to all the data in the detection programme for prostate cancer. Eur Urol Oncol 2021;10: study and takes responsibility for the integrity of the data and the accu- 731–9. racy of the data analysis. [4] Schoots IG, Padhani AR, Rouvière O, Barentsz JO, Richenberg J. Analysis of magnetic resonance imaging-directed biopsy strategies for changing the paradigm of prostate cancer diagnosis. Eur Urol Study concept and design: Morote, Borque-Fernando, Esteban. Oncol 2020;3:32–41. [5] Sathianathen NJ, Omer A, Harriss E, et al. Negative predictive value Acquisition of data: Triquell, Campistol, Servian, Abascal, Mendez, Planas. of multiparametric magnetic resonance imaging in the detection of Analysis and interpretation of data: Morote, Borque-Fernando, Esteban. clinically significant prostate cancer in the prostate imaging Drafting of the manuscript: Morote. reporting and data system era: a systematic review and meta- Critical revision of the manuscript for important intellectual content: analysis. Eur Urol 2020;78:402–14. Borque-Fernando, Esteban, Planas, Mendez, Trilla. [6] Wagaskar VG, Levy M, Ratnani P, et al. Clinical utility of negative multiparametric magnetic resonance imaging in the diagnosis of Statistical analysis: Esteban, Borque-Fernando, Morote. prostate cancer and clinically significant prostate cancer. Eur Urol Obtaining funding: Morote. Open Sci 2021;28:9–16. Administrative, technical, or material support: Morote. [7] Mazzone E, Stabile A, Pellegrino F, et al. Positive predictive value of Supervision: Morote, Trilla. Prostate Imaging Reporting and Data System version 2 for the detection of clinically significant prostate cancer: a systematic Other: None. review and meta-analysis. Eur Urol Oncol 2020;4:697–713. [8] Osses DF, Roobol MJ, Schoots IG. Prediction medicine: biomarkers, risk calculators and magnetic resonance imaging as risk Financial disclosures: Juan Morote certifies that all conflicts of interest, stratification tools in prostate cancer diagnosis. Int J Mol Sci 2020; including specific financial interests and relationships and affiliations rel- 20:1637. evant to the subject matter or materials discussed in the manuscript (eg, [9] Triquell M, Campistol M, Celma A, et al. Magnetic resonance employment/affiliation, grants or funding, consultancies, honoraria, stock imaging-based predictive models for clinically significant prostate cancer: a systematic review. Cancers (Basel) 2022;14:4747. ownership or options, expert testimony, royalties, or patents filed, [10] Steyberg R, Roobol-Bouts MJ, Kranse M, Schroder FH. Data storage received, or pending), are the following: None. device and method for determining the dependency of the risk for prostate cancer, device and method for indicating a risk for a disease in an individual. U.S. Patent Office, 8,087,576. Funding/Support and role of the sponsor: This study was supported by [11] Roobol MJ, Steyerberg EW, Kranse R, et al. A risk-based strategy the Instituto Carlos III (SP) and the European Union (project ref: improves prostate-specific antigen-driven detection of prostate PI20/01666). cancer. Eur Urol 2010;57:79–85. [12] Alberts AR, Roobol MJ, Verbeek JFM, et al. Prediction of high-grade prostate cancer following multiparametric magnetic resonance imaging: improving the Rotterdam European Randomized Study Appendix A. Supplementary data of Screening for Prostate Cancer risk calculators. Eur Urol 2019; 75:310–8. [13] Püllen L, Radtke JP, Wiesenfarth M, et al. External validation of Supplementary data to this article can be found online at novel magnetic resonance imaging-based models for prostate https://doi.org/10.1016/j.euros.2023.03.013. cancer prediction. BJU Int 2020;125:407–16. [14] Chen R, Verbeek JFM, Yang Y, Song Z, Sun Y, Roobol MJ. Comparing the prediction of prostate biopsy outcome using the Chinese References Prostate Cancer Consortium (CPCC) risk calculator and the Asian adapted Rotterdam European Randomized Study of Screening for [1] Mottet N, van den Bergh RCN, Briers E, et al. EAU-EANM-ESTRO- Prostate Cancer (ERSPC) risk calculator in Chinese and European ESUR-SIOG guidelines on prostate cancer—2020 update. Part 1: men. World J Urol 2021;39:73–80. 54 EUROPEAN UROLOGY OPEN SCIENCE 53 (2023) 46–54 [15] Petersmann AL, Remmers S, Klein T, et al. External validation of two 2012 updated Partin tables vs a European predictive nomogram for MRI-based risk calculators in prostate cancer diagnosis. World J organ-confined disease in prostate cancer. BJU Int 2014;113: Urol 2021;39:4109–16. 878–86. [16] De Nunzio C, Lombardo R, Baldassarri V, et al. Rotterdam mobile [25] Moonesingh SR, Bashford T, Wagstaff D. Implementing risk phone app including MRI data for the prediction of prostate cancer: calculators: time for the Trojan horse? Anesth Analg 2018;121: a multicenter external validation. Eur J Surg Oncol 2021;47:2640–5. 1192–6. [17] Morote J, Borque-Fernando A, Triquell M, et al. The Barcelona [26] Zattoni F, Marra G, Kasivisvanathan V, et al. The detection of predictive model of clinically significant prostate cancer. Cancers prostate cancer with magnetic resonance imaging-targeted (Basel) 2022;14:1589. prostate biopsies is superior with the transperineal vs the [18] Epstein JI, Zelefsky MJ, Sjoberg DD, et al. A contemporary prostate transrectal approach. A European Association of Urology-Young cancer grading system: a validated alternative to the Gleason score. Academic Urologists Prostate Cancer Working Group multi- Eur Urol 2016;69:428–35. institutional study. J Urol 2022;208:830–7. [19] Whitney J. Testing for differences with the nonparametric Mann- [27] Morote J, Borque-Fernando Á, Triquell M, Esteban LM, Trilla E. The Whitney U test. J. Wound Ostomy Continence Nurs 1997;24:12. true utility of predictive models based on magnetic resonance [20] Plackett RL. Karl Pearson and the chi-squared test. Int Stat Rev imaging in selecting candidates for prostate biopsy. Eur Urol Open 1983;51:59–72. Sci 2022;42:40–1. [21] Creelman CD, Donaldson W. ROC curves for discrimination of linear [28] Diniz MA. Statistical methods for validation of predictive models. J extent. J Exp Psychol 1968;77:514–6. Nucl Cardiol 2022;29:3248–55. [22] DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas [29] Strobl AN, Vickers AJ, Van Calster B, et al. Improving patient under two or more correlated receiver operating characteristic prostate cancer risk assessment: moving from static, globally- curves: a nonparametric approach. Biometrics 1988;44:837–45. applied to dynamic, practice-specific risk calculators. J Biomed [23] Vickers AJ, Elkin EB. Decision curve analysis: a novel method for Inform 2015;56:87–93. evaluating prediction models. Med Decis Making 2006;26:565–74. [30] Nandi A, Xhafa F. A federated learning method for real-time [24] Borque Á, Rubio-Briones J, Esteban LM, et al. Implementing the use emotion state classification from multi modal streaming. Methods of nomograms by choosing threshold points in predictive models: 2022;204:340–4.