Reducing prostate biopsies and magnetic resonance imaging with prostate cancer risk stratification

Abstract Objectives To recalibrate and validate the European Randomized Study of Screening for Prostate Cancer risk calculators (ERSPC RCs) 3/4 and the magnetic resonance imaging (MRI)‐ERSPC‐RCs to a contemporary Norwegian setting to reduce upfront prostate multiparametric MRI (mpMRI) and prostate biopsies. Patients and Methods We retrospectively identified and entered all men who underwent prostate mpMRI and subsequent prostate biopsy between January 2016 and March 2017 in a Norwegian centre into a database. mpMRI was reported using PI‐RADS v2.0 and clinically significant prostate cancer (csPCa) defined as Gleason ≥ 3 + 4. Probabilities of csPCa and any prostate cancer (PCa) on biopsy were calculated by the ERSPC RCs 3/4 and the MRI‐ERSPC‐RC and compared with biopsy results. RCs were then recalibrated to account for differences in prevalence between the development and current cohorts (if indicated), and calibration, discrimination and clinical usefulness assessed. Results Three hundred and three patients were included. The MRI‐ERSPC‐RCs were perfectly calibrated to our cohort, although the ERSPC RCs 3/4 needed recalibration. Area under the receiver operating curve (AUC) for the ERSPC RCs 3/4 was 0.82 for the discrimination of csPCa and 0.77 for any PCa. The AUC for the MRI‐ERSPC‐RCs was 0.89 for csPCa and 0.85 for any PCa. Decision curve analysis showed clear net benefit for both the ERSPC RCs 3/4 (>2% risk of csPCa threshold to biopsy) and for the MRI‐ERSPC‐RCs (>1% risk of csPCa threshold), with a greater net benefit for the MRI‐RCs. Using a >10% risk of csPCa or 20% risk of any PCa threshold for the ERSPC RCs 3/4, 15.5% of mpMRIs could be omitted, missing 0.8% of csPCa. Using the MRI‐ERSPC‐RCs, 23.4% of biopsies could be omitted with the same threshold, missing 0.8% of csPCa. Conclusion The ERSPC RCs 3/4 and MRI‐ERSPC‐RCs can considerably reduce both upfront mpMRI and prostate biopsies with little risk of missing csPCa.


| INTRODUCTION
The diagnostic work-up of men suspected of having prostate cancer (PCa) includes biopsy of the prostate, which is associated with a risk of infection (1%-9.6%) and an increasing risk of life-threatening sepsis (1%-4%) due to increasing fluoroquinolone resistance. [1][2][3][4] Multivariable risk calculators (RCs) that use clinical parameters to predict the likelihood of finding PCa on biopsy have been developed and can be used to omit biopsy when the predicted likelihood of cancer is low.
The RCs stemming from the Dutch arm of the European Randomized Study of Screening for Prostate Cancer (ERSPC) are some of the best known and validated multivariable RCs. The ERSPC RC3, for men without prior biopsy, and ERSPC RC4, for men with prior negative biopsy, can contribute to avoiding 20%-33% of unnecessary systematic transrectal ultrasound (TRUS)-guided biopsies. [5][6][7][8] The ERSPC RCs 3/4 use prostate-specific antigen (PSA), prostate volume, digital rectal examination (DRE) findings and previous biopsy history as predictors in logistic regression models that predict biopsy results. The North American Prostate Cancer Prevention Trial (PCPT)RC 2.0 is another well-known RC that stemmed from the placebo arm of the PCPT. 9 In a head-tohead comparison of the ERSPC RCs and the PCPTRC 2.0, as well as other current RCs, the ERSPC-RCs showed superior discrimination for detecting clinically significant PCa (csPCa) as well as the greatest net benefit, followed by the PCPTRC 2.0. 7,8,10 Multiparametric magnetic resonance imaging (mpMRI) of the prostate has emerged as an important investigation in PCa diagnostics due to its high negative predictive value and sensitivity in detecting csPCa 11,12 and is recommended before performing biopsy in the European Association of Urology (EAU) 2021 Prostate Cancer Guidelines. 13 RCs that incorporate information from mpMRI have been developed, including the MRI-ERSPC-RCs. [14][15][16] These outperform non-mpMRI RCs, 17,18 but at the cost of performing an mpMRI examination. Prediction models need to be calibrated and validated to the population on which they will be used before implementation. No RCs have been developed or validated for Scandinavian countries. The aim of this study was to facilitate a reduction of unnecessary mpMRI examinations and prostate biopsy procedures by calibrating and validating the ERSPC RCs 3/4 and MRI-ERSPC-RCs in a Norwegian university hospital cohort.

| Imaging
All mpMRI were obtained using a 3 T MRI system (Skyra; Siemens, Erlangen, Germany) without an endorectal coil. The imaging protocol included T2-weighted imaging in three planes, axial diffusionweighted imaging and dynamic contrast-enhanced imaging, in accordance with PI-RADS v2.0 19 recommendations. All mpMRI were interpreted prospectively by or under the supervision of expert uroradiologists (with 2-7 years of experience in prostate mpMRI interpretation). PI-RADS v2.0 and a standardized reporting template with a maximum of three lesions were used for all reports. All identified lesions were assigned a PI-RADS v2.0 score for each imaging sequence and an overall PI-RADS v2.0 score.

| Biopsy
After mpMRI, all men underwent either systematic TRUS-guided biopsy (n = 264), transrectal magnetic resonance-guided in-bore biopsies (MRGBs) (n = 35) or transperineal ultrasound-guided cognitively targeted biopsies (n = 4). Nine patients underwent both systematic TRUS biopsy and MRGB as part of another study (REC Central Norway Study identifier REK2013/1869). All patients received prophylactic antibiotics in accordance with national guidelines at the time. Systematic TRUS-guided biopsies were obtained under local anaesthesia and ultrasound guidance by a BK medical ultrasound system (Specto, BK medical, Herlev, Denmark) with a triplane transrectal probe. Biopsies were obtained in an extended core pattern consisting of two biopsies from the basis, mid-gland and apex of each side of the prostate and one core from the central zone of each side. The treating urologists were not blinded to the findings of the preceding mpMRI. Transrectal MRGBs were performed under lidocaine 2% gel (Xylocain, Aspen Nordic, Ballerup, Denmark) anaesthesia by one of two experienced radiologists.
Transperineal biopsies were obtained under general anaesthesia by the same radiologists, with cognitive ultrasound-guided targeting in the case of PI-RADS ≥ 3 on mpMRI. All histopathological specimens were analysed by dedicated uropathologists and graded according to International Society of Urological Pathology (ISUP) standards. 20 A biopsy was classified as positive for any PCa with any finding of cancer and csPCa defined as Gleason score ≥ 3 + 4 (ISUP grade group ≥ 2) in any number or length of biopsies. Indolent cancers were defined as Gleason 3 + 3 (ISUP grade group 1), independent of cancer core length.

| Statistical analyses
Missing DRE findings were imputed (MICE package for R, version 3.13.00). 21 Prostate volumes were calculated from T2 mpMRI by the ellipsoid formula (height Â length Â width Â π/6). The probabilities of detecting any PCa and csPCa on biopsy were calculated, using the ERSPC RC3 for biopsy naïve, the ERSPC RC4 for previously biopsied patients and the MRI-ERSPC-RCs, blinded to biopsy outcomes. Biopsy results were subsequently disclosed, and RCs re-calibrated based on the intercept in the large to account for differences in prevalence between the development and validation populations. Calibration slopes were calculated and visualized in calibration plots, both from the uncalibrated and re-calibrated RCs. 22

| Power calculations
On the basis of the framework of Snell et al., 24 we simulated the expected width of the 95% CIs of the calibration slope, the calibration-in-the-large and AUC using a prevalence of any PCa of 50% and a standard deviation of the linear predictor of 1.6. With a sample size of 300 men, we are powered to find a width of the 95% CI of 0.49 for the calibration slope, 0.56 for the calibration-in-thelarge and 0.09 for the AUC.

| Patient characteristics
A total of 473 patients underwent mpMRI during the inclusion period.
Seventy-seven patients with a prior history of PCa and 72 patients who did not undergo prostate biopsy after mpMRI were excluded, as were 19 patients where mpMRI was not performed prior to biopsy, and two patients with prostatic abscess as the reason for referral. For csPCa, the intercept in the large was 1.44 (95% CI 1.14-1.73) and the calibration curve slope 0.81 (95% CI 0.63-1.02). Calibration curves before and after recalibration shown in Figure 1.
DCA for the detection of csPCa showed a net benefit for the recalibrated ERSPC RCs 3/4 for all threshold probabilities >2% compared with the 'biopsy all' approach. For the detection of any PCa, net benefit compared with the 'biopsy all' approach was found for threshold probabilities >24% (Figure 3). approach, as well as compared with the ERSPC RCs 3/4 ( Figure 3).

| Sensitivity analyses
We performed sensitivity analyses by removing patients who received targeted biopsies by MRGB (n = 35) and transperineal cognitive targeted biopsies (n = 4); this did not alter the RC's discriminative ability significantly (data not shown). Choosing an optimal probability threshold for when to recommend mpMRI or prostate biopsy is difficult, 23