The True Utility of Predictive Models Based on Magnetic Resonance Imaging in Selecting Candidates for Prostate Biopsy.

Early detection of clinically significant prostate cancer (csPCa) has improved since the introduction of magnetic resonance imaging (MRI) and guided biopsies for this purpose; however, suspicion of PCa is still based on elevated serum prostate-specific antigen and/or abnormal digital rectal examination. At present, the decision to perform a prostate biopsy primarily depends on the Prostate Imaging-Reporting and Data System (PI-RADS) score for lesions observed on MRI. Prostate biopsy is typically avoided when negative MRI (PI-RADS <3) is reported owing to its high negative predictive value; nevertheless, uncertain scenarios remain [1]. Predictive models based on MRI (MRI-PMs) are appropriate tools for improving the selection of candidates for prostate biopsy when risk calculators (RCs) are available; however, external validation is essential to ensure accurate prediction [2].

Alberts et al [3] adjusted the Rotterdam MRI-RC using data for 961 men with PCa suspicion who were recruited in Düsseldorf and four Dutch cities and underwent systematic biopsies ± MRI-guided biopsies in the case of PI-RADS v1 ≥3 lesions. High-grade PCa (Gleason score ≥ 3 + 4) was detected in 35.9%. The recent validation of the Rotterdam MRI-RC in the PRECISION trial population required recalibration and model adaptation of the risk threshold to achieve proper predictions [4].

The Barcelona RC was designed, after MRI-PM development and external validation, using data for 2486 men with suspected PCa from the Barcelona metropolitan area, with csPCa (grade group ≥2) detection of 37.6% [5]. The Barcelona RC uses the same predictors as the Rotterdam MRI-RC in addition to PCa family history and PI-RADS v.2 (https://mripcaprediction.shinyapps.io/MRIPCaPrediction/). The Barcelona RC incorporates the new option of selecting the proper risk threshold and it is the first to analyse risk according to PI-RADS categories. The authors concluded that the global efficiency of RCs does not translate to a true utility for each PI-RADS category [5].

We performed a brief comparison of the Rotterdam MRI-RC and the Barcelona RC using data for 567 men from the Barcelona metropolitan area who underwent 3-T multiparametric MRI and 12-core TRUS systematic biopsy with or without two to four MRI-guided biopsies (in cases of PI-RADS ≥3). The csPCa detection rate was 40.9%. The area under the receiver operating characteristic curve was 0.866 (95% confidence interval [CI] 0.836–0.898) for the Rotterdam MRI-RC and 0.888 (95% CI 0.861–0.951) for the Barcelona RC (p = 0.016, Fig. 1A), with specificity for 95% sensitivity of 34.2% (95% CI 29.2–34.6%) and 57.6% (95% CI 52.1–62.9%), respectively (p < 0.001). The Barcelona RC exhibited a global net benefit over the Rotterdam MRI-RC (Fig. 1B). The Barcelona RC showed a net benefit over the Rotterdam MRI-RC for men with a PI-RADS 3 or 4 lesion; however, neither RC exhibited clinical usefulness for men with negative MRI or a PI-RADS 5 lesion (Supplementary Fig 1).

Fig. 1

(A) Receiver operating characteristic curves showing the efficacy of the Barcelona RC and the MRI-Rotterdam RC for detection of clinically significant prostate cancer detection. (B) Decision curve analysis for both models showing the net benefit over a biopsy-all strategy. MRI = magnetic resonance imaging; RC = risk calculator.

We suggest that locally developed and validated MRI-RCs are more efficient than external RCs. This is mainly because of variations in the incidence of csPCa and the mix of PI-RADS and population characteristics. Nevertheless, the true clinical utility of MRI-RCs must be analysed against each PI-RADS category.

The authors have nothing to disclose.

This study was supported in part by Instituto de Salut Carlos III (PI20/01666).