Development and validation of a nomogram based on biparametric MRI PI-RADS v2.1 and clinical parameters to avoid unnecessary prostate biopsies

Prostate cancer (PCa) is the second most common cancer in males worldwide [1]. Standard screening using prostate-specific antigen (PSA) levels has led to an overall reduction in mortality from PCa. Regrettably, PSA screening is also associated with the underdetection of clinically significant prostate cancer (csPCa) and the overdetection of clinically insignificant cancer [2]. Clinically insignificant cancers are cancers that are unlikely to progress or affect a man's life expectancy and therefore do not require immediate treatment. In contrast, csPCa exhibits greater aggressiveness and has a higher mortality rate. Prostate biopsy is the traditional diagnostic pathway in the detection of csPCa. Despite positive improvements in prebiopsy preparation, hematuria, hematospermia and infections remain the main adverse effects after the procedure. A noninvasive, easy-to-administer testing pathway that accurately diagnoses csPCa and ultimately avoids unnecessary biopsies is a major unmet need.

Risk-based patient selection for prostate biopsy has been used in daily clinical practice, either through empirical judgment or through the use of risk calculators. The use of multivariate risk calculators, such as the European Randomized Study of Screening for Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial Risk Calculator (PCPT-RC), has been demonstrated to avoid 20–35% of unnecessary biopsies in a number of external validation studies [3, 4]. Multiparametric magnetic resonance imaging (mpMRI) of the prostate has significantly improved the diagnostic accuracy for csPCa. The Prostate Imaging and Reporting Data System (PI-RADS) was first developed to standardize prostate MRI acquisition, data reporting, and interpretation, with version 1 published in 2012 and version 2.1 published in 2019 [5]. However, up to 24% of patients with a Gleason grade ≥ 2 may still be missed by mpMRI alone [2, 6]. Recent studies have shown that mpMRI can improve csPCa detection and risk stratification when incorporated into the available clinical parameters [7‐9]. Unfortunately, the full PI-RADS compliant protocol is time-consuming (~ 40 min), and expensive testing might be difficult to implement on a large scale, especially for active surveillance. It is noteworthy that the cohorts reported by Zhang et al. [10] and Tamada et al. [11] demonstrated that the diagnostic performance of biparametric MRI (bpMRI) was comparable to that of mpMRI by using PI-RADS v2.1 to detect csPCa. Theoretically, a predictive model based on this faster, cheaper and contrast-free bpMRI protocol may be an effective way to safely reduce unnecessary prostate biopsies.

Early diagnosis and timely treatment of csPCa can help improve the life expectancy of patients with PCa [12]. Although there are many different treatment options for PCa, aggressive PCa requires urgent, curative treatment [13]. However, active surveillance is a possible strategy for clinically insignificant PCa. Since mpMRI is part of the diagnostic pathway, it is necessary to use simple MRI-derived parameters to assess the aggressiveness of PCa. Previous studies have shown that quantitative MRI parameters can be used to assess tumor aggressiveness [14, 15]. However, due to inconsistent methods in different studies, heterogeneous results have been observed.

The objectives of this study were (1) to investigate the added value of the bpMRI PI-RADS v2.1 score for a clinical-only model to avoid unnecessary prostate biopsies and (2) to demonstrate whether there is a correlation between the simple bpMRI-derived score and PCa aggressiveness.

Biparametric MRI is increasingly being used to characterise prostate cancer. This study confirms that when bpMRI is incorporated into the prediction model, the model exhibits better fit and higher diagnostic accuracy, with fewer unnecessary biopsies compared to the clinical model. Furthermore, we demonstrated that the bpMRI PI-RADS v2.1 score was strongly correlated with the ISUP grade, which may give clinicians prebiopsy information about the aggressiveness of PCa. Finally, the results of sensitivity analysis demonstrated that our major findings were relatively reliable.

Remarkably, among the previously well-known risk calculators (ERSPC-RC and PCPT-RC), PSAd is not included. According to recent studies, PSAd not only predicted the outcome of the biopsy but was also a predictor of MRI equivocal lesions (PI-RADS score = 3) [24, 25]. In addition, Cuocolo et al. [26] demonstrated that PSAd derived from MRI correlated more significantly with PCa aggressiveness than the value measured from transrectal ultrasonography (TRUS). Therefore, it is not surprising that in our study, PSAd was one of the strong predictors in the clinical model. Similar to previously developed risk calculators [27, 28], our research also demonstrated that DRE findings are an independent predictor for csPCa detection. Overall, the clinical model achieved an AUC of 0.796 in detecting csPCa in the development cohort, which is consistent with previous studies [29, 30], suggesting that more efforts are needed to improve diagnostic accuracy.

In all cohorts of this study, the model based on the bpMRI-RADS v2.1 score performed better than the clinical model, as illustrated by the increase in AUC values. In addition, the clinical model was not calibrated as well as the bpMRI PI-RADS v2.1-based model. Importantly, with the increasing risk thresholds in DCA, the use of the bpMRI PI-RADS v2.1-based model can avoid more biopsies while maintaining higher csPCa detection rates compared to the clinical model, suggesting that the bpMRI PI-RADS v2.1-based model has higher value in reducing unnecessary biopsies. Understandably, the patient and the urologist can share the decision-making process to determine acceptable risk and biopsy thresholds to avoid missing csPCa [8]. In clinical practice, overtreatment of less invasive PCa decreases quality of life, but delayed treatment of csPCa increases treatment costs and mortality. Compared to mpMRI examination and biopsy, the bpMRI protocol is clearly a rapid, noninvasive, and less costly testing option, and the test trade-off can be considered low in clinical scenarios; therefore, the urologists would agree to perform 10 fewer bpMRIs to avoid one unnecessary biopsy with a risk threshold of 10–20%.

Novel risk tools based on clinical variables and additional genetic and/or protein-based biomarkers have been demonstrated to help avoid unnecessary biopsies, but they are laboratory dependent and expensive [27, 31]. In addition, in contrast to MRI-based stratification models, these risk models cannot determine the location or size of tumors within the prostate and therefore cannot be used to guide targeted biopsies. Multiple previous models that combine mpMRI findings with clinical variables have shown a 3–20% improvement in diagnostic accuracy [2, 18]. Recently, a small number of bpMRI-based nomograms have been developed [32‐34]. In contrast to our study, a study by Boesen et al. [34] used bpMRI results to build a predictive nomogram. However, in his research, the bpMRI score was based on PI-RADS v2, and the results lacked external validation. The bpMRI-based nomograms constructed by Lee et al. [35] achieved a 92% diagnostic rate for csPCa, but nearly 80% of these patients underwent primary biopsies, which may have led to an overestimation of the diagnostic accuracy of the model. The implementation of prebiopsy mpMRI in all men with suspected PCa constitutes a fundamental paradigm shift in PCa diagnosis and could impose a tremendous financial and resource burden on the health care system [36]. Since the bpMRI protocol represents a cost-effective procedure, this also accounts for its lesion identification in terms of high sensitivity [37], which may be helpful in low-risk patients who might be candidates for active surveillance.

The ISUP grade is a measure of cancer aggressiveness. Several MRI-based functional parameters and radiomic signatures have been developed for the assessment of the biological aggressiveness of PCa [38‐40]. However, no standardized imaging protocols have been available, and subjective measurements of the ROI depend on the experience and expertise of the radiologist, thus limiting the accuracy and reproducibility of the results. Previous studies concluded that the PI-RADS score can be used to predict lymph node involvement and extraprostatic extension [41] but only weakly correlates with the ISUP grade [42, 43]. In the present study, we demonstrated a strong correlation between the bpMRI PI-RADS v2.1 score and PCa aggressiveness, which is consistent with research by Pan et al. [32], suggesting that the bpMRI score can be used to predict the prebiopsy ISUP grade and potentially improve treatment planning.

Our study is not without limitations. First, this was a retrospective study, which may lead to patient selection bias. Further prospective, well-designed, large cohort studies are needed to confirm our findings. Second, the actual detection rate of csPCa may be underestimated compared to studies using template-mapped biopsies or whole-gland prostatectomy. Third, the data were interpreted by experienced radiologists at each institution using PI-RADS v2.1 for bpMRI; this may limit the generalizability of our results to less experienced institutions. Finally, assessing interobserver agreement for bpMRI PI-RADS v2.1 was beyond the scope of this study; however, van der Leest et al. [44] concluded that the interobserver agreement for biparametric MRI exceeded 90%.

In conclusion, a predictive model based on the bpMRI score and clinical parameters is an easily accessible tool for avoiding unnecessary biopsies, and furthermore, bpMRI scores derived from PI-RADS v2.1 can be used to determine the aggressiveness of PCa prior to biopsy.