Introduction

Prostate cancer (PCa) is the most common malignancy and the second leading cause of cancer-related deaths among men worldwide1. Over the past few decades, serum prostate-specific antigen (PSA)–based screening and subsequent prostate biopsy (PB) have become standard diagnostic practices, despite challenges such as overdiagnosis and overtreatment2,3.

Although transrectal ultrasonography (TRUS)–guided PB (TRUS-PB) has been in use since 19894, its false-negative rate ranges from 20 to 40%5. Consequently, managing patients with prior negative PB remains a significant challenge.

Magnetic resonance imaging (MRI) has emerged as a powerful tool to enhance diagnostic accuracy, particularly in patients with previous negative PB, by reducing unnecessary biopsies6,7,8. Combining MRI-targeted PB with systematic TRUS-PB has been shown to improve cancer detection9.

There is a consensus to perform MRI and subsequent MRI-targeted PB in patients with prior negative PB but persistent clinical suspicion of PCa. The decision to perform an MRI on these patients should consider the results of other biomarkers and the financial burden of the examination 10. Routine use of MRI could impose significant costs on patients and healthcare systems 7,11. Furthermore, the recommendation to conduct MRI-targeted PB after initial negative TRUS-guided PB remains questionable, particularly for patients with PSA < 10 ng/mL, due to low cancer-specific mortality12.

Detection rates for significant PCa on repeat PB following initial negative systematic and MRI-targeted PB range from 10 to 20%13,14,15,16. Therefore, determining appropriate MRI indications in patients with previous negative PB is essential. Although PSA and its derivatives, such as PSA density (PSAD) and the free-to-total PSA ratio (%fPSA), have been suggested as indicators for recommending PB17,18, studies evaluating their utility in minimizing unnecessary MRI remain limited. This study aims to assess the effectiveness of PSA and its derivatives in reducing MRI usage without comprising the detection of PCa in patients with prior negative biopsies.

Materials and methods

Data collection

This single-institution retrospective study was approved by the Institutional Review Board (IRB No. B-2410-f933-105), with a waiver of informed consent. At our institution, all male patients presenting with lower urinary tract symptoms undergo serum PSA testing and TRUS as part of their initial diagnostic workup. An additional serum-free PSA test is reimbursed by Korea’s national insurance when serum PSA levels exceed 2 ng/mL or PCa is suspected based on digital rectal examination or TRUS findings. Further MRI is performed following shared decision-making with the patient.

Between October 2015 and June 2024, 1,189 patients underwent repeated PB after serum PSA, free PSA testing, and MRI. Patients with a history of PCa, prostate surgery, prior PB, or use of medications affecting PSA levels (e.g., 5-α reductase inhibitors) were excluded. Ultimately, 251 patients with at least one previous negative PB were included in the study. A Gleason score ≥ 7 was classified as significant PCa.

A visible lesion on MRI was defined as one with a Prostate Imaging Reporting and Data System version 2 (PI-RADSv2) score of ≥ 3, which served as the target for MRI-targeted biopsy. Patients with MRI-visible lesions underwent MRI-targeted transrectal (TR) PB in addition to a systematic 12-core TRUS-PB. Each targeted biopsy included a minimum of two cores per lesion. Patients without visible MRI lesions underwent only systematic 12-core TRUS-PB.

The MRI/TRUS fusion-guided PB protocol used in this study has been described previously19. All patients’ prostate volumes were evaluated based on TRUS. All PB slides were reviewed by genitourinary pathologists with more than 10 yr of experience. The primary outcome was the predictive value of the Prostate Health Index (PHI) for PI-RADSv2 ≥ 3 lesions. Secondary outcomes included the diagnostic accuracy of PHI for overall PCa and significant PCa.

Statistical analysis

All statistical analyses were performed using the R statistical package (version 3.5.1; R Core Team, Vienna, Austria). Receiver operating characteristic (ROC) curves and areas under the curve (AUCs) were used to determine cutoff values for PI-RADSv2 in predicting overall and significant PCa. Cutoff values were defined using sensitivity analysis with the Youden index (sensitivity + specificity − 1). ROC curve analysis also determined optimal cutoff values for PSA and its derivatives in predicting MRI lesions. The DeLong test was used to assess statistical differences in AUCs. Multivariate logistic regression analyses were used to identify predictors for MRI lesions, overall PCa, and significant PCa on MRI-targeted PB cores. All p values were two-sided, with statistical significance set at p < 0.05.

Results

Table 1 presents the baseline demographic and clinicopathological characteristics of the patients. Among 251 patients, 127 (50.6%) patients had visible MRI lesions (PI-RADSv2 ≥ 3). Of these, 125 patients underwent MRI/TRUS fusion PB, while two patients with PI-RADSv2 ≥ 3 lesions and 124 patients without visible lesions underwent only systematic 12-core TRUS-PB. Patients with PI-RADSv2 < 3 lesions were older than those with PI-RADSv2 ≥ 3 (66.4 vs. 62.9 yr; p = 0.001). Higher PSA level (10.02 ± 9.31 ng/mL vs. 6.82 ± 5.39 ng/mL; p = 0.001) and PSAD level (0.26 ± 0.28 ng/mL2 vs. 0.15 ± 0.11 ng/mL2; p < 0.001) were observed in patients with PI-RADSv2 ≥ 3 (Table 1).

Table 1 Comparison of baseline characteristics of patients with prior negative prostate biopsy (n = 251) according to the presence of visible lesions (PI-RADSv2 ≥ 3) on multiparametric magnetic resonance imaging.
Full size table

ROC curve analysis identified PI-RADSv2 ≥ 4 as the optimal cutoff value for predicting overall PCa (AUC = 0.73, p < 0.001) and significant PCa (AUC = 0.81, p < 0.001; Fig. 1). The optimal cutoff values for predicting PI-RADSv2 ≥ 4 lesions were 11.87 ng/mL for PSA (AUC = 0.63, p < 0.001), 0.19 ng/mL2 for PSAD (AUC = 0.69, p < 0.001), and 18.76% for %fPSA (AUC = 0.63, p < 0.001; Fig. 2).

Fig. 1
Fig. 1
Full size image

Receiver operating characteristic curves of prostate-specific antigen (PSA) derivatives, including PSA density (PSAD) and the free-to-total PSA ratio (%fPSA), along with Prostate Imaging Reporting and Data System version 2 (PI-RADSv2) scores on multiparametric magnetic resonance imaging for predicting the presence of (A) overall prostate cancer and (B) clinically significant prostate cancer.

Fig. 2
Fig. 2
Full size image

Receiver operating characteristic curve of age, prostate-specific antigen (PSA), prostate volume, PSA density (PSAD), and the free-to-total PSA ratio (%fPSA) for predicting the presence of lesions on multiparametric magnetic resonance imaging with a Prostate Imaging Reporting and Data System version 2 (PI-RADSv2) score ≥ 4.

Multivariable regression analysis showed that older age (odds ratio [OR] = 1.04; 95% confidence interval [95% CI], 1.01–1.08; p = 0.016) and higher PSAD (OR = 154.44; 95% CI, 18.54–1,286.36; p < 0.001) were significantly associated with PI-RADSv2 ≥ 4 lesions (Table 2).

Table 2 Univariable and multivariable regression analysis for the prediction of lesions on multiparametric magnetic resonance imaging with Prostate Imaging Reporting and Data System score ≥ 4 in patients who underwent multiparametric magnetic resonance imaging.
Full size table

Among 125 patients who underwent combined MRI-targeted PB and systematic 12-core TRUS-PB, PSAD ≥ 0.19 ng/mL and PI-RADSv2 ≥ 4 were independent risk factors for overall PCa (OR = 2.90; 95% CI, 1.24–6.80; p = 0.014, and OR = 10.97; 95% CI, 3.08–39.09; p < 0.001, respectively) and significant PCa (OR = 3.27; 95% CI, 1.28–8.35; p = 0.013, and OR = 11.10; 95% CI, 2.44–50.46; p = 0.002, respectively; Table 3).

Table 3 Univariable and multivariable regression analysis for predicting overall prostate cancer and significant prostate cancer (≥ Gleason score 7) at magnetic resonance imaging targeted prostate biopsy cores.
Full size table

Table 4 shows the predictive values of PSA and PSA derivatives for PI-RADSv2 ≥ 3 lesions. Although PSA < 11.87 ng/mL had the highest AUC, significant differences were observed between PSA < 11.87 and PSAD < 0.19 ng/mL2 (p = 0.612) or %fPSA > 18.76% (p = 0.062). Avoiding MRI when PSA < 11.87 ng/mL would reduce MRI use by 45.8%, but 57.6% (19/33) of significant PCa cases could be missed. Using PSAD < 0.19 ng/mL2 and %fPSA > 18.76% as criteria to omit MRI would reduce MRI use by 38.6% and 25.1%, respectively, with a risk of missing 29.5% and 22.7% of significant PCa cases. When combining each cutoff value of PSA and its derivatives, there were no significant differences in the AUC compared with PSA < 11.87 ng/mL alone. However, recommending MRI only for patients with PSA < 11.87 ng/mL, PSAD < 0.19 ng/mL2, and %fPSA > 18.76% could reduce MRI utilization by 22.7%, with a minimal risk (9.1% or 3/33 cases) of missing significant PCa by omitting MRI-targeted PB.

Table 4 Comparison of area under the curve, sensitivity, and specificity to predict visible lesions on multiparametric magnetic resonance imaging (MRI) (PI-RADSv2 score ≥ 3) and risk of missing overall prostate cancer/clinically significant prostate cancer at MRI targeted biopsy cores according to each cutoff value of prostate-specific antigen (11.87 ng/mL), prostate-specific antigen density (0.19 ng/mL2), and %fPSA (18.76%).
Full size table

Discussion

Current guidelines recommend prostate MRI for patients with previous negative PB and ongoing suspicion of PCa20. A prior meta-analysis reported that the negative predictive value of MRI for detecting Gleason score ≥ 7 PCa was 90.8% when MRI-negative cases were defined as PI-RADSv2 < 321. The current guideline20 suggested that the cutoff value of Likert/PI-RADS ≥ 3 to recommend PB could reduce 30% of unnecessary PB while missing 11% of clinically significant PCa. For patients with previous negative PB, the incidences of clinically significant PCa were 5.6%, 13.6%, 33.6% and 70.3% in those with PI-RADS < 3, 3, 4, and 5 lesions22.

In this study, ROC curve analysis determined that the optimal PI-RADSv2 cutoff value for predicting both overall and significant PCa was 4, aligning with prior research findings22. Thus, we aimed to assess optimal PSA and PSA derivative cutoff values for identifying PI-RADSv2 ≥ 4 lesions on MRI. The cutoff values for predicting PI-RADSv2 ≥ 4 lesions were 11.87 ng/mL for PSA, 0.19 ng/mL2 for PSAD, and 18.76% for %fPSA. Among these markers, PSAD demonstrated the highest AUC.

Multivariate logistic regression analysis indicated that PSAD and older age were associated with the presence of PI-RADSv2 ≥ 4 lesions. After adjusting for other clinicopathologic factors, PSAD ≥ 0.19 ng/mL and PI-RADSv2 ≥ 4 lesions were identified as independent risk factors for overall and significant PCa on MRI-targeted biopsy. These results are consistent with prior studies highlighting PSAD and MRI suspicion scores as significant predictors of cancer detection23.

Although PSAD < 0.19 ng/mL2 had the highest AUC for predicting PI-RADSv2 ≥ 4 lesions, using this criterion to omit MRI could reduce MRI use by 38.6% but carried a 30.3% risk of missing significant PCa. Including %fPSA > 18.76% alongside PSAD < 0.19 ng/mL2 as a triage tool showed no significant change in AUC for predicting PI-RADSv2 ≥ 4 but reduced the risk of missing significant PCa to 9.1% while decreasing MRI use by 23.1%. Recommending MRI for patients with PSA ≥ 11.87 ng/mL, PSAD ≥ 0.19 ng/mL2, or %fPSA ≤ 18.76% resulted in a 22.7% reduction in MRI use, with a 9.1% risk of missing significant PCa.

The result of our study is comparable to those of previous studies. For overall population, the triage strategy utilizing PSA and its derivatives to recommend MRI could reduce 24.6–33.9% of MRI, with the risk of missing 4.2–9.5% of clinically significant PCa24. In addition, a recent study based on the Prostate Cancer Disease Observation (PRECISION) cohort25 reported that utilizing Rotterdam Prostate Cancer Risk Calculator (RPCRC) could reduce 34.9% of unnecessary MRI at the cost of missing 15.7% of clinically significant PCa. For patients with previous negative PB, utilizing RPCRC could reduce 51% of MRI at the cost of missing 10% of high-grade PCa24.

Although the risk of PCa in patients with previously negative PB is relatively lower than PB-naive men22, MRI and subsequent targeted PB should be determined based on patient shared decision making, considering the risk of missing clinically significant PCa.

The findings of this study should be interpreted cautiously due to several limitations, including the retrospective, single-institution design. Interobserver variability in MRI interpretation using PI-RADS remains a concern, recommending the agreement among well-experienced radiologists26. Although the radiologists in our institution are well-experienced over 20 years, the possibility of interobserver discrepancy highlights the need for multicenter studies with centralized MRI interpretation review system to validate our results. Additionally, novel biomarkers such as the PHI, PHI density27, and PCa antigen 3 gene28 were not evaluated. Considering that those novel biomarkers showed superiority to conventional PSA derivatives in prediction of PCa, the combination of such biomarkers into criteria might enhance the negative predictive value, minimizing the potential risk to miss clinically significant PCa. Further study including such novel biomarkers should be performed. Meanwhile, all biopsies in this study were performed via the TR route, which may have influenced the detection accuracy. Previous research suggests that transperineal (TP) biopsy offers equal or superior detection rates compared with TR biopsy and TP biopsy could potentially reduce unnecessary MRIs29,30. Furthermore, given the low cancer-specific mortality in patients with prior negative biopsies12, longer followup studies are needed to evaluate ultimate cost-effectiveness and survival analysis of clinical progression. Despite these limitations, this study is, to the best of our knowledge, the first to evaluate the use of PSA and PSA derivatives as a triage tool to reduce unnecessary MRI in patients with prior negative biopsy.

Conclusions

In patients with prior negative prostate biopsy, limiting MRI to cases with PSA < 11.87 ng/mL, PSAD < 0.19 ng/mL, and %fPSA > 18.76% can reduce MRI use by 22.7%, with a 9.1% risk of missing significant PCa. However, those criteria for MRI recommendation should be adjusted based on the other novel biomarkers or individual patient characteristics and preferences.