Prediction Factors for Detecting Clinically Significant Prostate Cancer in a PSA Gray Zone (4–10 ng/mL): A Systematic Review

Abstract

Prostate cancer (PCa) is one of the most commonly diagnosed malignancies among men worldwide. Prostate-specific antigen (PSA) testing has improved early detection; however, PSA levels within the so-called “gray zone” (4–10 ng/mL) remain a diagnostic challenge because of their limited specificity and the associated risk of unnecessary biopsies. In this clinical context, an important objective is the accurate identification of clinically significant prostate cancer (csPCa), defined as disease with a higher likelihood of progression or clinical impact. In recent years, several diagnostic approaches and risk prediction models have been proposed to improve csPCa detection in patients within the PSA gray zone. These models combine clinical parameters, PSA-derived indices, and imaging findings—particularly magnetic resonance imaging (MRI)—and, in some cases, incorporate advanced biomarkers or radiomic features. Nevertheless, considerable heterogeneity exists across studies with respect to predictor selection, model construction, and reported diagnostic performance. This systematic review aims to synthesize current evidence on the diagnostic characteristics and predictive models used to detect clinically significant prostate cancer in men with PSA levels between 4 and 10 ng/mL. For consistency, heterogeneous outcome terms used in the included studies (e.g., “probable csPCa”, “significant cancer”, “clinically important PCa”) were harmonized and analyzed under the unified term csPCa. By identifying the most consistently reported predictors and comparing univariate with multivariate approaches, this review seeks to support clinical decision-making and to highlight areas for future research in prostate cancer diagnosis within the PSA gray zone.

1. Introduction

Prostate cancer (PCa) is one of the most frequently diagnosed malignancies among adult men in Europe and North America and represents a histologically complex and multifocal disease [1,2]. Prostate-specific antigen (PSA) testing has enabled earlier detection of PCa and has been associated with a reduction in prostate cancer–specific mortality [3].

Despite these benefits, PSA screening for asymptomatic PCa remains challenging because of overdiagnosis and overtreatment [4,5]. Patients with PSA levels between 4 and 10 ng/mL are considered to fall within a “gray zone”, in which prostate biopsies may be unnecessary or may lead to overtreatment [6]. Conversely, individuals with clinically significant prostate cancer (csPCa) should be identified as early as possible [7].

Both PSA testing and magnetic resonance imaging (MRI) play important roles in PCa screening and in identifying appropriate candidates for prostate biopsy [8]. Individuals with PSA levels of 4–10 ng/mL may undergo unnecessary biopsies in approximately 70% of cases [9]. In addition to PSA-based parameters, such as PSA density (PSAD), the decision to perform a biopsy is influenced by multiple factors, including age, ethnicity, comorbidity risk, prior biopsy history, digital rectal examination (DRE), and prostate MRI [10]. To estimate prostate cancer risk, several prediction tools and risk calculators incorporating PSA-related and clinical variables have been developed and applied [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25].

Furthermore, prostate diagnostic testing aims to distinguish clinically significant from clinically insignificant prostatic adenocarcinoma. Evidence from large screening trials indicates that the majority of clinically insignificant prostate cancer cases do not progress to symptomatic disease or adversely affect men’s health during their lifetime [5,26,27]. Nevertheless, unnecessary biopsy procedures and the overdiagnosis of insignificant prostate cancer may lead to clinical and psychological burden, as well as biopsy- and hospitalization-related complications [28].

The European Society of Urogenital Radiology (ESUR) issued clinical recommendations for multiparametric magnetic resonance imaging (mpMRI) in 2012 and introduced a standardized reporting framework known as the Prostate Imaging Reporting and Data System (PI-RADS) [29,30]. PI-RADS version 1 was subsequently refined and updated to PI-RADS version 2.0. Using a five-point Likert scale, PI-RADS estimates the likelihood of detecting clinically significant prostate cancer (csPCa) in individual lesions [31]. An updated version, PI-RADS v2.1, was released in 2019 [32].

Similar to V2.0 [33], PI-RADS v2.1 recommends the use of a multiparametric MRI protocol incorporating T2-weighted imaging, diffusion-weighted imaging with corresponding apparent diffusion coefficient maps, and dynamic contrast-enhanced imaging. This standardized approach aims to improve lesion detection, localization, characterization, and risk stratification in patients with suspected clinically significant prostate cancer [34,35].

Over the past two decades, substantial progress has been made in the development of prediction techniques to support the diagnosis of prostate cancer. Multivariable models incorporating a range of clinical and demographic variables have consistently demonstrated superior clinical performance compared with individual predictors [22,23,25]. These approaches include nomograms [36], artificial neural networks [37], and risk calculators [38,39,40].

Despite these advances, there remains no clear consensus regarding whether existing prostate cancer prediction models significantly improve the diagnostic accuracy of PSA testing or which model provides the greatest clinical benefit. Moreover, many current models do not substantially reduce the proportion of unnecessary prostate biopsies. Further research is therefore required to define the most effective strategies for identifying appropriate candidates for biopsy [18].

To better understand how prostate magnetic resonance imaging and multivariate prediction tools may support individualized, risk-adapted diagnostic strategies, a systematic review focusing on models that predict clinically significant prostate cancer in men with PSA levels between 4 and 10 ng/mL is warranted. Although meta-analyses evaluating prostate cancer risk models have been previously published [41], to our knowledge, this is the first systematic review specifically addressing prediction models within the PSA gray zone.

2. Materials and Methods

2.1. Search Strategy

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A systematic literature search was performed using PubMed, ScienceDirect, and Scopus. The following search terms were applied: “grey zone AND multivariate AND MRI”, “PSA 10 AND multivariate AND MRI”, “combined clinical characteristics AND prostate cancer”, “multivariate AND high-risk prostate AND MRI”, and “PSA 10 AND logistic regression”. Studies published between 2010 and 2025 were considered eligible for inclusion.

2.2. Eligibility Criteria

Articles published in English between 2010 and 2025 with available full text were eligible for inclusion. Studies were included if they met the following criteria:

• Evaluation of predictors of clinically significant prostate cancer (csPCa);
• PSA levels between 4 and 10 ng/mL (gray zone);
• Use of multivariable models combining clinical characteristics;
• Application of logistic regression analysis;
• English-language publications;
• Publication period between 2010 and 2025.

For methodological consistency, all outcome definitions related to clinically significant prostate cancer reported in the included studies—such as “probable csPCa”, “early PCa”, “significant cancer”, and “clinically important PCa”—were harmonized and analyzed under the unified term csPCa.

Following the literature search and eligibility assessment, a total of 14 studies were included in the final qualitative synthesis (

Table 1. Characteristics of studies included in the systematic review.

Author (Year)	Inclusion Period	Number of Patients	Purpose	Methods
Wei et al., 2020 [11]	January 2015–October 2019	364	To predict clinically significant prostate cancer (csPCa) by combining PI-RADS v2 scores derived from biparametric MRI (bpMRI) with clinical indicators in men with prostate-specific antigen (PSA) levels of 4–10 ng/mL.	Univariate and multivariate logistic regression
Han et al., 2020 [13]	June 2010–May 2017	123	To compare the diagnostic performance of bpMRI alone versus bpMRI combined with PSA density (PSAD) in detecting csPCa in patients with PSA levels of 4–10 ng/mL.	Univariate and multivariate logistic regression
Liu et al., 2020 [12]	2014–2018	235	To develop a multivariate model using clinical parameters to predict prostate cancer (PCa) and reduce unnecessary biopsies among patients with PSA in the gray zone.	Univariate and multivariate analysis
Sun et al., 2019 [15]	January 2015–July 2019	335	To assess whether PI-RADS v2 and neutrophil-to-lymphocyte ratio (NLR) improve detection of csPCa in men with PSA < 10 ng/mL at first biopsy.	Univariate and multivariate analysis
Sasanka et al., 2019 [17]	April 2017–October 2018	104	To evaluate the diagnostic performance of free PSA percentage, PSA density, and PI-RADS v2 score alone and in combination for predicting csPCa in patients with PSA 4–10 ng/mL.	Univariate and multivariate analysis
Dwivedi et al., 2017 [19]	2009–2015	137	To develop an mpMRI-based risk score and statistical equation for predicting the risk of PCa in biopsy-naïve men with serum PSA levels of 4–10 ng/mL.	Risk score development using multivariate logistic regression
Lu et al., 2019 [20]	January 2014–December 2017	357	To improve detection of PCa by combining PI-RADS v2 and PSA-adjusted volume, particularly in patients with PI-RADS score 3 or PSA 4–10 ng/mL.	Univariate and multivariate analysis
Niu et al., 2017 [14]	January 2014–September 2015	225	To establish a predictive nomogram for high-grade PCa in patients with PSA 4–10 ng/mL based on PI-RADS v2, MRI-derived prostate volume, adjusted PSAD, and clinical parameters.	Univariate and multivariate logistic regression
Ng et al., 2013 [16]	April 2008–April 2013	230	To investigate the role of the Prostate Health Index (phi) in PCa detection in patients with PSA levels of 4–10 ng/mL undergoing first prostate biopsy in an Asian population.	Multivariate logistic regression
Qi et al., 2019 [21]	December 2015–March 2018	199	To develop and validate a radiomics model based on mpMRI for predicting PCa in patients with PSA levels of 4–10 ng/mL and reducing unnecessary biopsies.	Multivariate logistic regression
Fang et al., 2015 [22]	2011–2013	345	To identify characteristics and risk factors for positive biopsy outcomes and develop a risk stratification score in Chinese patients with PSA 4–10 ng/mL.	Univariate and multivariate analysis
Liu et al., 2018 [23]	January 2015–July 2017	308	To predict PCa using clinical parameters and reduce unnecessary biopsies among patients with PSA in the gray zone.	Univariate and multivariate logistic regression
Chai et al., 2025 [25]	June 2020–June 2024	327	To develop nomograms for predicting PCa and csPCa in patients with PSA < 10 ng/mL and PI-RADS v2.1 score ≤ 3.	Univariate and multivariate logistic regression
Ji et al., 2024 [24]	June 2000–February 2023	1196	To develop a risk prediction model using traditional diagnostic methods (PSA, digital rectal examination, transrectal ultrasound) in Asian patients with PSA levels of 4–10 ng/mL.	Univariate and multivariate logistic regression

).

2.3. Study Selection

Titles and abstracts were initially screened, and full-text articles were subsequently reviewed if inclusion criteria were met. Data were extracted regarding authorship, year of publication, geographic location, study design, study population, outcomes, and reported associations.

The literature search initially identified 1578 records (459 from Scopus, 837 from PubMed, and 282 from ScienceDirect). After removal of duplicates, 348 records were screened, resulting in the inclusion of 14 eligible studies. The main reasons for exclusion were failure to meet eligibility criteria based on title, abstract, or full-text review (Figure 1).

The included studies employed a range of approaches to predict csPCa in patients within the PSA gray zone, most commonly using multivariable models incorporating clinical parameters, imaging findings, and biomarkers.

3. Results

3.1. Results of Study Selection

The literature search identified a total of 1578 records (459 from Scopus, 837 from PubMed, and 282 from ScienceDirect). After removal of duplicates, 348 records were screened based on titles and abstracts. Following full-text assessment, 14 studies met the predefined inclusion criteria and were included in the final qualitative synthesis. The study selection process is illustrated in the PRISMA flow diagram (Figure 1).

3.2. Characteristics of the Included Studies

The 14 included studies were published between 2013 and 2025 and evaluated prediction models for clinically significant prostate cancer (csPCa) in men with PSA levels between 4 and 10 ng/mL. Study sample sizes ranged from 104 to 1196 participants. All studies used histopathological findings from prostate biopsy as the reference standard.

Most investigations employed multivariable logistic regression models incorporating combinations of clinical variables, PSA-related parameters, and imaging findings. Magnetic resonance imaging (MRI), particularly PI-RADS v2 or v2.1 scoring, was included in the majority of studies. PSA density (PSAD), prostate volume (PV), age, and digital rectal examination (DRE) were among the most frequently analyzed clinical predictors. An overview of predictor categories evaluated across studies is provided in

Table 2. Categories of predictors evaluated across the included studies.

Predictor Category	Variables Included
Demographic variables	Age, body mass index (BMI)
Lifestyle factors	Smoking status
Clinical variables	Digital rectal examination (DRE)
PSA-related variables	PSA/tPSA, %free PSA, PSA density (PSAD), prostate volume (PV)
Composite PSA indices	%p2PSA, Prostate Health Index (phi), f/t PSA ratio
Inflammatory markers	Neutrophil-to-lymphocyte ratio (NLR), lymphocyte-to-monocyte ratio (LMR)
Ultrasound-based features	TRUS findings, hypoechoic area
MRI modality	Biparametric MRI (bpMRI), multiparametric MRI (mpMRI)
MRI scoring systems	PI-RADS v2/v2.1
MRI-derived lesion characteristics	Lesion location, lesion volume, shape, border
Advanced imaging techniques	ADC, DW-MRI, MRSI, metabolite ratio

Abbreviations: PSA, prostate-specific antigen; PSAD, prostate-specific antigen density; PI-RADS, Prostate Imaging Reporting and Data System; MRI, magnetic resonance imaging; NLR, neutrophil-to-lymphocyte ratio; LMR, lymphocyte-to-monocyte ratio; TRUS, transrectal ultrasound; ADC, apparent diffusion coefficient; DW-MRI, diffusion-weighted magnetic resonance imaging; MRSI, magnetic resonance spectroscopic imaging.

.

3.3. Performance of Univariate and Multivariate Prediction Models

Across all included studies, multivariate models consistently demonstrated superior diagnostic performance compared with univariate predictors for the detection of csPCa in the PSA gray zone.

Table 3. Comparison of univariate and multivariate model performance for detecting clinically significant prostate cancer (csPCa) in the PSA gray zone ¹.

Study (Year)	Best Univariate Predictor (AUC)	Best Multivariate Model (AUC)	Key Variables Included
Wei et al. (2020) [11]	PI-RADS v2 (0.827)	PI-RADS v2 + prostate volume (0.882)	bpMRI, PI-RADS v2, prostate volume
Han et al. (2020) [13]	bpMRI (0.884)	bpMRI + PSAD (0.907)	bpMRI, PSA density
Liu et al. (2020) [12]	mpMRI (0.79)	mpMRI + PSAD (0.84)	mpMRI, PSA density
Sun et al. (2019) [15]	PI-RADS v2 (0.854)	PI-RADS v2 + NLR (0.876)	Age, %fPSA, DRE, NLR, PI-RADS v2
Sasanka et al. (2019) [17]	PSAD (0.968)	PSAD + PI-RADS v2 (0.827)	PSA density, PI-RADS v2
Dwivedi et al. (2017) [19]	MRSI (0.83)	MRI-based risk score (0.89)	MRSI, DW-MRI, PSA
Niu et al. (2017) [14]	PSAD (0.74)	Nomogram (0.85)	Age, PSA density, PI-RADS v2
Qi et al. (2019) [21]	mpMRI (0.726)	Radiomics model (0.956)	Multi-imaging fusion, age, PSA density
Dong Fang et al. (2015) [22]	mpMRI (0.753)	PAMD score (0.824)	Prostate volume, age, MRI, DRE
Chang Liu et al. (2018) [23]	PI-RADS v2 (0.855)	PI-RADS v2 + PSAD (0.93)	PI-RADS v2, PSA density

¹ csPCa: clinically significant prostate cancer. For each study, the highest reported area under the curve (AUC) from univariate and multivariate analyses is presented. All heterogeneous outcome terms used in the original studies were harmonized and analyzed under the unified term csPCa.

summarizes the best-performing univariate predictors and multivariate models reported in each study.

In univariate analyses, PI-RADS v2 scoring and PSAD were the most commonly reported predictors, with area under the receiver operating characteristic curve (AUC) values ranging from 0.74 to 0.97. However, univariate performance varied substantially between studies.

Multivariate models combining MRI-derived parameters with PSA-related and clinical variables achieved higher and more consistent AUC values, generally ranging from 0.82 to 0.96. Models integrating PI-RADS v2 scores with PSAD, prostate volume, age, or inflammatory markers demonstrated improved discrimination compared with MRI or PSA-based variables alone.

Several studies evaluated MRI-based strategies. Combinations of biparametric MRI (bpMRI) with PSAD showed diagnostic performance comparable to multiparametric MRI (mpMRI)-based models. Advanced imaging approaches, including radiomics-based and multi-imaging fusion models, achieved the highest reported AUC values.

Overall, the results indicate that multivariate diagnostic models incorporating clinical, biochemical, and imaging parameters outperform single-variable approaches for identifying csPCa in men with PSA levels between 4 and 10 ng/mL.

4. Discussion

The findings of this systematic review indicate that multivariate diagnostic models consistently outperform univariate predictors in the detection of clinically significant prostate cancer (csPCa) among men within the PSA gray zone. As demonstrated by the comparative analyses summarized in Table 3, models integrating MRI-derived parameters with PSA density, prostate volume, or additional clinical variables achieve superior diagnostic performance compared with isolated predictors.

Across the included studies, MRI-based approaches—particularly those incorporating PI-RADS v2 scoring—emerged as a central component of effective risk stratification. While individual variables such as PSA density or MRI findings alone often demonstrated moderate to high diagnostic accuracy, their incorporation into multivariable frameworks resulted in improved discrimination and a more balanced trade-off between sensitivity and specificity. This pattern supports the growing consensus that combined diagnostic strategies are more clinically informative than single biomarkers when guiding biopsy decisions in patients with PSA levels between 4 and 10 ng/mL.

Although one study reported exceptionally high univariate performance for PSA density alone, the majority of investigations demonstrated incremental diagnostic value when PSA-related metrics were combined with imaging findings in multivariate or model-based approaches, including nomograms and radiomics-driven models. These integrated strategies appear particularly useful for reducing unnecessary biopsies while maintaining robust detection of csPCa, especially when MRI-derived variables are incorporated.

Several studies also highlighted the potential role of biparametric MRI (bpMRI) as an alternative to multiparametric MRI (mpMRI). Multivariate models combining bpMRI with PSA density demonstrated diagnostic performance comparable to mpMRI-based models, with higher specificity reported in selected cohorts. These findings suggest that bpMRI-based strategies may offer a cost-effective and clinically feasible option for optimizing diagnostic pathways in patients within the PSA gray zone, particularly in resource-limited settings.

Recent investigations further expand the evidence base by exploring tailored risk stratification strategies in specific populations. For example, Ji et al. developed a prediction model based solely on traditional clinical variables, demonstrating good discriminatory ability, while Chai et al. focused on biopsy-naïve patients with low PSA and low PI-RADS scores, proposing nomograms that may assist decision-making in carefully selected clinical scenarios. Together, these studies underscore the flexibility of multivariate modeling approaches across diverse diagnostic contexts.

Despite these promising findings, several limitations should be acknowledged. Most included studies were retrospective and single-center in design, potentially limiting generalizability. In addition, reliance on systematic biopsy as the reference standard may introduce misclassification due to false-negative results. While biopsy remains the initial diagnostic standard in men suspected of having prostate cancer, it is well recognized that systematic biopsy may fail to detect anterior or small-volume tumors. Image-guided biopsy techniques, including MRI-targeted and MRI–ultrasound fusion approaches, have been proposed to address this limitation and improve diagnostic accuracy [42,43].

Future research should therefore prioritize large, prospective, multicenter studies employing standardized imaging protocols and harmonized outcome definitions. The integration of additional clinical variables, such as family history and genetic markers, may further enhance predictive performance. Although current multivariate models show considerable potential for improving csPCa detection within the PSA gray zone, further validation is required before their widespread incorporation into clinical guidelines.

5. Conclusions

This systematic review underscores the potential of risk stratification models to refine prostate cancer diagnostics within the challenging PSA gray zone. Integration of readily available clinical data including PSA levels, DRE findings, and TRUS parameters demonstrates promising predictive capabilities for identifying csPCa while minimizing unnecessary interventions. The models developed within this study offer a cost-effective approach to enhance diagnostic accuracy, particularly for populations in regions with limited access to advanced imaging modalities like MRI. By combining these traditional diagnostic methods, clinicians can better identify which patients would benefit most from further investigation. While our findings are specific to the studied population and diagnostic criteria, they highlight the potential of risk-adapted strategies in prostate cancer diagnostics and suggest that these data-driven tools may aid clinicians in identifying men at greatest risk while reducing unnecessary biopsies. Further validation and comparison with other models in diverse clinical settings are warranted to refine and generalize the application of our risk prediction model.

6. Limitations

Several limitations should be considered when interpreting this review. First, our search strategy, while comprehensive, may have missed relevant studies due to our focus on specific keywords (grey zone, PSA, multivariate, MRI, high risk). Second, we limited our search to three major databases (PubMed, ScienceDirect, Scopus) and did not perform hand-searching or explore the grey literature, potentially excluding relevant publications. Finally, we did not formally assess inter-rater reliability during data extraction, which could introduce bias. These limitations, coupled with our specific inclusion criteria, may limit the generalizability of our findings and highlight the need for further research using broader search strategies, more exhaustive data sources, and rigorous quality assessment methods.