Integrative analysis of blood biomarkers and clinical variables improves early detection of aggressive prostate cancer

Abstract

Aggressive prostate cancer (PC) represents a significant health concern. Conventional screening methods, primarily based on prostate-specific antigen (PSA) levels, lack specificity, leading to an urgent need for more accurate diagnostic tools. This study investigates whether integrating clinical and routine blood laboratory parameters can improve the early non-invasive prediction of aggressive PC. In a pilot study of 578 patients with suspicion of PC, 28 laboratory values alongside data on family history, diet, and lifestyle were analyzed. A logistic regression classifier was developed, with model performance evaluated using repeated k-fold cross-validation on the complete dataset (n = 282). Participants were categorized into healthy, moderate PC (ISUP 1–2), and aggressive PC (ISUP 3–5). Significant associations were found between PC aggressiveness and lower levels of androstenedione, dehydroepiandrosterone-sulfate (DHEA-S) and free PSA%, as well as higher levels of sex hormone binding globulin (SHBG). The integration of these serum markers with clinical parameters into a new multi-stage risk classifier significantly improved the predictive accuracy for aggressive PC, outperforming PSA-only methods. The integration of DHEA-S, androstenedione, and SHBG as widely available and cost-effective novel blood biomarkers offers a more targeted, non-invasive prediction of aggressive PC. This approach could reduce reliance on invasive prostate biopsies and expensive magnetic resonance imaging.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-98980-3.

Introduction

Prostate cancer (PC) is the second most prevalent cancer in men, representing 27% of all new cancer diagnoses in males¹. It is a major cause of mortality worldwide, with projections of the Lancet Commission on PC indicating a doubling of new cases between 2020 and 2040, particularly in low-income countries². Early diagnosis through screening can offer numerous benefits such as a higher likelihood of cure, less aggressive treatment options, reduced progression to metastatic stages, and improved quality of life^3–5. Prostate-specific antigen (PSA) screening is a widely used method for early detection of PC. Recent results from the Rotterdam Section of the European Randomized Study of Screening for PC indicate that PSA screening can substantially reduce long-term mortality rates⁶. However, PSA testing has limited specificity, frequently resulting in unnecessary biopsies and the detection of indolent PC⁷.

To address the limitations of PSA testing, the American Urological Association (AUA) guidelines recommend integrating risk calculators into the traditional decision-making process for PC screening⁸. The primary tool is the PSA blood test, and for cases with elevated PSA levels, the AUA guidelines advise using risk calculators that include PSA, free PSA%, age, ethnicity, and family history for personalized risk assessment. Clinicians are then encouraged to engage in shared decision-making with patients, discussing the value of additional diagnostic data from biomarkers or multiparametric magnetic resonance imaging (mpMRI) and considering the advisability of proceeding with a biopsy. Similarly, the European Association of Urology (EAU) recommends using risk stratification nomograms for patients with elevated PSA, including age, family history, digital rectal examination (DRE), prostate volume, and PSA density (PSAD)⁹. This method helps to identify men at low risk who may only require clinical follow-up, thereby reducing the need for expensive or invasive tests such as mpMRI and biopsy.

mpMRI and fusion-targeted biopsy of suspicious mpMRI lesions are advanced diagnostic methods for PC, significantly enhancing detection accuracy^10,11. High cost and restricted availability of mpMRI, especially in low-resource settings, limit its routine use as a screening tool¹². The economic and logistical challenges of MRI highlight the need for cost-effective and accessible biomarkers to facilitate broader implementation across diverse healthcare settings, improving early detection capabilities.

Recent research increasingly focuses on modifiable risk factors, such as lifestyle and diet, associated with PC risk^13,14. Metabolic changes linked to many diseases, including cancer, are reflected in human blood metabolomic patterns, which provide valuable insights into underlying physiological states¹⁵. Systemic information from blood parameters, such as C-reactive protein, amino acids, and glycated hemoglobin, has been predictive of multiple cancers¹⁶. Additionally, adrenal androgens influence the aetiopathogenesis of PC and are associated with aggressive PC^17,18.

The study aimed to improve non-invasive prediction of aggressive PC by integrating 44 laboratory and clinical parameters into a predictive model. Our findings demonstrate that incorporating these parameters after PSA testing significantly improved diagnostic accuracy. This approach could reduce unnecessary biopsies, decrease the number of costly MRIs, and more accurately predict PC aggressiveness, aligning with precision medicine principles. By validating a multifaceted diagnostic strategy over traditional PSA-only methods, our research supports a more nuanced approach to PC screening and decision-making.

Materials and methods

Study population

565 men who received a prostate biopsy at the University Hospital Heidelberg between June 2021 and December 2023 were recruited for supplementary clinical and, if a preoperative blood sample had to be taken and patient consent was given, laboratory values assessment. They were eligible for study inclusion in case of suspicion of PC with a PSA level ≥ 3 ng/ml and/or suspicious DRE and no previous treatment for PC. Additionally, 13 healthy men with a PSA < 2.5 ng/ml who presented to the Urology Clinic at Heidelberg University Hospital for benign conditions, were recruited to enlarge the healthy control group. Data was collected prospectively, and institutional review board approval was obtained (S-130/2021). Clinical information regarding family history (defined as having a first-degree relative with PC), lifestyle, etc. was assessed via questionnaire. Laboratory parameters included, among others, lipid metabolism, inflammatory markers, hormones, and obtaining informed consent, blood samples were collected via vein puncture between 8 and 11 AM in fasting patients. All blood samples were analyzed in the central laboratory of Heidelberg University Hospital, in addition to the standard preoperative lab values. The central laboratory of Heidelberg University Hospital has been accredited according to DIN EN ISO 15189 since 2005.

Prostate biopsy

Prebiopsy mpMRI was performed on 543 patients with suspicion of PC and eligibility for mpMRI. All mpMRI scans followed the PI-RADS recommendations, according to European Society of Urogenital Radiology guidelines¹⁹. Men underwent transperineal fusion-targeted biopsy using a UroNav system (Philips Invivo, Gainsville, FL, USA) of MRI-suspicious lesions and saturation biopsy adjusted to prostate volume, as previously described^20,21. Histopathological analyses were performed according to International Society of Urological Pathology (ISUP) standards. The participants were categorized into three risk groups: healthy patients (PSA < 2.5 ng/ml or no PC in biopsy), moderate PC (ISUP 1–2 PC), and aggressive PC (ISUP 3–5 PC)¹⁰.

Statistical analysis

In this study, 28 laboratory and 16 clinical parameters were analyzed to determine their impact on PC risk. Each parameter was initially tested for its ability to discriminate between the three predefined risk groups: healthy patients (n = 253), moderate PC with ISUP 1–2 (n = 229), and aggressive PC with ISUP 3–5 (n = 96) through statistical analysis. For continuous variables, an ordinary least square model was employed, fitting models for each parameter to predict the risk groups and subsequently testing the significance of the slope coefficients :

where is a risk category corresponding to healthy patients, ISUP 1–2 and ISUP 3–5 PC, and are intercept and slope coefficients for the k-th tested parameter .

For categorical variables the Chi-square test was used²². All p values were reported and subsequently adjusted for multiple comparisons using the Benjamini and Hochberg correction²³ with a family-wise error rate at 0.1 level. Complete results are provided in Supplementary Table 1, with significant findings highlighted in Fig. 1.

Fig. 1.

Statistically significant associations with the presence of moderate or aggressive PC. (A) Statistically significant association between laboratory parameters and the presence of ISUP 1–2 PC, ISUP 3–5 PC, or healthy patients. (B) Statistically significant associations between clinical and imaging parameters and the presence of ISUP 1–2 PC, ISUP 3–5 PC or healthy patients.

Risk modeling

Risk modeling was performed on 282 patients who had available data for clinical and laboratory parameters, showing a significant association with ISUP grade. The logistic classifier was developed in a staged approach:

1. Baseline model: Initiated with PSA, the standard marker for PC screening, to establish a benchmark for model performance.

2. Serum marker integration: Added laboratory measurements that showed a statistically significant association with PC risk, including Androstenedione, free PSA%, dehydroepiandrosterone-sulfate (DHEA-Sulfate), and sex hormone binding globulin (SHBG) to evaluate improvements in sensitivity and specificity.

3. Clinical parameter integration: Included clinical factors from international guidelines (age, family history, DRE results, prostate volume, and PSAD) to assess their combined predictive value.

4. MRI results: Added PIRADS scores to capture additional risk information provided by imaging.

5. Advanced baseline model: Combined MRI and PSA.

To evaluate multi-class performance of the model, we measured accuracy, F1 score, precision, and recall at each stage, providing a direct measure of performance across the three classes within a single unified framework. The model was evaluated using a repeated k-fold cross-validation (k = 10, repeated 10 times) approach, and all performance metrics were reported based on the test set (n = 282). We used L2 regularization to decrease overfitting probability with the default regularization strength (λ = 1) using scikit-learn models²⁴. The default behavior of the classifier (a one-vs-rest implementation with a 0.5 probability cut-off) was not modified. If a model’s predicted probability for a particular class exceeded 0.5, the patient was assigned to that class.

Interpretative analysis using SHAP values

To understand the influence of individual variables on predictive outcomes, SHAP (SHapley Additive exPlanations) methodology was applied, following the approach proposed by Lundberg and Lee²⁵. This method quantitatively measured the impact of each parameter included in the model, providing clear insights into how specific factors contribute to overall risk prediction.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used ChatGPT-4o in order to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Results

Patient cohort description

The study included a total of 578 patients. During the initial phase of the analysis, when statistically significant markers were selected, all available measurements were used without restricting to a fully overlapping cohort, maximizing the use of the data. Each laboratory parameter was represented by a minimum of 314 and a maximum of 353 measurements, while other parameters ranged from 314 to 578 measurements. A comprehensive list of these parameters and their respective counts is provided in Supplementary Table 1.

Following feature selection, the cohort was refined to include only patients with complete data across all parameters identified as significant (post-multiple hypothesis correction). This subset comprised 282 patients. Detailed descriptions of these patients and their corresponding data are presented in Table 1.

Table 1.

Patient characteristics (modeling cohort n = 282) in the healthy, ISUP 1–2 PC versus ISUP 3–5 PC groups for significant parameters.

Parameter	Unit	Healthy patients (n = 122)	ISUP 1–2 PC (n = 117)	ISUP 3–5 PC (n = 43)
Age	Years	63 (57–69)	66 (61–73)	72 (64.50–75)
PSA	ng/dl	5.75 (4.47–8.06)	5.81 (4.2–8.41)	10.11 (6.67–15.49)
Androstenedione	µg/l	70 (59–88)	66 (53–78)	58 (48–67)
Free PSA %	%	16 (0.30–25)	11 (0.24–20)	8 (0.18–17.50)
DHEA-sulfate	µg/ml	1.23 (0.76–1.75)	1.06 (0.69–1.47)	0.76 (0.54–1.14)
Sex hormone binding globulin	nmol/l	43 (32–53)	46 (35–58)	48 (35.5–62)
Prostate volume	ml	61.32 (45–81.97)	46.37 (34–66.45)	45 (33.94–60)
PSAD	ng/ml/ml	0.1 (0.07–0.13)	0.13 (0.08–0.19)	0.21 (0.16–0.35)
Positive family history	Count, %	12 (9.83)	14 (11.97)	7 (16.28)
Positive DRE	Count, %	10 (8.20)	72 (61.54)	21 (48.84)
PIRADS ≤ 2	Count, %	19 (15.57)	6 (5.13)	0 (0)
PIRADS 3	Count, %	44 (36.07)	23 (19.66)	1 (2.33)
PIRADS 4	Count, %	53 (43.44)	58 (49.57)	14 (32.56)
PIRADS 5	Count, %	6 (4.92)	30 (25.64)	28 (65.12)

For continuous parameters, median and IQR are provided, otherwise, count per group and percentage are shown.

Role of blood biomarkers in prostate cancer

Significant associations were found between hormonal markers and PC severity as classified by ISUP grades (Fig. 1). Lower levels of androstenedione and DHEA-Sulfate were significantly associated with more aggressive forms of PC (ISUP grades 3–5) (corrected p values 0.013 and 0.03, respectively). This finding supports the hypothesis that reductions in certain androgens may signal advanced disease.

The bioavailability of circulating androgens is largely determined by the concentration of SHBG. Accordingly, in our study higher levels of SHBG were associated with increased PC severity (corrected p value 0.037).

Higher PSA levels were positively correlated with PC severity (corrected p value 0.03). Additionally, a lower percentage of free PSA% indicated more aggressive cancer (corrected p value 0.025). These results reinforce the role of PSA and free PSA percentage as critical biomarkers in assessing PC aggressiveness.

We chose ISUP ≥ 3 as our primary definition of aggressive PC because most guidelines and studies identify ISUP 3 disease as having a higher risk of progression, earlier biochemical recurrence, and poorer long-term outcomes compared to ISUP 2 PC, requiring more intensive management²⁶. To address the variability in the definition of clinically significant PC, we also evaluated ISUP ≥ 2 as an alternative threshold. Under this cut-off, similar directional associations of clinical, laboratory, and imaging parameters with the presence of ISUP ≥ 2 PC were observed, although the effect magnitudes differed slightly (Supplementary Fig. 1).

Incorporating clinical and lab values into new risk models for non-invasive prostate cancer prediction

In developing a PC risk classifier, we included the five significant laboratory values, along with clinical features and imaging parameters, that were strongly associated with ISUP grades. Age, DRE results, PSAD, PIRADS scores, and prostate volume showed statistically significant associations with ISUP grade severity, all with corrected p values < 0.05. Although a family history of PC did not show a significant statistical association with ISUP grades (corrected p value 0.552), it was included in the model to adhere to AUA and EAU guidelines^8,9.

The performance of the risk classifier was evaluated through a multi-stage approach (Fig. 2): (1) Baseline model initiated with PSA, (2) Enhancement with laboratory biomarkers significantly associated with PC risk (Androstenedione, free PSA%, DHEA-Sulfate, and SHBG), (3) Integration of clinical factors (Age, family history, DRE results, prostate volume, and PSAD), (4) Integration of MRI results (PIRADS scores) and (5) Advanced baseline model with solely PSA and MRI as comparison. This stepwise model enhancement resulted in gradual improvements across all metrics from stage 1–4, including F1 score, precision, recall, and accuracy. The progression of median scores through each stage is depicted in Fig. 3A, with exact values provided in Supplementary Table 2.

Fig. 2.

Schematic overview of the multi-stage risk classifier. Step 1 begins with PSA screening, followed by the integration of four significant laboratory biomarkers (Step 2), the addition of clinical parameters (Step 3), and finally MRI findings (Step 4). Each step refines the risk assessment and guides decisions about further imaging or biopsy. DHEA dehydroepiandrosterone, DRE digital rectal examination, PSA prostate-specific antigen, PSAD prostate-specific antigen density, SHBG sex hormone binding globulin.

Fig. 3.

Risk modeling of PC. (A) Performance metric of a logistic classifier based on five sets of features: (1) PSA, (2) Integration of additional laboratory biomarkers, (3) Integration of clinical factors, (4) Integration of MRI results and (5) Solely PSA and MRI as comparison. (B) Feature importance for the three groups (healthy patients, ISUP 1–2 PC and ISUP 3–5 PC).

The regression coefficient distribution was analyzed to assess the prognostic significance of each parameter (Fig. 3B). PSAD emerged as a particularly influential factor in identifying healthy patients. Established diagnostic parameters like PSA, DRE results and PIRADS also showed high importance for non-invasive prediction of PC presence. However, with these parameters, coefficients for ISUP grades 1–2 and 3–5 PC often shared the same sign, indicating challenges in differentiating these two categories. Serum androgens displayed variable signs for ISUP 1–2 versus ISUP 3–5 PC, suggesting their potential impact on the model’s predictive capability.

Influence of individual variables on predictive outcomes using SHAP values

To understand the influence of individual variables on predictive outcomes for each patient, we employed SHAP analysis to obtain per-patient SHAP values. These values help to identify the specific factors contributing to each patient’s prediction (PC vs. no PC) and provide insight into the influence of each variable on the predictive outcomes. The model was developed to assist urologists in making informed decisions regarding prostate biopsies, especially when the complexity of individual patient profiles challenges clinical judgement. Values from two study patients are shown where biopsies didn’t reveal PC, and the model correctly predicted the absence of PC, demonstrating the practical utility of SHAP analysis as a transparent, data-driven foundation for clinicians (Fig. 4A,B).

Fig. 4.

SHAP analysis to obtain per-patient SHAP values. SHAP values for two patients of the cohort who had no PC in the prostate biopsy: blue color indicates a reduction of the probability of having PC, and orange color indicates an increased probability of having PC.

Discussion

This study aimed to improve the non-invasive prediction of aggressive PC by integrating clinical and laboratory parameters into a comprehensive risk model. By identifying novel, cost-effective serum markers, we sought to improve decision-making following PSA testing and reduce reliance on invasive procedures and expensive MRI use.

Our study revealed several results. First, levels of DHEA sulfate and androstenedione were significantly reduced in newly diagnosed aggressive PC compared to healthy controls. In line with these results, higher levels of SHBG were significantly associated with increased PC aggressiveness. These markers had a stronger association with aggressive PC detection than established variables advised for use in risk stratification nomograms, such as family history^8,9. Thus, androgens have potential as diagnostic markers for early detection of aggressive PC, offering a complementary diagnostic tool alongside PSA testing. Androgens were particularly effective in distinguishing between ISUP 1–2 and ISUP 3–5 PC, highlighting their potential in differentiating moderate from aggressive PC, e.g. for treatment approaches such as active surveillance²⁷. ISUP 1–2 PC patients may be offered active surveillance instead of radical therapy due to the indolent nature of their disease²⁸. Existing biomarkers expressed by all PC, including indolent tumors, offer limited potential to selectively detect high-grade disease²⁹. Novel biomarkers distinguishing high-grade from low-grade PC are urgently needed, aligning with guidelines emphasizing a focus on high-grade PC due to the indolent nature of low-grade PC⁸.

Second, integrating clinical findings and laboratory values into the decision-making process for recommending MRI scans has the potential to improve the selection of patients who benefit most from imaging. MRI remains time- and resource-intensive and may pose considerable costs for patients, apart from limitations in availability and inter-rater reliability. Barriers in healthcare delivery limit the wide adoption of MRI in real-world practice as a diagnostic test following PSA. The decision to use MRI and fusion biopsy can be supported by relatively inexpensive routine blood values and clinical parameters, minimizing unnecessary MRIs for PSA-positive patients with a low likelihood of PC. Commercially available blood values are straightforward to implement and less resource-intensive, allowing urologists to more selectively apply MRI and/or biopsy^30,31. Our results support implementing integrated decision-making frameworks in clinical routine to optimize MRI and biopsy use in PC diagnostics.

Third, the application of an explainable decision support concept in our study demonstrates an approach to guide urologists and patients in making informed decisions regarding prostate biopsies. A transparent and comprehensive assessment of each patient’s clinical scenario, including evaluation of clinical, lab values, and, if applicable, MRI findings, allows urologists and patients to make trustworthy and clinically justified decisions.

Our findings align with previous research, such as Severi et al., who found high levels of adrenal androgens like androstenedione and DHEA Sulfate associated with a reduced risk of aggressive PC³². Their study of 17,049 men reported that double concentrations of androstenedione were associated with almost half the risk of aggressive PC, and double concentrations of DHEA-Sulfate were associated with a 37% lower risk. Additionally, Travis et al. described an inverse association of the androstenedione concentration with advanced PC³³.

Our finding of higher SHBG levels in ISUP 3–5 PC complements previous results that proposed SHBG as a marker of aggressive PC in early-stage disease. Preoperative serum SHBG levels were an independent predictor of extraprostatic extension in localized PC, suggesting a role of SHBG in PC progression³⁴. A study by Price et al. demonstrated that germline genetic variations of androgen-related pathway genes are associated with serum androgen and SHBG concentrations and PC risk³⁵.

Routine blood test results harbor valuable information that may not be immediately apparent to clinicians. Podnar et al. demonstrated the feasibility of brain tumor diagnosis from routine blood tests using a machine learning predictive model with diagnostic accuracy comparable to imaging studies³⁶. Regular screening with blood tests for disease prevention is already recommended in many countries³⁷. This study’s results indicate the potential of simple clinical parameters combined with comprehensive laboratory predictors to estimate aggressive PC risk. Highlighting androgens as complementary diagnostic tools at minimal cost and emphasizing an integrated approach to MRI utilization could lead to more efficient screening strategies for aggressive PC.

Recent large-scale trials have demonstrated the high diagnostic accuracy of mpMRI in combination with targeted prostate biopsy, underscoring the essential role of MRI in PC detection^10,11. Our proposed model aims to complement MRI in clinical pathways. In many cases, especially in well-resourced settings, MRI remains a central tool for accurate lesion localization and biopsy guidance. However, the proposed integrated blood biomarker model may help triage patients who may benefit most from MRI, particularly where healthcare resources or imaging availability are limited. By using MRI for higher-risk patients identified by our multi-stage risk assessment, unnecessary imaging can be minimized and MRI resources can be allocated more efficiently, especially in regions where specialized radiologists and expensive equipment are less accessible. Despite its diagnostic advantages, MRI has important limitations. A major challenge is reader variability, as differences in radiologist experience and interpretation can lead to inconsistent diagnostic results, ultimately affecting accuracy. In addition, the ability of MRI to detect indolent lesions that may not require treatment contributes to the risk of overdiagnosis in PC detection. Overdiagnosis can lead to unnecessary interventions, increasing the burden on healthcare systems. MRI also relies on specialized infrastructure and trained personnel, which limits its widespread implementation, particularly in low-resource settings³⁸. Taken together, these challenges highlight the need for cost-effective and widely available diagnostic solutions, such as the proposed model, while recognizing that MRI remains essential in settings where adequate resources and expertise are available. The multi-stage blood biomarker model has the potential for wider real-world application, particularly in low-resource healthcare systems where access to MRI is limited by cost, availability or lack of specialist radiologists. By enabling cost-effective and widely reproducible blood tests to refine the decision-making process, our approach could enable more equitable screening and reduce both over- and undertreatment. Ensuring that MRI is used primarily for patients with a higher likelihood of aggressive disease would minimize unnecessary procedures and improve clinical efficiency across diverse healthcare settings.

A cost–benefit perspective is increasingly important, particularly in low-resource settings or in healthcare systems where MRI availability is limited. Although MRI is highly accurate in detecting PC, its high cost, infrastructure requirements, and need for expert radiologists can be significant barriers to widespread use in screening. Recent cost–benefit analysis studies support the integration of biomarkers early in the screening pathway. Jiao et al. showed that biomarker testing can reduce unnecessary MRI and biopsy procedures, resulting in significant cost savings. Using prostate cancer antigen 3, TMPRSS2:ERG gene fusion, or MyProstateScore as a reflex test has greater economic value than MRI or biopsying all men or not biopsying men with intermediate PSA levels³⁰. Takeda et al. demonstrated that a multimodal deep learning approach integrating imaging data and serum PSA in patients with PSA ≤ 20 ng/ml could effectively predict clinically significant PC, thereby reducing unnecessary biopsies³⁹. Their findings underscore that more accurate risk stratification can limit resource use, especially expensive or invasive procedures that may not be warranted in lower-risk cases. In addition, Teoh et al. conducted a Markov model analysis showing that incorporating the prostate health index (PHI) into routine screening pathways for Chinese men not only improved diagnostic accuracy but also reduced total healthcare costs and increased quality-adjusted life years compared with a PSA-only approach⁴⁰. Although the PHI was not evaluated in our current study, this example illustrates the broader point that adding accessible biomarkers, particularly those with high specificity for aggressive disease, can reduce the need for confirmatory MRI or biopsy in low-risk men, ultimately reducing the economic burden on patients and healthcare systems.

In line with these findings, our multi-stage risk classifier combines inexpensive routine blood markers (DHEA-S, androstenedione, SHBG, etc.) with clinical variables (PSAD, DRE findings, prostate volume) to assess which patients are most likely to have aggressive PC. By identifying those who require MRI or biopsy, the proposed model could streamline clinical workflows and reduce costs associated with over-examination. This paradigm holds particular promise in healthcare settings where MRI resources are limited or where cost considerations often limit access to advanced imaging. Although we did not perform a formal cost-effectiveness analysis in this pilot study, the growing body of evidence strongly suggests that biomarker integration into early PC screening may be both medically beneficial and economically advantageous. Future multicentre trials, including real-world cost analyses, are essential to confirm the generalizability of our findings and to identify the most cost-effective set of markers for widespread implementation in different healthcare systems.

Several promising additional biomarkers could complement the current diagnostic pathway for PC, including PSA derivatives and genetic markers⁴¹. Combining the PHI and the 4K score⁴², if available, with widely accessible and cost-effective clinical and lab values may offer a comprehensive and accurate approach for early non-invasive prediction of aggressive PC. Recognizing the low specificity of PSA as a challenge for PC screening programs, Srivastava et al. recently proposed a “more is more” approach in PC screening programs with the addition of biomarkers in a multistep approach³¹. This strategy could help to optimize biopsy decisions, reduce unnecessary biopsies, and improve patient outcomes.

Several limitations must be acknowledged. The pilot study’s sample size may limit the generalizability and statistical power of the findings. Validation in independent, larger cohorts is necessary to ensure the classifier’s robustness and generalizability. A major limitation of our study is the largely homogeneous, single-center European cohort, which limits the generalizability of our findings to more diverse populations. The incidence and progression of PC varies between ethnic groups⁴³. Evidence suggests that healthy black men have lower PHI and altered testosterone/estradiol ratios compared with other populations, which may influence screening performance in this high-risk population⁴⁴. Future studies should explore ethnicity-specific biomarker thresholds in large, diverse cohorts to address the known racial disparities in PC incidence and mortality. Prospective validation in multi-ethnic cohorts is therefore essential to confirm the observed associations of androstenedione, DHEA-S, and SHBG with aggressive PC. Additionally, the study’s nature did not allow determining causal relationships among variables. Longitudinal studies are necessary to establish relationships between hormone levels and PC development or progression. Lastly, the definition of PC that is aggressive and requires active treatment rather than active surveillance is controversial, with varying definitions of aggressive PC. As ISUP 1–2 PC patients can be offered active surveillance due to the indolent nature of the disease, histological assessment was used to stratify patients into moderate (ISUP 1–2) and aggressive (ISUP 3–5) PC^10,21. The lack of consensus on what constitutes a “clinically significant” PC is a challenge for the universal adoption of risk models. While some clinicians consider ISUP ≥ 2 PC to be significant, others emphasize ISUP ≥ 3 PC^10,45. In a recent commentary by Carlsson et al., the authors highlight how evolving definitions of clinically significant PC continue to pose challenges for both clinicians and researchers⁴⁶. The historical reliance on the Gleason score alone to determine significant PC has shifted to a more nuanced framework centered on the grade group system. Institutions incorporate additional criteria (e.g. tumor volume, PSA density, emerging biomarkers), leading to a patchwork of definitions of significant PC. From a modeling perspective, this variability in what constitutes a “clinically significant” disease can have a profound impact on model development, validation, and implementation. A tool optimized for one definition may not perform as well when applied to settings using alternative thresholds. Our analyses show that biomarkers retain their prognostic value even when different cut-offs are used to define significant disease. However, each threshold may slightly alter the operating characteristics of the model. In the future, more uniform standards or at least robust validation of predictive tools across multiple significant PC definitions will be essential. This step would not only streamline clinical decision-making and facilitate model comparisons but also ensure that clinicians and patients have a clearer, more consistent understanding of what truly represents a “clinically significant” disease that warrants intervention.

Conclusions

Incorporating serum markers DHEA-S, androstenedione, and SHBG into a novel risk classifier can improve early detection of aggressive PC. These widely available and cost-effective blood biomarkers could reduce reliance on invasive prostate biopsies and expensive MRI by providing a more targeted approach to non-invasive prediction of aggressive PC following PSA testing. Our pilot study lays the groundwork for larger-scale research to further explore the integration of androgens and SHBG in future risk stratification models for improved clinical decision making.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

AUA

American Urological Association

DHEA

Dehydroepiandrosterone

DRE

Digital rectal examination

EAU

European Association of Urology

ISUP

International Society of Urological Pathology

mpMRI

Multiparametric magnetic resonance imaging

PC

Prostate cancer

PHI

Prostate health index

PSA

Prostate-specific antigen

PSAD

Prostate-specific antigen density

SHBG

Sex hormone binding globulin

Author contributions

Conceptualization: M.G.; Data curation: A.R., O.L.; Formal analysis: O.L.; Funding acquisition: M.G., M.H.; Investigation: H.S.; M.G., O.L; Methodology: H.S.; M.G., O.L.; Project Administration: M.G.; Resources: M.G., M.H., O.S.; Software: O.L., O.S.; Supervision: M.G.; Validation: A.R., H.S., O.L.; Visualization: O.L.; Writing—original draft: M.G., O.L.; Writing—review and editing: A.R., H.S., M.H., O.S. All authors critically reviewed, edited, and added to the manuscript as well as approved the final version of the manuscript.

Funding

Open Access funding enabled and organized by Projekt DEAL.

This work was realized through support by the German Federal Ministry for Economic Affairs and Climate Action (funding # 01MT21004A) and the Dieter Morszeck Foundation.

Data availability

The data sets generated during and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

The study was approved by the ethical committee of the University of Heidelberg (Approval No. S-130/2021) and was performed in accordance with the Declaration of Helsinki.