Molecular Biomarkers for the Detection of Clinically Significant Prostate Cancer: A Systematic Review and Meta-analysis

Take Home Message The Prostate Health Index test had high diagnostic accuracy for the detection of clinically significant prostate cancer. Its incorporation in the diagnostic process could reduce the necessary biopsies. There is a lack of evidence on the effect of this test on clinical outcomes.


Introduction
Prostate cancer (PCa) is a major health problem, with approximately 1.4 million cases diagnosed worldwide each year [1]. It is the second most common cancer in males after lung cancer worldwide [2], and its prevalence increases with each additional year of age [3]. The mean age of PCa onset is 65 yr and the majority of PCa patients are diagnosed from then onwards, with the age group of 70-75 yr having the highest incidence rate [4].
PCa often progresses slowly and has a prolonged preclinical phase. Therefore, many men with PCa die from causes other than PCa and without evidence of pathological manifestation [3].
Traditionally, diagnosis and staging of PCa have been based on prostate-specific antigen (PSA) level, digital rectal examination (DRE), and transrectal ultrasound guided prostate biopsy.
Serum PSA measurement is the reference standard for the early detection of PCa. However, PSA level does not exclusively increase in malignant pathology, as high levels can also be observed in benign prostatic pathologies such as benign prostatic hyperplasia, prostatitis, other urinary tract infections, and even acute urine retention. Moreover, PSA cannot discriminate between indolent PCa (iPCa) and aggressive tumors.
Most men with positive screening results (elevated PSA levels or abnormal DRE) who undergo prostate biopsy will not have PCa. Approximately two-thirds of men with an elevated PSA level can expect a false positive test result [5]. Moreover, biopsy procedures are related to complications such as pain, bleeding, and sepsis, and the related consequences on the utilization of health resources. However, the most serious harm of PCa screening may be overdiagnosis, which may result in subsequent overtreatment [6]. Consequently, strategies to differentiate iPCa from aggressive tumors are necessary [1]. Current European Urological Association guidelines recommend the use of risk stratification tools, such as risk calculators and magnetic resonance imaging (MRI), and biomarker tests for the prediction of a positive prostate biopsy as reflex tests after an elevated PSA level [7].
Recently, there has been an expansion in the availability of new molecular blood and urine test biomarkers that can be used to support prostate biopsy decisions, providing more individualized risks for PCa, distinguishing between clinically significant PCa (csPCa) and iPCa, or predicting the prognosis of patients already diagnosed [8]. However, no consensus has been reached on the use of these tests in routine clinical practice.
The objective of the present study is to examine the diagnostic accuracy of the biomarker tests in the identification of patients with csPCa.

Evidence acquisition
A systematic review (SR) of the literature was carried out following the Cochrane Collaboration methodology [9] with reporting in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [10]. The prespecified protocol for this review was registered in PROSPERO (registration number CRD42021240638).

Data sources and searches
The following electronic databases were searched (from 2010 to March 1, 2021): Medline (Ovid platform) and EMBASE (Elsevier interface). The search strategy included both controlled vocabulary and text-word terms related to PCa and molecular biomarker. Searches were limited to the English and Spanish languages. The complete search strategy is available in Supplementary Table 1. We also examined the reference lists of included articles and, through search in Google Scholar, articles that referenced the included studies.

Selection criteria and study selection
Studies were eligible for inclusion if they fulfilled the following criteria: 1. Design: randomized or nonrandomized clinical trials (RCTs or non-RCTs) were eligible for inclusion. In the absence of such designs, cohort and case-control studies that performed an evaluation of the diagnostic validity of the tests were considered. 2. Population: adult men (!18 yr) with clinical factors that suggested csPCa comprised the study population. Studies with a heterogeneous group of patients (eg, patients with suspected PCa, either iPCa or csPCa) were included only if the results for patients meeting the inclusion criteria were reported separately. 3. Index tests: any blood or urine test based on biomarkers aimed at distinguishing csPCa from iPCa was performed. 4. Reference standard: it included alternative tests, biopsy, magnetic resonance, or usual care.  Two reviewers (D.I.-V. and A.A.C.) screened retrieved references independently and in duplicate, starting with titles and abstracts. The full texts of all articles deemed potentially relevant were then screened to confirm eligibility. Disagreements between the reviewers were checked by a third reviewer (T.P.-S.).

2.3.
Data extraction process and risk of bias assessment Data extraction and risk of bias (RoB) assessment were also conducted independently and in duplicate. Discrepancies were discussed and, when no consensus was reached, a third reviewer was consulted. Data extracted include general information, study design, sample characteristics, test details (biomarker and cutoff point), reference standard, and results. RoB was assessed using either the Cochrane Risk of Bias tools for RCT (RoB 2.0) [11], or the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) revised tool [12].

Assessment of publication bias
Potential publication bias was explored by constructing the Deeks asymmetry graphs and computing the Egger test [13], with the significance level set at 0.05, using metafunnel and metabias commands, respectively, in STATA version 16.

Data synthesis
We built 2 Â 2 tables summarizing true positive (TP), false positive, true negative, and false negative (FN) values to calculate sensitivity and specificity for detecting csPCa. Review Manager (RevMan, version 5.4.1., 2020; The Nordic Cochrane Center, The Cochrane Collaboration, Copenhagen, Denmark) was used to show the sensitivity and specificity measurements at the study level. Pooled estimates with 95% CIs were performed by bivariant random-effect metaanalyses using the midas command in STATA version 16 [14]. A continuity correction was used in trials that reported zero cells in a 2 Â 2 table (eg, when TP or FN is zero). Heterogeneity was assessed by visually analyzing forest plots and through the Higgins I 2 statistic [15]. Several sources of heterogeneity were anticipated, including the type of diagnostic test, cutoff point, definition of csPCa, and ethnic origin. When reported in studies, the effect using a subgroup analysis was explored.

Certainty of evidence assessment
An assessment of the certainty of evidence per outcome was performed for tests included in the meta-analysis, based on the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach. We developed evidence profiles and rated the overall certainty of evidence as high, moderate, low, or very low [16].

Evidence synthesis
The results of the literature search and study selection process are shown in Figure 1. Our search identified 2954 references, of which 336 studies were selected for full text assessment. Three of these could not be retrieved [17][18][19].
Sixty-five studies , reported in 69 articles , finally fulfilled the pre-established selection criteria. The list of studies excluded at the full-text level and the reasons for exclusion are provided in Supplementary Table 2.                  3.1.

Characteristics of included studies
The main characteristics of the selected studies are summarized in Tables 1 and 2.
Only diagnostic performance studies with an observational design were included: 43 prospective, 13 retrospective, two ambispective, two with a retrospective analysis of prospectively collected data, three case-control studies, and two nested case-control studies. Ten studies contained two [25,31,35,37,41,55,78,81,89] or three [40] different populations and data were treated as separate studies.

RoB in included studies
The RoB assessment is summarized in Figures 2A and 2B

Quality of evidence
The overall quality of the evidence for PHI and SelectMDx was considered low (Supplementary Tables 4 and 5 provide the evidence profiles, respectively).

Synthesis of results
Diagnostic accuracy results of selected studies are listed in Supplementary Table 6. Out of the 65 included studies, only 21 remained for a quantitative analysis for PHI and SelectMDx [36,42,49,57,78]. The results of all meta-analyses and subgroups analyses are available in Supplementary Table 7. 3.4.1.
Progensa PCA3. Cutoff points ranged from 5 to 35. For the cutoff point 15, the test yielded sensitivity ranging between 93% and 99%, and specificity of 37%. The cutoff point 20 showed sensitivity between 89% and 99%, and specificity of 51%. Finally, at the cutoff point 35, the sensitivity ranged between 62% and 71%, while the specificity increased to 59-66%. The AUC ranged from 0.59 to 0.83.
No subgroup analysis could be performed by ethnic origin.
Assuming a risk of suffering csPCa of !7.5%, the sensitivity was 95.5% and the specificity was 32.1%, whereas when a risk of 12% was assumed, the sensitivity decreased to 90.1% and the specificity increased to 53.5%. The AUC ranged from 0.72 to 0.87.

Publication bias
No publication bias was identified, except in the PHI analysis with a cutoff point between 30 and 35 (p = 0.02). The results of the Egger tests and Deeks asymmetry graphs are available in Supplementary Table 7 and Supplementary Figure 1, respectively.

Discussion
The assessment of molecular biomarkers for the detection of csPCa is based on the data derived from 65 studies  (N = 34 287), which evaluate their diagnostic accuracy in a population undergoing initial biopsy for suspected csPCa due to high PSA levels, family history, abnormal DRE, or altered multiparametric MRI. Quality of evidence for the tests included in the meta-analysis (PHI and SelectMDx) has been rated as low.
Approximately 77% of biopsies performed in men included in this SR did not yield a positive csPCa result. Furthermore, a 20% received a diagnosis of iPCa, placing them at risk of overdiagnosis, biopsy-related complications, and wasted health care resources, evidencing the need for better risk stratification.  Results of the assessed tests are measured on a continuous scale so that their behavior depends on where the cutoff point is set. However, information on established cutoff points was not provided in 25 of the included studies [22,27,32-34,37,38,41,46,48,55,58,59,62,69,75,76,80-82,87 ,89,91,93,94], and variability in terms of selected cutoffs is present among studies assessing the same test. Since the optimal 4Kscore, PCA3, and PHI cutoff points for the diagnosis of csPCa are not established, a comparison of diagnostic accuracy at different cutoff points was performed.
The results indicate that four analyzed tests (two urine tests and two blood tests) show an ability to identify !95% patients with csPCa: Progensa PCA3, with cutoff point 15; My-Prostate Score, using a cutoff point of >10; PHI, with any cutoff point between 15 and 30; and 4Kscore test, assuming a risk of csPCa of !7.5%. Using these tests and cutoff points, the ability to prevent unnecessary biopsies ranged between 14% and 37% [22,24,25,[28][29][30]40,42,47,51,64,65,67,72,73,77] which shows that theses could be useful as a noninvasive method of supporting the decision on whether or not the first prostate biopsy is necessary and, consequently, reducing the total number of unnecessary biopsies. However, it should be taken into account that only the results related to PHI are pooled effect estimates. The biomarkerbased tests considered in this SR (particularly, PHI) would be included with triaging purposes in the diagnostic pathway for patients with a high clinical suspicion of csPCa but negative MRI results, in order to prevent unnecessary biopsies.
Of the nine SRs on biomarker-based tests for the management of PCa published to date to the best of our knowledge [62,[96][97][98][99][100][101][102][103], only two focused on evaluating the use of these tests in discerning iPCa from csPCa and, consequently, in improving the decision-making process for first biopsies and treatment planning. Nevertheless, neither of them analyzes the available scientific evidence for all available tests, but rather for specific tests. Russo et al. [96] obtained for PHI and 4Kscore tests, sensitivity for the detection of csPCa of 93% and 87%, respectively, and specificity of 34% and 61%, respectively. Zappala et al. [101] evaluated the predictive precision of 4Kscore to discriminate between patients with and without csPCa, obtaining a pooled estimate for AUC of 0.80. Our results are consistent with these previous results.
One aspect to consider is the difference in test accuracy depending on the ethnic origin of patients. The evidence has shown possible improved performance of these tests in Asian populations, followed by Caucasian and African-American men. Confirmation of this finding would support the need for research on the best cutoff points based on patient ethnicity.
Available evidence exclusively consists of studies that evaluate the diagnostic validity of tests. However, using new tests with evidence of diagnostic utility does not directly imply improving decisions related to the diagnosis and treatment of PCa. Therefore, further research is needed to determine what effect the implementation of these tests would have on clinical decision making and patientimportant health outcomes (eg, complications, recurrencefree survival, cancer survival, morbidity, and quality of life).
The main limitation of the present review is the methodological differences among studies, mainly the diversity or the lack of information about cutoff values. Moreover, a subgroup analysis to explore this issue could not always be performed. Another potential limitation is the possibility that some studies have not been included because those are not written in English or Spanish or because those are not indexed in the consulted databases. However, to the best of our knowledge, our SR is the most extensive review carried out to date on the effectiveness of the incorporation of tests, based on biomarkers in samples of blood or urine, for the identification of patients at high risk of csPCa. Methodologically, the SR benefits from rigorous methods following the fundamental principles of transparency and replicability; a comprehensive search, a peer selection, data extraction, and RoB assessment; and an assessment of the certainty of evidence on the basis of a structured and explicit approach.

Conclusions
Our findings indicate that PHI has high diagnostic accuracy for csPCa detection, and its incorporation in the diagnostic pathway could reduce unnecessary biopsies. However, there is a lack of evidence on the effects on patient consequences, supporting the need for well-conducted testtreatment RCTs in which investigators allocate patients to receive a PHI test or a control diagnostic approach (no test), and measure patient-important outcomes. Based on the pooled sensitivity estimate for SelectDMx, it is possible that the use of this test for the identification of patients with csPCa is not the best option. Finally, according to the limited available evidence, it is not possible to reach a clear conclusion on the other tests evaluated.
Author contributions: Tasmania del Pino-Sedeño had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.   the manuscript (eg, employment/affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: None.
Funding/Support and role of the sponsor: The study was financed by the Ministry for Health of Spain in the framework of activities developed by the Spanish Network of Agencies for Health Technology Assessment and Services for the National Health System (RedETS).