Diagnostic ability of %p2PSA and prostate health index for aggressive prostate cancer: a meta-analysis.

The role of [-2]proPSA (p2PSA) based diagnostic tests for the detection of aggressive prostate cancer (PCa) has not been fully evaluated. We conducted a meta-analysis to evaluate the diagnostic performance of p2PSA/free PSA (%p2PSA) and prostate health index (Phi) tests for PCa and to evaluate their ability in discriminating between aggressive and non-aggressive PCa. A total of 16 articles were included in this meta-analysis. For the detection of PCa, the pooled sensitivity, specificity, and AUC were 0.86 (95% CI, 0.84-0.87), 0.40 (95% CI, 0.39-042) and 0.72 (95% CI, 0.67-0.77) for %p2PSA respectively, and were 0.85 (95% CI, 0.83-0.86), 0.45 (95% CI, 0.44-0.47) and 0.70 (95% CI=0.65-0.74) for Phi, respectively. In addition, the sensitivity for discriminating PCa between higher Gleason score (≥7) and lower Gleason score (<7) was 0.96 (95% CI, 0.93-0.98) and 0.90 (95% CI, 0.87-0.92) for %p2PSA and Phi respectively, and the specificity was low, only 0.09 (95% CI, 0.06-0.12) and 0.17 (95% CI, 0.14-0.19) for %p2PSA and Phi, respectively. Phi and %p2PSA have a high diagnostic accuracy rates and can be used in PCa diagnosis. Phi and %p2PSA may be useful as tumor markers in predicating patients harboring more aggressive disease and guiding biopsy decisions.

Prostate Cancer (PCa) is one of the most commonly diagnosed malignant genitourinary tumors among men in United States and other Western countries1. To date, prostate-specific antigen (PSA) is the most widely used and successful serum marker to detect PCa, however, the clinical value of PSA for early detection of PCa has been questioned because of its poor specificity, especially in men with serum PSA levels in the 2–10 ng/ml range2. It is well-known that PSA testing leads to overdiagnosis and overtreatment of indolent PCa that does not evolve into aggressive life-threatening cancers3. Thus, significant effort has been placed on the development of new serum markers that could complement PSA for early aggressive PCa diagnosis.

It has been suggested that measurement of the precursor PSA isoform [−2] proPSA (p2PSA) and its derivatives might offer improvement of PCa detection in men with a total PSA (tPSA) level between 2 and 10 ng/ml4. The isoform p2PSA is a promising marker for PCa detection, and compared to other isoforms, measurement of the p2PSA to free PSA (%p2PSA) and Prostate Health Index (Phi), which is calculated using the formula (p2PSA/fPSA × √tPSA) can better distinguish between PCa and non-PCa than tPSA and fPSA/tPSA (%fPSA)567. Values for %p2PSA and Phi were significantly higher in patients with PCa than patients with benign prostatic hyperplasia (BPH) and chronic histologic prostatic inflammation89.

Previous studies have also shown that %p2PSA and Phi were significantly associated with an increased probability of detecting aggressive PCa, as indicated by a Gleason score ≥758. Both %p2PSA and Phi gained attention as potentially very promising markers, that is, to be able to help predict aggressive PCa within a tPSA range of 2–10 ng/ml101112. But the ability of these two markers to distinguish between aggressive PCa and non-aggressive PCa remains unclear. Here, a systematic review was performed to evaluate the emerging role of %p2PSA and Phi in diagnosing PCa, and the ability of these markers to distinguish patients with and without aggressive PCa.

Results

Description of included studies

As shown in Figure 1, we identified 323 papers from a literature search and excluded 299 irrelevant articles. The remaining 24 potentially relevant studies were further assessed. Eight studies were further excluded because they contained duplicated populations68131415161718. Although the study by Lazzeri et al.6 was excluded due to partially duplicated data with the study of Lughezzani et al.19, the data on aggressive and non-aggressive PCa were available and extracted.

Figure 1

Flowchart of selecting process for meta-analysis.

A total of 16 papers were included in the meta-analysis571011192021222324252627282930, 13 papers reported on %P2PSA, and 12 papers reported on Phi. In addition, the study by Jansen7 included two different populations which were treated as two data sets, and the study by Scattoni20 contained initial and repeat biopsy groups which were treated as two separate studies. Finally, 14 studies were reported on %P2PSA and Phi. The detailed characteristics of each study are presented in Table 1. All studies presented the results of sensitivity and specificity, and most studies presented the results of ROC curve. The cutoff points were reported in seven studies for %p2PSA and nine studies for Phi. The data on the patients with and without aggressive PCa were extracted from four and five studies for %p2PSA and Phi, respectively.

Table 1

Characteristics of studies included in this meta-analysis for %p2PSA and Phi

First Author	Year	Sampling frame	Years of recruitment	Country	Ethnicities	Inclusion criteria	Patients age(y)	Assay method
Scattoni	2013	Prospective	2011.12–2012.5	Italy	Caucasian	PSA2–15 ng/ml, and/or positive DRE	67.5 ± 7.5	Beckman Coulter
Scattoni	2013	Prospective	2011.12–2012.5	Italy	Caucasian	PSA2–15 ng/ml, and/or positive DRE	67.5 ± 7.5	Beckman Coulter
Stephan	2013	Prospective, retrospective	2003–2012	Germany, France	Caucasian	PSA1.6–8 ng/ml,	64 (63–65)	Beckman Coulter
Perdonà	2013	Prospective	2010.3–2010.12	Italy	Caucasian	PSA<20 ng/ml, normal DRE	64.91 ± 7.37	Beckman Coulter
Ferro	2013	Prospective	N/A	Italy, Germany	Caucasian	Age ≥ 50,PSA2–10 ng/ml	60.4 ± 6.0	Beckman Coulter
Lazzeri	2013	Prospective	2011.3–2012.7	Italy, Germany, France, UK, Spain	Caucasian	Age ≥ 45,PSA2-10 ng/ml,and/or positive DRE	64.2 ± 7.5	Beckman Coulter
Ito	2013	Prospective	2004.4–2007.12	Japan	Asian	PSA2–10 ng/ml	66 (37–82)	Beckman Coulter
Ng	2013	Rertospective	2008.3–2013.3	China	Asian	Age ≥ 50,PSA4–10 ng/ml,normal DRE	65.9 (50–79)	Beckman Coulter
Lughezzani	2012	Prospective	2010.7–2011.7	Italy	Caucasian	PSA0.5–20 ng/ml, and/or abnormal DRE	64.3 ± 7.8	Beckman Coulter
Guazzoni	2011	Prospective	2010.3–2010.8	Italy	Caucasian	PSA2–10 ng/ml, normal DRE	63.3 ± 8.2	Beckman Coulter
Catalona	2011	Prospective,retrospective	2003.10–2009.6	USA	Caucasian	Age ≥ 50, PSA2–10 ng/ml, normal DRE	62.8 ± 7.0	Beckman Coulter
Jansen	2010	Rertospective	1994–1997.2	Austria	Caucasian	Age ≥ 50, PSA2–10 ng/ml	66(55–75)	Beckman Coulter
Jansen	2010	Rertospective	1993	Netherlands	Caucasian	Age ≥ 50, PSA2–10 ng/ml	69(50–77)	Beckman Coulter
Le	2010	Prospective	2007.4	USA	Caucasian	PSA2.5–10 ng/ml, normal DRE	65	Beckman Coulter
Sokoll	2010	Prospective	N/A	USA	Caucasian	PSA0.29–310.6 ng/ml	61.7 ± 8.6	Beckman Coulter
Stephan	2009	Rertospective	2002–2006	Germany	Caucasian	PSA0.26–28.4 ng/ml, normal DRE	62.1 ± 5.63	Beckman Coulter
Sokoll	2008	Rertospective	N/A	USA	Caucasian	PSA0.48–33.18 ng/ml	62.2 ± 8.2	Beckman Coulter
Catalona	2003	Rertospective	1995–2002	USA and Austria	Caucasian	PSA2–10 ng/ml	N/A	Research assay

Abbreviation: DRE, digital rectal examination; N/A, not available.

Abbreviation: DRE, digital rectal examination; N/A, not available.

Overall meta-analyses for %p2PSA and Phi

A total of 6,279 patients (2,722 PCa and 3,557 non-PCa) and 5,856 patients (2,474 PCa and 3,382 non-PCa) were included in the pooled analyses for %p2PSA (Supplementary table 1) and Phi (Supplementary table 2), respectively. The pooled sensitivity and specificity were 0.86 (95% CI, 0.84–0.87) and 0.40 (95% CI, 0.39–042) for %p2PSA respectively, and were 0.85 (95% CI, 0.83–0.86) and 0.45 (95% CI, 0.44–0.47) for Phi, respectively (Figure 2).

Figure 2

Forest plot of sensitivities and specificities of %p2PSA and Phi for the diagnosis of PCa.

(A) sensitivity for %p2PSA; (B) specificity for %p2PSA; (C) sensitivity for Phi; (D) specificity for Phi.

The ROC space showed a curvilinear trend of points and Spearman's correlation coefficient was 0.705 (P = 0.005) for %p2PSA and 0.744 (P = 0.002) for Phi. It was suggested that there was the existence of a threshold. Due to the presence of significant heterogeneity for sensitivity (I2 = 90.4%, χ2 = 135.15, P < 0.001), specificity (I2 = 96.1%, χ2 = 329.39, P < 0.001) for %p2PSA; and sensitivity (I2 = 91.5%, χ2 = 152.79, P < 0.001), specificity (I2 = 93.1%, χ2 = 188.66, P < 0.001), and diagnostic OR (Cochran-Q = 24.25, P = 0.0289) for Phi; the diagnostic indices were calculated using a random effect model. We used a summary SROC curve to aggregate data and obtained a symmetrical curve with an AUC of 0.72 (95% CI = 0.67–0.77) and 0.70 (95% CI = 0.65–0.74) for %p2PSA and Phi, respectively (Supplementary Figure 1).

Subgroup analysis

Significant heterogeneity for Phi (I2 = 46.4%, P = 0.0289) and low heterogeneity (I2 = 29.1%, p = 0.1448) were observed. Therefore, further subgroup analyses were performed to explore the heterogeneity on data stratified by ethnicity group, study design and QUADAS score. Subgroup analyses showed that a retrospective study design (I2 = 61.6%, p = 0.0231) and low QUADAS score (I2 = 63.7%, p = 0.0171) were responsible for the heterogeneity of %p2PSA (Table 2); and the prospective study design (I2 = 51.3%, p = 0.0368), Caucasian group (I2 = 52.5%, p = 0.0168) and low QUADAS score (I2 = 55.9%, p = 0.0201) were responsible for the heterogeneity of Phi (Table 3).

Table 2

Subgroup analyses of %p2PSA diagnostic value

Variable	Data sets	Sensitivity (95% CI)	Specificity (95% CI)	Diagnostic OR(95% CI)	I2, p-value	AUC (95% CI)
Total	14	0.86(0.84–0.87)	0.40(0.39–0.42)	4.11(3.46–4.87)	29.1%,0.1448	0.72(0.67–0.77)
Design
Retrospective	6	0.90(0.88–0.92)	0.30(0.28–0.33)	4.29(2.87–6.42)	61.6%,0.0231	0.96(0.91–1.01)
Prospective	7	0.76(0.73–0.79)	0.54(0.51–0.56)	3.85(3.15–4.71)	0.0%,0.6260	0.70(0.66–0.74)
Mixed	1	0.90	0.34	N/A	N/A	0.72(0.70–0.75)
Ethnicity
Caucasian	12	0.86(0.85–0.87)	0.38(0.37–0.40)	4.05(3.36–4.87)	36.1%,0.1016	0.71(0.65–0.77)
Asian	2	0.76(0.64–0.85)	0.55(0.50–0.60)	5.37(2.93–9.84)	0.0%,0.7095	0.50(0.50–0.50)
QUADAS
≥12	8	0.83(0.81–0.85)	0.44(0.42–0.46)	4.00(3.38–4.70)	0.0%, 0.7132	0.69(0.65–0.73)
<12	6	0.89(0.87–0.91)	0.34(0.32–0.36)	4.40(2.98–6.52)	63.7%,0.0171	0.78(0.78–0.89)

Abbreviation: QUADAS, Quality Assessment of Diagnostic Accuracy Studies; N/A, not available.

Table 3

Subgroup analysis of Phi diagnostic value

Variable	Data sets	Sensitivity(95% CI)	Specificity(95% CI)	Diagnostic OR(95% CI)	I2, p-value	AUC (95% CI)
Total	14	0.85(0.83–0.86)	0.45(0.44–0.47)	4.82(3.90–5.97)	46.4%,0.0289	0.70(0.65–0.74)
Design
Retrospective	3	0.92(0.89–0.95)	0.39(0.35–0.43)	6.05(3.98–9.18)	0.0%,0.4700	0.66(0.23–1.08)
Prospective	9	0.80(0.77–0.82)	0.52(0.49–0.54)	4.98(3.61–6.87)	51.3%,0.0368	0.71(0.66–0.76)
Mixed	2	0.86(0.84–0.88)	0.39(0.37–0.42)	4.04(2.73–5.97)	71.1%,0.0629	0.50(0.50–0.50)
Ethnicity
Caucasian	12	0.85(0.83–0.86)	0.44(0.42–0.46)	4.85(3.85–6.11)	52.5%,0.0168	0.69(0.64–0.74)
Asian	2	0.76(0.64–0.85)	0.56(0.51–0.61)	4.76(2.53–8.98)	4.1%, 0.3071	0.50(0.50–0.50)
QUADAS
≥14	5	0.89(0.87–0.91)	0.43(0.40–0.45)	5.44(4.29–6.89)	0.0%,0.8272	0.78(0.71–0.85)
<14	9	0.82(0.80–0.84)	0.47(0.45–0.49)	4.46(3.34–5.96)	55.9%,0.0201	0.67(0.62–0.71)

QUADAS: Quality Assessment of Diagnostic Accuracy Studies.

The ability of %p2PSA and Phi to discriminate between aggressive and non-aggressive PCa

A Total of 728 cases of PCa (292 with Gleason score ≥ 7 versus 436 with Gleason score < 7) and 1,157 cases of PCa (431 with Gleason score ≥ 7 versus 726 with Gleason score < 7) were included in the pooled analysis of %p2PSA (Supplementary table 3) and Phi (Supplementary table 4).

The pooled sensitivity for the detection of PCa with higher Gleason score (≥7) was 0.96 (95% CI, 0.93–0.98) and 0.90 (95% CI, 0.87–0.92), while the specificity was low, only 0.09 (95% CI, 0.06–0.12) and 0.17 (95% CI, 0.14–0.19) for %p2PSA and Phi respectively (Figure 3). The pooled AUC of SROC was 0.54 (95% CI,−0.52–1.61) and 0.67 (95% CI, 0.57–0.77) for %p2PSA and Phi respectively (Supplementary Figure 1). The accuracy of %p2PSA and Phi in discriminating aggressive PCa with non-aggressive PCa is summarized in Table 4.

Figure 3

Forest plot of sensitivities and specificities of %p2PSA and Phi for the diagnosis of aggressive PCa.

(A) sensitivity for %p2PSA; (B) specificity for %p2PSA; (C) sensitivity for Phi; (D) specificity for Phi.

Table 4

Diagnostic accuracy of %p2PSA and Phi for Gleason score ≥7 VS Gleason score<7

Variable	Data sets	Sensitivity (95% CI)	Specificity (95% CI)	Diagnostic OR(95% CI)	I2, p-value	AUC (95% CI)
%p2PSA	4	0.96(0.93–0.98)	0.09(0.06–0.12)	3.77(1.13–12.57)	43.5%,0.1507	0.54(−0.52–1.61)
Phi	5	0.90(0.87–0.92)	0.17(0.14–0.19)	3.06(1.61–5.84)	29.5%,0.2248	0.67(0.57–0.77)

Assessment of publication bias

We analyzed possible publication bias by generating funnel plots, and the shape of the funnel plots was symmetrical in the diagnosis of PCa suggesting the absence of publication bias. We also used the Deeks' funnel plot asymmetry tests (t = 1.86, P = 0.09 for %p2PSA and t = 1.37, P = 0.20 for Phi), results suggested that there was no publication bias in the meta-analysis (Figure 4).

Figure 4

Linear regression test of funnel plot asymmetry for %p2PSA (A) and Phi (B).

The statistically non-significant p value of 0.09 (A) and 0.20 (B) for the slope coefficient suggest symmetry in the data.

Discussion

Despite the wide and successful use of PSA blood testing, its use as a serum marker for PCa is far from ideal, due to the lack of specificity; it has only 20% specificity31. This review synthesized the current knowledge about early diagnosis of PCa with %p2PSA and Phi determination. The present meta-analysis indicated that %p2PSA and Phi had a high cancer detection rate. Most importantly, %p2PSA and Phi may be useful as tumor markers for high risk PCa detection.

PSA cannot accurately distinguish between benign and malignant pathology in the tPSA range of 2–10 ng/ml. In addition, more indolent PCas were detected, which will not develop into clinical significant PCas32. New markers that can accurately detect as well as differentiate patients with aggressive and non-aggressive PCa are needed. According to preliminary investigations and observational studies, %p2PSA and Phi may improve discrimination between men with and without PCa, and they are associated with aggressiveness of PCa, as indicated by Gleason score6101112. But some studies did not demonstrate a relationship between Gleason score and p2PSA level7. So we carried out a meta-analysis to further confirm the ability of %p2PSA and Phi in PCa patient's diagnosis, and derive a more precise estimation of their ability to predicate aggressive PCa.

To date, only one systematic review has been conducted to assess the ability of %p2PSA and Phi to diagnose PCa. Both %p2PSA and Phi improved the accuracy of PCa detection compared to PSA and %fPSA, particularly with the PSA range of 2–10 ng/ml33. However, only the papers published before December 2011 were included, and subgroup analysis or meta-regression were not performed to explore the sources of heterogeneity. In addition, the diagnostic accuracy of %p2PSA and Phi for differentiating aggressive PCa from non-aggressive PCa has not yet been summarized. In our meta-analysis, articles published in 2012, 2013 and 2014 were also included. To the best of our knowledge, this was the first attempt to synthesize existing literature to evaluate the diagnostic accuracy of %p2PSA and Phi for aggressive PCa.

Continued emergence of promising data indicate that %p2PSA and Phi are better parameters than tPSA and fPSA for the detection of PCa and aggressive PCa27. Based on synthesis of the results in our meta-analysis, the diagnostic test may be particularly useful in patients with diagnostically challenging PSA scores ranging 2–10 ng/ml, as sensitivity reaches 85%–86%, and specificity reaches 40%–45% for %p2PSA and Phi respectively. As shown previously, the PSA test had varying sensitivity and poor specificity, only about 20% for specificity31. Our results showed that %p2PSA and Phi test have potential to complement PSA screening due to their high specificity. We also observed that %p2PSA and Phi have nearly the same effect in diagnosing PCa. Both %p2PSA and Phi showed a high accuracy for detecting PCa (AUC of 0.72 and 0.70 respectively). SROC of our systematic review showed a curvilinear relationship between sensitivity and 1-specificity, and the Spearman's correlation coefficient was 0.705 (P = 0.005) and 0.744 (P = 0.002) for %p2PSA and Phi respectively, which may be due to a threshold effect.

The detection of aggressive PCa is the main concern of current discussions on the usefulness of PCa detection markers. Recent studies have shown that higher %p2PSA and Phi values were associated with increased probability of detecting Gleason score ≥ 7 PCa and were higher in aggressive cancers, so %p2PSA and Phi may be able to predict more aggressive PCa on prostatic biopsies5. Phi provided the best discrimination between aggressive and non-aggressive PCa than other markers5. According to our pooled results, the sensitivity for discriminating between aggressive and non-aggressive PCa was very high, 96% and 90% for %p2PSA and Phi respectively, which would miss a small number of aggressive PCa, but the specificity was relatively low, only 9% and 17% for %p2PSA and Phi respectively. The AUC for Phi (0.67) exceeded that of the %p2PSA (0.54) in discriminating aggressive versus non-aggressive PCa. Increased Phi levels and %p2PSA might be helpful in identifying patients harboring more aggressive forms of the disease. Some cutoffs were used for %p2PSA and Phi, at a threshold of 1.2 and 25.5 for %p2PSA and Phi respectively, about 4% of aggressive cancers would have been missed8. And these cutoffs did not allow significant improvement in differentiation between Gleason score 4 + 3 = 7 and 3 + 4 = 7 cancers5. Phi values did not differ with age and race1124, this suggests that Phi may be more prosperous to be used in a wider scope of all men, irrespective of race, age and background in predicting aggressive PCa. A limited number of studies were included, therefore it was difficult to draw a definitive conclusion for their ability to discriminate. Hence, further research should aim at determining the potential aggressiveness of PCa in patients with high %p2PSA and Phi values.

When interpreting the results, the heterogeneity caused by the different optimum cutoff points determining whether a test is positive or negative must be taken into consideration, since there is actually no consensus regarding the most appropriate level, in part, due to differences in study design and methodology. The cut-off points were only reported in 7 and 9 studies for %p2PSA and Phi respectively, and the cut-off values were quite different for all the studies. Increased Phi values were associated with increasing PCa detection rate, when the phi values were <25, 25–34.9, 35–54.9 and ≥55, the PCa detection rates were 11%,18.1%, 32.7% and 52.1% respectively11. The full clinical potential and the threshold of %p2PSA and Phi for diagnosing PCa and identification of aggressive disease need to be determined. If the maximum number of aggressive PCas were to be detected, it might be acceptable to not detect some of the less aggressive, clinically insignificant PCas.

The present meta-analysis has a number of limitations that must be taken into account. First, the preanalytic and analytic management of blood samples were different in the studies such as the case when archived serum was used in some studies, and different calibrations were used in the literatures, including Hybritech and WHO standardization. PSA and fPSA were measured using Hybritech Tandem assays except for one study. Second, the main bias of all the studies was that the inclusion criteria was based on the risk of having PCa, the final decision on whether to biopsy or not was based on elevated PSA values and not based on the %p2PSA or Phi, which would cause selection bias. Third, only one study was a randomized study, differences in study designs including prospective, retrospective and mixed, population characteristics, cutoff points and the number of cores obtained in prostate biopsies (rang, 6–24) may contribute to the heterogeneity. Fourth, another limitation was potential publication bias, since yet unpublished studies and non-English studies were excluded, however, we did not confirm the presence of publication bias. We should also note that the number of studies included for analysis for the purpose of discriminating aggressive and non-aggressive PCa was relatively low. Due to limited literatures, subgroup-analyses and meta-regression were not performed. Finally, the best strategy to correlate %p2PSA and phi with aggressiveness is to perform a correlation between preoperative %p2PSA, phi and whole gland sample after radical prostatectomy; unfortunately only the paper by Guazzoni et al12 focused on this topic. Due to the pathological difference between prostate biopsy and sample after radical prostatectomy, we did not include this study into our meta-analysis.

Due to the limitations mentioned above, the results of this meta-analysis should be interpreted with care. Nevertheless, our study was conducted at an appropriate time, because enough data was available to use meta-analytical methods. And we provided the most up-to-date information on this topic. However, it is necessary to conduct the large multi-center randomized-controlled studies or prospective trials with similar methodology to evaluate the ability of %p2PSA and Phi in predicting aggressive PCa in the future. In addition, similar comparison outcomes should be based on the same preanalytic and analytic management for blood samples, same calibration technology (Hybritech or WHO), same TRUS prostate biopsy techniques and biopsy cores or based on the pathology of prostatectomy specimen.

In conclusion, our meta-analysis suggested that both %p2PSA and Phi have high diagnostic accuracy rates and can be used in early PCa diagnosis. In addition, %p2PSA and Phi might be potential biomarkers in predicating aggressive PCa. In our opinion, if the data will be confirmed by further larger sample size studies, the two biomarkers might be helpful in guiding biopsy decisions.

Methods

Literature search and selection criteria

Our meta-analysis was performed following the guidelines concerning Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) statement34 (Supplementary Table 5). A systematic literature search on PubMed was conducted for articles published from January 2000 to March 2014. The strategy was to use and combine the following terms included in their titles, abstracts, or keywords lists: “prostatic neoplasm”, “prostate cancer”, “p2PSA”, “[−2]proPSA”, “[−2]proenzyme prostate specific antigen”, “Prostate Health Index”, “Phi”, “diagnosis”, “sensitivity“, and “specificity”. Additionally, we reviewed the reference list of each relevant paper by hand. Two researchers (W Wang and M Wang) independently reviewed all identified titles, abstracts and manuscripts to determine if a study was suitable for the meta-analysis.

Studies that met the following criteria were included in the meta-analysis: (1) case-control or cohort design; (2) diagnostic test about %p2PSA or Phi for PCa; (3) histological results were based on prostate biopsy; (4) data on sensitivity and specificity could be extracted. Non-English language papers or studies with only abstracts available for information extraction were excluded. When more than one article was published based on the same patient population or same project, only the most recent or complete report was used, to avoid overlapping between cohorts.

Data extraction and quality assessment

We collected information from each eligible study and entered the data into a structured database including: last name of the first author, year of publication, country of origin, study design, ethnicity, age range of participants, the tPSA and fPSA range, sample size, indication for biopsy, type of biopsy used, the number of biopsy-confirmed PCa and the values of true positive (TP), false-positive (FP), false negative (FN), true negative (TN), summary receiver operating curve (SROC) and cut-off points if available. Of these studies, Gleason score ≥7 was defined as high Gleason score and aggressive, and Gleason score <7 was defined as low Gleason score and non-aggressive. The study quality was assessed with Quality Assessment of Diagnostic Accuracy Studies (QUADAS)3536. Two authors extracted the data and undertook the quality assessment independently. If disagreements appeared, a third author helped to resolve the problem.

Data analysis and presentation

Based on the collected data, 2 × 2 contingency tables were created to calculate the indices of diagnostic validity, the main outcome data included pooled sensitivity (Se) and specificity (Sp), each value was determined with 95% confidence intervals. A Forest plot was used to display the results of the meta-analysis. Methodological heterogeneity was assessed during selection.

We determined the existence of a threshold effect, which is the primary cause of heterogeneity in test accuracy studies, defined as the use of different cut-offs or thresholds in different studies to determine a positive test result. The analysis of the diagnostic threshold was assessed using the graphic representation of “Se” vs “1-Sp” on a receiver operating characteristic (ROC) plane and also by calculating the Spearman correlation coefficient. The ROC plane provides a graphic representation of the pairs of Se and Sp; characteristically its points show a typical pattern of “shoulder arm” if the threshold effect exists. The results were synthesized and represented graphically in a forest plot. If there was evidence of a threshold effect, the studies were combined to create a SROC and we calculated an additional measurement of accuracy of the technique (Q*), and obtained the area under the curve (AUC). Possible sources of heterogeneity were assessed through chi-square, the null hypothesis that the studies are homogeneous was rejected with a p value of <0.05, and heterogeneity was quantified using the I2 statistic which was interpreted as low (25%–50%), moderate (51%–75%) and high (>75%) levels of heterogeneity. The Mantel-Haenszel fixed effect model was used if heterogeneity was not found; otherwise, the Dersimonian-Laird random effect model was used. In addition, subgroup analyses were also conducted by collection of sera (prospective, retrospective), Ethnicities (Caucasian and Asian) and QUADAS score. Potential publication bias was examined using Deeks' funnel plots37, with P < 0.05 for the slope coefficient indicating significant asymmetry. All meta-analyses were performed using MetaDisc (Version 1.4)38 and STATA (Version 10.0).