Study population

We evaluated 1372 men from October 2003 through June 2009 from 8 medical centers. The study cohort included men age ≥50 years of all ethnic backgrounds who met the following criteria: (1) no history of PCa, (2) non-suspicious digital rectal examination (DRE) findings, (3) pre-study PSA of 1.5–11.0 ng/mL (all PSA concentrations were re-tested in the Access Hybritech assay, and only those 2–10 ng/mL were included), (4) ≥6 core biopsy within 6 months of blood draw, and (5) a histologic diagnosis from prostate biopsy.

Exclusion criteria were: (1) treatment with medications that alter PSA levels or interventions such as transurethral resection of the prostate prior to blood draw, (2) acute prostatitis or urinary infection at blood draw, (3) a final Access Hybritech PSA value outside the 2.0–10.0 ng/mL range, (4) no blood draw or biopsy at the appropriate time interval, or (5) prior androgen-replacement therapy.

Seven men were excluded due to unevaluable tests from hemolyzed or lipemic samples or p2PSA duplicate results with >15% coefficient of variation at p2PSA concentrations ≤ 20 pg/mL, for which samples could not be retested. Finally, one site enrolled only men aged 55–75 years (our study enrolled men aged ≥ 50 years), and our study-specific sample storage limit (≤ 5 years) further limited the evaluable population to men aged 62–74. Because the age distribution from this site may not be representative of the target population, we performed separate analyses excluding and including these men.

The final study population of 892 men included: (1) 121 (13.6%) prospectively enrolled, (2) 743 (83.3%) prospectively enrolled under separate protocols, and (3) 28 (3.1%) retrospective samples. The study population included 706 (79.2%) initial biopsies, 159 (17.8%) repeat biopsies, and 27 (3%) with unknown history of prior biopsy. Each institution enrolled an approximately equal number of men with or without PCa, for a total of 430 (48.2%) men with PCa and 462 (51.8%) without. Participants and investigators were blinded to p2PSA results, and testing sites were blinded to individual clinical information.

Test Methods

Access Hybritech p2PSA, PSA, and fPSA assays were measured on the Beckman Coulter Access 2 Immunoassay Analyzer^***. Serum samples were collected and processed within 8 hours, then stored frozen at ≤−70°C prior to testing (≤5 years from the date of blood draw), conditions that allowed accurate measurement of phi.²¹ Samples were tested at one of 3 laboratories. PSA and fPSA assays were run using one-sample replicate. The p2PSA assay was run in duplicate (first replicate used for data analysis, consistent with the proposed product labeling) according to the testing protocol. Evaluation of the first replicate compared to the mean of duplicates using Passing-Bablock regression analyses showed no difference (Spearman R=0.9985). The p2PSA assay is a two-site immunoenzymatic sandwich assay using specific monoclonal antibodies and 6 calibrators from 0- 5000 pg/mL.

Statistical Methods

The minimum sample size was estimated as 295 patients without cancer to detect a 10% difference in specificity between phi and % fPSA at α = 0.05 and β = 0.10. In addition, a minimum sample size of 350 cancer patients was determined to accurately estimate sensitivity at 95% with a 95% confidence interval of ± <3%. The target sample size was then increased to 400 participants in each group.

The primary null hypothesis was that phi has no greater specificity than %fPSA at 95% sensitivity. This hypothesis was tested using bootstrap-based receiver operating characteristic (ROC) analysis.²² Briefly, 1000 datasets of benign and PCa patients were generated to repetitively sample the study population.^23–25 Differences in the specificity between phi and %fPSA at 95% sensitivity were calculated for the 1000 pairs of replicate datasets. The standard error of the difference in specificities was then estimated with adjustment for correlation between the results of the two tests. Finally, the bootstrap-estimated standard error was used to evaluate whether the difference in specificities is >0 assuming normal distribution of the differences. A one-sided statistical test was performed for this analysis. This method was also used to compare the specificities of phi and %fPSA at 90%, 85%, and 80% sensitivities.

The secondary null hypothesis was that the area under the ROC curve (AUC) for phi equals that of %fPSA. This hypothesis was tested by evaluating whether the difference between the estimated AUCs for the two tests equals 0 using empirical methods.^{26, 27} The standard error of the difference was calculated accounting for the correlation in AUCs as appropriate for comparison of paired data. The difference between the two estimated AUCs has been shown to have a Chi-square distribution with one degree of freedom. The AUCs for phi and %fPSA were also estimated for each prostate volume tertile to determine whether the observed trend in AUCs differed by prostate volume.

The validity of pooling data across sites was evaluated by fitting a logistic regression model with cancer status as the dependent variable, with phi (dichotomized at the estimated cutoff for 95% sensitivity) and site as independent predictors including interaction terms for site and phi. A statistically significant parameter estimates for this interaction terms was considered evidence of heterogeneity in phi performance by site.

Comparisons between participant subgroups were performed using the Wilcoxon Rank-Sum test for continuous variables and the χ2 test for categorical variables. Two-sided statistical tests were used on all analyses except as noted above, and statistical significance was defined as p<0.05. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, North Carolina).

Individual Patient Risk Assessment

A 25% PCa detection rate has been previously reported in men with PSA of 2.0–10.0 ng/mL.³ For this study, cancer patients were over-sampled by design, resulting in 48.2% of study participants with PCa. Since the proportion of PCa was determined by design, direct calculation of PCa probability would result in inflated estimates for detecting PCa. Therefore, to obtain more accurate risk estimates for PCa, we adjusted the proportion of PCa to 25% by repetitively sampling the study population 1000 times with each replicate dataset consisting of 462 (75%) benign and 154 (25%) cancer participants.^23–25 The mean probability of cancer in the bootstrapped datasets for each phi range was used as the point estimate, and bootstrap-estimated standard errors were used to calculate 95% confidence intervals. Likewise, relative risk estimates were calculated for each replicate dataset by dividing the probability of PCa in each phi range to that of phi 0–24.9. The mean relative risk and bootstrap-estimated standard errors were used to calculate the risk estimate and 95% confidence intervals. In addition, age-stratified probability estimates for PCa were calculated to determine whether observed trends persist in all age groups.

Receiver Operating Characteristic (ROC) Results

Figure 1 shows the sensitivity and specificity for all observed PSA, fPSA, p2PSA, %fPSA, and phi cutoffs in the 2.0–10.0 ng/mL PSA range. At a given sensitivity, phi demonstrated greater specificity than the other analytes (Table 2). At 95% sensitivity, the specificity of phi was 16.0% compared to 8.4% for %fPSA (P=0.015), 7.6% for p2PSA, 6.5% for PSA, and 3.5% for fPSA, rejecting the primary null hypothesis. Moreover, at lower sensitivities (90%, 85%, and 80%) for PCa detection, the specificity of phi was significantly greater than %fPSA (i.e., unnecessary biopsies possibly avoided: 26% vs. 18%, P= 0.036; 39% vs. 28%, P= 0.006; 45% vs. 37%, P= 0.031, respectively).

Open in a separate window

Figure 1

PSA, fPSA, [−2]proPSA, %fPSA, and Phi ROC Curves in the 2–10 ng/mL PSA Range Sensitivity × 1-Specificity for Sequential Cutpoints

The AUC for PCa detection was significantly greater for phi (AUC=0.703) than for %fPSA (0.648, P=0.004), fPSA (0.615), p2PSA (0.557), or PSA (0.525), rejecting the secondary null hypothesis.

Individual Patient Risk Assessment

Higher phi values were associated with an increased risk of PCa detection based upon the adjusted 25% proportion of PCa cases (Table 3). Of the study population, 25%, 33%, 30%, and 13% had phi values of 0–24.9, 25.0–34.9, 35.0–54.9, and ≥ 55.0, respectively. Compared to phi < 25.0, the relative risk of PCa detection on biopsy was 1.6-, 3.0-, and 4.7-fold higher at phi values of 25.0–34.9, 35.0–54.9, and ≥ 55.0, respectively. Overall, a phi≥ 55.0 was associated with a 52.1% probability of PCa.

Age and Probability of PCa

Higher phi values were also associated with higher bootstrapped risk estimates of PCa within each age group. The probability (and relative risk [RR]) of PCa ranged from 10.9% (phi 0–24.9) to 53.4% (phi ≥ 55) (RR 4.9) for the 50–59 age group, 12.5% (phi 0–24.9) to 54.5% (phi ≥ 55) (RR 4.4) for the 60–69 age group, and 5.8% (phi 0–24.9) to 44.8% (phi ≥ 55) (RR 7.7) for the > 70 age group.

Association of phi with Gleason Score

Phi also had a significant relationship with biopsy Gleason score (r=0.138, P=0.004). Among participants with PCa, biopsy Gleason score was <7 in 290 (67.6%) and ≥7 in 139 (32.4%) Compared to phi < 25.0, the relative risk of Gleason ≥ 7 PCa increased to 1.08 for phi values from 25.0–34.9, 1.15 for phi values from 35.0–54.9, and 1.61 for phi ≥ 55.0. The corresponding proportion of cancers with a Gleason score ≥ 7 increased from 26.2% to 28.2%, 30.1%, and 42.1% at phi values of 0–24.9, 25.0–34.9, 35.0–54.9, and ≥ 55.0, respectively (Cochran-Armitage test for trend, P=0.013) (Table 4). The AUC for phi (0.724) exceeded that of %fPSA (0.670) in discriminating between Gleason ≥ 4+3 vs. lower Gleason grade PCa or negative biopsies.

Relationship of TRUS volume and phi

The AUCs for phi exceeded those of %fPSA in all three prostate volume tertiles (≤38, 39–53, and ≥54cc): 1^st tertile: AUC 0.693 for phi vs. 0.614 for %fPSA; 2^nd tertile: 0.707 vs. 0.593; 3^rd tertile: 0.642 vs. 0.559.

Funding/Support:

This work was funded by Beckman Coulter Incorporated, Carlsbad, California; and supported in part by the National Institutes of Health/National Cancer Institute (NIH/NCI) Johns Hopkins Prostate SPORE Grant #P50CA58236, the Early Detection Research Network NIH/NCI Grant #U01-CA86323, and NIH/NCI U01 CA86323 to Dr Partin; NIH/NCI U24 CA115102 to Dr Chan; NIH/NCI U01CA113913 to Dr Sanda; the Urological Research Foundation, Northwestern-University of Chicago Prostate SPORE grant (NIH/NCI P50 CA90386-05S2), the Robert H. Lurie Comprehensive Cancer Center grant (NIH/NCI P30 CA60553), and Beckman Coulter Incorporated to Dr Catalona; the Mayo Clinic Prostate SPORE grant NIH/NCI CA091956 to Dr Klee.

Role of the Sponsor:

Funding for the study was provided by Beckman Coulter, Inc., which contributed to the design, collection and analysis of the study data. Beckman Coulter authors and the clinical investigators jointly developed the manuscript content.