Prostate cancer antigen 3 (PCA3) is a noncoding RNA that is highly overexpressed in prostate cancer (PCa) tissue and excreted in urine in patients with PCa.
To assess the clinical utility of urinary PCA3 in men at risk of PCa.
We retrospectively reviewed a cohort of 271 men (median age, 63 years) with elevated prostate-specific antigen (PSA), and/or strong family history, and/or abnormal digital rectal examination findings. Diagnostic sensitivity, specificity, positive and negative predictive values (PPV, NPV), positive and negative likelihood ratios (LR+, LR−), and diagnostic odds ratio (DOR), and area under the receiver-operating characteristic curves (AUC) were evaluated.
PCA3 score was a significant predictor of prostate biopsy outcome (P < .001). A PCA3 score of 30 was the optimal cutoff for our study cohort, with a diagnostic sensitivity of 72.7%, specificity of 67.5%, PPV of 47.1%, NPV of 86.2%, LR+ of 2.24, LR− of 0.40, and DOR of 5.55. At this cutoff score, the PCA3 assay could avoid 57.4% of unnecessary invasive biopsies in the overall study cohort and 70.3% in the subgroup with PSA level in the “gray zone” (4–10 ng/mL). A logistic regression algorithm combining PCA3 with PSA increased the AUC from 0.571 for PSA-only to 0.729 (P < .001). The logistic combined marker gained the ability to discriminate low-grade from high-grade cancers.
Our data suggest that PCA3 improves the diagnostic sensitivity and specificity of PSA and that the combination of PCA3 with PSA gives better overall performance in identification of PCa than serum PSA alone in the high-risk population.
P rostate cancer (PCa) is the most frequently diagnosed cancer in US men, with an estimated 180 890 new cases expected in 2016.1 This high incidence is largely due to the widespread use of prostate-specific antigen (PSA) as a screening and diagnostic test for PCa.2 PSA is an androgen-regulated serine protease encoded by the KLK3 gene. It is secreted exclusively by prostate epithelial cells, and the serum concentration is very low in men with a healthy prostate.3 Since normal prostate epithelium can produce PSA too, serum PSA level may be elevated not only in PCa but also in noncancerous conditions such as prostatic inflammation, benign prostate hyperplasia, and urinary tract infection.4 The current recommendation is that a PSA level of 4.0 ng/mL or greater is a basis for referral for further evaluation or biopsy.5 However, the positive biopsy rates are about 25% in men with a PSA level in the “gray zone” (4–10 ng/mL) and 50% in men with PSA greater than 10 ng/mL.6,7 Thus, false-positive PSA results lead to unnecessary biopsy, an invasive procedure whose associated risks include bleeding and sepsis.8,9 False-negative results also occur; 20% to 25% of patients with PCa have a PSA level less than 4.0 ng/mL.10 Therefore, more specific and sensitive biomarkers are needed for more accurate diagnosis of PCa.
Another disadvantage of PSA is its inability to discriminate low-grade, low-stage indolent tumors from more aggressive disease. Treatment of indolent cancer may negatively affect patients' life qualities by causing psychological distress, side effects, or complications.11,12 As most cases of newly diagnosed PCa (92%) are local or regional disease,1 biomarkers that can distinguish aggressive from less aggressive disease are needed to reduce overdiagnosis and overtreatment of indolent PCa.
Among the numerous novel biomarkers identified in an effort to overcome these weaknesses of PSA, prostate cancer antigen 3 (PCA3) is one of the most promising. Identified in 1999, PCA3 is a long noncoding RNA located on 9q21-22.13 PCA3 is PCa specific with overexpression in PCa but not in benign prostate tissue13,14 or other types of normal tissue or tumors.15 It is excreted in the urine of patients with PCa. In 2012, the Progensa urinary PCA3 assay (Hologic, Marlborough, Massachusetts) was approved by the US Food and Drug Administration (FDA) as an aid in decision making for repeated biopsy in men older than 50 years who have had 1 or more previous negative prostate biopsy results and for whom a repeated biopsy would be recommended by a urologist. Many studies have been performed hitherto to test the clinical utility of PCA3. Most of them showed that this noninvasive assay is useful in predicting prostate biopsy outcome.16–18 However, there are some controversies over the optimal PCA3 cutoff score to detect PCa and the ability of PCA3 to predict tumor aggressiveness.
The aims of this study were to evaluate the clinical performance of the urinary PCA3 assay in a clinical setting with men at high risk of PCa, to assess the correlation between PCA3 score and tumor aggressiveness, to achieve an optimal PCA3 cutoff score for PCa detection, and to provide a logistic regression model for better detection of PCa by combining urinary PCA3 score with serum PSA level and other demographic and clinical variables such as age, race, family history, digital rectal examination (DRE) findings, and prostate volume.
METHODS
Study Cohort
We retrospectively reviewed 304 PCA3 and PSA results from 287 men (median age, 63 years; range, 43–79 years) obtained during the period from January 2013 to March 2016 at The University of Texas MD Anderson Cancer Center (Houston, Texas). The institutional review board approval for this study was waived by our institute. This investigation was performed by following the Standards for Reporting of Diagnostic Accuracy (STARD) guidelines.19 These men represent a population at high risk of PCa, with elevated PSA level (>4 ng/mL) or increasing PSA dynamics, and/or strong family history of PCa, and/or abnormal findings on DRE, with or without a previous negative biopsy finding. For the 16 men who had more than 1 PSA or PCA3 result, the man's first result during the study period was used and the repeated results were excluded. Of the 287 men selected, 16 of the cases were excluded due to no biopsy results being available. The final study cohort included 271 men, of which 72 (27%) underwent initial biopsy (patients have never done biopsy before) and 199 (73%) underwent repeated biopsy (patients previously had 1 or more negative biopsy findings in a time range of 1–10 years).
Data Collection
Clinical and demographic data were obtained from the medical records. Urine specimens for PCA3 measurement were collected after DRE. PCA3 was assayed with the Progensa PCA3 kit (Hologic). PCA3 scores were reported as a ratio of urinary PCA3 mRNA to PSA mRNA multiplied by 1000. Serum specimens were collected before prostate examination and PSA was tested on a Tosoh automated AIA-2000 enzyme immunoassay analyzer (Tosoh Bioscience, South San Francisco, California). The PCA3 and PSA assays were tested concurrently or within 2 weeks. Diagnosis of PCa was confirmed by biopsy using 12 or more cores, which was performed within 1 to 6 months before or after the PCA3 and PSA tests. Prostate volume was determined by magnetic resonance imaging using the ellipsoid formula: V = H × W × L × π/6, where H is height, W is width, and L is length of the prostate in the 3-D image.
Statistical Analyses
The men's clinical characteristics were summarized by using descriptive statistics. Two-sample tests were conducted to determine whether the PCA3 score and PSA level were significantly different between men found to have PCa and those not found to have PCa. The area under the receiver-operating characteristic (ROC) curves (AUC) was quantified to evaluate the diagnostic performance of PCA3 score, PSA level, and the two combined. The performance characteristics of both PCA3 and PSA were evaluated by determining the diagnostic sensitivity and, specificity, positive and negative predictive values (PPV, NPV), positive and negative likelihood ratios (LR+, LR−), and diagnostic odds ratio (DOR) with 95% CIs. Multivariable logistic regression analyses were conducted to develop a model incorporating the 2 biomarkers and other risk factors such as age, race, family history, DRE findings, and prostate volume. Statistical analyses were performed by using R software (version 3.2.3, R Foundation for Statistical Computing, Vienna, Austria) and Medcalc (MedCalc Software, Ostend, Belgium).
RESULTS
Clinical Characteristics of the Study Cohort
The study cohort selection and biopsy results are summarized in Figure 1. Biopsy results revealed newly diagnosed prostate adenocarcinoma in 77 of the 271 men (28.4%). Of the 72 men undergoing initial biopsy, 9 (12.5%) had a positive biopsy result. Of the 199 men undergoing a repeat biopsy, 68 (34.2%) had a positive biopsy result, a significantly higher positivity rate than initial biopsy (P < .001). Since the sample size for positive initial biopsy result was not large enough for further analysis, all initial and repeated biopsy results were treated equally as a mixed biopsy data set.
Study cohort selection and biopsy results. Abbreviations: HGPIN, high-grade prostatic intraepithelial neoplasia; PCa, prostate cancer.
Study cohort selection and biopsy results. Abbreviations: HGPIN, high-grade prostatic intraepithelial neoplasia; PCa, prostate cancer.
The baseline demographic and clinical characteristics of the 271 men who underwent biopsy are summarized in Table 1 and are stratified by biopsy result (PCa versus no PCa). Differences in risk factors such as age, race, family history, and abnormal DRE findings did not reach statistical significance between the PCa group (n = 77) and the no PCa group (n = 194; all P > .05). Prostate volume, repeated biopsy, and PCA3 score were identified as significant univariable predictors of prostate biopsy outcome (all P ≤ .001). However, serum PSA level did not show significant association with biopsy outcome (P = .07).
Clinical Performance Evaluation of Urinary PCA3 Assay
Compared with men without PCa, those with PCa showed a significantly higher median PCA3 score (P < .001) but not a significantly higher PSA level (P = .07). Further analysis of the correlation between PCA3 score and biopsy result showed that the probability of a positive biopsy outcome increased as the PCA3 score increased, while positive biopsy results were associated with the entire range of PSA levels (Figure 2, A and B). Correlation analysis showed no significant correlation between PCA3 score and PSA level (Spearman rank correlation: r = 0.05; P = .44) or between PCA3 score and prostate volume (r = 0.03; P = .65). Men with a greater prostate volume (>30 mL) did not show a higher median PCA3 score than those with a smaller prostate volume (<30 mL; P = .41). Unlike serum PSA level, which differed significantly between the inflammation and noninflammation groups (P < .001), PCA3 score did not differ significantly between the 2 groups (P = .60) (Figure 3, A and B). To determine whether PCA3 score can predict tumor aggressiveness in terms of Gleason score, PCA3 scores from the men with a positive biopsy result (n = 77) were analyzed. No significant difference in PCA3 score was detected between the men with a low-grade tumor (Gleason score = 6, contemporary grading group 1, n = 25) and those with a high-grade tumor (Gleason score ≥ 7; contemporary grading group 2, 3, 4, and 5; n = 52; P = .12). Similar results were observed for PSA level (P = .13) (Table 2).
Percentage of biopsy results positive for prostate cancer at various ranges of prostate cancer antigen 3 (PCA3) (A) and prostate-specific antigen (PSA) (B). Values in parentheses indicate sample size.
Figure 3 Relationship of prostate cancer antigen 3 (PCA3) (A) and prostate-specific antigen (PSA) (B) with prostatic inflammation.
Figure 4 Receiver-operating characteristic curves for prostate cancer antigen 3 (PCA3) and prostate-specific antigen (PSA) in (A) the whole study cohort and (B) the subgroup with PSA level in the gray zone (4–10 ng/mL).
Percentage of biopsy results positive for prostate cancer at various ranges of prostate cancer antigen 3 (PCA3) (A) and prostate-specific antigen (PSA) (B). Values in parentheses indicate sample size.
Figure 3 Relationship of prostate cancer antigen 3 (PCA3) (A) and prostate-specific antigen (PSA) (B) with prostatic inflammation.
Figure 4 Receiver-operating characteristic curves for prostate cancer antigen 3 (PCA3) and prostate-specific antigen (PSA) in (A) the whole study cohort and (B) the subgroup with PSA level in the gray zone (4–10 ng/mL).
To assess the test accuracy of the PCA3 assay, AUC was quantified by using R software. The AUC for PCA3 was 0.72 (95% CI, 0.65–0.78), significantly greater than that for PSA, which was 0.57 (95% CI, 0.49–0.65; P = .007; Figure 4, A). To test the clinical utility of PCA3 in men with a PSA level in the gray zone (n = 174), the ROC curve was analyzed. The results showed that PCA3 had a greater AUC (0.70; 95% CI, 0.62–0.79) than PSA (0.53; 95% CI, 0.43–0.62; P = .004) in these men (Figure 4, B).
To characterize the diagnostic performance of PCA3, the sensitivity, specificity, PPV, NPV, LR+, LR−, and DOR were calculated at different PCA3 score cutoffs of 25, 30, and 35 (Table 3). The cutoff score of 25 was used for FDA approval, 30 was chosen by our ROC analysis, and 35 has been reported by other studies.18,20 To determine the optimal cutoff score, 2 different methods were used.21 The first was to calculate the square of the distance on the ROC curve [d2 = (1 − sensitivity)2 + (1 − specificity)2], where the smaller value was the better choice for cutoff. The second method was to calculate the Youden index [Youden index = (sensitivity + specificity) − 1], where the larger value was the better choice for cutoff. The results showed that the cutoff of 30 had the smallest d2 and largest Youden index (Table 3). Furthermore, considering that the aim of the PCA3 test is to reduce unnecessary invasive biopsies and overcome the overdiagnosis and overtreatment that results from the low specificity of PSA, a cutoff with higher specificity was needed. Among all 3 cutoffs compared, the cutoff of 30 gave the best performance in our study cohort, with a diagnostic sensitivity of 72.7%, specificity of 67.5%, PPV of 47.1%, NPV of 86.2%, LR+ of 2.24, LR− of 0.40, and DOR of 5.55. PCA3 score outperformed PSA level at a cutoff of 4.0 ng/mL, with a diagnostic sensitivity of 84.4%, specificity of 23.7%, PPV of 30.5%, NPV of 79.3%, LR+ of 1.11, LR− of 0.66, and DOR of 1.68.
To further evaluate the clinical utility of PCA3 score in preventing unnecessary biopsy, the following formula was used to calculate the probability of avoided biopsy (PAB): (number of false-positive PSA results – number of false-positive PCA3 results)/number of false-positive PSA results.22 The results showed that the PABs were 45.9%, 57.4%, and 62.8% at the cutoffs of 25, 30, and 35, respectively. In men with a PSA level in the gray zone, the PABs were 60.2%, 70.3%, and 70.3% at the cutoffs of 25, 30, and 35, respectively (Figure 5).
Probability of avoided biopsy (PAB) of prostate cancer antigen 3 (PCA3) in the whole study cohort and in the subgroup with prostate-specific antigen (PSA) level in the gray zone (4–10 ng/mL). Colors represent PCA3 cutoff scores of 25, 30, and 35. PAB is defined as the (number of false-positive PSA results – number of false-positive PCA3 results)/number of false-positive PSA results.
Figure 6 Receiver-operating characteristic curves for (A) the bivariable and (B) multivariable logistic regression models of combining prostate cancer antigen 3 (PCA3) score and prostate-specific antigen (PSA) level.
Probability of avoided biopsy (PAB) of prostate cancer antigen 3 (PCA3) in the whole study cohort and in the subgroup with prostate-specific antigen (PSA) level in the gray zone (4–10 ng/mL). Colors represent PCA3 cutoff scores of 25, 30, and 35. PAB is defined as the (number of false-positive PSA results – number of false-positive PCA3 results)/number of false-positive PSA results.
Figure 6 Receiver-operating characteristic curves for (A) the bivariable and (B) multivariable logistic regression models of combining prostate cancer antigen 3 (PCA3) score and prostate-specific antigen (PSA) level.
Logistic Regression Model
To improve the diagnostic accuracy of the PCA3 score, we performed a logistic regression analysis combining PCA3, PSA, and other PCa risk factors. By fitting a logistic regression model to the data, a PCA3-PSA combined marker was generated by computing −2.032 + 0.014×PCA3 + 0.067×PSA. The AUC for the combined biomarker was 0.73 (95% CI, 0.67–0.79), slightly higher than that for PCA3-only and significantly higher than that for PSA-only (P < .001; Figure 6, A). The combined biomarker was significantly associated with biopsy outcome (PCa versus no PCa, P < .001). Moreover, the combined biomarker had a significantly higher median value in men with a Gleason score of 7 or greater (median, 0.18) than in men with a Gleason score of 6 (median, 0.09; P = .04).
After fitting univariable logistic regression models for each covariate, prostate “volume” and “repeated biopsy” were identified as significant risk factors (P < .001 and P < .01, respectively). “Volume” and “repeated biopsy” were incorporated into the PCA3-PSA combined biomarker, and a new biomarker was generated by computing (−0.786 + 0.015×PCA3 + 0.077×PSA – 0.037×Volume + 0.944×Repeated Biopsy). In this analysis, 91 men with missing “volume” or “repeated biopsy” data were excluded. Addition of the 2 significant risk factors increased the AUC of the PCA3-PSA combined biomarker to 0.79 (95% CI, 0.72–0.86), significantly higher than the 0.68 (95% CI, 0.60–0.76) for PSA-only (P = .02; Figure 6, B). The multivariable marker demonstrated a significant association with biopsy outcome (PCa versus no PCa, P < .001). However, it did not show a significant association with tumor aggressiveness (P = .14), partially because of the missing data.
DISCUSSION
While the urinary PCA3 assay was approved in 2012 by the FDA to aid in decision making for repeated prostate biopsy, its clinical utility has not been fully elucidated. In the present study, we retrospectively investigated the clinical performance of urinary PCA3 assay in the population undergoing initial or repeated biopsy and at high risk of PCa with elevated PSA, and/or strong family history, and/or abnormal DRE findings.
In the univariable analysis, the classic risk factors such as age, race, family history, and abnormal DRE findings were not significantly associated with prostate biopsy outcome. In accordance with previous studies,16,23–25 PCA3 score was significantly associated with biopsy outcome, and the urinary PCA3 assay outperformed the serum PSA test, with a significantly greater AUC. Furthermore, our findings are consistent with those of other studies showing that PCA3 score is independent of prostate volume and PSA level.24,26,27 We also looked at the relationship between prostatic inflammation and PCA3 score, which has not been evaluated in many published studies. Vlaeminck-Guillem et al,28 in a small study of 38 men (mean age, 37.5 years), found PCA3 score to be independent of chronic prostatitis. In our study, PCA3 score was not affected by prostatic inflammation, either acute or chronic. Finally, in our ROC analysis, PCA3 had a greater AUC than PSA in the subgroup whose PSA level was in the gray zone. These data suggest that PCA3 is a useful, independent diagnostic tool for PCa, especially in men with elevated PSA due to an enlarged prostate or prostatic inflammation.
Since the optimal cutoff PCA3 score is still inconclusive, we compared the diagnostic performance characteristics of PCA3 at different cutoffs. In consideration of the clinical need for a more specific biomarker for PCa, a cutoff of 30 was chosen for our study cohort. At the cutoff of 30, PCA3 score reached the best balance of sensitivity and specificity and considerably better performance than PSA. Furthermore, use of PCA3 score at the cutoff of 30 could avoid 57.4% of unnecessary biopsies in the overall cohort and 70.3% in the subgroup with PSA level in the gray zone. These data suggest that the PCA3 assay can help reduce unnecessary biopsies triggered by elevated PSA. In a recent systematic review and meta-analysis of PCA3 utility in the diagnosis of PCa,18 a total of 46 clinical trials comprising 12 295 participants were analyzed. For the mixed biopsy studies, the pooled sensitivity, specificity, DOR, and AUC were 66%, 68%, 5.13, and 0.75, respectively. For the studies with a cutoff of 35, the pooled sensitivity, specificity, DOR, and AUC were 63%, 74%, 4.75, and 0.74, respectively. Compared with these data, our results are similar or better, further confirming that PCA3 is a reliable predictor of PCa.
It has been reported that incorporating PCA3 score with other biomarkers or risk factors such as PSA level, age, race, family history, DRE findings, and previous biopsy can improve diagnostic accuracy.23,24,29,30 In our bivariable logistic regression model combining PCA3 and PSA, the AUC was significantly higher than the AUC of PSA-only but not of PCA3-only. These data suggest that adding PCA3 to PSA can largely enhance the diagnostic value of PSA. Furthermore, the bivariable model was able to distinguish low-grade tumors (Gleason score = 6) from high-grade tumors (Gleason score > 7), which may help reduce overdiagnosis and overtreatment via better risk stratification of the disease. No consensus has yet been reached on the relationship between PCA3 and tumor aggressiveness, some studies claiming a significant correlation of PCA3 score with Gleason score16,31–35 and others finding no such correlation.24,25,36 Although PCA3 score alone did not distinguish between low-grade and high-grade tumors in our study, the bivariable model combining PCA3 and PSA gained the ability to distinguish the 2 groups. In the multivariable model including PCA3, PSA, prostate volume, and repeated biopsy, the AUC was increased significantly over that of the bivariable model, which suggests that incorporating these 2 clinical characters can even further improve diagnostic accuracy.
The major limitation of this study is the cohort size, especially the subgroup representing the population undergoing initial biopsy. Because of the small size of the initial biopsy group, a comparison between initial biopsy and repeated biopsy was not possible. However, the mixed biopsy population may better reflect the real patient population in clinical practice. Moreover, many observations on prostate volume and repeated biopsy were missing, which reduced the size of the cohort for multivariable logistic regression analysis. However, the usefulness of both the bivariable and multivariable models needs further internal and external validation before they can be used in the clinic.
In summary, our results demonstrate that the urinary PCA3 assay is a sensitive and specific diagnostic tool for PCa in men at high risk of developing the disease. It overcomes the shortcomings of PSA, which may be elevated in noncancerous conditions such as enlarged prostate and inflammation. Combining PCA3 with PSA improves performance accuracy in identification of PCa with better stratification of cancer aggressiveness, which would provide better guidance for more effective biopsy decisions and lead to a considerable reduction of overdiagnosis and overtreatment of low-grade PCa.