Key points

Introduction

Prostate cancer is the second most common cause of cancer death in US men. In 2015, an estimated 220,800 men will be diagnosed with prostate cancer, and there will be 27,540 prostate cancer–related deaths. The prevalence of prostate cancer increases with age. Of all new prostate cancer cases, only 0.6% are diagnosed among men younger than 44 years of age, with most cases being diagnosed at ages 65 to 74. Prostate cancer is generally asymptomatic until it has reached an advanced stage, a strong incentive to the widespread use of prostate-specific antigen (PSA)-based screening for early detection within the window of curability.

The goal of PSA screening is to test asymptomatic men and improve health outcomes by diagnosing the cancer at an early stage. A benefit of screening is a reduction in the proportion of advanced-stage cases at the time of diagnosis and a decrease in the prostate cancer–specific mortality rate. However, PSA screening has been controversial because of numerous limitations. Although higher PSA levels are a strong predictor of prostate cancer risk, the total PSA measurement is not specific for prostate cancer and is influenced by other factors, such as benign prostatic hyperplasia, prostatitis, and other benign conditions. Consequently, many men undergo unnecessary biopsies leading to the overdetection of some indolent tumors.

The Prostate Health Index (phi), approved by the US Food and Drug Administration (FDA) in June 2012, addresses many of the drawbacks associated with PSA screening. Its specificity is greater because phi is a combination of three different isoforms of PSA: total PSA, free PSA (fPSA), and [-2]proPSA, combined in the following mathematical formula: phi = ([-2]proPSA/fPSA) × √PSA. Phi is a simple blood test, but it outperforms any of its individual components for the identification of clinically significant prostate cancer. This article reviews the major studies on phi in prostate cancer detection and risk stratification.

Introduction

Prostate cancer is the second most common cause of cancer death in US men. In 2015, an estimated 220,800 men will be diagnosed with prostate cancer, and there will be 27,540 prostate cancer–related deaths. The prevalence of prostate cancer increases with age. Of all new prostate cancer cases, only 0.6% are diagnosed among men younger than 44 years of age, with most cases being diagnosed at ages 65 to 74. Prostate cancer is generally asymptomatic until it has reached an advanced stage, a strong incentive to the widespread use of prostate-specific antigen (PSA)-based screening for early detection within the window of curability.

The goal of PSA screening is to test asymptomatic men and improve health outcomes by diagnosing the cancer at an early stage. A benefit of screening is a reduction in the proportion of advanced-stage cases at the time of diagnosis and a decrease in the prostate cancer–specific mortality rate. However, PSA screening has been controversial because of numerous limitations. Although higher PSA levels are a strong predictor of prostate cancer risk, the total PSA measurement is not specific for prostate cancer and is influenced by other factors, such as benign prostatic hyperplasia, prostatitis, and other benign conditions. Consequently, many men undergo unnecessary biopsies leading to the overdetection of some indolent tumors.

The Prostate Health Index (phi), approved by the US Food and Drug Administration (FDA) in June 2012, addresses many of the drawbacks associated with PSA screening. Its specificity is greater because phi is a combination of three different isoforms of PSA: total PSA, free PSA (fPSA), and [-2]proPSA, combined in the following mathematical formula: phi = ([-2]proPSA/fPSA) × √PSA. Phi is a simple blood test, but it outperforms any of its individual components for the identification of clinically significant prostate cancer. This article reviews the major studies on phi in prostate cancer detection and risk stratification.

Phi as a predictor of biopsy outcome

A large prospective multicenter study of phi was initiated in the United States from 2003 to 2009, and ultimately enrolled 892 men with total PSA levels of 2 to 10 ng/mL and findings that were not suspicious for cancer on digital rectal examination (DRE). Participants underwent at least 10-core prostate biopsy, which was the initial biopsy in 79%, repeat biopsy in 18%, and unknown in 3%. The primary objective of the study was to compare the specificity of phi with percent free PSA (%fPSA) at 95% sensitivity for prostate cancer detection. The results showed that phi had significantly greater specificity at 95% sensitivity compared with %fPSA (16.0% vs 8.4%; P = .015). It was also more specific than total PSA. Similar patterns were observed at the 90%, 85%, and 80% sensitivity thresholds. On receiver operating characteristic analysis, phi outperformed both %fPSA (area under the curve [AUC], 0.703 vs 0.648; P = .004) and total PSA (AUC, 0.525). There was also a significant association between phi with the Gleason score on biopsy. Compared with the lowest phi category (scores of 0–24.9), men with the highest phi scores (>55) had a significantly higher risk of detecting any prostate cancer (relative risk, 4.7; 95% confidence interval [CI], 3.0–8.3), and Gleason score greater than or equal to seven disease on biopsy (relative risk, 1.61; 95% CI, 0.95–2.75).

A later study in this population examined the relationship of phi with clinically significant disease in greater detail. Specifically, among 658 men from the prospective trial undergoing initial or repeat prostate biopsy for a PSA level of 4 to 10, phi was a more accurate predictor of clinically significant prostate cancer on biopsy using a variety of different criteria for significant disease. On receiver operating characteristic analysis, phi had a higher AUC for Gleason score greater than or equal to seven (0.707) and Epstein significant disease (0.698) compared with its components PSA (AUCs, 0.551 and 0.549), %fPSA (AUCs, 0.661 and 0.654), and p2PSA (AUCs, 0.661 and 0.654), respectively.

The specificity of phi for clinically significant prostate cancer also was evaluated in biopsy-naive men from the National Cancer Institute Early Detection Research Network Clinical Validation Center cohort. Using Gleason score greater than or equal to seven disease on biopsy as the primary end point, de la Calle and colleagues compared phi with its component parts. The first cohort included 561 men from Harvard with a mean PSA of 6.5 ng/mL and abnormal DRE in 23.7%. Of these men, 20.3% were found to have Gleason score greater than or equal to seven disease on biopsy, and phi had an AUC of 0.82 for high-grade disease. Using a cutoff of 24 as the criterion for biopsy would have avoided 41% of unnecessary biopsies among men without prostate cancer and 17% of overdiagnosed cases. These results were compared with a validation population including 395 men from two other US institutions (Weill Cornell Medical College and University of Michigan), with a mean PSA of 5.9 ng/mL and abnormal DRE in 10.6%. In this cohort, 30.9% had Gleason score greater than or equal to seven disease on biopsy, and the AUC for phi was 0.78. Using a phi cutoff of 24 as the criterion for biopsy would have avoided 36% of unnecessary biopsies among men without prostate cancer, and 24% of overdiagnosed indolent cancers in the validation population.

Phi also has been evaluated prospectively in several European populations. Guazzoni and colleagues reported on 268 men with PSA levels of 2 to 10 ng/mL and negative DRE who were scheduled for extended prostate biopsy (18–22 cores) at a large academic center in Italy. The primary objective of the study was to compare phi with commonly used reference tests, including total PSA, %fPSA, and PSA density. Overall, 39.9% of the population was diagnosed with prostate cancer, and these men had a significantly higher phi (median, 44.3 vs 33.1; P <.001). At 90% specificity, phi had greater sensitivity (42.9%) than %fPSA (20.0%) or PSA density (26.5%). Predictive accuracy was higher for phi (AUC, 0.76) than for PSA density (61%), %fPSA (58%), and total PSA (53%). At 90% specificity, phi had greater sensitivity (42.9%) than %fPSA (20.0%) and PSA density (26.5%). The addition of phi to a multivariable model with age, prostate volume, and total and free PSA led to a significant gain in predictive accuracy (0.83 from 0.72; P <.001). Phi was also a significant independent predictor of high-grade disease.

Lazzeri and colleagues subsequently reported on a large prospective evaluation of phi in 646 men age greater than 45 years from five European countries. All of the men in this study were undergoing initial prostate biopsy with at least 12 cores for a PSA level from 2 to 10 ng/mL with or without an abnormal DRE. The primary objective of the study was to compare the performance characteristics of phi with total and free PSA for prostate cancer detection; a secondary end point was to examine the relationship to Gleason score on biopsy. The 264 (40.1%) men diagnosed with prostate cancer had a significantly higher median phi compared with those with negative biopsy (48 vs 32; P <.001). On multivariable analysis including total, free, and %fPSA, the addition of phi led to a significant 6.4% and 7.5% gain in predictive accuracy for overall and Gleason score greater than or equal to seven prostate cancer on biopsy, respectively. At 90% sensitivity, use of phi would have avoided 100 unnecessary biopsies (15.5%) while missing only 1.1% of aggressive cancers (compared with 7.5% missed using %fPSA).

Phi as a component of multivariable risk stratification

There has been a paradigm shift in prostate cancer decision-making from a one-size-fits-all approach using total PSA, as was done in the early 1990s, toward multivariable risk assessment taking into account individual patient characteristics. Such an approach is recommended by numerous contemporary clinical practice guidelines, such as the Melbourne Consensus statement.

Given the substantial international evidence showing the superiority of phi over PSA, several tools have been created that combine phi with other clinical risk factors to aid in prostate biopsy decisions. Lughezzani and colleagues reported a study including 729 men from a major tertiary referral center in Italy undergoing extended prostate biopsy (66.5% initial, 33.5% repeat). These men had total PSA levels ranging from 0.5 to 20 ng/mL, and 17.7% had a suspicious DRE. Similar to the previous studies, phi had superior predictive accuracy for biopsy outcome (AUC, 0.80) compared with %fPSA (AUC, 0.62) or total PSA (AUC, 0.51). The addition of phi to a multivariable model with age, prostate volume, DRE, and prostate biopsy history led to a statistically significant 7% gain in predictive accuracy. The authors created a nomogram combining these five variables, which had an AUC of 0.80. The nomogram was well calibrated for men at low to intermediate risk. Of note, using PSA or %fPSA in the nomogram instead of phi resulted in significantly inferior predictive accuracy (AUC, 0.73 and AUC, 0.75, respectively).

This nomogram was subsequently externally validated in an independent population of men undergoing initial or repeat greater than or equal to 12-core prostate biopsy at five European centers from the PRO-PSA Multicentric European Study Group described previously. This study included 883 patients with total PSA levels of 0.5 to 20 ng/mL, of whom 17% had positive DRE. As in the previous study, phi on its own had a higher AUC (0.68) for prostate cancer detection than total or free PSA. The phi-based nomogram had an AUC of 0.75, was well-calibrated in men at low to intermediate risk, and showed the greatest net benefit on decision curve analysis.

Researchers from Ireland also created a multivariable phi-based nomogram to aid in prostate biopsy decisions. From 2012 to 2014, 250 men aged greater than 40 years were referred to the Irish Rapid Access Clinic for prostate biopsy for an elevated age-specific PSA level or abnormal DRE. All of the men underwent at least 12-core prostate biopsy, and 112 (45%) had prostate cancer detected. Similar to other studies, phi as a stand-alone test had greater predictive accuracy for overall (AUC, 0.71) and particularly high-grade prostate cancer detection (AUC, 0.78) compared with total and free PSA. A multivariable model was then constructed including age, family history, DRE, previous negative biopsy, and either PSA or phi. The model using phi had an AUC of 0.77 for overall prostate cancer and 0.79 for high-grade disease, and significantly outperformed PSA-based model and the predictions of the online Prostate Cancer Prevention Trial risk calculator. In a subset of men undergoing repeat prostate biopsy, the phi-based multivariable model had an AUC of 0.85 for any prostate cancer, and 0.88 for Gleason score greater than or equal to seven disease. These studies confirm that phi is a useful addition to multivariable nomograms for initial or repeat biopsy to improve the accuracy of risk stratification. Phi has also been integrated into the Rotterdam Risk Calculator app, a multivariable prediction tool available on smartphones and tablets for easier use at the point of care.