Catalona WJ, Partin AW, Slawin KM, Brawer MK, Flanigan RC, Patel A, Richie JP, deKernion JB, Walsh PC, Scardino PT, Lange PH, Subong ENP, Parson RE, Gasior GH, Loveland KG, Southwick PC. Use of the Percentage of Free Prostate-Specific Antigen to Enhance Differentiation of Prostate Cancer From Benign Prostatic DiseaseA Prospective Multicenter Clinical Trial. JAMA. 1998;279(19):1542-1547. doi:10.1001/jama.279.19.1542
From the Division of Urologic Surgery, Washington University School of Medicine, St Louis, Mo (Dr Catalona); Department of Urology, The Johns Hopkins Hospital, Baltimore, Md (Drs Partin and Walsh and Mr Subong); Baylor College of Medicine, Houston, Tex (Drs Slawin and Scardino); University of Washington, Seattle (Drs Brawer and Lange); Loyola University Medical Center, Maywood, Ill (Dr Flanigan); School of Medicine, University of California at Los Angeles, (Drs Patel and deKernion); Harvard Program in Urology, Brigham and Women's Hospital, Boston, Mass (Dr Richie); Department of Research and Development, Hybritech Inc, San Diego, Calif (Mr Parson, Mss Gasior and Loveland, and Dr Southwick).
Context.— The percentage of free prostate-specific antigen (PSA) in serum has
been shown to enhance the specificity of PSA testing for prostate cancer detection,
but earlier studies provided only preliminary cutoffs for clinical use.
Objective.— To develop risk assessment guidelines and a cutoff value for defining
abnormal percentage of free PSA in a population of men to whom the test would
Design.— Prospective blinded study using the Tandem PSA and free PSA assays (Hybritech
Inc, San Diego, Calif).
Setting.— Seven nationwide university medical centers.
Participants.— A total of 773 men (379 with prostate cancer, 394 with benign prostatic
disease) 50 to 75 years of age with a palpably benign prostate gland, PSA
level of 4.0 to 10.0 ng/mL, and histologically confirmed diagnosis.
Main Outcome Measures.— A percentage of free PSA cutoff that maintained 95% sensitivity for
prostate cancer detection, and probability of cancer for individual patients.
Results.— The percentage of free PSA may be used in 2 ways: as a single cutoff
(ie, perform a biopsy for all patients at or below a cutoff of 25% free PSA)
or as an individual patient risk assessment (ie, base biopsy decisions on
each patient's risk of cancer). The 25% free PSA cutoff detected 95% of cancers
while avoiding 20% of unnecessary biopsies. The cancers associated with greater
than 25% free PSA were more prevalent in older patients, and generally were
less threatening in terms of tumor grade and volume. For individual patients,
a lower percentage of free PSA was associated with a higher risk of cancer
(range, 8%-56%). In the multivariate model used, the percentage of free PSA
was an independent predictor of prostate cancer (odds ratio [OR], 3.2; 95%
confidence interval [CI], 2.5-4.1; P<.001) and
contributed significantly more than age (OR, 1.2; 95% CI, 0.92-1.55) or total
PSA level (OR, 1.0; 95% CI, 0.92-1.11) in this cohort of subjects with total
PSA values between 4.0 and 10.0 ng/mL.
Conclusions.— Use of the percentage of free PSA can reduce unnecessary biopsies in
patients undergoing evaluation for prostate cancer, with a minimal loss in
sensitivity in detecting cancer. A cutoff of 25% or less free PSA is recommended
for patients with PSA values between 4.0 and 10.0 ng/mL and a palpably benign
gland, regardless of patient age or prostate size. To our knowledge, this
study is the largest series to date evaluating the percentage of free PSA
in a population representative of patients in whom the test would be used
in clinical practice.
MEASUREMENT of serum prostate-specific antigen (PSA) is widely used
as an aid in the early detection of prostate cancer.1
A limitation of PSA testing has been its relative lack of specificity within
the 4.0- to 10.0-ng/mL range, a diagnostic gray zone in which prostate cancer
is present in only 25% of patients. Most patients with prostate cancer and
a PSA level less than 10.0 ng/mL have early-stage disease, whereas more than
half of the patients with PSA levels above 10.0 ng/mL have advanced disease.1 Thus, the detection of prostate cancer in its potentially
curable stages requires the use of low PSA cutoffs for screening, which leads
to many unnecessary biopsies.
Prostate-specific antigen exists in multiple forms in serum and is predominantly
complexed to protease inhibitors; however, one form of PSA, free PSA, is not
bound to these proteins.2,3 Measurement
of PSA forms in serum helps discriminate between prostate cancer and benign
prostatic disease.4,5 For unknown
reasons, the percentage of free PSA is lower in serum samples from patients
with prostate cancer than in serum samples from patients with a normal prostate
or benign disease. Preliminary evidence also suggests that a lower percentage
of free PSA may be associated with a more aggressive form of prostate cancer.6- 9
Low percentages of free PSA are well established as indicators of prostate
but the selection of cutoffs for use in clinical practice is complicated by
the partial dependence of percentage of free PSA on patient age, prostate
size, and the total PSA level.10- 12,16,17
In the present study, we evaluated the ability of percentage of free
PSA to enhance the specificity of PSA testing in prostate cancer detection,
developed guidelines for use of percentage of free PSA in clinical practice,
and determined the relationships between percentage of free PSA and the histopathologic
features of the prostate cancers detected.
Serum samples were obtained from subjects meeting study entry criteria
at 7 university medical centers between July 1994 and December 1996; 91% of
the samples were prospective and 9% were from serum banks of recently evaluated
patients. Subjects were 50 to 75 years of age and had received no treatment
for prostatic disease at the time of the phlebotomy. All subjects had serum
PSA concentrations between 4.0 and 10.0 ng/mL and digital rectal examination
(DRE) findings that were not indicative of cancer (Table 1 and Table 2).
All subjects had undergone ultrasound-guided 6-sector needle biopsies of the
prostate and, thus, had a histologically confirmed diagnosis prior to determination
of free PSA concentrations. In this blinded study, pathologists did not have
access to percentage of free PSA values and laboratory scientists did not
have access to diagnoses. The study was performed in compliance with the requirements
of the respective institutional review boards.
Subjects were enrolled primarily through prostate cancer screening centers,
although subjects screened elsewhere and referred for treatment to the study
centers were also enrolled. This population of men with biopsies based solely
on elevated PSA levels represents the population of interest in which percentage
of free PSA would be used in clinical practice.
Serum samples were processed and refrigerated within 3 hours of blood
draw. If the serum sample was to be assayed within 24 hours after collection,
the specimen was stored at 2°C to 8°C. Specimens held for longer times
were stored at −70°C until analyzed.18
Concentrations of PSA were determined using the Tandem PSA and free PSA monoclonal
antibody assays (Hybritech Inc, San Diego, Calif).19
The same serum sample was used to determine both total PSA and free PSA concentrations,
as would be done in clinical practice. Total PSA testing was performed and
the sample was frozen. The free PSA concentration was measured after it was
determined that the subject met all study entry criteria. Multiple freeze-thaw
cycles do not affect total PSA and free PSA measurements, and both analytes
are stable under the conditions described herein.18
The nonparametric Wilcoxon test was used for comparisons between group
medians, and linear regression analysis was used to assess the relationships
among percentage of free PSA and PSA, age, and prostate volume. Percentage
of free PSA was calculated as the ratio of free PSA to total PSA multiplied
A multivariate logistic regression analysis,20
with the likelihood ratio χ2 test, was used to evaluate percentage
of free PSA, age, and PSA. Odds ratios (ORs) and 95% confidence intervals
(CIs) were derived from the logistic regression method using the Wald method.
Receiver operating characteristic (ROC) curves were generated for PSA
and percentage of free PSA, plotting sensitivity vs (1 − specificity).
We determined percentage of free PSA cutoffs that would detect 90% and 95%
of the cancers (sensitivity), as well as the corresponding percentages of
biopsies with negative findings that could be avoided (specificity) by using
each of these cutoffs. Areas under the ROC curves (AUC) were calculated for
the percentage of free PSA and PSA, and a z test
was used to measure differences.
Sample size requirements for the study were calculated to obtain 95%
CIs for all possible cutoffs along the percentage of free PSA ROC curve, so
that all sensitivity and specificity estimates would be within 5 percentage
points or less. This required nearly 400 subjects with cancer and 400 subjects
with benign prostatic disease.
However, since sample size requirements were calculated to address the
ROC curve analysis, the 50:50 ratio of patients with cancer to patients with
benign prostatic disease could not be used for a risk assessment analysis
(positive predictive value), which, unlike ROC curve analysis, is susceptible
to cancer prevalence. Cancer probabilities based on the 50:50 ratio would
inflate the risk estimates,11,15
whereas probabilities based on a 25:75 ratio would provide accurate risk estimates
appropriate for the group of men in whom this test will be used. Therefore,
it was necessary to statistically adjust the 50:50 ratio of subjects with
cancer to subjects with benign prostatic disease to a 25:75 ratio, since we
have previously shown that a 25% rate of positive biopsy results would be
obtained in a population of men with PSA values between 4.0 and 10.0 ng/mL
and normal findings on DRE.1,21
The bootstrap method22,23 was
used to repetitively sample the study population and adjust the proportion
of subjects with cancer from 50% to 25%. This random sampling process was
repeated 1000 times. Median cancer probabilities (risk estimates) and ORs
We enrolled 773 men with histologically confirmed diagnoses (379 with
prostate cancer and 394 with benign prostatic disease). Age and PSA values
were similar for the cancer and benign groups (Table 3). The study population was 86% white, 9% African American,
3% Hispanic, and 2% Asian.
Free PSA values ranged from 0.2 to 5.0 ng/mL, and percentage of free
PSA values ranged from 2% to 52%. Median percentage of free PSA values were
significantly lower in the cancer group (12%) than in the benign group (18%)
No differences were seen between retrospective and prospective samples
in the cancer group (P=.41) or benign group (P=.84). When controlling for patient age, no differences
were seen in percentage of free PSA values for cancer subjects across sites
(P=.11) or for benign subjects across sites (P=.17). No differences were seen between subjects screened
at the study centers vs subjects screened elsewhere and referred to the study
centers for treatment (P=.13). Presence or absence
of symptoms did not affect results (P=.12).
Percentage of free PSA (AUC, 0.72; 95% CI, 0.68-0.75) was significantly
more predictive of cancer than total PSA level (AUC, 0.53; 95% CI, 0.49-0.57)
for this population with total PSA values of 4.0 to 10.0 ng/mL (P<.001) (Figure 2).
Cutoffs for percentage of free PSA of 25% and 22% yielded 95% and 90%
sensitivity, respectively. Use of these cutoffs (ie, performing biopsies only
in patients with percentages of free PSA less than or equal to these cutoffs)
could have avoided biopsies in 20% and 29%, respectively, of the patients
with benign prostatic disease. Calculations, sample size, and 95% CIs for
these cutoffs are shown in Table 4.
There was an inverse relationship between percentage of free PSA and
PSA (r=−0.14, P<.001),
and a direct relationship between percentage of free PSA and age (r=0.34, P<.001). Although the first correlation
is not strong enough to affect the cutoffs within the 4.0- to 10.0-ng/mL range,
the latter correlation has a significant effect on the selection of cutoffs,
as shown in Table 5 and Figure 3. Because of this increasing trend
with age, cutoffs could be adjusted upward as age increases to maintain a
constant 95% sensitivity within each age decade (ie, age-specific cutoffs
would miss cancers in 5% of patients within each age decade: 50-59, 60-69,
and 70-79 years). Alternatively, a single cutoff (25% free PSA) may be used
across all age ranges; in this case, 5% of cancers were missed in subjects
aged 50 to 75 years, with more of the missed cancers occurring in older subjects
(median age, 68 years). The 25% cutoff detected 98% of cancers for subjects
aged 50 to 59 years, 94% for subjects aged 60 to 69 years, and 90% for subjects
aged 70 to 75 years.
Of the 379 subjects with cancer, 268 (71%) underwent radical prostatectomy.
Analysis (χ2 testing) showed a statistically significant increase
in probability of favorable pathologic findings (Gleason score<7, organ-confined
[stages T1 and T2], lymph nodes with negative test results for metastasis,
tumor volume ≤ 10% of gland) as percentage of free PSA increased (34% with
favorable findings for subjects with values ≤ 15% free PSA vs 70% for subjects
with values > 25% free PSA, P<.001). Thus, subjects
with cancer with values above the cutoff tended to have less aggressive disease.
Percentage of free PSA was an independent predictor of prostate cancer
(OR for 10-point decline in percentage of free PSA, 3.2; 95% CI, 2.5-4.1; P<.001) and contributed significantly more than age
(OR, 1.2; 95% CI, 0.92-1.55) or total PSA value (OR, 1.0; 95% CI, 0.92-1.11)
in this cohort of subjects with total PSA values between 4.0 and 10.0 ng/mL.
Prostate volume estimated by transrectal ultrasound was available for
695 subjects (90%). Percentage of free PSA increased with prostate volume
(r=0.55, P=.001). Using
a 25% free PSA cutoff, sensitivity was 99%, 93%, and 87% for subjects with
volumes of 30 cm3 or less, 31 to 50 cm3, and more than
50 cm3, respectively.
Table 6 shows the probability
of detecting cancer with needle biopsy, based on PSA and percentage of free
PSA results. The PSA results in this table were obtained from a prior multicenter
study evaluating the efficacy of total PSA for prostate cancer detection,1,21 and percentage of free PSA results
were obtained from the current study. It can be seen that rising PSA levels
increase the risk of detectable cancer. Percentage of free PSA can further
stratify risk for patients with PSA values between 4.0 and 10.0 ng/mL.
Figure 4 shows the probability
of cancer based on percentage of free PSA as well as patient age. The risk
of cancer was high (55%-56%) when percentage of free PSA was 0% to 10%, regardless
of age. The risk decreased as percentage of free PSA increased (eg, cancer
probability was 5%-9% when percentage of free PSA > 25%). Older subjects were
generally at higher risk than younger subjects.
This prospective study from 7 university medical centers is, to our
knowledge, the largest series to date evaluating the usefulness of percentage
of free PSA in patients with moderate elevations in PSA levels (4.0-10.0 ng/mL),
benign findings on DRE, and histologically confirmed diagnoses.
In prostate cancer detection programs, prostatic biopsy is routinely
recommended for patients with abnormal DRE results regardless of PSA level
(15% of subjects screened for prostate cancer, Table 1), and for patients with normal DRE results and PSA levels
higher than 4.0 ng/mL. Those patients with normal DRE results and PSA levels
of 4.0 to 10.0 ng/mL represent the diagnostic gray zone, in which the total
PSA value has identified the patients as high risk (25% cancer rate compared
with a 4% cancer rate for the general population of men older than 50 years1,21 ), but specificity could be improved
because 75% of biopsy findings are negative.
The current study shows that the use of percentage of free PSA in this
group of patients can enhance the specificity of PSA screening and decrease
the number of unnecessary biopsies. A cutoff of 25% or less free PSA (ie,
perform a biopsy for patients at or below this cutoff) would detect 95% of
cancers and spare 20% of patients with benign prostatic disease from biopsy.
Since the majority of patients with PSA levels between 4.0 and 10.0 ng/mL
have benign prostatic disease, this represents a substantial number of patients
who would avoid an inappropriate biopsy. With the widespread use of PSA screening,
improvements in specificity are desirable. A recent study found that the benefit-cost
equation of prostate cancer detection programs is very sensitive to changes
in test specificity, so that minor increases in specificity (≈5%) produced
marked reductions (≈50%) in net cost per individual screened.24
Only 9% of patients tested for total PSA value would be tested for percentage
of free PSA, but this represents 35% of all biopsies (Table 1). Prostate cancer screening generates controversy, in part
because of specificity concerns, but also because of a lack of long-term studies
evaluating the outcome of treatment. However, interim studies have shown that
PSA testing has contributed to a shift in detection of earlier-stage disease,
from the historically observed 70% rate of advanced cancer at diagnosis to
the currently observed rate of 30%.25 It was
recently shown that the prostate cancer death rate in the United States declined
between 1991 and 1995, providing the first evidence that early detection (probably
through transurethral resection of the prostate, since widespread PSA screening
did not begin until 1991) may decrease mortality rates.26
Medical policy will continue to be shaped as long-term studies progress, but
in the near term, screening test improvements can provide an immediate benefit
to patient care.
The large number of subjects in this study provides confidence in the
percentage of free PSA cutoffs determined for use in clinical practice. A
25% free PSA cutoff will produce 95% sensitivity (95% CI, 92%-97%) and 20%
specificity (95% CI, 16%-24%). In contrast, the cutoffs previously reported
in the literature have provided only preliminary estimates with wide CIs.
The AUC was significantly higher for percentage of free PSA (0.72) than
for total PSA (0.53), indicating that percentage of free PSA is more predictive
of cancer in patients with PSA levels of 4.0 to 10.0 ng/mL. However, in patients
with PSA levels of 0 to 50 ng/mL, for example, the AUC would be approximately
0.73 for PSA, because PSA is highly predictive of cancer when a wide range
of values is tested.1 This explains in part
why several early studies composed of patients with high PSA levels did not
find that percentage of free PSA was effective. Percentage of free PSA is
highly effective when used in patients with moderately elevated PSA values,
but it cannot improve the considerable specificity of total PSA values in
patients with high PSA concentrations. Factors such as this that may affect
percentage of free PSA studies are discussed in a detailed review article
by Woodrum et al.17
Percentage of free PSA values decreased as total PSA values increased.
This relationship was not sufficiently robust to affect the cutoff in a population
with PSA values of 4.0 to 10.0 ng/mL, but could affect cutoffs if percentage
of free PSA is useful outside of this range, as suggested by recent studies.16,27,28
The percentage of free PSA values increased as patient age increased.
Because of this relationship, age-specific percentage of free PSA cutoffs
could be used, but this approach resulted in more undetected cancers in younger
subjects. In contrast, a single 25% free PSA cutoff across all age groups
resulted in the highest sensitivity (98%) in younger subjects, those most
likely to benefit from early detection. Most of the missed cancers occurred
in older subjects. Clinically, this finding is advantageous because older
subjects (those with less than a 10-year life expectancy) are often not affected
by or treated for prostate cancer. For this reason, we recommend a single
cutoff (25% free PSA) for patients aged 50 to 75 years.
Other studies have also reported a relationship between percentage of
free PSA and age,11,12,16
although some studies composed of only healthy subjects have not found this
A significant positive correlation was found between percentage of free
PSA and prostate volume. This correlation provides insight into how percentage
of free PSA aids in distinguishing prostate cancer from benign prostatic disease.
Percentage of free PSA provides information on gland size and total PSA production
relative to each other. If PSA is elevated and only a small portion of the
total PSA is free PSA, the probability is high that the patient has a small
gland with cancer. If PSA is elevated and a large portion of the total PSA
consists of free PSA, the probability is high that the patient has a large
gland without cancer. Thus, use of a single 25% free PSA cutoff is recommended
regardless of the patient's prostate size. When a recommendation is made not
to perform a biopsy for patients above this cutoff, this is the group with
the lowest risk of cancer and the highest probability of benign prostatic
Thus, the cancers that would be missed, occurring in patients with a
percentage of free PSA value above the 25% cutoff, are found primarily in
older patients with larger glands. These patients also tended to have less
aggressive disease. As the percentage of free PSA increased, the probability
of favorable pathologic findings increased. The natural history of prostate
cancer shows it to be slow-growing; with annual screening, it would be possible
to monitor patients and perform biopsies only in patients with increasing
PSA levels or decreasing percentage of free PSA.30- 32
Our study confirms earlier reports of a relationship between the percentage
of free PSA and adverse pathologic features or cancer aggressiveness.6- 9,14
In contrast, other investigators did not find the percentage of free PSA to
be predictive of pathologic results, possibly because of the wide range of
PSA values in their study populations,33,34
small sample size, or differences in staging techniques or assay systems.
Our data were also analyzed to estimate an individual patient's probability
of having detectable cancer based on the percentage of free PSA (Table 6) and the subject's age (Figure 4). Lower percentages of free PSA
indicated higher risk, and older subjects were at higher risk than younger
subjects. The risk of cancer ranged from 8% (subjects with percentages of
free PSA >25%) to 56% (subjects with percentages of free PSA ≤10%). Thus,
percentage of free PSA may be used to stratify individual patients into low-risk
to high-risk populations to aid in biopsy decisions.
With this risk assessment approach, it may be possible to actually increase
cancer detection (sensitivity). Patients who have undergone 1 biopsy with
negative findings might be advised to undergo a second biopsy if the percentage
of free PSA indicates high risk (approximately 20% of cancers are missed on
the first biopsy specimen).21 Thus, cancers
that otherwise might be missed would be detected.
Our results confirm earlier reports showing a similar gradient of risk
for cancer associated with percentage of free PSA, age, and PSA.10,11,15,16
However, preliminary studies often contained a relatively high percentage
of subjects with cancer (39% to 44%)10,16
that was not adjusted to the known prevalence of detectable prostate cancer
(25%) in patient populations with PSA values between 4.0 and 10.0 ng/mL, producing
inflated risk estimates.
The protocol for the current study required nearly 400 subjects with
cancer and 400 subjects with benign prostatic disease to obtain reliable estimates
of cutoffs, sensitivity, and specificity in the ROC curve analyses. It should
not be interpreted that the 50% proportion of subjects with cancer in this
study means that the population from which they were selected is not representative
of a screening population. Subjects with cancer were generally more difficult
to enroll because they represented only 25% of the population from which they
were selected. However, if enrollment had been limited to this 25% proportion,
an inadequate number of subjects with cancer would have been available for
analyses by variables such as age, total PSA, and prostate volume. By enrolling
nearly 400 patients in each disease group, it was possible to perform both
ROC curve and risk assessment analyses using statistically valid methods;
prevalence can be adjusted downward using the bootstrap method, but reliable
cutoffs cannot be estimated if the sample size is inadequate.
These study results were obtained using the Hybritech Tandem free PSA
and total PSA assays, both of which have been approved by the Food and Drug
Administration for use as an aid in prostate cancer detection. Recent studies
have shown that percentage of free PSA cutoffs and clinical performance differ
when various combinations of free PSA and total PSA assays from different
manufacturers are used.35- 37
Mean percentage of free PSA values from identical serum samples may have a
2-fold increase using different assay combinations.36
We conclude that the use of free PSA measurements can reduce unnecessary
biopsies in patients with PSA levels of 4.0 to 10.0 ng/mL who are undergoing
evaluation for prostate cancer, with a minimal loss of sensitivity in detecting
cancer. These results apply to the Hybritech Tandem PSA assays but may not
apply to assays of other manufacturers.