Download Medically Significant Concentrations of Prostate

CLIN. CHEM. 36/3, 515-518 (1990)
Medically Significant Concentrations of Prostate-Specific Antigen in Serum Assessed
Larry H. Bernstein,’Resser A. Rudolph,3Marguerite
M.
Pinto,’ NicholasVlner,2 and Howard Zuckerman2
We used the method of Rudolph et al. (Clin Chem 1988;
34:2031-8) to find information in the data from correlated
determinations of acid phosphatase (PAP, EC 3.1.3.2; DuPont aca) and prostate-specific antigen (PSA, Hybritech).
We described there how we assign medical decision limitsfor
two or more correlated variables and convert the database to
a binary coded message, allowing separation of a selected
disease class with minimum error. The decision point, analogous to a percentile upper limit on the ordered values of
each variable in the reference group, satisfies the maximum
entropy constraints of reference, producing a minimum entropy for the binary coded patient database. We found
maximum entropy decision points at PAP = 0.75 U/L and
PSA = 22.8 cg/L. Patients with PSA values exceeding 22.8
g/L had no benign prostatic disease except for five patients
with benign prostate hyperplasia (BPH) with adjacent colon
carcinoma (95.3), BPH with infarction (27.6), BPH (23.4
28.1), or acute prostatitis (34.6). We consider PSA exceeding
22.8 gIL as indicative of carcinoma of the prostate, stage C
or 0, in the absence of disconfirming evidence. Another
decision value for PSA is 11.3 zg/L. This bounds the region
between 11.3 and 22.8 /.L9IL, where the frequency of BPH is
1.5 times that for aderiocarcinoma. At PSA <11.3 jzg/L there
is a high frequency of BPH. PSA concentration is not
correlated with prostatic size (mass) or with prostatitis. A
metastatic carcinoma is as likely to be nonprostatic as
prostatic when the PSA concentration is <11.3 g/L.
AdditionalKeyphrases:cutoff values
cancer
prostate
acid phosphatase
Since the description of prostate-specific antigen (PSA), a
prostatic epithelial glycoprotein and kallikrein acting as a
serine protease on the seminal vesical coagulum (1, 2),
several papers have demonstrated its value as a potential
marker in serum of patients with prostatic cancer.4 This
test was found in early studies to have a mean concentration in serum proportional to the clinical stage of carcinoma (3). Also, the measured concentrations of PSA correspond to the disease-free interval after treatment (4).
Moreover, PSA is more sensitive than prostatic acid phosphatase (PAP; EC 3.1.3.2) for detection of recurrent carcinoma and for monitoring therapy, because the concentration of PSA increases before there are clinical signs of
relapse (4-6).
In this study we examined the use of PSA and PAP
measurements
in patients with benign prostatic hyperplasia (BPH), prostatitis, and adenocarcinoma of the prostate,
to determine what antigen concentrations in serum really
are helpful in making therapeutic decisions. Despite the
sensitivity of PSA for finding a high-risk population who
may have prostate cancer, particularly in those undergoing
evaluation for cancer (primary site unknown), the common
occurrence of BPH and prostatitis in the same population
with increased concentrations of PSA makes it necessary to
define the relationship of PSA in serum to the common
diseases of the prostate and adenocarcinoma. Moreover,
because of this sensitivity, the use of disease-specific ranges
for concentrations of PSA in serum does not improve the
use of this test in medical decision making. We present an
alternative
optimization of the reference interval, based on
the segregation of patients with high risk from those with
intermediate or low risk by using the correlation of PSA
and PAP increases with carcinoma of the prostate, with
distant metastasis, by the method of Rudolph et al. (7).
Materials and Methods
Assay method. PSA and PAP were determined as a
prostate test-panel for diagnostic evaluation of hospitalized
patients. PSA was radioimmunoassayed (Hybritech, San
Diego, CA) and reported in nanograms per milliliter. PAP
was assayed enzymatically with the aca analyzer (DuPont
Instruments, Wilmington, DE) and reported in IUB units
(U) per liter.
Population
study. For evaluation
of diagnostic cutoff
values for the respective methods, we used values for
specimens from a sample of the population evaluated during a year without separation of a nondiseased population
for study. The population evaluated consisted of 255 men
over 50 years of age and comprised 304 visits (Table 1).
These patients had prostatitis, BPH, prostatic and nonprostatic carcinoma, urinary obstruction, or back pain with no
symptoms referable to the urinary tract. Table 1 defines
this population with respect to prostate carcinoma by stage
and other diseases. This population was divided into those
with high and low PSA concentrations
after unstageable
disease was removed from consideration.
The low-PSA group consisted of 214 patients, 45 of them
with prostate carcinoma, 20 of whom were in stage C or D.
Of the remaining 169 patients, 96 had BPH averaging 147
g (range 3-215 g); 11 had other benign diseases (two with
bladder calculi); nine had prostatic carcinoma after surgery
(Gleason scores 4-8); and 23 had nonprostatic carcinoma,
nine of them urothelial. Nine of the patients who had stage
Table 1. StatistIcs for Prostate Car clnoma Study
Population
81
Percentof total
68.2
31.8
4
17
1.5
6.7
9
5
3.5
C
D
46
PatIent class
Benign
Departmentsof’ Pathologyand2 Surgery (Divisionof Urology),
Bridgeport Hospital, Bridgeport, CT 06610.
3Department of Pathology,Mercy Hospital, Elwood,IN.
4Nonstandard abbreviations: PSA, prostate-specificantigen;
PAP, prostatic acid phosphatase;and BPH, benign prostatic hyperplasia.
Received August 8, 1989; acceptedDecember26, 1989.
Carcinoma
Al
B1
A2/B2
No.
174
1.9
18.1
Percentof class
4.9
21.0
11.1
6.1
56.8
CLINICAL CHEMISTRY, Vol. 36, No. 3, 1990 515
C or D carcinoma of the prostate (Gleason scores 5-8) had
their PSA concentrations measured after treatment.
The 41 patients in the high-PSA group included five
patients without prostate carcinoma (Table 2). Those patients with prostate cancer were classified by stage and by
Gleason’s score.5 Of 29 patients who had benign disease, 12
had two or more PSA determinations.
Analysis. We used the method of Rudolph et al. (7) to
determine which values for PAP and PSA in serum divide
the population into reference and disease subsets, similar
to the definition of an upper limit of normal. Using only the
population data at hand, we determined the decision limits
for PSA and PAP that optimally classified patients with the
least misclassification of adenocarcinoma when prostatitis
and BPH contribute mostly to misclassification error rates.
We also carried out such standard statistical studies as
ANOVA2,Kruskal-Wallis
analysis, and histogram plots,
using “stsc statgraphics” software (stsc, Rockville, MD).
Group-based
reference. The optimum decision levels for
PSA and PAP are measured by using Shannon entropy. The
basic measure of missing information
is as described (3):
H
=
Table 2. Falsely Positive Serum Test Results
PSA, nglmL
PAP, U/L
Diagnosis
6.72
0.7
95.3
28.1
23.4
27.6
35.5
BPH
BPH
BPH
BPH
Prostatitis
3.1
14.69
0.87
a
P1 log2(l/P,) + P2 log2 (11P2)#{149}#{149}#{149}
+ P, log2 (1fP)
0.7
11.3
PSA
where H is entropy (the average information of a message
set), and P is the probability of a choice from a set of
messages. The results referred to these decision values
(cutoff limits) were converted to a binary code. We used the
Bernoulli trial to determine maximum entropy reference. In
the reference or normal population, the entropy of all possible binary patterns approaches 2 bits as the number of
observations increases (7). Because no information
is present in this population, the measured uncertainty is maximum and the probability distribution is flat. The probability
distribution of the binary patterns is not flat when there is
effective information in the data, which results in a drop in
entropy. We create a flat probability distribution by randomly shuffling the variables and destroying the information in the data. The distribution is not flat if information is
present because of correlation in the data. The frequency
distributions we refer to are an all-negative pattern, 00, an
all-positive pattern, 11, and intermediate patterns, 01 and
10. The difference between the measured entropy of the
database and the randomization maximum entropy reflects
the effective information present. This difference, an effective information maximum at PSA = 22.8 ng/mL, is plotted
in Figure 1.
The value for any variable at which this difference is
greatest is the optimum decision point for the variable used
in the classification. The decision point is analogous to a
percentile upper limit on the ordered values of each variable
in the normal reference group, and the values associated
with these points serve the same function as traditional
StageAl is the finding of no palpabletumor (Al is the finding
of carcinomain three or fewer microscopicfields).Stage Bi is the
finding of a palpable nodule with no extensionand no symptoms.
(Stages A2 and B2 have diffuse or multinodular involvement
without extension beyond prostate.) Stage C is invasion of the
capsulewith or without involvement of seminal vesicles.Stage D
is local extensionwith pelvic nodeinvolvementand symptoms or
distant metastasis. The PSA range was 18.1 to 1340 ng/mL for
high-PSA subjectswith stage D carcinoma.
Gleason’sscorewas usedin assessingthe grade of the tumor by
pathological examination. A Gleason’sscore >7 is expected to
occurwith stage C or D carcinoma-as we found in this study.
516 CLINICAL CHEMISTRY, Vol. 36, No. 3, 1990
228
54.6
&.‘,izO
Fig. 1. Informationcontent(bits)vs PSA concentration
The arrow indicatesthe PSA concentrationgMng maximuminformation
normal
limit values. In using the Bernoulli test we calculate the entropy and the relative frequencies of the binary
patterns produced. The decision levels so obtained satisfy
the maximum
entropy constraints of reference and also
produce a minimum
entropy for the binary coded patient
database.
Results
The method to which we refer (7) does not depend on
a normal range for a low-prevalence population. It allows us to define and differentiate patients with
carcinoma in a disease population, when in nondisease PSA
values are <3.5 nglmL, but there is a large disease population with nonmalignant
diagnoses at higher values for
PSA. To eliminate this problem, we use the mutual dependency of the PSA and PAP to effectively
identify
the
malignant disease population, using the entire population
for evaluation.
The PAP had a single decision value, 0.75 U/L. The PSA
has two major decision values, 54.6 and 22.8 ng/mL, and
additional information peaks at 11.3 and 3.5 ng/mL.
The decision value at PSA 22.8 ng/mL mainly separates
adenocarcinoma of the prostate-stages
B, C, and D-from
a subpopulation with a 60.6% frequency of BPH and prostatitis. At 54.6 ng/mL, we eliminated
four of five patients
with benign PSA increases within the population >22.8
nglmL who are selected for prostatic adenocarcinoma (Table 2). A PSA value of 11.3 to 22.8 ng/mL may be equivocal.
It is not possible to distinguish carcinoma of prostate from
other diagnoses in this range by use of any PSA value. The
median PSA value for 13 patients with other disease is 14.4
ng/mL.
In addition, we reviewed 94 specimens from patients in
the benign group with disease other than carcinoma of the
prostate and found that 71 (75.5%) had PSA values <10
ng/mL and, of 17 cases of nonprostatic carcinoma, all had
PSA <5 ng/mL. We used a decision value of 22.8 ng/mL,
which defines the upper confidence limit for distinguishing
establishing
the disease population of interest in the same way as upper
limits of normal, to cluster the population with a PAP of
0.75 UIL as a covariant test. We did this by using the
binary equivalent assignments, 0 or 1, to each of the paired
values of PSA and PAP, based on whether the assigned
decision value is exceeded or not, and examining the
frequency distribution of carcinoma against the classification formed. The result was the pattern shown in Table 3.
Only one of 13 patients with above-normal PAP without
increased PSA had adenocarcinoma of the prostate. Consequently, a value for PSA >22.8 ng/mL, with or without
increased PAP, is required to identify significant risk of
adenocarcinoma of the prostate, and at such values it
appears usually to be associated with stage C or D disease.
Table 4, a reorganized Table 3, defines the occurrence of
binary class patterns
with respect to carcinoma of the
prostate by stage compared with nondisease. The table
allows the estimation of 87.8% sensitivity, 79% specificity,
a negative predictive value of 97.1%, and a positive predictive value of 44.4% for PSA >22.8 ng/mL.
Tables 5 and 6 list the PAP and PSA statistics for
positive and for negative tests by patient class. Of the 20
falsely negative PSA assays, all were stage C or D. The
median of PSA values was <7 nglmL (range from 0.19 to
19.6 ng/mL), and at least nine were post-treatment values.
We subsequently removed the patients with stages Al,
A2, or B adenocarcinoma or in remission with normal PSA
and PAP values and recombined these with the other three
patterns to make a population of 133 patients, none of
whom was identified as part of the traditional normal
population, and we studied this subpopulation to examine
within-disease reference intervals. Each patient was classified according to the following staging criteria: 1-nondisease or stage Al adenocarcinoma (40); 2-stage Bi (19);
3-stage A2/B2 (10); 4-stage C (2); 5-stage D (62). In
addition, each patient was classified according to the following criteria: 1-PSA 22.8 ng/mL (80); 2-PSA >22.8
ng/mL (53). Analyses of variance of PSA and KruskalWallis analysis by ranks for stage and for class were
significant (P <0.0001). Discriminant analysis for class by
PSA and stage was significant (P <0.00001), with eigenvalue, 0.4307; canonical correlation, 0.5487; Wilks
Lambda, 0.6990; and chi-square,
46.558. Figure 2 shows
the means and confidence levels for PSA classification
of
the two groups by stage and by class, respectively. The PAP
test could not separate the groups by stage. PSA sufficed for
classifying patients with or without stage as a cofactor, but
the means and confidence intervals for class 1 were somewhat less than for stage 1 because of patients in class 1 who
belonged to higher stages. This represents patients who
were treated with stable remission or nonprogression.
Discussion
Here we have used the method of Rudolph et al. (7) to find
information in the data based on the mutual dependence of
Table 4. DistrIbution of Cases In Binary Classes
Patientclass
Benign
Malignant
AliBi
A2/B2
C
0
00
01
10
11
157 (90.2)b
1(0.57)
12(6.9)
43 (53.1)
8(9.9)
2(2.5)
16 (76.2)
1 (4.8)
1 (4.8)
8(88.9)
1(11.1)
4(80)
15(32.6)
6(13)
1(2.2)
a PAP/PSA.bPercent of classin parentheses.
-
-
4(2.3)
28 (34.6)
3 (14.3)
1(20)
24 (52.2)
Table 5. StatistIcs for Positive Prostate Markers by
DIsease Categories
PAP(+)
PatIent class
PSA(+)
No.
No.
Benign
16
9.2
5
2.9
Carcinoma
30
37
36
43.2
Al/Bi
4
19
4
A2/B2
C
O
0
0
1
1
20
1
19
11.1
20
30
65.2
25
54.3
Table 6. StatistIcs for Negative Prostate Markers, by
Disease Categories
PAP
PatIent class
Benign
Carcinoma
PSA (-)
(-)
No.
158
No.
90.8
169
97.1
51
17
53
81
45
17
55.5
81
C
9
4
100
80
8
4
o
21
88.9
80
34.8
Al/Bi
A2/B2
45.6
-
16
Class
350
280
E
-
210
C
U
0
140
0.
70
0
2
3
Ciassification
4
5
Vector
Fig. 2. Means and confidenceintervalsfor PSA by disease stage
(solid bars) and by classassignment(cross-hatched bars) basedon
entropydecisionvalues
Table 3. Frequency Distribution of Binary Patterns
Pattern
Frequency
00
10
230
01
11
13
12
46
ProbabIlIty
0.7616
0.0397
0.0464
0.1523
Information,
bits
0.2992
0.4841
0.6896
1.1031
the PAP and PSA values for serum. The PAP maximum
entropy decision point at 0.75 U/L and PSA at 22.8 ng/mL
were used to separate patients according to the finding of
adenocarcinoma and stage of disease, usually greater than
B. When PSA exceeded 22.8 ng/mL, there usually was no
benign prostatic disease, except for five patients with BPH
CLINICAL CHEMISTRY, Vol. 36, No. 3, 1990 517
or prostatitis. The distinction was almost absolute for PSA
>54.6 ng/mL. The maximum entropy decision point at 54.6
ng/mL was consistent with studies showing mean PSA
concentrations of 223 ng/mL with confidence intervals from
141 to 305 ng/mL for stages B, C, or D adenocarcinoma.
This is consistent with findings of stage C and D adenocarcinoma having PSA >85 ng/mL (8,9).
Another decision point occurs at 11.3 ng/mL. This defines
the equivocal region between 11.3 and 22.8 ng/mL, where
the incidence of carcinoma of the prostate is 39%. At <11.3
ng/mL there is a high frequency of BPH, and when PSA is
<3.5 ng/mL there usually is no disease. A PSA value <10
ng/mL has no information. Further, at this concentration a
metastatic carcinoma is as likely to be nonprostatic as
prostatic. These studies establish the importance of a PSA
value >22.8 ng/mL as a basis for careful evaluation for
adenocarcinoma of the prostate. Further, these studies reinforce the findings of Seamonds et al. (8) suggesting the
sensitivity of PSA values >11.3 ng/mL, despite the inconclusiveness of such a finding because of the high frequency of
BPH and prostatitis. Another diagnostic test in combination
with PSA other than PAP would be necessary to relieve the
uncertainty
when PSA is between 11.3 and 22.8 ng/mL.
518 CLINICAL CHEMISTRY, Vol. 36, No. 3, 1990
References
1. Wang MC, Valenzuela LA, Murphy GP, Chu TM. Purification
of a human prostate specific antigen. Invest Urol 1979;17:159-63.
2. Lilja H. A kallikrein-like serine protease in prostatic fluid
cleavesthe predominant seminal vesical protein. J Clin Invest
1985;76:1899-903.
3. Kuriyszna M, Wang MC, Papsidero LI), et al. Quantitation of
prostate specific antigen in serum by a sensitive enzyme immunoassay. Cancer Res 1980;40:4658-62.
4. Killian CS, Yang N, Emrich U, et al. Prognostic importance of
prostate specific antigen for monitoring patients with stages B2 to
Dl prostate cancer. Cancer Res 1985;45:886-91.
5. Kuriyama M, Wang MC, Lee CI, et al. Use of prostate specific
antigen in monitoring prostate cancer. Cancer Res 1981;41:38746.
6. Killian CS, Erorich U, Vargas FP, eta!. Relative reliability of
five serially measured markers for prognosis of progression in
prostate cancer. J Natl Cancer Inst 1986;76:179-85.
7. Rudolph RA, Bernstein LH, Babb J. Information induction for
predicting acute myocardialinfarction. Clin Chem 1988;34:20318.
8. SesmondsB, Yang N, Anderson K, et al. Evaluation of prostate
specific antigen and prostatic acid phosphatase as prostate cancer
markers. Urology 1986;28:472-9.
9. Stainey TA, Yang N, Hay AR, eta!. Prostate-specific antigen as
a serummarker for adenocarcinoma of the prostate. N Engl J Med
1987;317:909-16.