Study highlights insensitivity of PSA screening
BMJ 2005; 331 doi: https://doi.org/10.1136/bmj.331.7508.67-a (Published 07 July 2005) Cite this as: BMJ 2005;331:67All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Ian Thompson concludes from his study (JAMA 2005;294: 66-70) that
“there is no cutpoint of PSA with simultaneous high sensitivity and high
specificity for monitoring healthy men for prostate cancer, but rather a
continuum of prostate cancer risk at all values of PSA”. However, one
might be interested in the question of whether a particular PSA value
supports a clinical suspicion of having prostate cancer or speaks against
it.
With decreasing PSA values not only sensitivity increases but
specificity decreases and there is a point where we don’t gain anymore
positive support for the probability of prostate cancer. This is where the
likelihood ratio (LR) changes from >1 to <1. This point is where the
slope of the tangent of the ROC curve equals 1, i.e. at PSA levels around
2 ng/mL, as can be deduced from the ROC of the study.
According to Bayes’ theorem the post-test odds are equal to the pre-
test odds multiplied by the LR of a diagnostic test. The LR is the ratio
of the likelihood of a test outcome given diseased cases over the
likelihood of the test outcome given non-diseased cases. It can easily be
calculated from the sensitivity and specificity of a test: LR for positive
results = sensitivity / (1-specificity) and LR for negative results = (1-
sensitivity) / specificity. In quantitative diagnostic tests, the LR can
be calculated for each level or range of test results. In a ROC curve, it
equals the slope of the tangent at the point to the curve that corresponds
to a cutpoint identical to the quantitative test result (Choi BC,
Am.J.Epidemiol. 1998;148:1127-32).
In healthy men, there is a certain pre-test probability for prostate
cancer, depending mainly on the age. By screening with PSA tests one likes
to get information that modifies this probability. According to the data
of Thompson’s study, PSA values >2 ng/mL increase the probability and
PSA values <2 ng/mL lower the probability. As far as can be estimated
from the data published in the study, PSA values of between 5 to 10 ng/mL
increase the pre-test odds of prostate cancer vs. no prostate cancer by a
factor of about 4 (= slope of the ROC curve at the lower left end).
Competing interests:
None declared
Competing interests: No competing interests
The study by Thompson et al. provides evidence of a large number of
prevalent prostate cancer which are not detectd by PSA elevation. This is
of no surprise. We know from autopsy studies the prevalence of prostate
cancer in men over 50 to be 30% or higher, whereas the better detection
rates reported in aggressive screening experiences reach at most 4% at
first round. Moreover, we know that screening leaves behind a substantial
number of cancers as, in spite of its very high detection rate at
prevalent screening, cancer detection rate remain high (often "as" high)
at repeat screening.
But we also know that a substantial proportion (at least 50%) of prevalent
cancers in the male prostate, including those detected at screening, are
actually insignificant ("latent", or "ovediagnosed" if detected at
screening), that is not bound to surface as clinical entities, nor to
threaten life, over lifetime. Thus we should make a distinction between
"latent" and potentially life threatening "aggressive" prostate cancers.
Assuming both as PSA false negatives if they are associated to low PSA
levels is not fully correct: screening is aimed at reducing prostate
cancer mortality by detecting cancer which would kill the patient unless
early detected, not at detecting "latent" cancers, which are irrelevant as
far as mortality is concerned: the latter would be better ignored by
screening, as their detection causes only unnecessary anxiety and the side
effects of unnecessary treatment.
Unfortunately we are not able to identify "overdiagnosed" cancers, which
might be even of high grade and/or locally advanced, as its irrelevance to
mortality might be due to intercurrent death from other causes, not only
to its growth speed.
The fact that we may miss a large number of prostate cancers in the low
PSA range is not an argument in favour or against prostate cancer
screening. Pap smear misses at least 50% of high grade dysplasias, FOBT
misses the majority of adenomas, mammography misses at least one in four
cancers, but they all have a substantial impact on specific cancer
mortality.
We know for sure from screening practice that we are detecting prostate
cancer at least 10 years in advance with respect to its symptomatic onset,
and this is a sufficient reason to further explore screening efficacy. The
question whether such a huge diagnostic anticipation will translate into
mortality reduction, or just into undesired knowledge of irrelevant non-
aggressive cancers, will be answered by ongoing randomized trials.
Mortality has decreased since the "opportunistic" screening era in
population statistics, but it is almost impossible to disentagle the
effect possibly due to screening and that certainly due to improved
(medical) treatment. Moreover, since the average detection lead time is
ten years or more, any effect on mortality figures is expected more than a
decade after the peak of incidence due to screening detection.
I do not agree that the study by Thompson et al. adds much to prove or
disprove screening efficacy, though I certainly agree that it confirms the
need for great caution, and the substantial unethicity of opportunistic
screening in absence of any reliable scientific evidence of a favourable
cost-benefit balance.
Competing interests:
None declared
Competing interests: No competing interests
Multi-level Likelihood Ratios and Post-Test Probabilities for Prostate Cancer According to Levels of Prostate-Specific Antigen
In reference to the article by Thompson et.
al.,1 reporting the operating characteristics of the
Prostate-Specific Antigen (PSA) in the placebo group of the Prostate Cancer
Prevention Trial, I would like to present multi-level likelihood ratios (LR)2
and post-test probabilities for a range of pre-test probabilities. These were
calculated from the published report (Table 3) for any cancer (n = 1225) versus
no cancer (n = 4362). This will augment the report by informing on the decrease
in diagnostic uncertainty of prostate cancer if PSA is performed on healthy
males over 54 years with initial PSA values of 3.0 ng/ml or lower.
Multi-level Likelihood Ratios and Post-Test Probabilities
for Prostate Cancer According to Levels of Prostate-Specific Antigen (PSA)
PSA (ng/ml)
Prostate Cancer
LR
Post-Test
Probabilities for Prostate Cancer Given Pre-Test Probabilities
Present
N (%)
Absent
N (%)
10%
25%
50%
75%
90%
≤ 2.0
581 (47.4)
3162 (72.5)
0.65
6.78
17.90
39.55
66.25
85.48
2.1 - 4.0
393 (32.1)
930 (21.3)
1.51
14.41
33.55
60.23
81.96
93.17
4.1 - 6.0
195 (15.9)
205 (4.7)
3.39
27.34
53.03
77.21
91.04
96.82
6.1 - 8.0
35 (2.9)
39 (0.89)
3.20
26.20
51.58
76.17
90.55
96.64
> 8.0
21 (1.7)
26 (0.59)
2.88
24.22
49.95
74.20
89.61
96.28
The multi-level likelihood ratio is less than 1 only for
values of PSA ≤ 2.0 ng/ml.
It is greater than 1 in all other categories with the highest being 3.39 for
values between 4.1 and 6.0 ng/ml. However, neither is
it low enough at values of PSA ≤ 2.0 ng/ml nor high enough at values great than 2.0 ng/ml to change the pre-test probability in the risk of
prostate cancer appreciably.
Of note is that the study sample consisted of health males
over the age of 54 years (mean 62 years) with initial PSA values of ≤ 3.0
ng/ml. The
proportion of self-reported Afro-Americans, a group with a greater risk of
prostate cancer was only 0.2% (n=11). Also, the above re-presentation of data
does not include those without prostate biopsy (n = 2988, 34.8%) in the placebo
group (n = 8575). However, those who underwent biopsy were more likely to be
older whites with a family history of prostate cancer, with PSA values > 4.0
ng/ml and an abnormal digital rectal examination.
Presentation of data in the form of multilevel likelihood
ratios and post-test probabilities will inform clinical decision making. PSA is
a commonly performed test and is frequently used by clinicians along with
digital rectal examination to decrease the diagnostic uncertainty of prostate
cancer. However, as seen in the above
re-presentation, little, if any is added to the pre-clinical probability of
prostate cancer by testing for PSA alone, on healthy middle-aged males.
As highlighted by the authors, great efforts should be
undertaken to inform clinicians and public about the absence of a “normal” PSA.
Especially worrisome is the false re-assurance that could be imparted by a low
value of PSA. The change in pre-test probability of prostate cancer is minimal
even at extremes of PSA.
Research efforts should focus on identifying better
biomarkers for prostate cancer.
References:
1.
Thompson IM, Ankerst DP, Chi C, et. al.,Operating Characteristics
of Prostate-Specific Antigen in Men With an Initial
PSA Level of 3.0 ng/mL or Lower. JAMA. 2005;294:66-70.
2. Sackett Philadelphia:
DL, Straus SE, Richardson WS et.al., Diagnosis and Screening, In Evidence-Based Medicine: How to Practice and Teach EBM (2nd ed.,
pp 76-80).
Churchill Livingstone.
Competing interests:
None declared
Competing interests: In reference to the article by Thompson