Damned statistics.
David Canzi
Bio/technology:
The test recommended is ELISA(29), which cannot be considered
specific. In Russia, in 1990, out of 20,000 positive screening
tests 'only 112 were confirmed' using the WB as a gold standard. In
1991, of approximately 30,000 positive screening tests, only 66
were confirmed(30).
These statistics make ELISA look really bad, but some important
information is missing: How many people were tested using ELISA in
order to get 20000 positives in 1990 and 30000 positives in 1991?
I looked up the reference for these numbers, a letter to the Lancet by
Alexander Voevodin. What I found was this: In 1990, of 20.2 million
ELISA tests done, 112 were confirmed positive, and 20,000 were false
positives. In 1991, of 29.4 million ELISA tests, 66 were confirmed
positive and 30,000 were false positives.
In both 1990 and 1991, a large number of Russians were tested using
ELISA, and only 1/1000th of those tested had false positives. If you
use a test even this accurate on a population where only about 1 out
200000 people have HIV, you would expect about 200 false positives for
every true positive. Voevodin notes this phenomenon at the start of
his second paragraph: "Such huge numbers of false positives are
predictable for mass screening for HIV in low prevalence populations,
..."
All the relevant numbers needed to judge the accuracy of ELISA were
there in Voevodin's letter, and the figures show that ELISA is very
accurate. Somehow, with the real truth staring them in the face,
Papadopulos-Eleopulos and her coauthors managed to omit from their
paper just the figures that had to be omitted so that a statistical
fallacy could drastically mislead their audience about the accuracy of
ELISA.
Reference:
Voevodin, A. 1992. HIV screening in Russia. Lancet 339:1548.
--
David Canzi
KENNETH%BR...@pucc.princeton.edu
>(stuff deleted)
>These statistics make ELISA look really bad, but some important
>information is missing: How many people were tested using ELISA in
>order to get 20000 positives in 1990 and 30000 positives in 1991?
>All the relevant numbers needed to judge the accuracy of ELISA were
>there in Voevodin's letter, and the figures show that ELISA is very
>accurate. Somehow, with the real truth staring them in the face,
>Papadopulos-Eleopulos and her coauthors managed to omit from their
>paper just the figures that had to be omitted so that a statistical
>fallacy could drastically mislead their audience about the accuracy of
>ELISA.
>
low prevalence situations (which are by far the most common in epidemiology)
is _not_ what one would expect from its specificity, no matter how good it
is. That's why confirmatory exams are needed in almost every conceivable
situation, not just AIDS.
It also shows that medical technology has its limitations too.
Yours,
Ken
David Canzi
>in Bio/technology:
>The test recommended is ELISA(29), which cannot be considered
>specific. In Russia, in 1990, out of 20,000 positive screening
>tests 'only 112 were confirmed' using the WB as a gold
>standard. In 1991, of approximately 30,000 positive screening
>tests, only 66 were confirmed(30).
>These statistics make ELISA look really bad...
These statistics demonstrate how bad ELISA testing is when
it is used as a DIAGNOSTIC tool. It is only useful as a blood
supply SCREENING test because it errs tremendously on the
side of safety by allowing a high number of false positives
in proportion to the number of actual positives.To tell an
individual that he or she is infected with HIV based on a
positive ELISA test is criminal. According to this data, the
odds AGAINST a person being Western Blot positive following
a positive ELISA test are at least 179 to 1 and can be as
high as 455 to 1. If that person tested WB+ ,it still would
not be proof of HIV infection.
Many countries still, however, use only an ELISA test for
determining if an individual is HIV antibody positive.
>I looked up the reference for these numbers, a letter to
>the Lancet by Alexander Voevodin. What I found was this:
>In 1990, of 20.2 million ELISA tests done, 112 were confirmed
>positive, and 20,000 were false positives. In 1991, of 29.4
>million ELISA tests, 66 were confirmed positive and 30,000
>were false positives.
>In both 1990 and 1991, a large number of Russians were tested
>using ELISA, and only 1/1000th of those tested had false
>positives.
Do you think giving a death sentence to 50,000 people is
acceptable when only 178 might be at risk?
>...the figures show that ELISA is very accurate.
If we were talking about inanimate objects perhaps-but we
are talking about people. Since the HIV/AIDS hypothesis is
based only on a weak correlation, and since many countries
only use ELISA testing which gives a high number of false
positives compared to possibly true positives, that correlation
disintegrates. Inaccurate testing is just one of the many
dangers of the HIV/AIDS hypothesis.
David Canzi
NO! I DID NOT WRITE THAT.
--- - --- --- ----- -----
Somehow, somebody's response to an article of mine has been mistakenly
posted under my name. And, frankly, I greatly resent having my name
associated with the content of that article.
I think I can guess the real author. So,
In article <14...@sci.med.aids>, John Lauritsen probably wrote:
>>Consider this excerpt from an article by Papadopulos-Eleopulos
>>in Bio/technology:
>
>>The test recommended is ELISA(29), which cannot be considered
>>specific. In Russia, in 1990, out of 20,000 positive screening
>>tests 'only 112 were confirmed' using the WB as a gold
>>standard. In 1991, of approximately 30,000 positive screening
>>tests, only 66 were confirmed(30).
>
>>These statistics make ELISA look really bad...
>
>These statistics demonstrate how bad ELISA testing is when
>it is used as a DIAGNOSTIC tool. It is only useful as a blood
>supply SCREENING test because it errs tremendously on the
>side of safety by allowing a high number of false positives
>in proportion to the number of actual positives.
>in proportion to the number of actual positives.To tell an
>individual that he or she is infected with HIV based on a
>positive ELISA test is criminal. According to this data, the
>odds AGAINST a person being Western Blot positive following
>a positive ELISA test are at least 179 to 1 and can be as
>high as 455 to 1.
This is a crock. 179 to 1 is not a lower limit, nor is 455 to 1 an
upper limit. These are just, I assume (without checking your
arithmetic), the odds from the two sets of tests Voevodin reported,
which were done on populations where the prevalence of HIV is very
low.
When the figures Voevodin supplied that Papadopulos-Eleopulos et al
omitted are taken into account, they show ELISA falsely positive for
only 1 out of 1000 uninfected people. Consider applying this to
different populations:
(1) If you apply ELISA to a population that is 0.01% infected,
about 0.11% of that population tests positive, and there
are about 10 false positives for each true positive.
(2) If you apply ELISA to a population that is 0.1% infected,
about 0.2% test positive, and there is one false positive
per true positive.
(3) If you apply ELISA to a population that is 1% infected,
about 1.1% test positive, and there is one false positive
for every 10 true positives.
(4) If you apply ELISA to a population that is 10% infected,
about 10.1% test positive, and there is one false positive
for every 100 true positives.
Do you see the pattern here? At higher levels of prevalence, the false
positives become a small fraction of the set of positive tests. And at
high levels of prevalence, the number of people testing positive is
almost the same as the number who are actually infected.
Now I understand HIV has a fairly high prevalence in Africa. In
addition, if ELISA is used as part of an African AIDS definition, I
suspect it would be applied mainly to people who already have clinical
symptoms of immune deficiency, and so the population ELISA is applied
to would have a higher prevalence of HIV than the general African
population.
>>In both 1990 and 1991, a large number of Russians were tested
>>using ELISA, and only 1/1000th of those tested had false
>>positives.
>
>Do you think giving a death sentence to 50,000 people is
>acceptable when only 178 might be at risk?
When you don't have logic and evidence on your side, you use righteous
indignation, eh?
There is no rational reason to expect the results of applying ELISA to
selected sick Africans to resemble the results of using ELISA on random
healthy Russians.
>>...the figures show that ELISA is very accurate.
>
>If we were talking about inanimate objects perhaps-but we
>are talking about people.
Fine-sounding, but empty, rhetoric. We use the same laws of
mathematics for counting people as we use for counting apples.
I'm on vacation as of, well, some hours ago, but I couldn't leave
without doing something about having garbage posted under my name.
See you in July.
--
David Canzi