I was thinking about these HIV figures from South Africa, so
I wondered if there might not be a way to quantify the real
number.
So let's give it a try. Let's assume there are four components,
the actual number (the outcome), the assumed prevalence,
representativeness, and the use of a single ELISA versus the
use of multiple ELISAs and Western Blot, which was how the
orignal 99.9% accuracy was arrived at.
assumed prevalence = 25% ; that's 25% of the entire population,
a number based on tests of pregnant women at antenatal clinics
representativeness factor = 11.4/19 ; we know that the HSRC did a
representative study, which led to the estimation of 11.4% when
the original estimation of national HIV infection was 19%, which was
based on an unrepresentative sample.
11.4/19 = 0.6
single elisa test's effect = (20+66)/2 ; journalist Rian Malan did
research and found studies where original single elisas, when
checked against multiple elisas and western blot testing, were
as low as 20% true postitives and as high as 66% true postitives.
If we take the average of that, you get (20+66)/2 or 44% or a
factor of 0.44.
Real infection rate = 25% x 0.6 x 0.44 = 6.6%
The HSRC study described:
http://www.independent.co.za/index.p...3345189C308684
Rian Malan's article for Rolling Stone:
http://www.whatisaids.com/rollingstone.htm
Alex
PS, as the elisa's effectiveness was viewed against western blot, it is possible
that the actual number of HIV infections is even lower, as there are several
pathogens that can cause a false positive result in western blot.
