The accuracy of any test is usually described using the ideas of 'sensitivity' and 'specificity'. However, the predictive value of the test - which is what most people would mean by its 'accuracy' - also depends on how common or rare HIV infection is in the population that is being tested.

Sensitivity

Sensitivity is the ability to detect all true positive results. This is most important for an initial screening test and it is where basic ELISA tests are strongest.

A basic ELISA test can be very sensitive, meaning that if antibodies are present, it will almost certainly detect them. It is also cheap to carry out on large numbers of samples, which makes it very suitable for initial screening. Commercial tests typically include a range of antigens to cover HIV-1 (group M) and HIV-2 and, increasingly, HIV-1 group O as well.

WHO currently recommends (in 2001) that ELISA kits should not be used unless they can detect 100% of known positive samples in a reference panel of 203 HIV-1 and/or 60 HIV-2 samples. For Rapid Tests, they require 99 per cent sensitivity, reflecting the different use to which they tend to be put (telling individuals their status, rather than screening blood supplies).

The main limitation on the sensitivity of antibody tests is that it takes the body a while to mount an immune response to any infection.  In the first few weeks, the antibodies may be at low levels in the blood and may not bind as well as antibodies produced later in the course of infection. These limitations have been partly overcome by improving the antigens used in HIV ELISA tests, but will always remain.

People who are infected with HIV will mostly test positive on modern antibody tests within six weeks of infection, some in as little as one week, though the possibility of window periods of up to three months or, exceptionally, even longer, remains.

Specificity

Specificity is about the number of negative sera correctly identified as negative. WHO does not recommend the use of tests with a specificity of less than 95 per cent when tested on a panel of African sera, but in practice most commercial tests are much better than this. One example is a test described as '98.8 per cent specific'.

What this means is that 1.2 per cent of samples - or twelve in every thousand - will test positive when they are not.

Such a test must be used with another test, using different antigens, to confirm it. That second test need not be any more specific (though it helps if it is) to reduce the chance of a double-false-positive result to a vanishingly small level.

The most common way to do this has been to use a Western Blot test or an IFA test. Another way is to use other ELISA tests based on different antigens or to use a competitive ELISA test.

Predictive value

If the test mentioned in the previous example were used in a population such as UK blood donors - where the level of HIV infection has been around 1 in 100,000 donations - then, for every true positive result, there would be 1.2/0.01 = 1,200 false positive results.

Consider, however, the opposite case where the true level of HIV in a population is one in three. Here, there will be 33.3/1.2 = approximately 28 true positive cases for every false positive test.

It is important to confirm the test result in both cases, but if the people being tested are, for example, women about to give birth who have not received previous antenatal care, then in the second case one might not wait for confirmation before deciding to offer antiretroviral treatment to prevent HIV transmission to their babies. In the first case, all 1,200 donations would be kept out of the blood supply used in hospitals.

This sensitivity has a price - giving some false positive results. In the older kinds of test, as we have seen, these could happen because non-HIV proteins were included in the system. In the newer systems there is still a trade-off. Tests that are too precise can miss genuine HIV antibodies. This would, for example, be a disaster for a blood transfusion service. In practice, other tests are needed to confirm apparently positive results.