Tuesday, July 06, 2004

HIV/AIDS: revised prevalance rates

UPDATE: The big announcement is that 5 million people were newly infected with HIV last year ... which UNAIDS says is the highest one-year increase since the beginning of the epidemic. Less publicized has been the news that the prevalence rates have been revised downwards ... UNAIDS says that about 38 million people are infected. Until now, experts had estimated that about 40 million were infected. From the UNAIDS report ...
UNAIDS and WHO have revised their global estimates of the number of adults living with HIV, particularly in the sub-Saharan region. These new estimates are the result of more accurate data from country surveillance, additional information from household surveys, and steady improvements in the modelling methodology used by UNAIDS, WHO and their partners. This has led to lower global HIV estimates for 2003, as well as for previous years. Although the global estimates are lower, this does not mean the AIDS epidemic is easing off or being reversed. The epidemic continues to expand. (source)
I have been scanning the report and it doesn't seem to show how the numbers changed! I'm going with the Associated Press' report that estimates are down 2 million.

I wrote the following (about what might have brought about this revision) 12 hours ago ...

----

I caught the end of a BBC World Service report about (I think) how UNAIDS is changing the way it calculates HIV/AIDS prevalence rates.

An update of the country estimates, published biannually by UNAIDS and WHO, is due to be released sometime this week. I believe I heard the BBC reporter say that the numbers are going to be noticably (or significantly) lower because of the change in methodology.

While I haven't found any published stories like the one I heard on the radio, a few minutes of rooting around the UNAIDS website led me to this, the UN's "Epidemiology Reference Group".

In May 2004, the group released a report titled: "Technical report and recommendations: Methods and software to produce HIV/AIDS estimates in the era of population-based prevalence surveys." I think that means ... how can we make sure our estimates are as close to reality as possible.

The paper itself is slightly, but only slightly, less jargon-y than the title. And here is what I got out of it ...

This is uber simplified ... to calculate HIV prevalence rates, researchers test those pregnant women who show up at health clinics -- antenatal clinic attenders (ANC) -- and based on the number of women who test HIV +ve, researchers extrapolate the HIV rate for the whole population.

There are many problems with this method ...
[...] rural populations are poorly covered by ANC-based surveillance systems and the bias tends to be in one direction - overestimation. Remote rural populations have a lower probability of being included in surveillance systems and generally such populations have lower HIV prevalence than less remote rural populations.

It is likely that urban ANC clinics provide a better picture of prevalence among all pregnant women in urban areas. However, the location of the urban sites may not be representative of the whole urban population (e.g. in low income area). Also, urban
ANC attendees may be less representative of the population living in the catchment area of the clinic, than rural ANC attendees are, because more women attend private health care in urban areas compared to rural. [Pg. 5]
And a couple of other problems with how surveys are done ...
Refusal to participate and absence from population based surveys may significantly bias estimates of prevalence in the population. In some but not all cases this appears to be confounded with movement of individuals. [Pg. 6]

[...] In some countries problems with the specificity of HIV test kits in field conditions has led to overestimates of HIV prevalence. Quality control programmes should aim to pick up on these problems. [Pg. 7]
The paper recommends ways to correct for these problems ... and I read through them but I'm not confident in my ability to summarize them. If you're interested, read the report for yourself!