UNDER STRICT EMBARGO UNTIL 6 JULY 2004, 1100 GMT
Understanding the latest estimates of the global
AIDS epidemic -- July 2004
New AIDS Estimates & Trends
? At the global level, the number of people living with HIV continues to grow - from 35
million in 2001 to 38 million in 2003. Just under 5 million people became infected
with HIV in 2003, more than any year before.
? Despite the fact that the new global estimate is lower than the previously published
estimate of 42 million in 2002, the actual number of people living with HIV has not
decreased, rather the epidemic continues to grow – due to new infections.
? The current estimates are more accurate due to the availability of more
comprehensive country surveillance data and improved methods for estimating HIV
? For the first time, UNAIDS is publishing revised HIV prevalence rates for previous
years, allowing for a better understanding of the epidemic‘s trends. Specifically, this
report features HIV estimates for 2001 and 2003.
? With 60% of the world‘s population, Asia is now home to some of the fastest-growing
AIDS epidemics in the world. As of end 2003, an estimated 7.4 million people are
living with HIV in the region compared to 6.6 million in 2001. This is primarily due to
sharp increases in HIV infections in China, Indonesia and Viet Nam, which together
make up close to 50% of Asia‘s population.
? As of end 2003, in sub-Saharan Africa, an estimated 25 million people were living
with HIV compared with just under 24 million in 2001. Of the nearly 3 million AIDS
deaths globally in 2003, 2.2 million (or 75%) were in sub-Saharan Africa.
? Although HIV prevalence rates in sub-Saharan Africa have stabilized over the past
few years, the actual number of people infected continues to grow because of
population growth. The stabilization of prevalence rates is due to increasing AIDS
deaths and continuing high numbers of new infections.
? With the wide availability of antiretrovirals in high-income countries, high-risk
behaviour is on the rise, leading to an increase in new HIV infections. For example,
in North America there were 950,000 people living with HIV in 2001 compared with
one million in 2003. In Europe, there were 540,000 people living with HIV in 2001
and 580,000 in 2003.
1. On what data do UNAIDS and WHO base their HIV prevalence estimates?
The precise numbers of people living with HIV, people who have been newly infected or
who have died of AIDS are not known. Achieving 100% certainty about the numbers of
people living with HIV globally, for example, would require testing every person in the
world for HIV every year—which is logistically impossible. But we can estimate those numbers by using other sources of data. UNAIDS/WHO estimates are based on all pertinent, available data—including surveys of
pregnant women, population-based surveys (conducted at the household level), sentinel
surveillance among populations at higher risk of HIV infection, case reporting, vital
registration systems (the official recording of births and deaths), as well as other
Different sets of data are used to calculate estimates of HIV prevalence for so-called
generalized (high-level) and concentrated (low-level) epidemics.
In countries with generalized epidemics, estimates of HIV prevalence are primarily based on blood samples from pregnant women in antenatal clinics (ANC). In the
absence of population-based surveys that include testing for HIV antibodies, sentinel
surveillance of women attending antenatal clinics generally provides the best available
estimate of HIV prevalence in the population.
1.1 How can we be confident that antenatal clinic-derived data
provide such good estimates?
Studies have shown that high proportions of women in most of the
highly-affected countries have access to antenatal clinic services.
In addition, women visiting these services are generally healthy,
which avoids the sampling bias inherent in testing sick individuals.
Where possible, estimates derived from antenatal clinic data have
been compared at local level with HIV prevalence data acquired in
population- based surveys. Such validation exercises have
concluded that estimates based on antenatal clinic sentinel
surveillance provide a good approximation of HIV prevalence
among adults aged 15-49 (men and women combined) in the local
For countries with low-level or concentrated epidemics, HIV estimates are based on
sentinel surveillance among key populations who are at higher risk of HIV exposure—
such as injecting drug users, sex workers, or men who have sex with men.
Countries with concentrated epidemics sometimes have additional sources of data which can help refine estimates. For example, in the Russian Federation, all pregnant women are
tested for HIV, which means that estimates of the number of children infected through
mother-to-child HIV transmission can be especially accurate. Likewise, in countries such as
Argentina and Brazil, which have extensive voluntary counselling and testing programmes,
case reports can make estimates more precise.
2. Why are the new global HIV/AIDS estimates lower than those published before?
The number of people living with HIV is not actually lower. The epidemic continues to
grow. The estimates have been revised as a result of improved modelling methodologies
and better country surveillance data.
The latest estimate for the number of people living with HIV incorporates improvements
in the estimation of the status of the epidemic, particularly in sub-Saharan Africa. New
data from many countries in that region, both from expanding surveillance systems and
from population-based surveys have enabled UNAIDS/WHO to adjust and refine
estimates of national prevalence:
2.1 What are the main adjustments that have led to the latest
More comprehensive information and improved methodologies
have enabled further refinements to HIV/AIDS estimates—for
? Improved surveillance has shown that HIV prevalence in
rural areas is lower than previously estimated, and that the
differences between infection levels in rural and urban areas
are greater than previously thought. This has been
especially notable in countries such Burundi, Ethiopia,
Kenya, Rwanda and Zambia, where expanded HIV
surveillance systems and national surveys have provided
new data in remote rural areas. Such improvements in data
collection and analysis will continue to enhance our
understanding of the epidemic.
? A major downward adjustment was made in the estimate of
people living with HIV in Zimbabwe. Figures released in
2003 placed national adult HIV prevalence in Zimbabwe at
25%, while it had been estimated at 34% at the end of 2001.
Unfortunately, this did not correspond to a real decline of 9%
prevalence. The 2001 estimate was based on antenatal
data that included a significant proportion of testing
irregularities. The corrected, lower national HIV prevalence
estimate for Zimbabwe means that the total number of
people living in Zimbabwe with HIV is currently estimated to
be several hundred thousand fewer than at the end of 2001.
This, too, has contributed to the ―lower‖ HIV prevalence
estimates for sub-Saharan Africa overall (and globally).
Steady improvements in the modelling methodology and better data from country
surveillance therefore are enabling UNAIDS/WHO to develop more accurate estimates.
These advances have led to ―lower‖ global HIV/AIDS estimates—both for the current
year and for past years.
The latest estimates cannot be compared directly with estimates published in previous years. Nor should these latest estimates be compared directly with those UNAIDS/WHO will publish in the years to come. Why not? Because the assumptions, methodologies and data used to produce the estimates are gradually changing, thanks to ongoing enhancement of our knowledge of the epidemic. Comparing the latest estimates with those published in previous years is liable to yield misleading conclusions.
In a nutshell, the latest estimates—for the current year and for past years—will tend to
be more accurate and reliable than those produced in previous years, since they are based on improved methods and more data than earlier estimates.
3. If the current estimates are lower than those arrived at in recent years, how can UNAIDS/WHO say that the AIDS epidemic continues to grow?
When we apply the latest data and improved methods to previous years, it is clear that there have been steady increases in the number of people living with HIV, as well as in the number of AIDS deaths.
The number of people living with HIV continues to grow in several regions, including in sub-Saharan Africa, despite a stabilization of the prevalence rate. Asia, as well as Eastern Europe and Central Asia, continue to experience expanding epidemics, with the number of people living with HIV growing year by year.
In sub-Saharan Africa, the estimated number of people living with HIV grew very rapidly from the mid-1980s through the mid-1990s, with continued but slower growth since that time. There has been a continued increasing trend in the number of people living with HIV.
4. If UNAIDS and WHO are now saying that earlier estimates were too high, how can they be sure that the current estimates are any better?
UNAIDS and WHO, together with experts from national AIDS programmes and research institutions, regularly review and update the estimates as improved knowledge about the epidemic becomes available. They also draw on advances made in the methods for deriving estimates. These improved methods, more and better data, and new estimation tools are enabling a better understanding of the epidemic. This is especially true for certain highly-affected countries.
We are confident, therefore, that the latest estimates provide reasonably accurate indicators of the level and scope of the AIDS epidemic and that they are of sufficient quality to support public health policy-making and planning. This does not mean that the estimates cannot be improved, but rather that higher levels of precision are not a prerequisite for sound decision-making and action.
5. Why are UNAIDS and WHO publicizing ranges of HIV and AIDS estimates? The ranges reflect the degree of uncertainty associated with estimates and define the boundaries within which the actual numbers lie.
In earlier UNAIDS/WHO reports, we reported point estimates (for example, fixing HIV
prevalence in country X at 12.5%). In addition, we also published the ranges of
uncertainty around those point estimates, depending on the quality of the data that had
yielded the estimates. This was done because all estimates fall within a range of
Because the quality of data varies from country to country, the ranges of uncertainty
surrounding our estimates can widen or narrow depending on the country. As well,
presenting point estimates might have encouraged a false sense of precision,
notwithstanding the fact that ranges of uncertainty were also provided.
Improved methods, enhanced data and new estimation tools are enabling a better
understanding of the degrees of uncertainty that surround HIV and AIDS estimates. This
is part of an ongoing process of improving estimates and developing appropriate
ranges—all of which are vital for effective HIV/AIDS planning and programming at national and regional levels.
UNAIDS and WHO are confident that the actual numbers of people living with HIV,
people who have been newly infected or who have died of AIDS lie within the reported
6. If UNAIDS and WHO claim the current estimates are more accurate, why are the
ranges for some countries so large?
The ranges reflect the degrees of uncertainty around HIV estimates in particular
countries. Accordingly, the ranges vary, depending on the quality of HIV data available in
We now know more about the epidemics in certain countries, due to improvements in
data gathering. This allows us to fix comparatively narrow ranges for their estimates.
However, other countries have not yet achieved such improvements, and the ranges for
their estimates are therefore wider.
Four factors determine the extent of the ranges around the HIV estimates:
(i) The HIV prevalence level. Ranges tend to be smaller when HIV prevalence is
higher. Thus the bounds around the best estimate of adults living with HIV in
Botswana are relatively small (310,000 – 340,000) while they are much wider
in a lower prevalence country such as Senegal (21,000 – 83,000).
(ii) The quality of the data. Countries with better quality data have smaller ranges
than countries with poorer quality data. The ranges for Asia and the Pacific
are comparatively broad—which reflects the fact that HIV surveillance of key
populations (such as injecting drug users, sex workers and men who have
sex with men) is relatively poor in most countries in that region. In general,
the ranges for sub-Saharan Africa are narrower, because of recent
improvements in the collection and interpretation of HIV data in that region.
Uganda and the Democratic Republic of Congo both have very similar
estimates of adult prevalence. However Uganda, which has a very strong
surveillance system, has a much smaller range around its estimate of adult
prevalence (2.8% - 6.6%) than does the Democratic Republic of Congo (1.7%
- 9.9%), where the quality of surveillance is weaker when judged by a
standard set of criteria.
(iii) The number of steps or assumptions used to arrive at an estimate. The more
steps and assumptions, the wider the range is likely to be (since each step
introduces additional uncertainties). For example, ranges around estimates of
adult HIV prevalence are smaller than those around estimates of HIV
incidence among children, which require additional data on the probability of
mother-to-child HIV transmission. The latter are based on prevalence among
pregnant women, the probability of mother-to-child HIV transmission, and
estimated survival times for HIV-positive children. There is therefore greater
uncertainty in these estimates than for adult prevalence alone.
(iv) The type of epidemic (generalized or low-level/concentrated). Ranges tend to
be wider in countries with low-level or concentrated epidemics than in
countries with generalized epidemics. Why? In low-level or concentrated
epidemics, one needs to estimate both the numbers of people in the groups
at higher risk of HIV infection and HIV prevalence rates in those groups.
7. Do UNAIDS and WHO have any way of checking the accuracy of the estimates?
There are several ways in which the estimates can be checked and validated.
Processes of triangulation (drawing on other sources of independent data) are used to
assess estimates. In countries with generalized epidemics, estimates of deaths derived
from vital registration systems or censuses can be used to detect changes in the age
patterns of mortality over time. Those changes can help shape independent estimates of
deaths due to AIDS. Such an approach has been used in Zimbabwe and South Africa,
and the results have contributed to improved estimates of HIV.
Similarly, estimates of the number of children orphaned by AIDS in countries with
generalized epidemics (drawn from sentinel surveillance of HIV) have been compared
with estimates from household questionnaires to check for bias and to improve
estimates of HIV prevalence and AIDS mortality.
Large-scale household surveys also make it possible to assess and improve the
accuracy of HIV estimates. This is because they can provide countrywide data on HIV
prevalence for both sexes including in remote rural areas that are rarely covered by
sentinel surveillance systems. Data from several such studies (for example, in the
Dominican Republic, Kenya, Niger, South Africa, Zambia and Zimbabwe) have led to
important refinements in the national estimates for 2003. The data have also enabled
adjustments to be made in the assumptions about differences in HIV prevalence
between urban and rural areas, and between males and females.
In addition, assumptions and data sources used in making estimates are being reviewed
at country-level. For example, HIV prevalence data from a recent household survey in
Kenya found lower prevalence than expected. (Previous estimates had relied on sentinel
surveillance of pregnant women attending antenatal clinics.) A review concluded that adult HIV prevalence in Kenya had been overestimated. Why? Earlier estimates had assumed too small a ratio of females with HIV to males with HIV. In Kenya, this sex ratio was found to be almost 2:1—larger than in most other countries, where the average ratio
was 1.3:1. (The numbers of males in Kenya with HIV had therefore been overestimated.) The estimate was also adjusted because sentinel surveillance did not adequately cover remote rural areas. The review led to significant adjustments in the national estimate for Kenya.
8. How confident are UNAIDS and WHO about the estimates of the number of
people who die of AIDS each year?
Estimates of adult AIDS mortality are based on several assumptions and sets of data—
including estimates of the numbers of adults and children who are HIV-infected, and estimations of survival times for adults and children infected with HIV.
In some countries with generalized epidemics, estimates of deaths derived from vital
registration systems or census rounds can provide additional information. Those data can be used, for example, to gauge changes in age patterns of mortality over time and thereby provide an independent estimate of deaths due to AIDS. This approach has been used in Zimbabwe and South Africa, where the results have helped to improve estimates of HIV/AIDS. However, in most countries with generalized epidemics, coverage of vital registration is too low to provide useful information on AIDS mortality. Generally, mortality estimates are most reliable in countries where prevalence is declining or has been steady for some time. This is because AIDS mortality in a given year is largely dependent on prevalence levels 5-10 years earlier.
Estimating adult new infections and mortality in countries with low-level or concentrated
epidemics is more difficult. Some at-risk groups are likely to have a different background mortality, in other words they are more prone to other causes of death. (for example, injecting in drug users are vulnerable to fatal drug overdoses and other life-threatening hazards). As well, some people tend to drift out of such at-risk groups. All this can have substantial effects on patterns of mortality. Unfortunately, country-specific data on mortality and on changes in risk behaviour are seldom available. However, some countries with low-level/concentrated epidemics have well-functioning vital registration systems that include the cause of death. The information documented in those systems has been used to help refine the mortality estimates in countries such as Brazil and Mexico.
9. Has the epidemic peaked in sub-Saharan Africa?
In sub-Saharan Africa, adult HIV prevalence appears to have stabilized. However, it is important to realise that a stable prevalence is only possible if HIV-associated deaths are replaced by new infections. Thus, a stable prevalence in sub-Saharan Africa still represents over 2 million new infections each year.
The number of people living with HIV in the region rose dramatically in the late 1980s and 1990s, and was still growing in 2003, although at a slower rate. This slower growth is a result of a peak in new infections which occurred in the mid-1990s and a rapid
increase in the annual number of people who die of AIDS.
It is now clear that across most of sub-Saharan Africa (including parts of southern Africa),
HIV prevalence among pregnant women visiting antenatal clinics has been roughly level
for several years—albeit at very high levels in Southern Africa. This apparent ‗levelling
off‘ of HIV prevalence has been interpreted by some observers as an indication that the
AIDS epidemic might have reached a turning point in sub-Saharan Africa. Unfortunately,
available evidence does not offer grounds for such conclusions.
Even though HIV prevalence rates have stabilized in this region the actual number of
people infected continues to grow because of population growth. Applying the same
prevalence rate to a growing population will result in increasing numbers of people living
9.1 What might be causing the apparent stabilization of HIV
prevalence in sub-Saharan Africa?
Two factors are causing the apparent stabilization of prevalence
rates observed in much of the region: AIDS mortality rates and
HIV incidence. High and, in some countries, rising rates of AIDS
mortality and continuing high HIV incidence offsetting this mortality
are the cause of this appearance of levelling off. In Zambia, for
example, national HIV prevalence appears to have stayed
relatively stable for the past 8–10 years. Since it is estimated that
close to 80,000 people living in Zambia have been newly infected
annually over that period, overall prevalence has remained
roughly level because AIDS has killed as many people each year.
HIV prevalence might therefore appear stable, but it hides a
persistently high number of annual, new HIV infections and an
equally high number of AIDS deaths.
But we are not, unfortunately, witnessing a decline in this region‘s
epidemic. In the absence of effective interventions, the epidemic
will continue to wreak havoc in these countries.
9.2 There is no such thing as the “African” HIV/AIDS epidemic
It is important to remember that there is not one, typical ―African‖
HIV epidemic. In six countries, adult HIV prevalence is below 2%,
while in six other countries it is over 20%. These extreme
differences in prevalence levels fall roughly into geographically
separate areas. Of the seven countries of southern Africa
(Bostwana, Lesotho, Namibia, South Africa, Swaziland, Zambia,
Zimbabwe) all have prevalence above 17%, with Botswana and
Swaziland having prevalence above 35%. In West Africa, HIV
prevalence is much lower with no country having a prevalence
above 10% and most having prevalence between one and five
percent. Adult prevalence in countries in Central and East Africa
falls somewhere between these two groups, ranging from 4% to
The countries of the region also differ in the time course of their
epidemics, with epidemics starting earlier in East and Central
Africa and much later in countries in southern Africa. Extreme
cases of this difference include Uganda (where adult HIV
prevalence peaked in the early 1990s) and Madagascar (which
detected virtually no HIV infection among pregnant women during
the 1980s and 1990s but where adult prevalence has now
increased to 1.7%).
10. Is the percentage of women infected with HIV rising globally?
No, this estimate is roughly the same as it was in 2001. UNAIDS estimates that almost
half (48%) of people living with HIV globally are women. However, the numbers of
women living with HIV globally are rising. This is because the overall numbers of people
living with HIV are rising. By far the majority of women living with HIV are in sub-Saharan
Africa. In sub-Saharan Africa, young women are much more likely to be HIV-infected
than young men – a ratio of over 30 to 10.
The concentration of HIV infected children in sub-Saharan Africa reflects the estimate
that 57% of HIV infected adults in that region are women. This is very different from all
other regions of the world where far less than half of adults living with HIV are women.
However, in all these other regions the percentage of women among those living with
HIV is increasing.
11. Are more young people becoming infected with HIV?
UNAIDS estimates that of the 38 million people living with HIV globally, 10 million are
young people, aged 15-24 years. Since more people overall are becoming infected with
HIV, the numbers of young people living with the virus are therefore increasing.
12. Which are the more accurate: antenatal clinic- or national population-based
In countries with generalized epidemics, estimates of HIV prevalence have primarily
been based on blood samples left over from syphilis tests of pregnant women in
antenatal clinics (or ―sentinel surveillance‖). Until very recently, these have provided the
best available estimates of HIV prevalence in the population.
However, national population-based or household surveys are increasingly becoming
available. Such surveys have the potential to improve the accuracy of estimates of HIV
because they can provide countrywide data on HIV prevalence for both sexes including
samples from remote rural areas rarely covered by sentinel surveillance systems.
Population survey data have been used to help refine the estimates for several countries
in the UNAIDS/WHO 2003 estimates (including the Dominican Republic, Kenya, Niger,
South Africa, Zambia and Zimbabwe). They have also enabled the improvement of
assumptions about urban-rural and sex differences in HIV prevalence that are used to
determine HIV estimates in other countries in the same region.
Both antenatal clinic and population-based survey data, though, have advantages and
National population-based surveys, on the one hand, capture a much wider
representation of the general population than do antenatal clinics. They can yield
information on HIV prevalence among men and non-pregnant women, and they can
provide better coverage of rural populations than antenatal clinic-based surveillance.
On the other hand, the fact that some respondents refuse to participate or are absent
from the household adds considerable uncertainty to survey-based HIV estimates.
(Indeed, non-response rates ranged from 24% to 42% in the recent surveys carried out
in African countries.) The estimates can be adjusted if the basic characteristics of the
non-responders can be discerned. The problem is that the survey itself cannot measure
the possible association between a person‘s absence or refusal to participate, and that
person‘s HIV status. It might be that a person‘s refusal to participate or absence from the
household is correlated with a stronger likelihood of HIV infection. (For example, the
association of mobility with HIV infection may affect the findings of household surveys.
Mobile men, who generally have higher levels of HIV infection, are less likely to be found
at home for these surveys. This is especially important in countries with high levels of
mobility and/or migration, and for surveys with a high proportion of absentees.)
Meanwhile, antenatal clinic data form the basis for HIV estimates that rest on a set of
assumptions that may not apply equally well to all countries and at all stages of the
epidemic. (It is assumed, for example, that HIV prevalence among pregnant women is
roughly the same as in the adult population overall, that the ratio of women with HIV to
men with HIV is 1.3:1, and that adult survival time is roughly nine years. Assumptions
about age distribution of HIV infections are also factored in.) In addition, most antenatal
clinic-based surveillance systems have limited geographical coverage, which can lead to
significant variations in the quality of the national estimate of HIV prevalence across
With the exception of South Africa, surveillance systems often select clinics located in
urban or peri-urban areas, both for ease of access and because these clinics serve a
larger number of pregnant women and can yield sufficient sample sizes during data
collection. Often this leads to few data being available from pregnant women in rural
areas. This bias can be corrected, but it introduces a further layer of uncertainty (the
extent of the differences between HIV prevalence levels in urban and in rural areas).
Nevertheless, antenatal clinic-based data are especially useful for gauging HIV trends
over the years. National household surveys help fill out our picture of the epidemic.
Conducted at three-to-five-year intervals, such surveys can serve as valuable
components of surveillance systems and can help improve estimates of the levels and
trends in HIV prevalence.
All in all, there is no gold standard for HIV surveillance. All HIV estimates need to be
assessed critically—whether they are based on a national survey or on sentinel
surveillance data. Using all available data to arrive at HIV estimates ensures the best
13. Which are the more accurate sources of data: sentinel surveillance or case