Secondary Logo

Share this article on:

Country-specific estimates and models of HIV and AIDS: methods and limitations

Schwartländer, Bernharda; Stanecki, Karen A.b; Brown, Timc; Way, Peter O.b; Monasch, Roelandd; Chin, Jamese; Tarantola, Danielf; Walker, Neffa

Epidemiology and Social: Original Papers

Objective: This paper presents the methods used to calculate the end of 1997 country-specific estimates of HIV and AIDS produced by the UNAIDS/WHO Working Group on Global HIV/AIDS and STD Surveillance. The objective of this exercise was to improve estimates on HIV/AIDS by using country-specific models of HIV/AIDS epidemics. The paper describes and discusses the processes and obstacles that were encountered in this multi-partner collaboration including national and international experts.

Methods: The 1997 estimates required two basic steps. First, point prevalence estimates for 1994 and 1997 were carried out and the starting year of the epidemic was determined for each country. The procedures used to calculate the estimates of prevalence differed according to the assumed type of the epidemic and the available data. The second step involved using these estimates of prevalence over time and the starting date of the epidemic to determine the epidemic curve that best described the spread of HIV in each particular country. A simple epidemiological program (EPIMODEL) was used for the calculation of estimates on incidence and mortality from this epidemic curve.

Results: Regional models that were used in previous estimation exercises were not able to capture the diversity of HIV epidemics between countries and regions. The result of this first country-specific estimation process yielded higher estimates of HIV infection than previously thought likely, with over 30 million people estimated to be living with HIV/AIDS. The application of survival times that are specific to countries and regions also resulted in higher estimates of mortality, which more accurately describe the impact of the epidemics. At the end of 1997, it was estimated that 11.7 million people worldwide had died as a result of HIV/AIDS since the beginning of the epidemic.

Conclusion: This exercise is an important step in improving understanding of the spread of HIV in different parts of the world. There are, however, shortcomings in the current systems of monitoring the epidemic. Improvements in HIV surveillance systems are needed in many parts of the world. In addition, further research is needed to understand fully the effects of the fertility reduction as a result of HIV, differing sex ratios in HIV infection and other factors influencing the course and measurement of the epidemic.

From the aUnited Nations Joint Programme on HIV/AIDS, Geneva Switzerland; bUS Bureau of the Census, Washington DC, USA; cEast-West Center, Honolulu, Hawaii, USA: dWorld Health Organization, Geneva, Switzerland; eSchool of Public Health, University of California, Berkeley, CA, USA; and fFrançois-Xavier Bagnoud Center for Health and Human Rights, Harvard School of Public Health, Boston, MA, USA.

Correspondence to: Bernhard Schwartländer, UNAIDS, 20 Avenue Appia, CH 1211, Geneva 27 Switzerland.

Received: 10 June 1999; revised: 8 September 1999; accepted: 16 September 1999.

Back to Top | Article Outline


In June 1998, The Joint United Nations Programme on HIV/AIDS (UNAIDS) and the World Health Organisation (WHO) published new country-specific estimates for HIV infection. Working with national governments and research institutions, the most recent data from a number of sources were compiled and analysed. This information was used to estimate the number of men, women, and children living with HIV and AIDS at the end of 1997, as well as the number of deaths and children orphaned by AIDS in each country.

The 1997 estimation was the first such exercise by UNAIDS and WHO since 1995, when WHO‚s Global Programme on AIDS (GPA) estimated end-1994 HIV prevalence for all countries[1,2]. Regional and global estimates since then have been made annually by projecting 1994 prevalence data forward on the basis of regional models (produced for sub-Saharan Africa, Latin America and the Caribbean, eastern Europe and central Asia, east Asia and the Pacific, North Africa and the Middle East, western Europe, North America, south and Southeast Asia and Australasia) that were thought to hold true for all major geographical areas where the epidemic had been documented. The underlying assumption behind these regional models was that the different countries within a given region, although at different prevalence levels, would follow roughly the same basic pattern in HIV spread.

New information from surveillance and population-based studies, however, has called this assumption into question. In sub-Saharan Africa, for example, the HIV/AIDS epidemics in the eastern parts of the continent started to level off in the mid-1990s after spreading at a dramatic pace since the late 1970s. In some countries of western Africa, levels of HIV have stayed relatively low for more than a decade. In the southern part of the continent, a massive and rapid spread started only in the late 1980s and early 1990s[3].

The models used in the previous estimates were based on regional models[1,2]. Although point prevalence estimates in 1994 were made for each country, the model of the course of the epidemic and the resulting estimates of incidence and mortality were performed at a regional level with regional models. These regional models did not capture such diverse patterns of spread and may not have produced valid estimates on regional and global levels. Country-specific models should yield more accurate figures and are also more useful for informed decision making in countries. Although the limitations of regional models were recognized, there was not enough information available in 1995 to produce country-specific models with a reasonable level of confidence.

GPA took important steps to address this problem by improving the data collection and surveillance systems in countries through the introduction and support of sentinel surveillance[4]. In the sentinel surveillance system, countries test blood collected for other purposes from high-risk or other sentinel populations. Data from these groups serve as guides to the current level of the epidemic in the population. For most countries, data from attendees of sexually transmitted infection clinics were tested along with blood collected from other high-risk populations [men who have sex with men (MSM), sex workers, intravenous drug users (IDU)]. In countries where HIV prevalence was assumed to be spread through heterosexual contacts, blood from attendees of antenatal clinics is the primary source of information on HIV prevalence in the general population. These efforts led to a much improved knowledge base in many countries, allowing for the development in 1997 of the first country-specific estimates of HIV prevalence for 86 countries in sub-Saharan Africa, Asia, Latin America and the Caribbean.

This paper describes how these country-specific estimates were developed and the assumptions on which they were based. It also discusses the limitations of the approach and how these estimates can be improved in the future.

Back to Top | Article Outline

Estimation and modelling procedures

The 1997 estimates required two basic steps. First, the year the epidemic began and point prevalence estimates for 1994 and 1997 were made for each country. The procedures used to arrive at the estimates of prevalence differed according to the assumed stage of the epidemic and the data available on the epidemic in each country. The second step involved using these three (and sometimes more) estimates of prevalence over time and the starting date of the epidemic to determine the epidemiological curve of prevalence that best described the epidemic in each country. A simple epidemiological program developed by WHO (EPIMODEL) [5]was used to develop the epidemiological curve, and with that and additional assumptions (e.g. progression rates from infection to death) one was able to derive estimates of incidence and mortality. The specific procedures followed will be described in detail below.

Back to Top | Article Outline

Estimating prevalence for 1994 and 1997

The sources of information that can be used to estimate the number of people living with HIV/AIDS depend on the type and pattern of the epidemic and the quality of the surveillance systems. We describe below how these two characteristics were combined to form groups of countries for estimation purposes.

Back to Top | Article Outline

Developing countries

Two sets of procedures were used to estimate prevalence in developing countries, based on the nature of the HIV epidemic. One set of procedures was used for countries in which HIV is concentrated in sub-populations at higher risk of infection (countries with concentrated epidemics)[6]. These sub-populations, which vary from country to country, may include sex workers, IDU, MSM, and others. A second set of procedures is used for countries in which HIV has spread significantly within the general population (countries with generalized epidemics). In this type of epidemic, the primary mode of transmission is through heterosexual contacts.

For developing countries with concentrated epidemics, estimates were made by adding together the number of individuals assumed to be infected in each identifiable sub-population at risk in that country and the number of infected individuals outside these groups (e.g. estimates of the numbers of infections among men and women in the general population). This required two types of information: estimates of the size of each sub-population and estimates of the HIV prevalence in each of these population groups. These estimates were drawn from routine surveillance information, special studies if available, and estimates provided by national AIDS programmes. The quality of the estimate is therefore dependent upon the quality of data on the size of and HIV prevalence in specific sub-populations.

Overall prevalence was then produced by calculating the number of infected individuals in each at-risk population (prevalence rate times population) and adding the number of infected individuals in the general population. For prevalence in the general population, primarily data from antenatal clinics (with other sources, such as blood donors, tuberculosis clinic attendees) were used to calculate the estimates (see procedures described in the section on generalized epidemics below). If there appeared to be a significant overlap between high-risk groups and the measures for the general population, adjustments were made on a country-specific basis. The female-to-male ratio was estimated by reviewing existing studies in the countries or by extrapolating forward the trends in the sex ratio from existing data on AIDS cases. For countries in sub-Saharan Africa, a 1_:_1 ratio for women to men was used. For other countries, the ratio depended on the unique transmission pattern of the country.

For developing countries with generalized epidemics, prevalence estimates were based primarily on surveillance data collected from women attending antenatal clinics. The prevalence estimate for women was then used to estimate the prevalence among men, based on the assumed female-to-male ratio. Although surveillance in antenatal clinics samples only women, in most developing countries it provides the best available estimates of HIV prevalence within the general population. This is especially true for developing countries where fertility is very high and the rate of contraceptive use is low. In most developing countries, antenatal services are among the most utilized health services, provide the broadest levels of coverage, and do not have as many confounding problems as do other potential surveillance sites (e.g. voluntary blood donors, general health clinics).

Separate estimates were made for two groups of clinics, based on their location. One group included clinics located in ‚major urban areas‚. Major urban areas usually included the capital city and other major urban agglomerations. The other group included clinics located ‚outside major urban areas‚, defined as all other clinics. The second group included clinics located in periurban and rural settings. Although imperfect, the distinction represents an attempt to take into account indications that the HIV epidemic often varies widely between rural and urban areas within a single country. Of course, variation in prevalence also occurs within major urban areas and outside urban areas in a country, because these areas differ in many factors that can affect the spread of the epidemic. However, the data are not at a level of specificity to capture these differences. To account for these differences, adjustments were made and will be described below.

Initially, unadjusted prevalence rates from antenatal clinic surveillance sites for major urban and outside major urban areas were applied to total adult population estimates for the respective areas. For countries with multiple sites, median prevalence rates calculated separately from all sites in major urban areas and outside major urban areas were used as the initial estimates of prevalence. Medians were used instead of means, because the surveillance sites were not representative of all antenatal clinics in either major urban or outside major urban areas. By using medians, the effects of outliers in the sample of sites were reduced. These initial prevalence estimates were then adjusted for representativeness. This adjustment usually took the form of lowering the estimated prevalence outside major urban areas, because surveillance sites in this group were more often periurban than rural and may thus overestimate the prevalence for large parts of the rural populations.

A Delphi-type procedure was used to make this adjustment. In this procedure experts from UNAIDS, WHO, country programmes and research institutions who had knowledge of the country were asked to adjust the prevalence rates for major urban areas and outside major urban areas, on the basis of their judgement about the extent to which the surveillance sites were representative of the general population. The group then discussed the various adjustment factors and decided on a consensus adjustment that would be used. This adjustment factor was then applied to the adult populations for major urban and outside major urban areas to determine the overall prevalence estimates.

For the countries in sub-Saharan Africa, a 1_:_1 ratio of men to women was assumed, to arrive at the estimates. For some Asian countries that have seen significant general population spread, (e.g. Cambodia and Thailand) higher male-to-female ratios were used. This ratio reflects the imbalance in male-to-female ratios produced in rapid epidemics in which most of the early transmission occurs as a result of commercial sex or among sub-populations that are predominately male (e.g. IDU, MSM). These ratios were established by referring to existing studies, by soliciting recommendations from national AIDS programmes, and by reviewing existing AIDS case reports.

Back to Top | Article Outline

Industrialized countries

For industrialized countries, UNAIDS and WHO did not attempt to develop estimates on HIV prevalence or model the epidemic. There were two reasons for this decision. First, in most industrialized countries national expert groups have produced their own estimates for HIV and AIDS, based on extensive national surveillance systems. Second, the models used for the exercise described in this paper cannot currently take into account important parameters of the response to the epidemic in these settings. In most industrialized countries, prevention efforts began early. In some, effective treatments either for the prophylaxis of opportunistic infections or antiretroviral agents have been widely and increasingly used over time. These interventions have altered the ‚natural‚ course of the HIV epidemic in ways that cannot be taken into account in a simple modelling process. The end-1997 country estimates for industrialized countries were thus exclusively based on information provided by national authorities.

Back to Top | Article Outline

Countries with inadequate data

In some 32 out of 182 countries, available data were so limited that it was not possible to produce models of the epidemic. For most of these countries, the 1994 prevalence rates were very low. The 1994 prevalence rates published by WHO/GPA [1,2] were used for these countries and applied to the 1997 population as conservative estimates.

Back to Top | Article Outline

Review of 1994 estimates

The above procedures were used to make revised estimates of prevalence in 1994 for all countries for which sufficient data were available. These revised 1994 estimates were then compared with those produced previously by WHO[1,2,7]. The purpose of recalculating the 1994 estimates was to update those estimates with more complete data and with information showing the trends in HIV infection since 1994. The procedures used here were basically the same as those used by WHO for the 1994 estimates. The difference is that when the 1994 country estimates were made, data for 1994 were rarely available, and only limited data were available for 1993. Many additional data sources are now available for the years 1993 and 1994, and subsequent trends can be checked for consistency with the 1994 estimate. The purpose of reviewing the original 1994 estimate was to ensure that in fitting the epidemiological curve to both the 1994 and 1997 estimates, the 1994 estimates would be made on the basis of the best information available. By using the currently available surveillance information, the 1994 estimates could be checked and improved if necessary.

For the majority of countries, we found small or no differences between the original projected 1994 estimates and new 1994 estimates based on more complete and more recent data, providing partial validation of the approach. In a few countries, however, differences between the original and the new estimates of 1994 prevalence were considerable. In particular, the original 1994 estimates tended to underestimate HIV prevalence in several countries in southern Africa (e.g. Botswana, Zimbabwe, South Africa) and in one country, Uganda, the original estimates were too high. The differences for the countries in southern Africa are easy to explain. In 1994 the epidemics in this part of the continent were still in their very early phases, and it was therefore not at all clear from the available data at that time that explosive epidemics would develop as is known today. Similarly, in Uganda the epidemic started to turn around in the early/mid 1990s. This change in the epidemic was not to be foreseen in the surveillance data available in 1994/1995.

Back to Top | Article Outline

Validation of estimates

The estimates of prevalence for 1994 and 1997 were sent to national AIDS programmes for review and approval. The national programmes were asked to review the prevalence estimates and the data (surveillance site data) that were used to calculate their estimates of prevalence. Minor adjustments were made for a few countries where valid information suggested such alterations. For a few countries, the prevalence estimates were raised or lowered because the national programmes had data or information that were not available during the estimation procedure. Such information included the results of national consensus workshops that had been held in a series of countries in 1997/1998 including national and international experts.

Back to Top | Article Outline

Estimation of AIDS deaths and other epidemiological parameters using EPIMODEL

After point prevalence estimates for 1994 and 1997 had been established and confirmed, a simple modelling program was used to calculate numbers of HIV infections, AIDS cases and AIDS deaths over time. Modelling was carried out using the EPIMODEL version 2.1 developed by WHO[5]. It is worth noting that this software was created to assist in developing estimates of various aspects of the epidemic from existing data. It was not designed for making projections. The model requires several pieces of information: point prevalence in a chosen year; the year ‚extensive spread‚ of infection started; the most likely shape of an epidemiological curve of HIV prevalence over time; progression rates (e.g. from HIV infection to AIDS, AIDS to death for adults and children); age structure of the HIV-infected population and age-specific fertility rates; ratio of women to men infected; and the rate of mother-to-child transmission. The remainder of this section describes how each of these parameters were determined and applied to prepare the 1997 estimates.

Back to Top | Article Outline

Establishing the first year of extensive _spread of HIV

The starting year of ‚extensive spread‚ was determined from a review of existing research and data from the countries. The date of ‚extensive spread‚ was defined as the date on which indigenous spread of HIV began in a country. This concept was used instead of date of first reported infection, because in many countries the epidemic started well before the first case of HIV or AIDS was reported (e.g. countries in sub-Saharan Africa). On the other hand, in some countries, individual cases of infection were reported well before the epidemic begins (e.g. countries in central Asia). These cases would not define the beginning of extensive spread. For countries where little or no data existed, information from neighbouring countries was used.

Back to Top | Article Outline

Assumptions about rates of disease progression and vertical transmission

Although data are still scarce, recent research indicates that progression rates from HIV infection to AIDS and then to death (survival time) differ between the industrialized countries and developing countries. Data on progression to AIDS survival times from infection are ideally derived from cohort studies. However, individuals included in specific studies may not be representative of the total infected population. This is especially true in the least developed countries, where participation in a study or research project is often associated with receipt of HIV-specific care that is not available to the majority of infected individuals.

After extensive review of the literature and expert consultations, it was decided to apply different progression rates from HIV infection to death for adults and children in different regions[8-12]. For adults, two progression curves from HIV infection to death were used with a median of 11 years (‚slow‚) and 9 years (‚fast‚). Both assumed the progression from AIDS to death to be a median of one year. The slower rate of 11 years represents the natural history in western countries with strong health systems before the era of treatment[10,11]. The ‚fast‚ rate was used for countries that have weaker healthcare systems or reduced access to healthcare. This rate corresponds closely to survival rates found in a study in rural Uganda[13,14]. The two resulting adult survival curves are shown in Fig. 1(a).

Fig. 1.

Fig. 1.

Studies have found that perinatally infected children fall into two subclasses: a smaller group of rapid progressors who mostly develop AIDS in the first year of life, and a larger group of children with HIV who progress more slowly to AIDS[15-17]. These studies have also seen substantial variation in the overall rates of progression between developing and industrialized countries. A review of the available research suggested the use of three progression curves for children: fast, medium, and slow for countries with different levels of healthcare and child mortality[12,18]. Accordingly, three curves were developed on the basis of a double Weibull model, one component for rapid progressors and another for slower progressors, as described in Downs[16]. The fast progression curve assumes that 84% of children develop AIDS within 5 years of birth; the medium progression curve assumes 65% and the slow progression curve assumes 46%. Furthermore, it was assumed that most children die within the first year after developing AIDS for all groups. The survival curves for the three paediatric progression rates are shown in Fig. 1(b).

For each country, the specific paediatric and adult progression rates were based on judgements about access to and quality of healthcare. For paediatric rates, infant and child mortality rates were used as the best proxy measures to determine the progression rate. For adults, there was a complex set of proxy measures. Measures of maternal mortality, life expectancy, and access to healthcare were among the measures reviewed to determine the appropriate adult progression rate. For sub-Saharan Africa the fast progression curves were used for both children and adults. For Latin America and the Caribbean the slow progression curve was applied for adults and the medium progression curve for children. For Asia and the Pacific, the picture is less uniform and a country-specific mix of the progression curves for adults and children described above was developed. In the wealthy Asian state of Singapore, for instance, the country model assumed that half the adults living with HIV will survive for at least 10 years before developing AIDS, whereas approximately 46% of infected children will have developed AIDS by the age of 5 years (both slow adult and paediatric progressions rates). In neighbouring Indonesia, where per capita GNP was less than a tenth of Singapore‚s, even before the current economic crisis, healthcare is more limited and survival is shorter. The model there assumes that half of Indonesian adults infected with HIV will have developed AIDS after 8 years (a fast adult progression rate) and nearly two-thirds of infected children will have developed AIDS by their fifth birthday (and a medium child progression rate). The decision on which progression rate to use for children in Asian settings was based on infant mortality rates. Countries in the region with comparatively high infant and child mortality (e.g. Papua New Guinea) were assumed to follow the fast paediatric progression rate, whereas those with very low rates (e.g. Hong Kong) were assumed to follow the slower rate. For most countries in Asia, the medium rate of progression was used.

Region- and selected country-specific rates were also developed for vertical transmission. In the absence of widely available prophylactic antiretroviral treatment or elective caesarean section, the level of breast-feeding mainly influences differences in the vertical transmission rate. Breast-feeding is likely to be linked to knowledge in the general population about the vertical transmission of HIV, individuals‚ knowledge of their sero-status, access to obstetric care and the accessibility of and ability of the mother to provide safe, sanitary substitute milk, all of which vary by region. The basic rate for mother-to-child transmission was set at 35% (practically assuming 100% breast-feeding and no interventions), but lower transmission rates were used based on the judged availability of testing, the prevalence of breast-feeding, and other medical/therapeutic interventions. This yielded three rates, 25, 30 and 35%. In sub-Saharan Africa, where breast-feeding is the norm and access to obstetric care is lower, the rate used was 35%. In Latin America, where HIV counselling and testing was introduced comparatively early and more mothers were therefore aware of their HIV status, a rate of 25% was used. As was the case with progression rates, intra-regional variations in Asia required the use of all three rates.

Back to Top | Article Outline

Age-specific information about population _and fertility

Modelling the vertical transmission of HIV and orphaning of children by AIDS requires the age distribution of HIV-infected women and age-specific fertility rates. Regional age distributions for HIV-infected women were derived from studies in which such information was available and were applied to neighbouring countries without such information. Age-specific fertility rates and population figures were taken from the database of the United Nations Population Division for each country. These data were used along with estimates of prevalence among women and the rate of vertical transmission, to produce estimates of the number of infected children.

Back to Top | Article Outline

Fitting an epidemic curve through point prevalence estimates

EPIMODEL provides two approaches to fit an epidemiological curve of estimated prevalence over time: a follow-curve scenario and a population growth-rate scenario. In the follow-curve scenario, one specifies parameters related to a gamma function of the prevalence of infections. The two parameters relate to the slope of the curve (rate of increase) and the current location on the curve (time into the epidemic). The population growth-curve scenario allows one to specify directly the growth rate in prevalence. For countries in which the point prevalence increased between 1994 and 1997, the follow-curve scenario was used. For countries in which the point prevalence did not change or decreased between 1994 and 1997, suggesting a stable epidemic, the population growth-rate scenario was used. These two scenarios are explained below.

Back to Top | Article Outline

Follow-curve scenarios

The assumption used for the 1997 estimates (as well as the original and new 1994 estimates) is that the distribution of HIV infection over time will follow a gamma function during the early period when the epidemic is growing. Experience has shown that such functions are quite powerful in fitting epidemics in their increasing phases until they level off, especially if interventions are limited or do not have a major impact on the spread of HIV. For countries still experiencing growing epidemics in the 1994 to 1997 time-frame, EPIMODEL was applied to identify gamma curves that matched observed prevalence trends. The slope of the curve in each country was determined by fitting the prevalence observed at a minimum of three points in time: beginning year of extensive spread, 1994 and 1997. In countries where time series of data for additional years were available from directly comparable surveillance data sets, they were matched to the trends predicted by the curve, and the curve parameters were adjusted to reflect more accurately the observed trends.

Back to Top | Article Outline

Population growth-rate scenarios

For some countries, especially in the south American region, there was little increase or a decrease in point prevalence estimates between 1994 and 1997, indicating that the epidemic was stable. For these countries, a population growth rate scenario was followed. In this growth rate scenario in EPIMODEL, a gamma curve was fitted through available data points during the growth phase of the epidemic and the year in which prevalence reached a relatively stable level was established. For subsequent years, instead of following the curve used during the growth phase, EPIMODEL asks the user to enter a steady growth rate for prevalence. EPIMODEL then adjusts HIV incidence from that year forward as needed to maintain the prevalence growth rate specified. Frequently, the growth rate is similar to the growth rate of the adult population. In Brazil, for example, there has been no evidence of an increase in the HIV prevalence rate since 1994. Therefore the prevalence growth rate was set at the average yearly rate of population growth, resulting in a slightly increased absolute number of people living with HIV/AIDS.

Back to Top | Article Outline


Globally, an estimated 30.6 million persons were infected with HIV/AIDS at the end of 1997. As shown in Table 1, 21 million of the people infected with HIV/AIDS were in countries in sub-Saharan Africa (69% of the global total). There were 5.8 million infections in south and Southeast Asia (19%), 1.6 million in Latin America and the Caribbean (5%), 190_000 in eastern Europe and central Asia (0.6%), 420_00 in east Asia and the Pacific (1%), 210_000 in North Africa and the Middle East (0.7%), and 1.3 million in the countries of western Europe and North America (5%).

Table 1

Table 1

Eight countries were estimated to have over one million people infected with HIV/AIDS (see Table 1). India had the highest number of infected persons (4.1 million), followed by South Africa (2.9 million), Ethiopia (2.6 million), and Nigeria (2.3 million). Four other countries in sub-Saharan Africa, Kenya, Mozambique, Tanzania and Zimbabwe, each had between one and two million people estimated to be living with HIV/AIDS. Among countries outside of sub-Saharan Africa and excluding India, only the United States, with an estimated 820_000 infections, and Thailand, with an estimated 780_000, had HIV-positive populations approaching one million.

Average prevalence rates among adults were estimated to be very high in sub-Saharan Africa (7.4%). No other region approached this level in adult prevalence rate. The countries in the Caribbean had the next highest prevalence rate at 1.8%. The other regions all had adult prevalence rates well below 1%.

Among the 45 countries in sub-Saharan Africa, Zimbabwe and Botswana were estimated to have over a quarter of their adult populations infected, and another 11 countries had estimated adult prevalence rates of between 10 and 20%. Outside this region, 11 countries had adult prevalence rates estimated to be above 1%. In Asia, Cambodia (2.4%), Thailand (2.2%), and Myanmar (1.8%) had the highest prevalence rates. In the Caribbean, Haiti (5.2%), the Bahamas (3.8%), Barbados (2.9%), and the Dominican Republic (1.9%) all had adult prevalence rates above 1%. In South and Central America, Guyana (2.1%), Belize (1.9%), Honduras (1.5%), and Suriname (1.17%) had adult prevalence rates above 1%. No other countries had adult prevalence rates above 1%.

Whereas sub-Saharan Africa was home to most infected people, the spread and growth of the epidemic has been quite dramatic in many parts of the world. Figure 2 show the proportional change in adult prevalence rate between 1994 and 1997. As expected, many of the countries showing the greatest proportional growth are in southern Africa. However, substantial growth in prevalence is also shown in many countries in Asia and among the countries of the former Soviet Union.

Fig. 2.

Fig. 2.

The estimated cumulative number of deaths caused by HIV/AIDS globally had reached 11.7 million people by the end of 1997. Of this total, 9.6 million (82%) had occurred in the countries of sub-Saharan Africa, south and South-east Asia (730_000), Latin America _(470_000), and North America (420_000); each accounted for between 4 and 6% of the cumulative global deaths.

The estimated number of deaths caused by HIV/AIDS during 1997 followed the same general geographical distribution. In sub-Saharan Africa 1.8 million people were estimated to have died as a result of HIV/AIDS during 1997. This is almost 80% of the estimated global total of 2.3 million deaths during 1997. The estimated number of new deaths caused by HIV/AIDS in North America and western Europe were substantially lower than in previous years as a result of the widespread use highly effective antiretroviral therapies.

The number of children estimated to be living with HIV/AIDS globally was 1.1 million. Of these, 960_000 (87%) were in countries in sub-Saharan Africa. It was also estimated that 590_000 children were newly infected with HIV/AIDS during 1997. Again, countries in sub-Saharan Africa accounted for almost 90% of all new infections among children (530_000). The preponderance of new infections in these countries was the result of the high HIV prevalence among women, near-universal breast-feeding, high fertility rates, and the lack of access to treatments to reduce vertical transmission.

Back to Top | Article Outline


The quality of the estimates of HIV/AIDS at the end of 1997 are clearly only as good as the data used to make the estimates and the appropriateness of the assumptions made about disease progression and vertical transmission. It is worth noting, however, that HIV sentinel surveillance systems are generally rather extensive when compared with surveillance systems for other communicable diseases. However, even with high quality surveillance data, several issues related to how the data from these sites were used to produce estimates of the general population are open to question.

One limitation of the estimates presented here concerns the estimates of prevalence in countries with concentrated epidemics. In these countries, estimates were made on the basis of what was often very weak data regarding infection rates in marginalized risk groups (e.g. IDU, MSM). Few countries have systematic sampling from these populations, and therefore the estimates of prevalence rates are less precise as are the estimates of the size of the denominators (size of the population).

For countries with generalized epidemics, estimates of HIV prevalence can also be biased. One major assumption was that data from antenatal clinics could be used to estimate prevalence in the adult population. This assumption may not hold for several reasons. First, antenatal clinics may not be representative of all adult women. They contribute data only for recently sexually active women who are fecund and using no effective contraception. For younger age groups, in which high proportions of the population may not yet be sexually active and where condom use in response to HIV prevention programmes is not uncommon, HIV prevalence measured at antenatal clinics may tend to overestimate infection levels in the general population. However, because HIV infection lowers fertility progressively over time[20,21], antenatal data may tend to underestimate the HIV prevalence rate among older women, because those who are infected with HIV are less likely to become pregnant.

Studies comparing sentinel surveillance data with population-based data in mature epidemics confirm that these distortions are common. However, even with these problems, antenatal data across the entire 15-49-year-old age range often provide remarkably robust estimates of HIV levels in the general population in these epidemics[21,22].

A second issue relates to the use of ANC data to estimate prevalence among men in sub-Saharan Africa. In the generalized epidemics in this region, we have assumed that there are an equal number of men and women infected with HIV/AIDS. However, the ratio of men to women could easily vary by 5% or more in different countries[23]. In fact, research has suggested that the ratio of infected women to infected men may vary with the age of epidemic. In the early years, men account for a higher proportion of all infections. However, as the length of the epidemic increases, the ratio of infection becomes more equal and even reversed so that more women than men are infected. Also, there are often large differences between prevalence rates of men and women at different ages. In the 15-19-year-old age range, more women are infected than are men. At older age ranges (40-44 years) more men are infected[21,24,25]. If the assumption of overall equal levels of infection among women and men does not hold true, the impact on estimates is potentially significant. If women account for fewer than half of those infected, current methods would underestimate the true population prevalence in sub-Saharan countries. This is because the estimates are based on the assumption of a 1_:_1 ratio of women to men among infected individuals. If women account for more than 50% of all those infected, current methods are likely to overestimate the prevalence.

A third issue relates to the use of sentinel surveillance sites from major urban areas and outside major urban areas as the only geographical breakdowns in calculating country prevalence. For the purpose of calculating the estimates, we have assumed that once outside the major urban areas, HIV prevalence is almost uniform. In fact, there may be systematic differences in periurban and rural areas, and this difference is not being captured. This problem was partly addressed by the Delphi-type process that was used to adjust for unrepresentative samples of sites from outside major urban areas. However, the potential for error remains. To eliminate this problem, surveillance sites would need to provide coverage of both periurban and rural areas in a country. A related issue is the question of the overall representativeness of sentinel sites. In some countries they were initially chosen in areas believed to be of higher risk or showing higher prevalence, as an early warning system. This would tend to overestimate the prevalence. However, in most of the world, especially in many African countries where the epidemic is most severe, sentinel sites have expanded significantly and provide fairly broad coverage in urban areas, although rural coverage remains limited. This issue was also taken into account to the extent possible by the team working on the estimates.

An additional set of limitations or caveats to the 1997 estimates relates to the rates used for the progression of the disease. Two inputs that can have a great impact on the estimates of AIDS incidence and deaths are the annual progression rates from infection to AIDS and from AIDS to death. For the 1997 estimates, the mean time from infection to death in sub-Saharan Africa was set at 9 years. This rate corresponds closely to survival rates found in an ongoing study in rural Uganda[13,14]. The fact that survival times in industrialized countries, before HIV-specific treatment (including prophylaxis of opportunistic infections) but with access to better healthcare in general[10,11], are approximately 11 years supports the estimate of 9 year survival in those countries with the least resources.

However, if this estimate on progression from infection to death is too short, it could have a substantial effect on the estimates of death and incidence. If people survive longer, incidence and deaths will be lower, making the current estimates too high. Conversely, if the survival time were less than 9 years, the estimates of incidence and death would be underestimated. The age of infection has been shown to be a determinant of progression from HIV to AIDS and death[26,27]. In many of the most heavily infected countries of sub-Saharan Africa, infection occurs at a much earlier age for women than for men, with the peak age for infection being in the early 20s for women[27]. Even though the population-based cohorts referred to above do include HIV infections in the youngest age groups, the modelling process may benefit from age-group-_specific survival times between men and women.

Another issue related to estimating AIDS-related mortality is the interaction between HIV infection, susceptibility to opportunistic infections, and the background mortality rate. The current estimates of AIDS mortality make no correction for background mortality. In developing countries, adult mortality can be as high as 0.5% per year, so it is likely that some people infected with HIV will die from other causes not related to their infection each year. With HIV infection lasting between 8 and 10 years, on average; during this period up to 5% of those infected may be expected to die from other causes while infected with HIV. This does not have an impact on estimates of HIV incidence in our models, but it does mean that estimates of mortality given here reflect ‚people dying with HIV‚ rather than ‚people dying from AIDS‚.

As indicated in the previous section, there are many obvious needs for improved surveillance and for research that can provide answers to the questions about the assumptions that underlie the estimates of HIV and AIDS. Research that addresses the issues such as how representative antenatal clinic data are for the general adult population, or data on regional or country-specific survival times with HIV could improve greatly the quality of the HIV estimates. Additional data on how the ratio of males to females infected changes with the type and period of the epidemic would also lead to improved estimates.

There is also a need to develop other routine data sources that can be used to calibrate the estimates made from surveillance site data. One possible source is mortality data drawn from country databases. Currently, WHO and UNAIDS are working on comparing predicted mortality drawn from the 1997 HIV estimates to mortality data collected routinely in countries. Although this approach may provide a way to test and refine the estimation procedures in countries with good mortality data, it is more difficult in countries where there are poor systems of death registration[28]. Unfortunately, many of the countries most severely affected by the HIV/AIDS epidemic also have the poorest systems of mortality registration. In these countries, census data are often the only source of mortality data. However, the techniques used in census collection are often very poor at capturing data on household mortality[29]. Although there are techniques that can be used to help attenuate these problems (e.g. the use of sibling histories to capture mortality), the effectiveness of using these data to estimate HIV prevalence remains to be determined. In an ongoing effort to address these issues, UNAIDS and WHO in collaboration with the Population Division of the United Nations, have established a reference group that will focus on the measurement of AIDS-related mortality.

Another issue addressed by the reference group relates to how best to measure and present the degree of precision associated with these and future estimates of HIV/AIDS. In this report, most numbers have been rounded so as not to imply a level or precision that is unwarranted. However, even these rounded numbers (e.g. 30.6 million infected worldwide) may imply a level of precision that is not true. One way to present this lack of precision is to calculate and report confidence intervals around the estimates. However, the methods for calculating these intervals are not clear. For whereas statistical uncertainty and confidence intervals may be calculated, the effects of the representativeness of sentinel surveillance sites and assumptions about female to male ratios, among other factors, are not easily translated into statistically correct confidence intervals. A second approach would be to report a range of values, placing the best estimates within upper and lower bounds, based on best case and worst case assumptions. Further work is under way to develop procedures that can portray the level of precision of estimates of HIV/AIDS.

Back to Top | Article Outline


Although accurate global estimates and the tracking of regional prevalence trends are essential components of advocacy for international HIV/AIDS efforts, accurate country-specific estimates are equally important for national advocacy, planning, and evaluation purposes. At present, the capacity in most countries of the world to collect and undertake careful analysis of existing data sets and put them into a consistent local modelling framework is limited. This leaves many countries dependent upon external consultants for their estimates, and limits their ability to modify or determine the implications of these estimates for their future prevention and care needs. To address these issues, the UNAIDS/WHO Working Group on Global HIV/AIDS and STD Surveillance is currently working with many partners to expand existing country epidemiological and behavioural data sets. The working group is also developing new models using the techniques described in this paper, which can be applied by national programmes or co-operating institutions, and should help to build the capacity to make estimations and projections within national programmes. Improvements in both the surveillance systems in countries and our knowledge of the epidemic will allow refinement of the methodologies and increased precision of future estimates.

Back to Top | Article Outline


The authors would like to express their appreciation to the many people who contributed to these country-specific estimates of HIV/AIDS. First and foremost are the National AIDS programme managers and epidemiologists and the members of the regional and global MAP Network. In addition, John Stover of the Futures Group, Policy Project, and Gilles Poumerol, Fernando Zacharias, Jai Narain, Puru Shresta, Paloma Cuchi and Emil Asamoah-Odei, from the WHO regional offices, provided country-specific data and reviewed the work. Ryuichi Komatsu and colleagues of the United States Census Bureau worked on the HIV/AIDS database. Veronique Batter and Françoise Hamers prepared the fact sheets and databases for the European countries. Finally, Ties Boerma, Basia Zaba, Geoff Garnett, Tony Burton, Stefano Lazzari, and Joshua Salomon gave invaluable advice and reviewed the data.

Back to Top | Article Outline


1. Burton AH, Mertens TE. Provisional country estimates of prevalent adult human immunodeficiency virus infections as of the end of 1994: a description of the methods. Int J Epidemiol 1998, 27:101-107.
2. World Health Organisation. Provisional working estimates of adult HIV prevalence as of the end of 1994, by country. Wkly Epidemiol Rec 1995, 70:355-57.
3. Tarantola D, Schwartländer B. HIV/AIDS epidemics in sub-Saharan Africa: dynamism, diversity and discrete declines? AIDS 1997, 11 (Suppl. B):S5-S21.
4. Chin J. Public health surveillance of AIDS and HIV infection. Bull WHO 1990, 68:529-536.
5. Chin J, Lwanga SK. Estimation and projection of adult AIDS cases: a simple epidemiological model. Bull WHO 1991, 64:399-406.
6. United Nations Programme on HIV/AIDS and World Health Organisation. Guidelines on second generation surveillance of HIV infection. UNAIDS/WHO; forthcoming.
7. Zaba B, Gregson S. Measuring the impact of HIV on fertility in Africa. AIDS 1998, 12 (Suppl. 1):S41-S50.
8. Mulder DW, Nunn AJ, Wagner HU, Kamali A, Kengeya-Kayondo JF. HIV-1 incidence and HIV-1 associated mortality in a rural Ugandan population cohort. AIDS 1994, 8:87-92.
9. Hendriks J, Medley GF, van Griensven GJ, Coutinho RA, Heisterkamp SH, van Druten HA. The treatment-free incubation period of AIDS in a cohort of homosexual men. AIDS 1993, 7:231-239.
10. Rutherford GW, Lifson AR, Hessol NA, et al. Course of HIV-1 infection in a cohort of homosexual and bisexual men: an 11 year follow-up study. BMJ 1990, 301:1183-1188.
11. Turner BJ, Eppes S, McKee LJ, Cosler L, Markson LE. A population-based comparison of the clinical course of children and adults with AIDS. AIDS 1995, 9:65-72.
12. Morgan D, Malamba SS, Maude GH, Okongo MJ, Wagner H-U, Mulder DW, Whitworth JA. An HIV-1 natural history cohort and survival times in rural Uganda. AIDS 1997, 11:633-640.
13. Nunn AJ, Mulder DW, Kamali A, Ruberantwari A, Kengeya-Kayondo J-F, Whitworth J. Mortality associated with HIV-1 infection over five years in a rural Uganda population: cohort study. BMJ 1997, 315:767-771.
14. Auger I, Thomas P, De Gruttola V, et al . Incubation periods for paediatric AIDS patients. Nature 1988, 336:575-577.
15. Commenges D, Alioum A, Lepage P, Van de Perre P, Msellati P, Dabis F. Estimating the incubation period of paediatric AIDS in Rwanda. AIDS 1992, 6:1515-1520.
16. Downs AM, Salamina G, Ancell-Park RA. Incubation period of vertically acquired AIDS in Europe before widespread use of prophylactic therapies. AIDS 1995, 9:297-304.
17. Ryder R, Nsa W, Hassig S. Perinatal transmission of HIV-1 to infants of seropositive women in Zaire. N Engl J Med 1989, 320:1637-1642.
18. United Nations Secretariat, Department of Economic and Social Affairs, Population Division. World Population Prospects: The 1998 Revision. Diskettes United Nations, New York.
19. Whitworth J. Reduced fertility associated with HIV-1: the _contribution of pre-existing sub-fertility. Proceedings from HIV, STD and Infertility: Past Trends and Current Monitoring Problems. Arlington, Virginia, 14-15 December 1998.
    20. Gray RH, Wawer MJ, Serwadda D, et al. Population-based study of fertility in women with HIV-1 infection in Uganda. Lancet 1998, 351:98-103.
    21. Fylkesnes K, Ndhlovu Z, Kasumba K, Mubanga Musonda R, Sichone M. Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia. AIDS 1998, 12:1227-734.
    22. Berkley S, Naamara W, Okware S, et al. AIDS and HIV infection in Uganda - are more women infected than men? AIDS 1990, 4:1237-742.
    23. Carre N, Deveau C, Belanger F, et al. Effect of age and exposure group on the onset of AIDS in heterosexual and homosexual HIV-infected patients. AIDS 1994, 8:797-802.
    24. Fontanet A, Messele T, Dejene A, et al. Age and sex-specific HIV-1 prevalence in the urban community setting of Addis Ababa. AIDS 1998, 12:315-322.
    25. Kahindo M, Nyang J, Chege J. Muticentre study on factors determining the differential spread of HIV in Africa - preliminary results of the Kisumu site study (biomedical data). 2nd National HIV/AIDS/STDs Conference. Nairobi, Kenya, 28-30 October 1998.
    26. Rosenberg PS, Goedert JJ, Biggar RJ. Effect of age at seroconversion on the natural AIDS incubation distribution. AIDS 1994, 8:803-810.
    27. US Bureau of the Census, Population Division, International Programs Center. HIV/AIDS Surveillance Data Base. Washington, DC: January 1997.
    28. Garenne ML, Madison M, Tarantola D, Zanou B, Aka J, Dogore R. Mortality impact of AIDS in Abidjan, 1986-1992. AIDS 1996, 10:1279-786.
    29. Timaeus IM. Measurement of adult mortality in less developed countries: a comparative review. Population Index 1991, 57:552-568.

    AIDS; HIV; modelling; prevalence estimates

    © 1999 Lippincott Williams & Wilkins, Inc.