John Stover

The Futures Group International

Prepared for

AIDS in Development: The Role of Government

Château de Limelette, 17-19 June 1996

Revised February 1997

Ever since AIDS was recognized as a critical global health problem that would lead to increased adult and child mortality there have been debates about the demographic impact of AIDS. There has been speculation that the impact of AIDS might be so large as to cause negative population growth rates in some countries. Much of this debate centered on Africa, where HIV prevalence rates are the highest.

Some researchers concluded that:

Others found that

A review of the different approaches to estimating the demographic impact of AIDS conducted in 1993 (Stover, 1993) found that most researchers used similar methodologies. The opposite conclusions where due to different views about future levels of HIV prevalence in Africa. Those who felt that prevalence would continue to increase to very high levels found severe demographic impacts while those projecting more moderate prevalence levels in the future found that the demographic impact would be significant but would not lead to negative population growth in the African context.

In 1993 and 1994 three major institutions released country-specific population projections that, for the first time, included the impact of AIDS. None of these studies projected negative population growth in Africa as a result of AIDS. However, other results were quite different, especially for certain countries.

• The United Nations included the impact of AIDS in its 1992 and 1994 editions of World Population Prospects (United Nations, 1993 and United Nations, 1995) and will include AIDS in its new 1996 projections. In the 1994 revision the UN reported that:
  • AIDS would reduce the projected population of 15 countries in sub-Saharan Africa by about 4 percent (12 million) people by 2005.
  • The additional number of deaths due to AIDS would reach 9.7 million by 2005.
  • The most severely affected countries, Zimbabwe and Zambia, would have seven percent fewer people in 2005 than they would have without AIDS.
  • The average annual population growth rate in Zambia for the 2000-2005 period would be reduced from 2.4 percent without AIDS to 0.5 percent with AIDS.
  • Life expectancy for Zambia and Zimbabwe would be reduced by 22 percent by 2000-2005, from 59.7 to 46.5 for Zambia and from 65.9 to 51.1 for Zimbabwe.

• The United States Bureau of the Census included the impact of AIDS in its 1994 and 1996 sets of population projections for the countries of the world (McDevitt, 1996; Jamison, 1994). Its projections showed a much more dramatic impact of AIDS on population growth than those of the United Nations. The Census Bureau reported that:
  • By 2010, 66 million fewer people are expected in the 23 countries with the most severe epidemics.
  • Life expectancy in Botswana would be reduced by 50 percent (from 66 to 33). For all 23 countries, life expectancy would be 20 percent lower in 2010 than it would be without AIDS.
  • Population growth rates would remain positive in all countries but would be reduced significantly due to AIDS. For all 23 countries, the rate of natural increase in 2010 is projected to be 1.6 percent per year, rather than the 2.2 percent that would be projected without AIDS.

• The World Bank included the impact of AIDS in its 1994-95 report (Bos, 1994). Its projections show a smaller impact for AIDS than either the Census Bureau or the United Nations. The World Bank found that:
  • The population of sub-Saharan Africa would be reduced by about 9 million people by 2005 from what it would have been without AIDS.
  • Life expectancy in Uganda would be reduced by at most 15 percent (from 52 to 44 years) by 2005.
  • Annual population growth rates for all of sub-Saharan Africa would decline by at most 0.15 percent.

Why are the results so different? At first glance it may seem that the differences are due to different reporting dates and countries. For example, one institution reports figures for 2005 while another reports for 2010. However, upon closer inspection, it becomes clear that large differences exist even when projections are compared for the same year and the same countries.

This paper will first describe the different approaches to incorporating AIDS into demographic projections. Next it will compare the results at both the country and regional level. Then it will examine the reasons for the differences. Finally it will present some thoughts about the best set of assumptions for future projections.

The Population Division of the Department for Economic and Social Information and Policy Analysis of the United Nations Secretariat prepares the official population projections of the UN every two years. The 1992 Revision was the first to incorporate AIDS [UN, 1993] . The impact of AIDS was included for all countries with an estimated adult HIV prevalence rate of more than one percent. Estimates of HIV prevalence were provided by the Global Programme on AIDS (GPA) of the World Health Organization. As a result, AIDS was incorporated into the projections for 15 countries, all in sub-Saharan Africa: Benin, Burkina Faso, Burundi, Central African Republic, Congo, Côte d’Ivoire, Kenya, Malawi, Mozambique, Rwanda, United Republic of Tanzania, Uganda, Zaire, Zambia and Zimbabwe. For the 1994 Revision [UN, 1995] the projections for Thailand also included the impact of AIDS.

The UN population projections are made using a standard cohort component projection model developed by the UN called ABACUS. The AIDS projections were prepared using GPA’s Epi Model [Chin and Lwanga, 1991] and then added to ABACUS through modified death rates [UN, 1994]. Epi Model projects the past and future course of an AIDS epidemic based on three key assumptions: the year in which HIV infection first became widespread, the number of people alive with HIV infection in the current year, and the shape of the infection curve. The model allows the user to select a curve type to describe cumulative HIV infections over time. The UN projections assume a gamma curve (a type of S-shaped curve). A gamma curve is fitted to two points: zero infections the year before HIV infection became well established in a core group and the current estimate of infections. The user decides where on the gamma curve the current year lies. If the user decides that the epidemic is still in its early stages, then the point representing the current year would be placed in the early part of the S-curve, leaving the most rapid increase in infections to occur in the future. If the user decides that HIV incidence is currently at its peak, then the current year estimate would be placed right in the middle of the S-curve. Similarly, if it is assumed that the epidemic has reached the endemic stage, then the current year estimate would be placed near the top of the S-curve. Thus, the assumption about the current stage of the epidemic, largely determines the future projection. The UN projections assume that the peak incidence rate, the middle of the gamma curve, is reached 12 years after the beginning of the epidemic. Figure 1 illustrates how EpiModel might be used to project future incidence and prevalence in Kenya if the number of infections is estimated to be one million in 1994.

Epi Model also requires an assumption about adult and child incubation periods. The incubation period is the number of years from infection with HIV until the development of AIDS. This is usually described as the percentage of people newly infected with HIV that develop AIDS in each subsequent year.

Once these assumptions are made, the Epi Model projection process follows these steps:

1. Read the number of adults alive with HIV infection from the gamma curve for a particular year
2. Calculate the number of new adult HIV infections required to reach the total number alive with HIV infection in that year (by subtracting the number of infections in the previous year from the number of infections in the current year and adding the number of AIDS deaths during the past year)
3. Use the assumption about the incubation period to determine when people with a new HIV infection will develop AIDS and die
4. Calculate the number of child deaths from AIDS based on assumptions about the crude birth rate, the perinatal transmission rate of HIV and the incubation period for child infections.

From these steps, it can be seen that the assumptions about the average length of the incubation period and the length of the period from AIDS until death are also quite important. If the incubation period is assumed to be relatively short, then people infected with HIV will die soon, thus requiring a higher rate of new HIV infections to achieve the assumed number of people alive with HIV infection at any given time. A longer incubation period means that people infected with HIV live longer, thus requiring fewer new infections to achieve the same number of total HIV infections at any given time.

Epi Model is used to determine the number of AIDS-related deaths by year. These deaths are then distributed by age and sex according a typical pattern of AIDS deaths. For sub-Saharan African countries this pattern typically shows roughly the same number of male and female deaths with the largest number of deaths in the age group 35 to 45 for males and about five years younger for females. This pattern of AIDS deaths is then added to the deaths calculated from all other causes by the ABACUS model to produce the final population projections.

From this brief description, it is apparent that five assumptions are key to estimating the demographic impact of AIDS:

1. the year in which HIV infection became widespread
2. that the peak rate of HIV incidence will occur 12 years after HIV infection becomes widespread
3. the number of HIV infections in the current year
4. the duration of the incubation period
5. the perinatal transmission rate.

In East and Central Africa, HIV infection is generally assumed to have become widespread in the late 1970s and early 1980. Thus 1980 is a typical value for the first year of the epidemic. This implies that HIV incidence would reach a peak for most countries in the early 1990s.

The number of HIV infections in the most recent year was taken from GPA estimates. These estimates are based the results of surveys of various population groups that test blood to determine HIV infection. Most of these surveys are one time studies undertaken by different research groups. In some cases, sentinel surveillance systems provide annual estimates of HIV infection among certain population groups.

The UN estimates use the assumption that the average length of time from HIV infection until AIDS is about 10 years. In addition, the perinatal transmission rate is assumed to be 30 percent.

Until 1994, the World Bank prepared population projections for all the countries in the world for use in World Bank projects and analyses. The most recent projections [Bos, 1994] included the impact of AIDS. The World Bank approach differs from the UN approach, primarily in the method it uses for projecting the number of new HIV infections. These projections are based on a model prepared by Rodolfo Bulatao [Bos and Bulatao, 1992]. The Bulatao model simulates the spread of HIV through a population based on the behavior and characteristics of different population sub-groups and assumptions about key epidemiological parameters, such as the probability of transmitting the virus in a single unprotected contact. The model considers HIV transmission through heterosexual and homosexual contact, blood transfusions, infected needles, and perinatal transmission. This model was used to simulate a variety of epidemics that included a range of typical values for starting HIV prevalence in 1990 and included various levels of interventions as well. Then, regression analysis was used with the results to develop a set of equations to project future levels of HIV prevalence based on the 1990 level. A similar set of equations was developed to project life expectancy at age 10 as a function of adult HIV prevalence. The demographic projections involve the following steps:

1. Levels and patterns of mortality in the absence of AIDS are projected and the most appropriate Coale-Demeny life table is selected
2. HIV prevalence in future years is projected on the basis of estimated prevalence in 1990 and the assumed starting year of the epidemic. Incidence is assumed to decline by 50% each year after 2005.
3. The number of years of life expectancy lost due to AIDS is projected using the regression equations relating adult prevalence to life expectancy.
4. The number of years of life expectancy lost is subtracted from the no-AIDS trends until 2020-2025. After that period, AIDS mortality is assumed to decline to zero by 2050.

The 1994-95 projections were based on estimates of adult HIV prevalence by country prepared by GPA [Chin, 1991]. The projections assume an incubation period that averages 10 years and a perinatal transmission rate of 30%.

The US Census Bureau prepares demographic projections for all the countries of the world every two years. The most recent projections, prepared in 1996, include the impact of AIDS [McDevitt, 1996; Jamison, 1994]. Like the UN and World Bank projections, the Census Bureau used an external AIDS model to determine the number of AIDS deaths and then incorporated these estimates in to its urban/rural demographic projection model. Census used the iwgAIDS model, a complex simulation model of the spread of HIV through a population as a result of the behavior of various population sub-groups [Stanley, 1991]. This model was used to simulate three different African epidemics. In the low scenario HIV prevalence among adults aged 15-49 increases slowly reaching only about 5 percent after 45 years. In the medium scenario prevalence increases to about 17 percent after 35 years and then stabilizes. In the high scenario prevalence increases to about 37 percent after 45 years. This pattern is illustrated in Figure 2.

In order to project HIV prevalence for an individual country, estimates of prevalence for two historical years were prepared. The rate of increase in prevalence between these two years was compared to the rates of increase in the three simulated scenarios for the corresponding stage of the epidemic. This comparison was used to interpolate a new prevalence curve from the simulated scenarios that matched the historical experience. This interpolated curve provided the prevalence projection. Prevalence is assumed to peak in 2010, with no new infections occurring after that date. A similar interpolation procedure was used to determine age and sex-specific mortality rates.

The scenarios developed using the iwgAIDS model assumed an incubation period with an average duration of 7.5 years and a perinatal transmission rate of 39 percent. Initial year HIV prevalence assumptions were based on examination of the AIDS database [US Census Bureau, 1992].

Population projections incorporating AIDS were made for all countries with HIV prevalence above 5 percent for adults 15-49 in urban areas. This criterion selected 19 African countries (Botswana, Burkina Faso, Burundi, Cameroon, Central African Republic, Congo, Côte d’Ivoire, Ethiopia, Kenya, Lesotho, Malawi, Nigeria, Rwanda, South Africa, Tanzania, Uganda, Zaire, Zambia, Zimbabwe), Guyana and Haiti. Brazil and Thailand were also included for other reasons.

In 1995 John Bongaarts of The Population Council prepared a set of projections showing the demographic impact of AIDS by geographic region [Bongaarts, 1995]. These projections differ from the others discussed here because they are not country-specific and they present results for all the regions of the world, not just Africa. These projections are included in this paper because they produce similar results to those prepared by the UN and World Bank and can be used to illustrate the effects of different assumptions on the future projections more easily than the projections involving multiple countries.

Bongaarts starts with the GPA estimates of adult HIV incidence rates from 1980 to 1995 by major region of the world. He then makes several assumptions about the future course of incidence. In the medium projection, he assumes that incidence remains at the 1995 level through 2005.

Once HIV incidence is known, the Bongaarts model calculates new AIDS cases using an assumed incubation period with a median length of 9.5 years. The median time from AIDS to death is assumed to be 0.5 years. Child mortality due to AIDS is calculated from the crude birth rate and perinatal transmission rate.

This section compares the various projections by examining several key indicators. Areas of similarity are pointed out, but the focus is on describing the differences and, later, on the reasons for those differences. It should be noted, however, that all four sets of projections agree on one key point: AIDS will not cause negative population growth in any country in sub-Saharan Africa.

The following comparisons use the 1994 projections from each organization unless stated otherwise. (The World Bank did not produce projections in 1996 and the United Nations projections for 1996 have not yet been published.)

The demographic projections produced by the four different organizations discussed above are similar in many respects but there are also some striking differences. Before examining the projected demographic impact of AIDS it is interesting to compare the demographic projections without the influence of AIDS. Table 1 shows the projections of total population size from the US Census Bureau, the United Nations and the World Bank from 1990 to 2025 in the absence of AIDS. There is good agreement, although not perfect, on the estimate of population size in 1990. Even by 2010, the three projections show reasonably close agreement for most countries. The projections do diverge more by 2025, largely due to differences in assumed rates of fertility decline. For the entire set of 13 countries the UN and Census Bureau differ by only 4 percent by 2025. Even the World Bank projections are only 12 percent lower than the Census projections in 2025.

Table 2 presents a comparison of the population projections including the impact of AIDS. It is interesting to note that, in the aggregate, the agreement is good. The World Bank and Census Bureau differ by only one percent and the UN is only 16 percent higher by 2025. However, the aggregate figures mask some large differences at the country level. There are particularly large differences for the Congo. Kenya, Zimbabwe and the Central African Republic. Of course, the interpretation of these differences is complicated by the fact that they are the combined result of differences in demographic and AIDS assumptions.

For the 1996 projections, there are changes for some countries, but the results are quite similar in the aggregate. For all the countries listed in Table 2, the UN projections for 2020 are 16 percent higher than those from the US Census Bureau.

Table 3 presents the net change in the population projection as a result of AIDS. This change will result from AIDS deaths to adults and children as well as a reduced number of births due to a smaller reproductive population. None of the projections assumed any connection between HIV prevalence and fertility rates, so the only affect on births is through a reduction in the number of reproductive age women.

Table 3 shows very large differences in both 2010 and 2025 between the US Census Bureau projections and the UN and World Bank projections. With only a few exceptions, the differences between the Census Bureau projections and those by the World Bank and UN are larger than the differences between the UN and World Bank. Furthermore, there is a consistent pattern, the Census Bureau projection show a much larger effect of AIDS on population size than either the UN or World Bank. In fact, the Census projections show an impact two to three times larger.

Table 1. Comparison of Population Projections with No AIDS (Millions)

Table 2. Comparison of Population Projections with AIDS (Millions)