NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Jamison DT, Feachem RG, Makgoba MW, et al., editors. Disease and Mortality in Sub-Saharan Africa. 2nd edition. Washington (DC): World Bank; 2006.

Cover of Disease and Mortality in Sub-Saharan Africa

Disease and Mortality in Sub-Saharan Africa. 2nd edition.

Show details

Chapter 6Population and Mortality after AIDS

.

The acquired immune deficiency syndrome (AIDS) affects population size and composition in several ways. In particular age groups, deaths increase directly from AIDS and may also increase indirectly, as orphans, for instance, face higher mortality risks. Fertility can be affected, not only biologically but also from changes in sexual behavior. Communities may be weakened, and migration may alter the geographic distribution of the population.

This chapter draws on the work of agencies that produce global population projections to discuss the overall population effect of AIDS in Sub-Saharan African countries. It does not attempt to elucidate all the mechanisms involved—each of which deserves separate attention. Instead, the chapter focuses on the broad demographic impact. Decades after the start of the epidemics, estimates of this impact generally ignore all subtle and indirect effects and cover only the additional mortality directly from AIDS. Even this impact is highly uncertain. There is general agreement on substantial impact but no consensus on how substantial and long lasting it is.

Approach and Data

The projections of demographic impact to be considered come from three agencies: the Population Division of the United Nations (UN), the U.S. Census Bureau, and the World Bank. Each agency has produced population projections for most countries for some time. Up to 50 Sub-Saharan countries or territories are covered, although the smallest ones are left out in some data series. For at least a decade, these agencies have explicitly incorporated the effect of AIDS in selected countries. In general, mortality due to AIDS is added to a life table for a given country, and then overall mortality is projected into the future together with other vital parameters in order to project population. Possible effects of AIDS on fertility (for example, Zaba and Gregson 1998) and migration, except for indirect effects of mortality change, are not modeled.

The projections vary not only because of the mortality assumptions but also because of assumptions about initial population size and composition and fertility and migration trends. These differences are not detailed here. A recent description and results are in United Nations 2003a, 2003b, 2004; for the other two sets of projections, recent descriptions are difficult to find, but one might consult earlier descriptions in McDevitt 1999, Bos and colleagues 1994, and National Research Council 2000. Results are in Stanecki 2004 and World Bank 2004.

Although the approach to incorporating AIDS mortality in the projections is similar across agencies, important details are different. Agencies differ in how initial levels of mortality are estimated and projected and how AIDS mortality is projected.

United Nations Population Division

The Population Division of the UN seeks to apply country life tables where available but mostly uses UN model life tables. These are projected with reference to life expectancy, with gains expected to diminish as life expectancy rises. In the medium scenario (the only one considered here), the annual gain is 0.4 years for males and 0.5 years for females from a life expectancy level for both sexes combined of 60 years, but the gain shrinks to 0.1 years for males and 0.2 years for females when life expectancy reaches 75 years.

AIDS mortality is added to the life tables for 38 of the 50 countries or territories in Sub-Saharan Africa. The remainder, with reported low prevalence of the human immunodeficiency virus (HIV) or no data from UNAIDS, are island countries, except for Mauritania, Niger, Senegal, and Somalia. A criterion of HIV prevalence no higher than 1 percent appears to be used in excluding countries, but the size of the affected population is also taken into account. (This is noticeable mainly outside Sub-Saharan Africa, because AIDS mortality is also added for China.)

To add AIDS mortality, a separate projection of HIV/AIDS in the population is necessary. The procedure begins with reported levels of HIV prevalence among adults. With appropriate assumptions, the UN back-calculates the incidence of HIV and distributes cases by sex and age. From the size of the infected population and the remaining, susceptible population, the UN estimates the trend in subsequent infections. It derives mortality associated with AIDS given assumptions about the progression of the infection and adds additional mortality from perinatal infections. Life tables are adjusted for the additional mortality.

The Population Division has described the procedure for estimating AIDS mortality, which follows the recommendations of the Joint United Nations Programme on HIV/AIDS (UNAIDS) Reference Group on Estimates, Modelling and Projections (UNAIDS Reference Group 2002), in World Population Prospects: The 2004 Revision (United Nations 2004, 136–79). Calculations start with estimates of HIV prevalence over time, mainly estimates for pregnant women from antenatal clinics. The future trend of AIDS mortality is basically an extrapolation, with appropriate adjustments, from this trend. HIV prevalence estimates as of 2001 are taken from UNAIDS Reference Group 2002. For Djibouti, Gabon, Guinea, and Liberia, UNAIDS (2002) reports no estimate, and the UN Population Division makes its own. The source of estimates prior to 2001, used to establish a trend, is not specified, but is presumably the database maintained by the U.S. Census Bureau and similar sources. Although an attempt is made to match the UNAIDS (2002) estimate for 2001, fitting a trend sometimes leads to an HIV prevalence figure that varies from it, by a maximum of 2 percentage points (for Botswana) or 12 percent (for Uganda).

U.S. Census Bureau

The U.S. Census Bureau follows a similar but not identical procedure (Stanecki 2004). Usually, a life expectancy target is set in the future without considering AIDS, and logistic curves represent the trend to this target, with appropriate life tables generated to match it. AIDS mortality is added for 37 Sub-Saharan countries. In contrast to the countries covered by the UN Population Division, the Census Bureau does not introduce AIDS-specific deaths for Equatorial Guinea, The Gambia, and Sudan but does cover Niger and Senegal. The criterion for inclusion is national HIV prevalence above 1 percent (as estimated by UNAIDS for 1999), HIV prevalence of 5 percent or more in low-risk urban populations, or a prevalence trend that suggests the latter level will soon be reached.

Instead of modeling the AIDS epidemic in each country separately, the bureau takes a shortcut. It defines five epidemic scenarios, progressively more severe, ranging from a "low" to a "super high" epidemic, each starting at an indefinite date. These scenarios are developed using a program labeled iwgAIDS (Stanley et al. 1991), which models the spread of HIV infections, the development of AIDS, and subsequent deaths in a population. Each scenario incorporates the effect of increased condom use and shows HIV prevalence eventually plateauing and declining.

Each country is assigned to a scenario, or more precisely to an interpolated scenario between an assigned pair of these five. The assignment is made by matching the reported trend in HIV prevalence in urban areas to scenario trends. Total country prevalence is then used to determine dates for the epidemic. It is assumed that HIV prevalence will peak in 2010 and AIDS mortality will decline to zero by 2070. AIDS deaths are therefore added to the life table, from the projected epidemic, up to 2070, at which time the life table reverts to what it would be without the epidemic.

Uganda is treated as a special case, because of a reported decline in prevalence from the mid-1990s, which violates the assumption of a 2010 peak. Separate epidemics are modeled: a high one up to 1995 and a low one from 2005 on, with an interpolated epidemic in the intervening period.

The initial HIV national prevalence levels used are similar to those adopted by UNAIDS, relying heavily on reported infections among pregnant women. In this, the bureau projections resemble those of the UN Population Division. However, the future trend imposed on prevalence is unusual. The year 2010 represents a later peak to the epidemics than projected by the UN, in most cases. Peak incidence, in 55 percent of the Sub-Saharan cases in the UN projections, is before 1995, and in 95 percent of the cases before 2000. Although prevalence tends to peak later than incidence, this is unlikely to adequately account for the long delays between UN incidence peaks and Census Bureau prevalence peaks.

World Bank

World Bank mortality projections begin with life tables that already incorporate AIDS. However, these life tables, developed by the World Health Organization (WHO) for 2000, are constructed in a familiar way, with AIDS mortality being added to mortality from all other causes.

The WHO takes estimates of child mortality (5q0) and adult mortality (45q15) and expands them into life tables for each country, using the Brass logit approach (Murray et al. 2000; Lopez et al. 2001). The child mortality estimates were selected from survey and census estimates after systematic review of available statistics (Ahmad, Lopez, and Inoue 2000). The adult mortality estimates were more problematic. For Africa, plausible national estimates in the literature prior to 1990 (to exclude the effect of AIDS) were used, supplemented where necessary with UN estimates (United Nations 1998), and projected forward to 1999 and, later, 2000.

To add AIDS mortality, the WHO begins with reported HIV prevalence (based, as usual, on pregnant women) and works backward to HIV incidence and then forward to obtain an overall AIDS mortality level. (The researchers caution, based on preliminary work for Zimbabwe, that "these models may overestimate the level of the epidemic" [Salomon, Gakidou, and Murray 1999, 8], a point we return to below.) An age pattern of deaths, based on limited data from Tanzania, South Africa, and Zimbabwe (Lopez et al. 2001), is imposed, and AIDS mortality is then added to the life table.

In projecting mortality forward from the initial life tables taken from the WHO, the World Bank does not attempt to model the HIV/AIDS epidemics in any way, but simply assumes that their effect will gradually be reduced to zero by 2020 at the latest (depending on the country), at which time country mortality trends resume a standard pattern used in the Bank's projections.

Comparisons

Counterfactual projections that include no effect of AIDS are useful in assessing impact. Only the UN Population Division provides detailed results for such a scenario. The bureau has run such scenarios and provides brief descriptions but quite limited numerical information on the results (Stanecki 2004). The World Bank has no such scenario. Most comparisons to be made, therefore, will rely on the UN no-AIDS scenario.

It is not obvious from the approach taken in each projection set which should show the greatest current impact of AIDS on population and mortality. For future impact, however, a possible order emerges. Because the Census Bureau assumes a relatively late peak to the epidemics, one might expect greater demographic impact in its projections. Because the World Bank assumes a fairly direct decline in mortality impact, with the elimination of this impact by 2020 instead of the Census Bureau's projection of 2070, one might expect these projections to show the least future impact. This depends crucially on specific parameters, however, and the results of comparisons need not be uniform across countries.

Effects on Population Size, Decomposed

Population trends taking AIDS into account are illustrated for 2000–50 in figure 6.1, which compares results from the three agencies. An additional line represents the special UN scenario with no AIDS mortality. For Sub-Saharan Africa as a whole, the three projections of total population are well below the no-AIDS scenario practically from 2000 and fairly close to each other up to about 2015. By 2020, some divergence appears, with the Census Bureau projection 0.1 percent above the UN projection and the World Bank projection 2.5 percent below it. Divergence increases as the projections lengthen, so that by 2050 the Census Bureau projection is 8.6 percent above the UN projection and the World Bank projection is 9.4 percent below. This is the reverse of the relative impact one would expect from comparing methodologies.

Figure 6.1

Figure 6.1

Projected Population, Sub-Saharan Africa and Three Selected Countries (thousands) Sources: Stanecki 2004; United Nations 2004; World Bank 2004.

Three countries are shown in figure 6.1: Nigeria, with the largest regional population and an HIV prevalence rate intermediate for the region (5.4 percent infected of adults 15 to 49 years old, as reported by UNAIDS [2004] for 2003); Benin, with low HIV prevalence, for the region, of 1.9 percent in 2003; and Swaziland, with one of the highest HIV prevalence rates, for the region or the world, of 38.8 percent in 2003. For these individual countries, greater divergence in projections is evident than for the region as a whole, some of it clearly not due to AIDS mortality. The Census Bureau, for instance, projects a larger population than in the no-AIDS projection for Nigeria by 2035, as well as a larger population for Benin all the way back to 1990. Although AIDS does affect population size, its effects are entangled with, and may be overwhelmed by, assumed change (or lack of change) in other parameters, particularly fertility.

Fertility assumptions are indeed quite variable across the projection sets. The UN projection and the no-AIDS scenario, but not the other two projections, have similar total fertility rates by design. World Bank estimates of total fertility in Sub-Saharan countries for 2000–05 range from 21 percent below to 23 percent above UN estimates, with a tendency to be lower rather than higher. Census Bureau estimates, in contrast, range from 37 percent lower to 53 percent higher than UN estimates, with a decided tilt toward being higher. The variation tends to increase, in percentage terms, for later years, at least up to around 2025. Higher bureau and lower Bank fertility projections appear to counteract the effect on projected population of longer-lasting epidemics for the Census Bureau than for the Bank.

Ideally the effect of fertility assumptions would be discounted by running parallel scenarios with and without AIDS, as the UN has done. Such scenarios are not available, at least not in sufficient detail, for the other two sets of projections, but we do attempt to separate the effects of the fertility and mortality assumptions. We decompose the ratio of populations from two separate projections into multiplicative factors representing differences in the assumed base population and in fertility, mortality, and migration assumptions.1 This decomposition does not specifically identify AIDS impact, but it at least helps us separate out the fertility and mortality effects.

Table 6.1 shows the results for 2020 and 2050. The comparison between the UN projection and the no-AIDS scenario (also from the UN) has some interesting points. The projected total population for 2020 is 89 percent of the projection without AIDS (as the first "Total" column shows). Therefore, the overall reduction in population due to AIDS is 11 percent by 2020, 17 percent by 2050. Most of this is due to mortality, but other effects also contribute. Part is due to differences in the base population, the population assumed for the start of the projection in 2000. The UN estimates that AIDS had already produced some demographic effect before 2000, reducing regional population by over 2 percent. The largest such effect across countries is 9 percent, for Zimbabwe.

Table 6.1. Effects of Demographic Factors in Producing Differences between Projections.

Table 6.1

Effects of Demographic Factors in Producing Differences between Projections.

Births also have an effect—a positive one in the UN projection, in contrast to the no-AIDS scenario. The effect is small for the region as a whole, at 1.4 percent in 2020, rising to 2.6 percent by 2050. It is, however, positive for each country where some effect of AIDS is modeled. Since the same total fertility is assumed country by country in the UN projection and its no-AIDS scenario, this result deserves some explanation.

Crude birth rates differ between the projections, as figure 6.2 illustrates. For the region as a whole, the difference is relatively slight, but for countries severely affected by AIDS, the difference can be much larger, as in the projection for Botswana. AIDS raises the crude birth rate because of its effect on the age structure. Figure 6.3 shows the distribution by age group of women of reproductive age in Botswana. Whereas, without AIDS, the age distribution is projected, over 50 years, to gradually become rectangular over the range of 15 to 49 years, with AIDS it remains skewed toward those under 35. This translates into a higher crude birth rate because fertility is higher among women under 35 years. The fertility advantage of younger women is assumed to decline over time, but the decline does not entirely offset the relative increase, when AIDS is taken into account, in proportions of younger women. The effect is entirely distributional and does not involve any assumptions about the biological impact of HIV, which, if it were taken into account, might moderate the result somewhat. AIDS does not increase the number of births, but these results indicate that it does increase the ratio of births to the population.

Figure 6.2

Figure 6.2

Projected Crude Birth Rate, Sub-Saharan Africa and Botswana Sources: Stanecki 2004; United Nations 2004; World Bank 2004.

Figure 6.3

Figure 6.3

Distribution of Women of Reproductive Age in the UN Projection and the No-AIDS Scenario, Botswana Source: United Nations 2004.

The Census Bureau also shows a generally positive births effect. Part of this, as for the UN, may be due to the effect of AIDS on the age structure, but a larger part is simply higher assumed fertility than in the no-AIDS scenario. As expected, the births effect is negative for the World Bank.

The direct effect of mortality on population size in the projections is of course negative and much larger in the absolute than the effect of fertility, implying a reduction in population size for the region, if the UN projection is compared with the no-AIDS scenario, of 10 percent by 2020 and 8 percent by 2050. For the region as a whole, the mortality effect is roughly comparable, if the U.S. Census Bureau projections or the World Bank projections are compared with the same standard. We expected the Census Bureau to show a stronger mortality effect than the World Bank, given the assumption of longer-lasting epidemics, but the reverse is the case, except when the most affected countries are considered.

Figure 6.4 shows the mortality effects for individual countries—with the most affected countries toward the lower left—when assessed for 2020 and for 2050. Although the agencies generally agree about which countries are more affected by AIDS mortality, their estimates of the degree of this effect vary, particularly for the most affected countries. For countries where the UN estimates that AIDS mortality reduces population in 2020 by at least 20 percent, the Census Bureau estimates that the population reduction on average is 5 percentage points less, and the World Bank estimates that it is 8 percentage points less. For countries where the UN estimates a population reduction of at least 20 percent by 2050, the Census Bureau estimates an average reduction that is 3 percentage points less, and the World Bank a reduction of 13 percentage points less.

Figure 6.4

Figure 6.4

Mortality Effects on Population, Relative to the No-AIDS Scenario, in Different Projections Sources: Calculated from Stanecki 2004, United Nations 2004, and World Bank 2004.

These comparisons involve not only AIDS mortality but mortality generally. It is likely, however, that the differences have to do mainly with variation in AIDS mortality than in mortality from all other causes, although this cannot be established. At any rate, the results fail to substantiate any general agreement about the demographic effect of AIDS mortality.

Life Expectancy

Looking at other mortality indexes does not help further specify the AIDS effect but does provide additional perspectives on the severity of the impact. Figure 6.5 shows some projected life expectancy trends. For Sub-Saharan Africa as a whole, the UN Population Division indicates that by 2000–05 life expectancy had already fallen 8.8 years short of what it would be without AIDS and that the relative deficit will grow to a maximum of 10.4 years by 2010–15 before beginning to shrink slowly. The Census Bureau suggests that the current deficit is slightly smaller, and the World Bank suggests it will shrink slightly faster. For the region as a whole, there is not great disagreement about life expectancy trends.

Figure 6.5

Figure 6.5

Projected Life Expectancy, Sub-Saharan Africa and Three Selected Countries Sources: Stanecki 2004; United Nations 2004; World Bank 2004.

For individual countries, however, the differences among projections can be substantial. The UN projections start (in 2000) with a much higher estimate of life expectancy for Nigeria than the other two projections, a gap that is generally maintained over time. For Benin, the Census Bureau projects a substantial fall in life expectancy, unlike the other two agencies. For Swaziland, the decline in projected life expectancy up to about 2010–15 is greatest for the UN, which has it falling to 30 years. The decline is smallest for the World Bank, for which the minimum, in 2005–10, is 43 years. Subsequent trends for Swaziland are also quite different: the UN projects a long, slow recovery; the World Bank sees a quick, rapid recovery; and the Census Bureau finds a delayed but accelerating recovery.

How far life expectancy falls depends on current estimates of HIV prevalence. Figure 6.6 shows the maximum fall in life expectancy by country, when each set of projections is contrasted with the no-AIDS scenario. In the UN projection, a country loses four years of life expectancy when HIV prevalence is 2.5 percent among adults, and a little more than one additional year of life expectancy for every 1 percentage point rise in HIV prevalence.2 The U.S. Census Bureau projects a similar result (when the comparison is made to the UN no-AIDS scenario), and the World Bank projects a somewhat less serious effect, particularly at high HIV levels. The greatest effect on life expectancy is most often seen in the period 2015–20. For the UN, the period may be five years earlier or later, but for the Census Bureau, there is less temporal variation. For the World Bank, however, the period of maximum deficit is 2005–10, and this is fairly constant across countries.

Figure 6.6

Figure 6.6

HIV Prevalence and Maximum Loss in Life Expectancy in Alternative Projections Sources: Calculated from Stanecki 2004, United Nations 2004, and World Bank 2004.

The crude death rate is affected similarly. The maximum increase in the rate is about 7 per 1,000 people when HIV prevalence is 2.5 percent, rising to 25 per 1,000 when HIV prevalence is 35 percent. The greatest increase in the crude death rate tends to be a few years earlier than the greatest loss in life expectancy.

The decline in life expectancy affects women more than men. Worldwide, in the period 2000–05, women had a 4.3 year advantage in life expectancy. In Sub-Saharan Africa, their advantage would have been 3.1 years without AIDS and would have stayed essentially constant up to 2050 (in the no-AIDS scenario). Because of AIDS, their advantage has fallen to 1.9 years, according to the UN projection, and will fall further to 0.6 years for most of the period 2010–25, before recovering slowly, to barely more than 1.5 years by 2050. The Census Bureau shows less of a change, with the female advantage at 2.2 years in 2000–05, falling under two years during 2005–15, and recovering more quickly to about four years by 2050. (The World Bank does not provide these data.)

Country by country, projected trends in the female advantage can look strikingly different (figure 6.7). Although there are some commonalities in country patterns, the UN and the Census Bureau appear to disagree about many aspects of these trends: what the female advantage is initially, when and how steeply it declines, and if and when it begins to increase again. The Census Bureau generally assumes an earlier decline in the female advantage than the UN but a somewhat shallower decline at lower HIV prevalence levels. In both sets of projections, the female advantage falls by a year for every 4 to 5 percentage point increase in initial HIV prevalence. This can turn the female advantage into a disadvantage.

Figure 6.7

Figure 6.7

Female Advantage in Life Expectancy, Sub-Saharan Africa and Three Selected Countries Sources: Calculated from Stanecki 2004 and United Nations 2004.

Where HIV prevalence reaches 20 percent or more, female life expectancy usually falls, at some point in the projection, at least two years below male life expectancy. Most of these countries are in southern Africa, or in the slightly broader Southern African Development Community (SADC). The UN always shows some recovery by 50 years, although in the worst case, Botswana, the female disadvantage, at 4.5 years, is still extreme at the end of this period. The Census Bureau, in contrast, shows either a quick and substantial recovery, as is shown in figure 6.7 for Zimbabwe, or an apparently unarrested fall, as for Namibia, where the female disadvantage exceeds seven years by 2050. Long-term trends, particularly where HIV prevalence is high, are a particular area of disagreement.

The changes in life expectancy, and in the female advantage, are the result, of course, of rising adult mortality. The estimates of adult mortality (45q15) used in the projections are available only for the UN. These show the substantial rise in mortality over projected trends without AIDS, as well as the way female adult mortality catches up with and, in severe cases, passes male mortality (figure 6.8). The increases in the risk of death between ages 15 and 60, relative to the situation without an AIDS epidemic, can be substantial. If HIV prevalence in 2001 is 10 percent, the maximum increase in risk over the following two decades is, on average from a quadratic regression, 29 percentage points for males and 34 percentage points for females. If HIV prevalence is 30 percent, the increase in risk is 65 percentage points for males, 75 percentage points for females.

Figure 6.8

Figure 6.8

Adult Mortality (45q15) by Sex in UN Projections, Sub-Saharan Africa and Lesotho Source: United Nations 2004.

Population by Sex and Age

The shrinking (and occasionally disappearing) female advantage in life expectancy is projected to alter the sex ratio, although for national populations the effect is not large. For a handful of southern African or SADC countries, the sex ratio may rise 10 to 20 percentage points in the UN projection and slightly more in the Census Bureau projection. As a result, the highest sex ratio in these projections would be about 120 males per 100 females. This is not unprecedented. In 2000, five countries—all in the Persian Gulf and none particularly affected by AIDS—had higher ratios, up to 190 males per 100 females. National sex ratios approaching 120 could be destabilizing, depending on the flexibility, or lack of it, of particular cultures.

At specific ages, sex ratios may become more extreme. Figure 6.9 illustrates the situation for Zimbabwe, one of the countries most affected by AIDS mortality. Over decades, the imbalance between males and females resulting from higher female AIDS mortality produces a hump in the sex ratio, which, in this case, reaches about 230 in the year 2030 for those age 50 to 54 years. This hump, as it rises, gradually moves toward older and older ages and eventually begins to subside. Having almost two-and-a-half males for every female at older ages is unusual, but whether it is a serious issue depends on the culture.

Figure 6.9

Figure 6.9

Sex Ratios by Age from UN Projections, Zimbabwe Source: United Nations 2004.

Where age patterns are involved, the dependency ratio might be a particular concern. A low dependency ratio, implying a relatively large workforce, could be an economic advantage, but AIDS impedes progress toward a lower ratio by increasing adult mortality more than infant and child mortality. How great the effect is, though, is difficult to tell. In the countries most severely affected by AIDS, the dependency ratio might rise from 50 or 60 dependents per 100 adults of working age to 70 or 80 dependents. Comparison of the UN projection with the no-AIDS scenario suggests such a rise for various countries in southern Africa. However, the exact level of the dependency ratio will be heavily influenced by future fertility, about which there appears to be little consensus (figure 6.10). The apparent effect of AIDS is easily swallowed up by differences in assumed fertility trends. Although AIDS does delay the movement—typical in the demographic transition—toward lower dependency, a stalled fertility decline is at least as effective an impediment.

Figure 6.10

Figure 6.10

Dependency Ratio, Sub-Saharan Africa and Three Selected Countries Sources: Stanecki 2004; United Nations 2004; World Bank 2004.

Initial Assumptions

Is it possible to sort out the differences in projections and come to some reasonable conclusions about the likely demographic impact of AIDS? There is no way to determine in advance which forecasts will be most accurate in the decades to come. The only useful approach to assessing the relative adequacy of the different projections, therefore, would be to look at their methodology, especially the way they model the effects of HIV/AIDS. These models, however, are quite complex, not completely transparent, and beyond the scope of this chapter to critique in detail. We therefore settle for examining some of the initial assumptions that go into these projections, particularly assumptions about adult mortality and HIV prevalence. These two sets of parameters undergird projections of AIDS impact.

For the World Bank projections, calculations of adult mortality can be made from the WHO life tables for 2000 (Lopez et al. 2002). Similar calculations can be made from UN survivorship ratios; for 2000, we average estimates for the periods 1995–2000 and 2000–05. (Similar data are not available from the Census Bureau.) UN figures are plotted against WHO figures by country in figure 6.11.

Figure 6.11

Figure 6.11

Consistency among Alternative Estimates of Current Adult Mortality (45q15) Sources: Timaeus and Jasseh 2004; United Nations 2004; Lopez et al. 2002.

The UN and WHO adult mortality estimates generally agree. On average, UN estimates are only 1 percent lower than WHO estimates. (The differences may be slightly greater at high mortality levels.) Figure 6.11 also shows, however, another set of estimates of adult mortality, derived by Timaeus and Jasseh (2004) from sibling histories and incorporating recent Demographic and Health Surveys (DHS) data. These estimates, which cover seven countries, appear consistently lower than those from the WHO and the UN. (The only exception is for Zimbabwe, where the WHO estimate, but not the UN estimate, is slightly lower.) The gap appears to be substantial. The probability of dying between ages 15 and 60, averaged without weighting across the seven countries, is between 51 and 57 percent in the WHO and UN estimates (depending on which estimates and which sex is involved). The averages for Timaeus and Jasseh's estimates, in contrast, are 44 percent for males and 34 percent for females.

Somewhat more countries can be compared for 1995, for which year Timaeus and Jasseh (2004) cover 20 countries. These estimates are mostly well below the UN estimates for 1995 (estimated as the average for 1990–95 and 1995–2000). The average probability of dying between ages 15 and 60 across the 20 countries is 36 percent for males and 29 percent for females, according to Timaeus and Jasseh (2004). According to the UN, the respective averages are 50 percent for males and 42 percent for females.

The Timaeus and Jasseh (2004) estimates cannot be considered an entirely reliable standard because of the limited data on which they are based and the assumptions necessary in their calculation, such as the assumption of a common age pattern of mortality increase across countries (see also chapter 4 in this volume). Nevertheless, the consistently lower levels of adult mortality they determine may suggest some overstatement in the UN and WHO mortality estimates, which could be tied to an overestimation in the modeling of the mortality impact of AIDS.

Without getting into the intricacies of these models, we can look at the estimates of HIV prevalence with which the modeling starts. These are generally UNAIDS estimates for 2001. The UN estimates, as noted earlier, begin with these estimates but may be adjusted in order to fit the prevalence trend. The Census Bureau maintains the database that UNAIDS uses in making its estimates and reports similar figures, at least for countries for which it has estimates. The source of the WHO estimates is not specified but is presumably the same.

A problem with each of these projections is the 2001 UNAIDS estimates themselves. In 2004, UNAIDS (2004) revised these estimates, changing almost twice as many downward as upward. The changes were sometimes large. UNAIDS (2002) had previously reported "low" and "high" estimates to bracket their main estimates for 2001. In 10 cases, the new estimates were lower than the previous "low" estimates. In only two cases were they higher than the previous "high" estimates. The adjustments may have been made to ensure consistency with new estimates for 2003. These were generally lower than the earlier 2001 estimates, which might have suggested epidemics on the decline. Instead, by lowering 2001 estimates, UNAIDS can speak of "stabilization" in HIV prevalence in the region, which it attributes to a balance between deaths and new infections.

Figure 6.12 compares UNAIDS estimates of HIV prevalence in 2003 with the earlier estimates for 2001 (UNAIDS 2002, 2004). Among the reductions, 15 countries have 2003 rates that are at least 1.5 percentage points lower than the earlier 2001 estimates, three of them at least 5 percentage points lower. Among the increases, only one country (Swaziland) has a 2003 rate at least 1.5 points higher than in 2001. Between 1999 and 2001, in contrast, increases and decreases in prevalence were more balanced, with increases being slightly more likely and somewhat larger.

Figure 6.12

Figure 6.12

HIV Prevalence in 2001 and 2003 Source: UNAIDS 2002, 2004.

If the epidemics are indeed at a plateau or past their peak, that may be uncomfortable for the agency projections. The Census Bureau projections are inconsistent with this idea, since they assume that peak prevalence is not reached until 2010. Whether the UN projections are consistent with plateauing epidemics is not clear. The UN reports that peak incidence for the epidemics—not peak prevalence—was reached on average about 1994, which could be consistent with some decline in prevalence by 2001. Country by country, however, there is no relationship between an earlier assumed incidence peak for the UN and the amount of apparent prevalence decline in the UNAIDS estimates. Only the World Bank projections, which effectively assume that AIDS-related mortality is already in decline, would be consistent with the interpretation of epidemics in at least early decline in the region.

The correction of 2001 HIV prevalence rates is also uncomfortable for projections, which relied on the earlier rates. The correction seems to have some basis. Table 6.2 compares UNAIDS estimates with estimates from DHS data for five countries. Based mainly on antenatal clinic reports, the earlier UNAIDS estimates were typically higher than the DHS estimates. Among these countries, rather high rates for Kenya and Zambia were brought into the range of the DHS estimates, and rates for the other three countries were also lowered.

Table 6.2. Estimates of HIV Prevalence among Adults Age 15 to 49, 2001–03(percent).

Table 6.2

Estimates of HIV Prevalence among Adults Age 15 to 49, 2001–03(percent).

The survey estimates are meant to cover the adult population more comprehensively than sentinel surveillance systems based on special populations, especially pregnant women. Survey estimates do have their own problems, such as nonresponse, for which adjustments are attempted. The survey reports suggest one important reason why earlier UNAIDS estimates could have been too high. In four out of five cases, male prevalence is clearly lower than female prevalence. The exception, in Burkina Faso, occurs at a low prevalence level. This is in keeping with the typical pattern expected in primarily heterosexual epidemics, where male infections are initially more frequent than female infections, but the reverse eventually becomes the case. Given that UNAIDS applies rates for pregnant women to the population as a whole, earlier overestimates are certainly plausible, particularly at higher HIV prevalence levels. Further downward adjustments in prevalence may be needed as more surveys become available. They will, of course, come too late for the current set of population projections, all of which—if this interpretation is correct—start out from too-high levels of HIV prevalence and therefore AIDS deaths.

Revised Projections

These projections may be revised, in later rounds, to allow for slightly lower current adult mortality overall and, particularly in the most affected countries, either lower or declining HIV prevalence levels, if further research confirms either to be the case. Pending such revision, precise conclusions about the demographic impact of AIDS in the region cannot safely be drawn. The revisions to the projections might have to be substantial in some cases. For Kenya, for instance, the UNAIDS estimate of HIV prevalence for 2001 was 15.0, and the current set of projections depend on this estimate. The 2003 estimate fell to 6.7, consistent with DHS results. The 2001 estimate implies that the maximum reduction in life expectancy—relative to the no-AIDS scenario—would be 17.7 years, using the equation in note 2. The 2003 estimate implies a maximum reduction half as large, at 8.7 years.

To get a further, although only preliminary, idea of how projections might change in the most affected countries, we draw on projections recently prepared for the Central Statistical Offices (CSOs) in Zambia and Botswana (Bulatao 2003a, 2003b). The Zambia projection adopts HIV prevalence estimates from the 2001–02 DHS rather than from UNAIDS. The DHS data suggest that HIV prevalence among all adult women is only 75 percent of that among pregnant women, and that male prevalence is still lower. These assumptions are adopted for the neighboring country of Botswana, for which DHS data are not available. Botswana, with a relatively good health system for the region, has monitored the proportion of males infected, based on clients at counseling and testing centers. The proportion of males infected is, indeed, lower than that for women. Surveillance data on the age distribution of both men and women infected in Botswana also match quite closely the distributions in the Zambia DHS.3

Other procedures for these CSO projections follow World Bank methodology. However, in place of WHO life tables incorporating AIDS mortality, Coale-Demeny (1983) life tables without AIDS are used, so that AIDS mortality can be specifically modeled. The modeling follows the usual procedures, involving calculations of HIV incidence from prevalence and estimation of subsequent deaths, relying on procedures previously described (Bulatao and Bos 1992) but with curves fit to prevalence trends rather than specific calculations, from sexual behavior, of transmission probabilities.

The Zambia projection begins with a 1980 census, in order to incorporate the effects of an epidemic initially recognized about 1984, and progresses to 2025. The Botswana projection begins with a 1981 census and goes to 2031. The assumed total fertility trends were constructed from census and survey data, some of them only locally available, but are generally in the range of those that the UN, the Census Bureau, and the World Bank assume for the country.

Figure 6.13 compares the projected population with the previous projections. The effect of AIDS still seems to be large, relative to the UN no-AIDS scenario, but for both countries, these CSO projections give higher population figures than those from the international agencies. Which of the preceding projections the CSO projections are closest to is determined more by fertility assumptions than the modeling of the AIDS effect, since the World Bank has the highest fertility trend for Botswana and the Census Bureau for Zambia. By 2025, the CSO projections give a Zambia population 6 to 32 percent larger than the other projections, but still almost 20 percent below that in the no-AIDS scenario. For Botswana at the same date, the CSO projection is 10 to 30 percent higher than the others but almost 30 percent below the no-AIDS scenario. Like the World Bank, but unlike the UN and the Census Bureau, the CSO projections show no population decline.

Figure 6.13

Figure 6.13

Alternative Population Projections, Zambia and Botswana (thousands) Sources: Stanecki 2004; United Nations 2004; World Bank 2004; Bulatao 2003a, 2003b.

If we look instead at projected life expectancy (figure 6.14), the differences from the other projections stand out as sharply. For Zambia, the CSO projection shows levels 5 to 10 years higher than the other projections for the period 2000–25, although still about 10 years short of what the no-AIDS scenario shows. For Botswana, all three agency projections show a similar decline up to about 2000, followed by rapid divergence. The CSO projection shows no further decline beyond 2001–06, so that about 2025 life expectancy is 15 to 25 years higher than in the other projections, although almost 15 years below the no-AIDS scenario. Adjusting projections to incorporate somewhat lower survey estimates of HIV prevalence could therefore have a substantial effect. In these severely affected countries, it reduces the deficit in life expectancy for 2025 by anywhere from one-third to two-thirds.

Figure 6.14

Figure 6.14

Alternative Projections of Life Expectancy in Zambia and Botswana Sources: Stanecki 2004; United Nations 2004; World Bank 2004; Bulatao 2003a, .

Conclusion

The demographic impact of AIDS in Sub-Saharan Africa is substantial, but its precise dimensions remain largely a matter of conjecture. Sophisticated models of the course of HIV/AIDS epidemics have been developed and linked with population projection approaches that have proved relatively reliable in the past (National Research Council 2000). However, these projection models have used inadequate data on mortality and HIV/AIDS.

The regional projections, part of the global work done by three agencies, the UN Population Division, the U.S. Census Bureau, and the World Bank, show the population of Sub-Saharan Africa falling below what it would be without AIDS, by 11 to 13 percent by 2020 and by anywhere from 10 to 25 percent by 2050. For individual countries, the population shortfall is projected as high as 34 to 48 percent by 2020 and 48 to 68 percent by 2050. That there will be a shortfall relative to what it would have been had HIV/AIDS not invaded the region is certain. That it will be of these dimensions is not. Each of the projections of countries affected by HIV/AIDS starts with UNAIDS estimates for HIV prevalence in 2001, some of which were clearly overestimates. For how many this is true cannot yet be told.

Between 2002 and 2004, UNAIDS generally revised downward its estimates for 2001 adult HIV prevalence in Sub-Saharan Africa. DHS data show sharply lower estimates of adult prevalence in three countries, and UNAIDS made matching adjustments for two (and somewhat smaller adjustments for the rest). For countries where HIV prevalence is high, initial overestimates could have been due to the practice of generalizing prevalence estimates for pregnant women to the adult population. Prevalence among males tends to be lower than among females, particularly at higher levels.

Unfortunately, the population projections available rely on the earlier estimates of HIV prevalence. This could lead to overstated base adult mortality, as well as to an overstated future trend in AIDS deaths. Comparing projections that rely on survey-based HIV estimates suggests that, in severely affected countries, the reduction in population due to AIDS may be only 50 to 75 percent of what the agency projections show and that life expectancy may be reduced, at the maximum, by a third to two-thirds of what agency projections show.

If the agency projections are not, at present, a reliable guide to future populations in areas affected by HIV/AIDS, they nevertheless illustrate patterns of demographic impact. The effect on population does appear larger when countries are strongly affected by HIV/AIDS. However, even in these cases, AIDS is not necessarily the dominant factor in future population trends nor the main source of uncertainty about them. Fertility is at least as important. Differing judgments about future fertility trends can produce at least as great variation in future population as the variation between situations with and without an AIDS epidemic.

Fertility itself is affected by HIV/AIDS. Younger women are more likely to be infected than older women, but their eventual deaths leave gaps in the age structure that persist into and become increasingly evident at older ages. Crude birth rates actually rise slightly as a result, because younger, more fertile women continue to be replenished in the age structure. Unfortunately, this compositional effect is the only effect on fertility that can be seen in current projections. Biological and behavioral mechanisms that may lead to lower fertility are insufficiently understood to be modeled across countries.

The effect of HIV/AIDS on mortality, the primary mechanism by which it affects population growth in agency projections, appears substantial but may be overblown. Life expectancy is projected to fall a maximum of one year below its expected path for every one percentage point increase in initial HIV prevalence. With UNAIDS adjusting its estimates of HIV prevalence downward, life expectancy should probably be expected not to decline as much. For Kenya, for example, UNAIDS reduced its estimate of 2001 HIV prevalence from 15 to 8 percent. The projections show maximum loss in years of life ranging from 15 to 21 years, but a smaller loss of 8 to 14 years would seem more likely.

Given that the HIV/AIDS epidemics in Sub-Saharan Africa are mainly heterosexual, the rise in mortality affects women more than men. The female advantage in life expectancy will shrink and could turn into a female disadvantage. This effect too depends on levels of HIV prevalence, and the projections may overstate the change. Effects on the sex ratio, as well as on age structure, are more substantial at high HIV prevalence levels, but even in these cases they do not appear to produce patterns more extreme than those seen in other populations.

Each agency projection will undoubtedly be revised in the future to take adjusted HIV prevalence levels into account. How they came to adopt HIV prevalence levels that were too high is an interesting question. The agencies devote some effort to ensuring that they use the best possible estimates of current fertility and mortality. Perhaps they have not spent as much effort carefully considering the accuracy of input data on HIV/AIDS and have been insufficiently critical of UNAIDS reports. Even with the adjustments made to previous HIV prevalence estimates, there remain some that probably deserve closer scrutiny.

There also remain other challenges to future rounds of these projections. Although data on HIV prevalence are gradually improving, what path the epidemics will take on the downslope remains conjectural. How therapies and vaccines may help, what behavior changes can be anticipated, and what the implications for infections and deaths will be are matters about which little if any empirical information can be adduced. In this area projections are, at best, educated guesses.

Another challenge will be managing the complexity of the methodology to project AIDS mortality. Complexity need not itself be a problem, but where it disguises a poverty of data and lends an inappropriate air of authoritativeness to results while concealing the calculations on which they are based, it can easily mislead. Projections need to convince not only with sophisticated models but also with a good appreciation of historical patterns and trends. More transparency would probably make it easier not just to detect the weaknesses of the projections but also to better appreciate what they actually contribute.

References

  1. Ahmad O. B., Lopez A. D., Inoue M. The Decline in Trends in Child Mortality: A Reappraisal. Bulletin of the World Health Organization. 2000;78(10):1175–91. [PMC free article: PMC2560617] [PubMed: 11100613]
  2. Bos, E., M. T. Vu, E. Massiah, and R. A. Bulatao. 1994. World Population Projections, 1994–95 Edition: Estimates and Projections with Related Demographic Statistics. Baltimore: Johns Hopkins University Press.
  3. Bulatao, R. A. 2001. "Visible and Invisible Sources of Errors in World Population Projections." Paper presented at the annual meeting of the Population Association of America, Washington, DC, March.
  4. ———. 2003a. "Population Projections for Botswana up to 2031." Paper prepared for the Central Statistical Office, Gaborone, Botswana.
  5. ———. 2003b. "Population Projections for Zambia, 2000–2025." Paper prepared for the Central Statistical Office, Lusaka, Zambia.
  6. Bulatao, R. A., and E. Bos. 1992. "Projecting the Demographic Impact of AIDS." Policy Research Working Paper 941, World Bank, Washington, DC.
  7. Coale, A. J., and P. Demeny, with B. Vaughn. 1983. Regional Model Life Tables and Stable Populations. 2nd ed. New York: Academic Press.
  8. Lopez, A. D., O. B. Ahman, M. Guillot, B. D. Ferguson, J. A. Salomon, and C. J. L. Murray. 2002. World Mortality in 2000: Life Tables for 191 Countries. Geneva: WHO.
  9. Lopez, A. D., J. Salomon, O. Ahman, C. J. L. Murray, and D. Mafat. 2001. "Life Tables for 191 Countries: Data, Methods and Results." Global Program on Evidence for Health Policy, Discussion Paper 9, WHO, Geneva.
  10. McDevitt, T. M. 1999. World Population Profile: 1998. Report WP/98. Washington, DC: U.S. Census Bureau.
  11. Murray, C. J. L., O. B. Ahmad, A. D. Lopez, and J. Salomon. 2000. "WHO System of Model Life Tables." Global Program on Evidence for Health Policy, Discussion Paper 8, WHO, Geneva.
  12. National Research Council. 2000. Beyond Six Billion: Forecasting the World's Population. Washington, DC: National Academy Press.
  13. Salomon, J. A., E. E. Gakidou, and C. J. L. Murray. 1999. "Methods for Modelling the HIV/AIDS Epidemic in Sub-Saharan Africa." Global Program on Evidence for Health Policy, Discussion Paper 3, WHO, Geneva.
  14. Stanecki, K. A. 2004. The AIDS Pandemic in the 21st Century. U.S. Census Bureau, International Population Reports WP02-2. Washington, DC: U.S. Government Printing Office.
  15. Stanley, E. A., S. T. Seitz, P. O. Way, P. D. Johnson, and T. F. Curry. 1991. "The IwgAIDS Model for the Heterosexual Spread of HIV and the Demographic Impacts of the AIDS Epidemic." In The AIDS Epidemic and Its Demographic Consequences. New York: United Nations.
  16. Timaeus I. M., Jasseh M. Adult Mortality in Sub-Saharan Africa: Evidence from Demographic and Health Surveys. Demography. 2004;41(4):757–72. [PubMed: 15622953]
  17. UNAIDS. 2000. Report on the Global HIV/AIDS Epidemic–June 2000. Geneva: UNAIDS.
  18. ———. 2002. Report on the Global HIV/AIDS Epidemic 2002. Geneva: UNAIDS.
  19. ———. 2004. Report on the Global AIDS Epidemic, July 2004. Geneva: UNAIDS.
  20. UNAIDS Reference Group on Estimates Modelling Projections. Improved Methods and Assumptions for Estimation of the HIV/AIDS Epidemic and Its Impact: Recommendations of the UNAIDS Reference Group on Estimates, Modelling and Projections. AIDS. 2002;16:W1–14. [PubMed: 12045507]
  21. United Nations. 1998. World Population Prospects: The 1998 Revision. New York: United Nations.
  22. ———. 2003a. World Population Prospects: The 2002 Revision. Vol. 1: Comprehensive Tables. New York: United Nations.
  23. ———. 2003b. World Population Prospects: The 2002 Revision. Vol. 2: Sex and Age Distribution of the World Population. New York: United Nations.
  24. ———. 2004. World Population Prospects: The 2002 Revision. Vol. 3: Analytical Report. New York: United Nations.
  25. World Bank. 2004. World Development Indicators 2004. Washington, DC: World Bank.
  26. Zaba B., Gregson S. Measuring the Impact of HIV on Fertility in Africa. AIDS. 1998;12(Suppl. 1):S41–50. [PubMed: 9677188]

If Pt is the projected population to time t and Pt* is the projected population with which it is being compared, then Pt/Pt* = P0/P0* e(bb*)t e(−d+d*)t e(mm*)t, where P0 is the base population for the projection and b, d, and m represent crude birth, death, and net migration rates. Bulatao (2001) uses a similar decomposition in assessing projection accuracy.

The equation, derived from the UN projection, is: Loss in e0 = −1.5 − 1.08 × HIV adult prevalence in percent in 2001 (R2 = 0.97).

Preliminary results from the 2004 Botswana AIDS Impact Survey II, conducted after these projections were completed, appear to confirm the need to use substantially lower adult HIV prevalence levels in projections. They show adult HIV prevalence at 25.3 percent, well below the 37.3 percent that UNAIDS reports.

Footnotes

1

If Pt is the projected population to time t and Pt* is the projected population with which it is being compared, then Pt/Pt* = P0/P0* e(bb*)t e(−d+d*)t e(mm*)t, where P0 is the base population for the projection and b, d, and m represent crude birth, death, and net migration rates. Bulatao (2001) uses a similar decomposition in assessing projection accuracy.

2

The equation, derived from the UN projection, is: Loss in e0 = −1.5 − 1.08 × HIV adult prevalence in percent in 2001 (R2 = 0.97).

3

Preliminary results from the 2004 Botswana AIDS Impact Survey II, conducted after these projections were completed, appear to confirm the need to use substantially lower adult HIV prevalence levels in projections. They show adult HIV prevalence at 25.3 percent, well below the 37.3 percent that UNAIDS reports.

Copyright © 2006, The International Bank for Reconstruction and Development/The World Bank.
Bookshelf ID: NBK2299PMID: 21290660
PubReader format: click here to try

Views

  • PubReader
  • Print View
  • Cite this Page

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to pubmed

Related citations in PubMed

See reviews...See all...

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...