**3. Methods**

The method for reconstructing under-five mortality trends has been explained in details in other documents [Garenne & Gakusi, 2004]. In brief, age specific death rates are computed

 1'Purchasing Power Parity' corrects for the value of a common basket of goods across countries. 'Constant value' of the dollar corrects for inflation overtime.

The Impact of HIV/AIDS on the Health Transition Among Under-Five Children in Africa 241

already by itself for 95% of the population of this group. Islands have a very low prevalence of HIV, followed by the other groups in the order presented in Table 1. Average seroprevalence ranged from 1.5% in the Sahelian group to 17.4% in the Southern group, a major difference in terms of potential impact on child mortality. Note that if the Southern group is the wealthiest and the Sahelian group among the poorest, there is no linear relationship between income and HIV prevalence. For instance the Eastern group has almost the same income level as the Sahelian group, but five times more HIV, whereas the Islands

To illustrate the rationale of the calculations, one could firstly present aggregate values. According to the UNAIDS database, some 5.1% of adults of both sexes were infected by year 2005 in sub-Saharan Africa. This corresponds to about 6.1% women aged 15-49 infected, and to about 7.3% pregnant women infected. The coefficients used for deriving these numbers were taken from the African DHS surveys with data on HIV seroprevalence, found in 20 countries. The differences are due to higher infection rates among young women compared with young men, and to higher infection rates at the peak of fertility (around age 30 years) compared with younger and older women. Among the babies delivered by these women, some 1.8% will become infected, and 60% of them will die before age 5, so that AIDS mortality will be about 11 per 1000 live births. Compared with an average under-five mortality of 123 per 1000 in year 2005, this leads to an estimate of about 9% of deaths of under-five children attributable to HIV/AIDS. Of course, this is a rough estimate for the whole continent; however it provides an order of magnitude for the effect of HIV/AIDS on overall mortality levels. Since there is a strong interaction between level of mortality and HIV prevalence (the countries the most affected by HIV are also those with the lowest mortality), the formal calculations by country are likely to be somewhat different (see

The mortality decline has been steady for the continent as a whole since 1950. (Table 2, Figure 2) For this group of 38 countries, the under-five death rate was estimated at 346 per 1000 in 1950, 229 per 1000 in 1970, 166 per 1000 in 1990, and 123 per 1000 in 2005. The pace of mortality decline averaged -2.1 per cent per year from 1950 to 1970, somewhat less (-1.6 per cent per year) from 1970 to 1990, and -2.0 per cent per year from 1990 to 2005, the period where HIV spread and hit these countries the hardest. Overall, HIV/AIDS did not change radically the speed of the mortality decline, which remained at an average level between 1990 and 2005. This decline was even somewhat faster than between 1970 and 1990, a period of turmoil for many countries, and of long lasting economic recession. [see Garenne & Gakusi, 2006b, Gakusi & Garenne, 2007 for more details on the impact of political and

Mortality levels and trends differed quite significantly among the six groups of countries. Firstly, the levels at baseline differed: countries from the Southern and the Islands groups had much lower levels of mortality in 1950 as well as in 1970. By 1990 the situation was different, since Madagascar underwent a major rise in mortality for about 13 years for reasons other than HIV/AIDS. By 1990, Southern Africa had from far the lowest mortality, but this favourable trend reversed dramatically because of HIV, so that mortality in 2005 was much higher than in 1990, with an average rate of increase of +2.5 per cent per year.

group is the second wealthiest, but has the lowest HIV prevalence.

**4.2 Basic calculations and order of magnitude** 

**4.3 Overview of mortality trends by area** 

below).

economic crises]

from maternity histories, by calendar year, for each DHS survey conducted in sub-Saharan Africa, unless access to data was restricted. When several surveys were available for the same country, they were merged by adding events (deaths) and exposure periods (personyears at risk) for the same year and the same age groups. Trends were fitted on monotonic periods with a Linear-Logistic model. Changes in trends were tested using standard T-tests, and only those changes significant at P< 0.05 were kept for final analysis. For the early years, the DHS data were sometimes supplemented with other sources, such as census data or data from other sample surveys. All together, under-five mortality trends were reconstructed for all selected countries, year by year from 1950 up to 2005.

Estimating HIV/AIDS mortality among under-five children was completed using a basic model. Firstly, trends in HIV seroprevalence among pregnant women were estimated by year and country using the UNAIDS database, in its latest edition [UNAIDS, 2008]. Second, a standard mother to child transmission rate was assumed, at 25%, consistent with UNAIDS recommendations. Third, a standard AIDS mortality schedule was applied to infected children, so that 60% were assumed to die before age 5, which is consistent with empirical data, and with the UNAIDS recommendations. [UNAIDS, 2002]

Once the database was constructed by year and country, countries were grouped into 6 major areas, selected for their different profiles of HIV infection: Sahelian countries, Coastal West-Africa, Central Africa, Eastern Africa, Southern Africa, and Islands. The details of these countries are shown in Figure 1.
