Eric O. Udjo Bureau of Market Research University of South Africa P. O. Box 392 UNISA 0003

1. A RE-LOOK AT RECENT STATISTICS ON MORTALITY IN THE CONTEXT OF HIV-AIDS WITH PARTICULAR REFERENCE TO SOUTH AFRICA Eric O. Udjo Bureau of Market Research University of South Africa P. O. Box 392 UNISA 0003 South Africa Paper presented at The 3rd Africa Symposium on Statistical Development, 3-7 December 2007, Accra, Ghana

2. Since the out break of the HIV epidemic in the 1980s, statistics on HIV prevalence, incidence, number of AIDS cases, AIDS related mortality, infant mortality rate as well life expectancy at birth have been given by various researchers and organisations. There are regional variations in the HIV epidemic globally and within the African continent According to UNAIDS (2006), Sub-Saharan Africa is the worst-affected region in the world, almost 64% of all people living with HIV globally live in sub-Saharan Africa. Background 2

3. There is a gradient in the level of HIV prevalence moving from Western to Central, Eastern and Southern Africa with Southern Africa having the highest HIV prevalence in sub-Saharan Africa and in the world. Among the worst affected countries in sub-Saharan Africa and in the world are Botswana, Lesotho, Namibia, Swaziland, Zimbabwe and South Africa. Apart from Zimbabwe, these countries are in Southern Africa Table 1 shows recent statistics related to HIV/AIDS, infant mortality rate as well as life expectancy at birth for selected region/countries from various sources . Background (contd) 3

4. Table 1: Adult (aged 15-49) HIV Prevalence and mortality in selected sub-Saharan African countries

5. Sources: a United Nations. 2006. World population prospects: The 2006 revision population database. Population Division, United Nations, New York. b Bureau of Market Research (2007). Population and household projections for South Africa by province and population group, 2001-2021. BMR, University of South Africa Research Report no 364, Pretoria. c UNAIDS. 2006. 2006 Report on the global AIDS epidemic. UNAIDS, Geneva. d Mahomva A. et al. 2006. HIV prevalence and trends from data in Zimbabwe, 1997-2004. Sexually Transmitted Infections, 82(Suppl 1): i42-i47. e Population Reference Bureau. 2006. 2006 World population data sheet. PRB, Washington DC. f UNDP 2005. Human development index report 2005, UNDP, New York.

6. The following stand out very clearly from the table: Low level of HIV prevalence in a western African country (Ghana) and high levels of HIV prevalence in the other selected countries located in Eastern and Southern Africa. The disparity in population based and surveillance HIV prevalence rates within the same country especially in Botswana and South Africa. The apparently low levels of life expectancy at birth apparently due to AIDS in the selected countries located in Eastern and Southern Africa. Background (contd) 6

7. Researchers have often been baffled about the disparities in the levels of HIV prevalence in Western and Southern Africa and various explanations have been offered but this is not the focus of the present paper. Regarding the disparity between antenatal and population based HIV prevalence, Boisson et al; Boerma et al, have highlighted the potential biases in HIV prevalence among pregnant women. These include representativeness, reduced fertility in HIV-1-infected women, selection of sexual activity and absence of contraceptive use, and under-representation of smaller rural sites in surveillance systems. Background (contd) 7

8. Recent statistics on HIV prevalence from population based surveys as seen in Table 1 suggest that HIV prevalence in many countries may not be as high as earlier estimated and projected as antenatal data tend to over estimate the prevalence of adult HIV prevalence. Turning to the apparently low levels of life expectancy at birth apparently due to AIDS, the life expectancies shown in Table 1 are inconsistent with the infant mortality rates shown for these countries. With the seemingly low levels of life expectancy at birth, one would expect much higher levels of infant mortality rates. Table 2 and Figure 1 illustrate the relationship of infant mortality and life expectancy at birth. Background (contd) 8

9. Table 2. Selected life expectancies at birth and corresponding infant mortality rates from conventional model life tables.

10. The following stand out very clearly from the Table 2 and Figure 1: Very low levels of life expectancies at birth are associated with very high levels of infant mortality rates. This raises a question among others: Are the estimated life expectancies in the context of AIDS for Botswana, Zimbabwe, Lesotho, Namibia and South Africa as given in Table 1 reliable? Aside the above, the model life tables used in modeling HIV/AIDS and its impact do not depict the characteristic ‘hump’ in the mortality curve at young adult ages in the population with generalised HIV/AIDS epidemics. Background (contd) 10

11. Source: : Coale-Demeny, 1966

12. The following stand out very clearly from the Table 2 and Figure 1: Very low levels of life expectancies at birth are associated with very high levels of infant mortality rates. This raises a question among others: Are the estimated life expectancies in the context of AIDS for Botswana, Zimbabwe, Lesotho, Namibia and South Africa as given in Table 1 reliable? Aside the above, the model life tables used in modeling HIV/AIDS and its impact do not depict the characteristic ‘hump’ in the mortality curve at young adult ages in the population with generalised HIV/AIDS epidemics. Background (contd) 12

13. Estimates of life expectancy at birth (both sexes) for South Africa in 2005 given by ASSA2003 (see the EXCEL file of ASSA2003) is 51.2 years. But the estimate of life expectancy at birth (both sexes) from ASSA2000 cited in Vass (2003) is 46 years for the same period . Dorington et al’s (2005) estimate of life expectancy at birth (both sexes) for South Africa in 2006 is 50.8 years and for the same period, they estimated infant mortality rate (both sexes) of 48 per thousand for the same period. Using the EPP and SPECTRUM packages, Rehle and Shisana (2003) estimated life expectancy for South Africa in 2005 as 45.2 years and infant mortality rate (both sexes) as 56.2 years for the same period. It would appear from these estimates that life expectancy at birth plummeted from the aboutf 65 years in 1996 (see Udjo, 2006) to about 50 years in 2005, a drop of over 10 years within 10 years. Overview of life expectancy at birth an infant mortality rate for South Africa 13

14. This study provides a critical look at recent statistics on infant mortality rates and life expectancies at birth in the context of HIV/AIDS in parts of Southern and Eastern Africa with particular reference to South Africa. Objective 14

15. 2001 Census (a) Childhood mortality: Children ever born, children surviving (b) Adult mortality: Orphanhood. 2007 Community Survey (a) Sample size: 947,331 individuals from 250,348 households (b) Childhood mortality: Children ever born, children surviving (c) Adult mortality: Orphanhood. Death registrations (a) Registered deaths 2004 (b) Statistics South Africa’s 2004 mid-year population estimates Data 15

16. 2001 Census & 2007 Community Survey Data Childhood mortality: Brass methods: qx = Diki (1) α = Yx – Yxs ; (2) where Yx = 0.5loge {(1-lx)/lx}; (3) And Yxs = 0.5loge {(1-lxs)/lxs}. (4) The standard life table used in is INDEPTH model life tables for sub-Saharan Africa Methods 16

17. 2001 Census & Community Survey Data Adult mortality: Brass methods: lB+N/lB=WN(5PN-5) + 5PN (1-WN) (5) α = 0.5loge (1 + (NPB/– 1/lsB)/(1-NPB)) (6) Methods (contd) 17

18. Life tables One-parameter life table: Yxc,a = αc,a + βYsxm,f (7) Hybrid life table: Y5m,f = α m,f + βYs,5m,f (8) and Yh,60m,f = α m,f + βYs,60m,f (9) General level of mortality: α m,f = Y5m,f - βYs,5 m,f (10) Methods (contd) 18

19. Mortality from Death Registrations Brass Growth Balance method: Linear relationship of deaths and agedistributions expressed as: N(a)/N(a+) = r + D(a+)/N(a+) (11) N(a)/N(a+) = r + k(D(a+)/N(a+)) (12) Methods (contd) 19

20. Childhood mortality and life expectancy See Figure 2 RESULTS 20

23. The 2001 census suggests moderate levels and rising mortality in recent years in male children. The levels and trend in mortality among male children appear probable. The trend among female children is inconsistent with the trend for male children and appear to suggest underreporting of dead female children among younger mothers during the census. Figure 3 based on the 2007 community survey gives the impression of marked decline in childhood mortality in 1977 and marked increase two years later. The sex differential is odd as it suggests higher female than male childhood mortality. As an indication of the level of life expectancy at birth after taking account of errors in the data, the reports on child survival alone suggest life expectancy at birth decreasing from about 60 years in 1985 to about 54 years in 2001 and 56 years in 2006 among male children. Life expectancy at birth however should not be based on child survivorship alone but should take into account adult mortality. RESULTS (contd) 23

24. Adult mortality and life expectancy See Figure 4 RESULTS (contd) 24

26. The estimates derived from respondents aged 30 years and above regarding adult male mortality imply steep decline in the more recent periods. Estimates derived from the younger respondents are questionable as the implied male female difference in life expectancies are too high (adult male mortality about 11 years less than adult female mortality at age 15 in 1993). It would appear from this improbable difference that while adult male mortality was over reported in some degree by the younger respondents, adult female mortality was underreported: absentee effect for paternal orphanhood and adoption effect for maternal orphanhood. Figure 5 shows the report of adult mortality from the 2007 community survey. RESULTS (contd) 26

28. The community survey suggests mortality has been increasing since 1996 but the levels of adult male mortality appear too high relative to adult female mortality. They are even higher than those derived from the 2001 census for comparable time periods. It would appear therefore that the reports on paternal orphanhood from the community survey over states adult mortality relative to the 2001 census. In view of this, the “best estimates” of levels of adult mortality were based on the average levels of orphanhood derived from the reports of respondents aged 15-24 in the case of paternal orphanhood and respondents aged 30-40 in the case of maternal orphanhood in the 2001 census For the period 2006, the trend in the levels of adult female mortality in the most recent years derived from the community survey was extrapolated to 2006 RESULTS (contd) 28

29. Combining childhood and adult mortality Life expectancy at birth is more correctly estimated by combining information from childhood and adult mortality because “a single parameter is not normally enough to describe the variation in mortality that is found in different population” (Brass, 1971). The results suggest a life expectancy at birth of 61 years in 2001 and 56 years inin South Africa in 2006 both sexes combined. The corresponding estimated infant mortality rates are 67 per thousand live births in 2001 and 55 per thousand in 2006 both sexes combined. These estimated levels of life expectancies at birth are consistent with the levels of infant mortality rates. RESULTS (contd) 29

30. Mortality based on death registrations See Figure 6 as illustration RESULTS (contd) 30

32. The results of the application of the Growth Balance method suggest a completeness of registered deaths in 2004 of approximately 66.5% and 92.5% for males whose reported age at death were 0-44 and 45 and over respectively. For females, completeness of registered deaths was approximately 58% and 85.8% for females aged 0-34 and 35 and above respectively. Life expectancies at birth resulting from the life tables after splicing the slopes resulting from the adjustments for under registration of deaths were 52.8 years for males and 57.1 years for females in 2004. These values were of the same levels of magnitude with those estimated from the community survey though the infant mortality rates from the death registration were lower than those from the community survey. See Table 3. RESULTS (contd) 32

33. Table 3: Estimated Mortality in 2001, 2004 and 2006

34. Discussion and conclusion Indirect estimation of mortality including population projections in countries with high HIV prevalence should incorporate standard life tables that take into account HIV/AIDS. The mortality schedules depicted by the use of conventional model life tables do not depict the characteristic ‘hump’ in the mortality curve at young adult ages due to increased AIDS deaths in populations with generalised HIV/AIDS epidemic. It is not clear how current models estimate life expectancies at birth in populations with generalised HIV/AIDS epidemic but the estimates are inconsistent with the infant mortality rates given for these populations. 34

35. Discussion and conclusion (contd) In the case of South Africa, the results of this study appear to suggest that these models over estimate mortality in the context of AIDS. This may probably be the case in other populations like Botswana, Lesotho, Swaziland, Zimbabwe and Namibia. Further studies are required to confirm this. 35

36. Thank you! 36