Effects of HIV on methods of child mortality estimation

Desired Result

Introduction

All estimation methods for child mortality based on mother’s reports on the survival of their children are subject to selection biases. Although migration and selective non-response may introduce bias, the greatest threat to child mortality estimates based on reports of women arises from a generalized HIV/AIDS epidemic. Vertical transmission of HIV from mother to child during pregnancy, delivery and through breastfeeding in the first few months of life increases the risk that the child will be HIV-positive (HIV) by as much as 35 per cent in the absence of antiretrovirals (ARVs), and over 60 per cent of HIV+ children will die before their 5th birthday in the absence of treatment with ARVs (Schneider, Zwahlen and Egger 2004; Todd, Glynn, Marston et al. 2007). Since the mothers also suffer elevated mortality risks, the deaths of many of these HIV+ children, particularly those born 5 years or more before interview, will not be reported. Overall child mortality will therefore be under-estimated, whether using direct child mortality estimation or indirect child mortality estimation.

Effects on direct child mortality estimation

Only one known analysis of the magnitude of bias in direct child mortality estimates has been carried out using real data rather than simulations. Hallett, Gregson, Kurwa et al. (2010) use data from a prospective open cohort in Manicaland, Zimbabwe to measure the bias introduced by deaths of HIV+ mothers. The cohort was interviewed between July 1998 and February 2000, with follow-up interviews at three and five years. From 1998 to 2005, HIV prevalence in the study population fell from 22 per cent to 18 per cent. In the final round in 2005, a full birth history was collected from surviving women, and U5MR was estimated for the period 1998 to 2005, a seven-year period as opposed to the usual DHS five-year period. The direct estimates were then compared to true values, adding back the child mortality experience of women who had died before 2005. The bias, calculated as the estimates from surviving mothers divided by the estimates for all mothers, was 6.7 per cent for the IMR and 9.8 per cent for U5MR. Hallett, Gregson, Kurwa et al. (2010) also developed a model of bias, which they applied to Zimbabwe and six other countries with moderate or high HIV prevalence for the period 1980 to 2015. They did this using UNAIDS prevalence data and DHS estimates for pre-epidemic periods. The model indicates that bias in direct estimates increases with (a) the duration of the epidemic; and (b) the time before survey of the estimate. Conversely, bias in direct estimates decreases with the level of background, non-HIV child mortality.

Walker, Hill and Zhao (2012) developed a simple cohort component projection model, separating births into those to HIV-negative (HIV-) mothers (where the children are assumed not to be infected), those to HIV+ mothers but where the children are not themselves infected, and those to HIV+ mothers where the children are infected at birth or subsequently. The first two streams are assumed to experience background mortality (from model life tables) whereas the third stream is assumed to experience a probability of dying by age 5 of 62 per cent. HIV+ mothers are then aged forward to the date of a survey allowing for their excess mortality, and the U5MR estimated from reports of surviving women is compared to that which would have been observed had all women survived to the survey. No adjustment is made for prevention or treatment. As with the analysis by Hallett, Gregson, Kurwa et al. (2010), the extent of bias depends on the HIV prevalence and its past trajectory, the level of background under-5 mortality and the time period before the survey to which the estimate refers. It is thus not possible to provide a simple way to assess the magnitude of bias. However, as a general guide, some estimates of bias for countries collecting birth histories around the middle of the last decade, before ARV treatments were widely available, are shown in Table 1 for time periods 1-5, 6-10 and 11-15 years before each survey. It is important to remember, however, that bias is a function of non-HIV child mortality, which is not easy to estimate, and HIV prevalence, which is usually estimated with error.

Table 1: Estimates of bias for estimates of U5MR for periods 1-5, 6-10 and 11-15 years before each survey: selected sub-Saharan African countries.

Country

DHS Year

Approx. HIV prevalence 2005 (%)

Assumed background

U5MR

Estimated bias by period before survey (%)

 

 

 

 

1-5

6-10

11-15

Cote d’Ivoire

2005

4.6

125

4.0

6.6

3.3

Kenya

2003

7.1

75

8.0

14.1

6.7

Lesotho

2004

23.4

75

13.2

15.7

2.1

Namibia

2006-07

15.3

50

13.7

22.7

10.4

Zambia

2007

15.0

150

6.8

13.9

13.0

Zimbabwe

2005

18.0

75

16.6

31.4

25.6

Source: Hill, Walker and Zhao (2012)

Bias is highest for the period 6-10 years before the survey, exceeding 10 per cent if the HIV prevalence has exceeded 5 per cent. It is important to remember that ARV use to prevent mother-to-child transmission and to extend survival times will have a quick effect on reducing bias for the most recent time period, but bias for past time periods will persist for a decade or more after effective therapy is introduced.

Effects on indirect child mortality estimation

Few studies have explored the impact of HIV on indirect estimates of child mortality. HIV will affect the accuracy of indirect estimates not only because of the association between mortality of children and that of their mothers but also because of the effects of HIV on age patterns of child mortality, and its implications for approaches that infer fertility patterns from observed parity ratios. Child mortality risks can also no longer be assumed to be independent of the age of the mother. However, in one respect, indirect estimates may be less affected by selection for maternal survival than direct estimates, because the analysis is carried out by age group. Mothers under the age of 25 are unlikely to have died from HIV/AIDS, so reports of child survival based on age groups 15-19 and 20-24, and even 25-29, may be little biased by HIV. However, it is the age groups 15-19 and 20-24 that are most biased by other selection effects, so this may not be a huge help.

Ward and Zaba (2008) estimate likely bias from HIV on indirect child mortality estimates given a stable (constant incidence and mortality effects) epidemic. Their model shows that the bias for estimates based on women under age 30 for adult prevalence of 10 per cent or less will not exceed 5 per cent, and even for prevalence up to 30 per cent will scarcely exceed 10 per cent. These findings are reassuring. Of course, the HIV epidemic has been anything but stable, rising sharply across many countries to around 2000 and declining both in prevalence and impact since then. However, the dynamic of the epidemic will tend to reduce bias below the levels estimated by the Ward and Zaba model.

Mutemaringa (2011) compares indirect estimates derived from DHSs for Zimbabwe, Kenya, Lesotho, Malawi, Namibia and Zambia to direct estimates from the same surveys. The author confirms that the bias primarily arises from the survivorship correlation. The bias in the estimate based on reports of women aged 25-29 is in three cases out of six less than 5 per cent, although in two cases, Zimbabwe and Namibia, the bias exceeds 20 per cent. The bias of estimates based on reports of women aged 30-34 and 35-39 generally exceeds 20 per cent, and in Kenya and Namibia exceeds 30 per cent.

The conclusion we draw from these analyses is that estimates of child mortality derived from reports of women aged 25-29 concerning their children ever borne and surviving will not be greatly affected by even a generalized HIV epidemic. Drawing on the pattern of bias by HIV prevalence found by Ward and Zaba (2008), the child mortality estimate obtained by standard analysis of a summary birth history could be adjusted upwards by three points per thousand for every 10 percentage points of HIV prevalence.

References

Hallett TB, S Gregson, F Kurwa, G Garnett et al. 2010. "Measuring and correcting biased child mortality statistics in countries with generalized epidemics of HIV infection", Bulletin of the World Health Organization 88(10):761-788. doi: http://dx.doi.org/10.2471/BLT.09.071779

Mutemaringa T. 2011. "Impact of HIV on estimates of child mortality derived using the summary birth history (CEB/CS) method." Unpublished MPhil thesis, Cape Town: University of Cape Town.

Schneider M, M Zwahlen and M Egger. 2004. Natural history and mortality in HIV-positive individuals living in resource-poor settings. http://www.epidem.org/Publications/unaids%20HQ_03_463871%20final.pdf

Todd J, JR Glynn, M Marston, T Lutalo et al. 2007. "Time from HIV seroconversion to death: a collaborative analysis of eight studies in six low- and middle-income countries before highly active antiretroviral therapy", AIDS 21(Suppl 6):555-563. doi: http://dx.doi.org/10.1097/01.aids.0000299411.75269.e8

Walker PN, K Hill and F Zhao. 2012. "Child mortality estimation: Methods used to adjust for bias due to AIDS in estimating trends in under-five mortality", PLoS Medicine 9(8):e1001298. doi: http://dx.doi.org/10.1371/journal.pmed.1001298

Ward P and B Zaba. 2008. "The Effect of HIV on the Estimation of Child Mortality Using the Children Surviving / Children Ever Born Technique", Southern African Journal of Demography 11(1):39-73. 

Author

Hill K

Suggested citation
<p>Hill K</p> . 2013. Effects of HIV on methods of child mortality estimation. In <p>Hill K</p> (eds). Tools for Demographic Estimation. Paris: International Union for the Scientific Study of Population. https://demographicestimation.iussp.org/content/effects-hiv-methods-child-mortality-estimation. Accessed 2024-03-19.