Zaba, Basia*; Whitworth, Jimmy†; Marston, Milly*; Nakiyingi, Jessica†; Ruberantwari, Anthony†; Urassa, Mark‡; Issingo, Raphaeli‡; Mwaluko, Gabriel‡; Floyd, Sian*§; Nyondo, Andrew§; Crampin, Amelia§

HIV has increased adult mortality rates in sub-Saharan Africa^1,2 and probably child mortality as well. Because HIV prevalence levels are high and are still increasing in many countries, any effect of AIDS on child mortality is likely to persist for several decades.³ However, direct evidence for the impact of HIV on child mortality is relatively weak because most of the affected countries lack complete vital statistics.

The improvements in child mortality observed since the 1960s began to falter in many countries even before the AIDS crisis, possibly because of restrictions on health spending imposed as a condition of World Bank economic support and because of warfare, leading to stagnation in the development of primary healthcare.⁴ Thus, it is difficult to disentangle the contribution of HIV simply by monitoring time trends. Furthermore, the techniques for measuring child mortality through retrospectively reported birth histories collected in household surveys no longer yield reliable estimates⁵ because relatively few HIV-infected mothers survive to report their children's deaths.⁶ Many hospital-based studies^7–11 have provided information on mortality patterns of infected children, but such studies cannot show the impact of HIV at the community level.

Community-based measurements of HIV status of infants are difficult. More expensive procedures are needed because standard serological tests detect the presence of maternal HIV antibodies in children who are not themselves infected. However, it is possible to use data that measure child survival classified by the HIV status of the mother. Such data can gauge the population level impact of HIV on child mortality, even when the HIV status of the child is not known.

METHODS

Two prospective cohort studies in Kisesa (Tanzania)¹² and Masaka (Uganda)¹³ and a retrospective cohort study from Karonga (Malawi)¹⁴ are described in detail elsewhere. Separate analyses of child mortality have been undertaken for each of these data sets, exploiting the strengths of each and allowing for their weaknesses. To pool data for a meta-analysis, we applied common definitions and limited our investigation to those risk factors collected by all 3 studies.

Some time segments of a mother's life cannot be unambiguously classified as HIV positive or negative, even for women undergoing repeated serological tests. These segments include time lived after the last negative HIV test by mothers who never have a positive HIV test; time between the last negative test and the first positive test for seroconverters; and time lived before the first positive HIV test by women who never had a negative HIV test. We grouped these segments and classified them as “ambiguous” for the purposes of statistical analysis.

To measure the impact of a mother's death (from any cause) on the survival of her children, we identified the time period 1 year before her death (the period of her terminal illness) and the year after her death. The period of children's lives before these 2 years, and all the time lived by children whose mothers survive the whole of the study period, is referred to as “living with a healthy mother.” If a child survives the death of the mother by a year or more, this time segment is referred to as “mother died long ago.” We assume the child's experience of the crisis period is transitory. Mortality measures used for analyzing the impact of maternal death are single-year age-specific death rates, rather than conventional infant and child mortality rates which are cumulated risks of dying from birth.

Life table techniques were used to compute age-specific and cumulative mortality risks for infants and children. Confounding by various risk factors was adjusted by using a piece-wise exponential hazards model. The regression analyses allowed for both fixed covariates (such as study, sex, mother's HIV status at birth) and time-varying covariates (child's age, calendar year, mother's age and survival). We chose the covariates on the basis of their being identified as proximate determinants of child mortality in the literature^10,15 and being available in a similar format in each of the constituent studies. All covariates were incorporated in the multivariate model. Population-attributable fractions were calculated to measure the impact of HIV and the impact of maternal mortality at the population level.¹⁶

RESULTS

Mortality Risks in Children of HIV-Positive Mothers

Table 1 shows the proportions of children dying by age 1 and age 5, classified by HIV status of mother at the child's birth. Figure 1A shows the age-specific mortality rates, comparing children of HIV-positive and HIV-negative mothers. The rate ratio is highest in infancy, at 3.4 (95% CI = 2.6–4.9), and declines steadily with age.

Table 1

Figure 1

Mortality Associated With Orphanhood and Maternal Survival

Because of the frailty of infants and very young children, far greater numbers of child deaths are observed than deaths of mothers. However, the longer the follow-up, the more closely balanced the numbers of maternal and child deaths become. Small numbers at risk in some categories produce wide confidence intervals, but the results are highly suggestive and warrant grouping children of terminally ill mothers with recent orphans for further analysis. Age-specific mortality rates for children classified by mother's survival status are shown in Table 2 and Figure 1B. Recent or impending death of the mother is strongly associated with mortality of children younger than age 3, with rate ratios greater than 6 (95% CI = 4–8); after age 3, the impact is harder to measure, given the small numbers involved.

Table 2

A comparison of Figure 1A and Figure 1B suggests that the child mortality differential as the result of maternal survival is larger than that of maternal HIV status. However, compared with children of HIV-negative mothers, children of HIV-positive mothers are far more likely to have their mother die, with a risk ratio of 12, (95% CI = 8–20) over the follow-up times of these studies, which range, on average, from 5 to 13 years (data not shown).

Combined Effects of Maternal Survival and HIV Status

Our analysis of the combined effects of maternal survival and HIV status was limited to child-years lived at ages younger than 5 because data for this age range were available from all 3 studies. Table 3 shows the distribution of births, child-years of exposure and mortality rates, categorized by those variables that are common to all 3 studies and known to influence child mortality.

Table 3

The importance of age is highlighted: mortality declines rapidly after infancy. Boys have slightly higher mortality than girls, as is found in most populations. The current age of the mother has an important effect, with children of teenage mothers highly disadvantaged. Calendar year of observation also is important, with higher mortality risks observed in the past. The experience of mother's death or terminal illness also places children at very high risk of death. However, if a child survives for a year or more after the mother's death, there is no indication of further excess risk (though children orphaned for more than a year contributed only 73 person-years of observation). Children whose mothers have never been tested or whose mothers’ HIV status is ambiguous suffer higher mortality than children of HIV-negative mothers. Table 4 shows the results of the multivariate analysis. For 2 variables, (child's age 3+ and mother's age 20+) the categories used in Table 3 gave similar mortality rates and so were combined.

Table 4

After allowing for confounding, all the hazard ratios are substantial and in the expected directions. After controlling for age of child, maternal death appeared to have a more severe impact immediately after the event (hazard ratio = 5.2; 95% CI = 3.0–8.5) than in the year before (3.2; 2.0–5.3). Because the accuracy of dating of events is questionable, the 2 periods were grouped together in the final model to avoid making artificial distinctions between closely spaced events.

The final model was run separately for HIV-positive and HIV-negative mothers to investigate any interaction between mother's HIV status and survival. For uninfected mothers, the hazard ratio associated with mother's death was 3.3 (95% CI = 1.2–8.9); for infected mothers, the hazard ratio was 3.8 (2.3–6.5). Overall, grouping HIV-positive and negative mothers together for estimation of maternal mortality effect was therefore justified because of the similarity in the outcome and the increased power of the grouped analysis.

The regression model of Table 4 also was rerun separately for each study site. The differences that emerged in the hazard ratios had overlapping confidence intervals, confirming that pooling of the data was justified (data not shown).

Population Level Effects

Based on cumulative mortality risks in the pooled dataset in which HIV prevalence is 11%, the population-attributable fraction of infant mortality caused by maternal HIV infection is 19.4%, and for child mortality, it is 15.7%. The corresponding figures for the impact of maternal mortality are 2.4% and 2.5% for infant and child mortality, respectively. In the case of maternal mortality, the exposed populations are defined as children who lose their mothers within one year or within 5 years of birth; they form 0.3% and 2.3%, respectively, of the pooled dataset.

DISCUSSION

This analysis shows that, after allowing for geographical differences in levels of background mortality and considering child's sex and mother's age, the mortality rates of children born to HIV-positive women are almost 3 times those of children born to HIV-negative women. Age-specific death rates show mortality differences by HIV status of mother persisting into the third year of life; above this age, it is difficult to draw strong conclusions because of the relatively small numbers of children who died after the age of 3.

In addition, when mothers die, the mortality rates of their children in the 2 year period centered on the mother's death are more than tripled, irrespective of the mother's HIV status. Presumably this increased risk is caused by the disruption of breastfeeding in infants and by the inability of the mother to care for her children when she herself is subject to a terminal illness. Because children of HIV-positive mothers are far more likely to experience maternal death, they face a double jeopardy with respect to childhood mortality risks.^17,18

The years before and after a mother's death are a high-risk period for children. A disproportionate number of deaths occur in this relatively short period, particularly when the age of the child is allowed for. Previous studies may have failed to detect excess mortality in orphans because they have not considered the time since the death of the mother.¹⁹ In analyzing the effects of a mother's death on child survival, it is also important to allow for the period of her terminal illness; in 101 cases where both the mother and child died, the child died first in 82 instances.

Although the hazard ratios associated with maternal death are higher than those associated with maternal HIV infection, maternal death is more frequent for young children of HIV-infected mothers. Altogether, 2.6% of the children born into the 3 study populations experienced the death or terminal illness of their mothers: 1.0% of the children born to HIV-negative mothers and 21.8% of the children born to infected mothers. The relatively low proportion of children experiencing a maternal death or terminal illness means that the population-attributable fractions of infant and child mortality caused by maternal mortality (2.4% and 2.5%) are lower than the population attributable fraction caused by HIV (19.4% and 15.7%, respectively). These figures should be considered when developing policies to reduce HIV infection. Efforts to prevent mother-to-child transmission of HIV with nevirapine must be supported because they offer the most realistic chance of reducing the impact of HIV infection in children throughout sub-Saharan Africa; nonetheless, this approach will do nothing to reduce the effect of maternal mortality because nevirapine prophylaxis²⁰ is given with sole aim of preventing transmission of HIV from an infected mother to her child. The reduction of maternal mortality requires the provision of antiretroviral therapy to mothers, which is likely, in turn, to have a beneficial effect on infant and child survival. Preventing HIV infection in women in the first place would not only eliminate vertical transmission, but also prevent a large fraction of current deaths among mothers.

The adverse effect of maternal death may be limited to the first 2 years of life. We have firm evidence only about the excess mortality risk in children up to the age of 3, because after that age only 30 person-years exposure to maternal death were observed. The high mortality rate in HIV-infected children means that they will form a steadily declining fraction of the cohort born to HIV-infected mothers and, thus, we would expect the excess risk associated with maternal HIV status to decline with age. On the other hand, the mortality of the remaining HIV-infected children might be expected to eventually rise with age in the same way that mortality of HIV-infected adults rises with duration since infection. It will be useful to repeat this analysis when data become available from the continuing follow-up in these studies.

In populations without a tradition of record keeping, errors in ascertaining birth and death dates may obscure details of age-specific mortality patterns. Retrospective data from Karonga may have been slightly affected by omissions and dating errors. The prospective studies minimize such errors, with annual (Masaka) and half-yearly (Kisesa) follow-up, although the errors are not reduced sufficiently to permit decomposition into neonatal and postneonatal mortality.

Our identification of the relatively short period in which children of dead or dying mothers face extremely high mortality risks may have policy implications. Help for families with seriously ill mothers may be most useful before the mother's death and in the period immediately afterward. Orphans whose mothers died more than a year ago may not need help to such an extent. The concentration of infant and child deaths around the time of mother's terminal illness and death suggests that enforced, hurried weaning may be one precipitating factor. Provision of the healthy alternative feeding might benefit families with seriously ill mothers. A small scale observational trial of such an intervention would be useful.

Having a dead or dying HIV-negative mother carries as serious (short-term) consequences for child mortality as having a healthy HIV-positive mother. When HIV prevalence is low, the proportion of orphans whose mothers were HIV- negative is likely to be relatively large. In the pooled dataset, where HIV prevalence is around 11%, 38% of orphans younger than 5 had had an HIV-infected mother; thus mothers of the majority of orphans had been HIV negative. This implies that assistance for families with dead or dying mothers should be provided irrespective of the mother's HIV status.

REFERENCES