Maternal HIV status associated with under-five mortality in rural northern Malawi: A prospective cohort study
Abstract
Background
Under-five mortality is decreasing but with little change in neonatal mortality rates. We examined the effect of maternal HIV-status on under-five mortality and cause of death since widespread availability of antiretroviral therapy in rural Malawi.
Methods
Children born in 2006-2011 in the Karonga demographic surveillance area were included. Maternal HIV-status was available from HIV sero-surveys. Age-specific mortality rate ratios for children born to HIV-positive and HIV-negative mothers were obtained by fitting a Poisson model accounting for child clustering by mother and adjusting for potential confounders. Cause of death was ascertained by verbal autopsy.
Findings
There were 352 deaths among 6913 under-five singleton children followed for 20,754 person-years (py), giving a mortality rate of 17.0‘/1000py overall, 218/1000py (16.5/1000 live births) in neonates, 20/1000py (17.4/1000 live births) in post-neonatal infants and 8/1000py in 1-4 year-olds. Comparing those born to HIV-positive and HIV-negative mothers, the rate ratio, adjusted for child age, sex, maternal age, parity and drinking water source was 1.5 (95%CI 0.6-3.7) in neonates, 11.5 (95%CI 7.2-18.5) in post-neonatal infants and 4.6 (95%CI 2.7-7.9) in 1-4 year-olds. Birth injury/asphyxia, neonatal sepsis and prematurity contributed >70% of neonatal deaths, while acute infections, malaria, diarrhoea and pneumonia accounted for most deaths in older children.
Conclusions
Maternal HIV status had little effect on neonatal mortality but was associated with much higher mortality in the post-neonatal period and among older children. Greater attention to HIV care in pregnant women and mothers should help improve child survival but broader interventions are needed to reduce neonatal mortality.
Introduction
Background
Child mortality remains one of the biggest problems being faced by countries in sub-Saharan Africa.1 This has been compounded by high HIV prevalence in the region with an estimated 34 million people living with the virus,2 the majority of whom are women of child bearing age.3 The proportion of under-five deaths attributable to HIV in Malawi is estimated to be around 13%.2
In Malawi, with an HIV prevalence of between 10-14%, about one million adults and children are estimated to be infected.4 HIV prevalence among women in the reproductive age group is estimated at 13%,5 and before widespread use of antiretroviral therapy (ART), about 30,000 newborns were infected through mother-to-child transmission every year,6 prenatally, during delivery or through breast feeding. Maternal HIV-status also impacts on child mortality indirectly, through death or illness of the mother as a result of HIV/AIDS.7-11 Between 2005 and 2010, Malawi’s under-five child mortality was estimated at 112/1000 children nationally.5
Although child mortality is higher in children born to HIV-positive mothers than in children born to HIV-negative mothers,8-10, 12-13 Malawi’s child mortality has continued to decline5 despite high HIV prevalence. We have previously presented data on child mortality from Karonga District demographic surveillance site (DSS) up to 2006.14-15 Here we present under-five mortality and causes of death from 2006-2012, in the era of widespread antiretroviral therapy (ART), focussing on the contribution of maternal HIV status.
Methods
Study setting
The demographic surveillance area is located in the southern part of Karonga district. The total population is about 35,500 individuals, of whom 6,250 (18%) are children under five years of age.16 The population earns its living mostly through subsistence farming and fishing from the lake. The area has an adult HIV prevalence of approximately 9%.17 The community is served by a rural hospital, headed by a clinical officer, and 4 health centres, 3 government run and 1 private. At the time of the study there were two ART clinics available to residents of this area, one 70 km away from the study area which opened in July 2005, and another at the rural hospital within the study area which opened in September 2006.18 Nevirapine has been available for use in prevention of mother-to-child transmission since 2002. ART was recommended for women in WHO stage 3 or 4 or with CD4 counts of <250 (increased to 350 in 2011).
Demographic surveillance and verbal autopsy
Information about the population is collected and updated through a continuous registration system. This uses village volunteers known as key informants who each month record and report births and deaths, including stillbirths. The information reported by the key informants is complemented by a yearly house-to-house census to update household composition and migration.18 Each key informant has no more than 40 households under their responsibility.
The key informants report births and deaths to project interviewers at monthly meetings. Birth reports are immediately followed-up by interviewers, who go to the households to register the baby and record information such as date of birth, gender and parents’ information. In this area, about 80% of deliveries take place in health facilities. Following a death report, a medically-trained interviewer visits the deceased’s household to fill an age-appropriate verbal autopsy form. The form is similar to the INDEPTH verbal autopsy tool, which is an adaptation of the WHO questionnaire. The visit is done as soon as possible after a mourning period of two weeks. Repeat visits are arranged if no-one is found at home during the first visit.
The verbal autopsy forms are reviewed independently by two clinicians who assign a cause of death from a pre-defined hierarchical list. If there is disagreement the forms are reviewed by a third clinician. Full details of the study methods can be found elsewhere.16, 18-20
Socioeconomic surveys
Socioeconomic information such as level of maternal education and household head’s occupation, household’s source of drinking water, type of dwelling and main source of income was obtained from a socioeconomic survey that runs concurrently with the house-to-house annual re-census.
HIV sero-surveys
Since September 2007, HIV counselling and testing has been offered door-to-door to all individuals aged 15 years and older in the demographic surveillance area. Four rounds of HIV sero-surveys have been conducted from 2007 to 2011, with a participation rate of 55-60% testing at each round.17 In addition to voluntary testing and counselling, individuals are also asked about previous HIV-testing, including the timing and the results of the most recent test, and also about ART use if the participant reports that they are HIV-positive.17
Study Population
All children born between September 2006 and December 2011 who were born to a resident of the study area were included in the study. Children were defined as exposed to HIV if they were born to a mother who (a) had a positive HIV test result before the child was born or (b), who tested HIV-positive within one year from the child’s birth (since vertical transmission of HIV can also occur during breast feeding). A child was defined as not exposed to HIV if born to a mother who tested negative at any date after the child was born or who tested HIV-negative up to three years before the child was born, and did not have a positive test within a year of the birth. Children born to mothers whose HIV status was unknown at the time of the child’s birth were classified as unknown HIV exposure.
Statistical analysis
First we assessed the distribution of baseline characteristics by child exposure to maternal HIV status using chi-squared test of association. Children with unknown exposure status were included in the analysis as a separate category. Person years of observation (py) were calculated and used as the denominator in the calculation of mortality rates. Person years of observation (time at risk) were calculated from the child’s date of birth or in-migration to the earliest of the date of their death, the date of loss to follow-up, or the end of the study (31st December 2012) or 5 years of age. For children who had moved out of the study area and came back, only the time they were in the study area was considered in the calculation of person years.
Because mortality rates and causes of death change rapidly with age, analyses were stratified by age, separating neonatal (0-28 days), post-neonatal (29-365 days) and child (1-4 years). We calculated rates, and rate ratios, using Poisson regression. We allowed for clustering by mother using random effects or robust standard errors. The results were very similar: the results using robust standard errors are presented. We also calculated the proportion of deaths attributable to HIV among HIV-exposed children (the attributable risk percent), and in the whole population (the population attributable fraction, PAF), to measure the impact of maternal HIV on child mortality.
In model building, maternal HIV status was included as the main exposure while child age within the age groups and gender were included as a-priori confounders. Adding the other variables one at a time, starting with the variables most strongly associated with the outcome we checked if the RR and or the standard error changed or stayed the same. A change in RR of 10% or more was considered enough evidence of confounding and the variable was left in the model, otherwise it was not included. A noticeable increase in the standard error was evidence of multicollinearity and the variable was dropped from inclusion in the model. Maternal vital status was not included in the multivariable models because of small numbers.
In the multivariable analysis, only records with complete data were included. Data were analysed using STATA version 13 (Stata Corp, College Station, TX, USA). We also checked if there are differences in stillbirth rates between HIV positive and HIV negative mothers. Stillbirth rates were calculated per 1000 live births.
Ethics approval
The study received ethical approval from the National Health Sciences Research Committee of Malawi (protocol number 419) and the Ethics Committee of the London School of Hygiene and Tropical Medicine (protocol number 5081).
Results
Over the period of the study there were 7208 live births to 5266 mothers in the demographic surveillance area. Overall there were 401 deaths before the age of 5 in 21,357 person years of exposure, giving a rate of 18.8/1000py. The mortality rate among singletons was 17.0/1000py (95% CI 15.3-18.8) and among twins/triplets was 78.3/1000py (95%CI 62.6-82.8). Since being a member of a multiple birth was in itself such a strong risk factor for mortality, these children were not included in further analysis. This excluded 295 births and 49 deaths. There was no association between maternal HIV status and twin birth.
During the period of analysis, a total of 115 stillbirths were recorded, giving a stillbirth rate of 15.7/1000 live births. There was a higher proportion of stillbirths born to HIV-positive mothers compared to HIV-negative mothers: 10 were to HIV-positive mothers (representing 3.6% of the births to these mothers) and 91 to HIV-negative mothers (representing 2.1% of births to HIV-negative mothers.
The remaining analysis concentrates on the 6913 live born singleton children. They had 5152 mothers: 3500 (68%) contributed one child, 1544 (30%) contributed two and 108 (2%) contributed more than two. Of these children, 5698 (82.4%) were delivered at a health facility, 918 (13.3%) at home, and 297(4.3%) at a traditional birth attendant. Overall 5,862 (84.8%) were HIV unexposed at birth of whom 250 (4.3%) died; 335 (4.9%) were HIV exposed at birth of whom 55 (16.4%) died; while for 716 (10.4%), maternal exposure to HIV was unknown of whom 47 (6.5%) died.
Table 1 shows the distribution of baseline characteristics and their association with maternal HIV status. HIV-positive mothers tended to be older, more educated, have had at least one previous pregnancy, live nearer to a clinic or the hospital, and to be more likely to come from households with a source of income other than farming, better dwelling construction and access to tap water.
HIV status was missing if the mother was not seen during a sero-survey, or refused to be tested or if the gap between negative and positive tests made the status at birth uncertain. Of the 285 HIV-positive mothers, 52/285 (19%, who were mothers to 65 children) were recorded as being on full ART at the time of the child’s birth, 203/285 (71%, 234 children) were not on ART, and ART use was not known for 32/285 (11%) mothers (36 children). Of the 65 children born to HIV positive mothers on ART, 50 (77%) were born after their mother had been on ART for 6 or months and 13 (23%) were born to mothers who had only been on ART for less than 6 months. The extent of missing data on the other variables is shown in table 1.
Table 2 shows child mortality rates and rate ratios by each of the risk factors and by age group. We found no difference in neonatal mortality rate by HIV status of the mother while observing a very strong association between child mortality and maternal HIV status at birth among post-neonates and children aged 1-4 years old. The neonatal mortality rate was 218/1000py (95%CI 181-261), (16.5/1000 live births), post-neonatal mortality 20/1000py (95%CI 17-24), (17.4/1000 live births), and mortality at age 1-4 years 8/1000py (95%CI 9-13). Overall, the mortality rate was 14/1000py (95%CI 12-15) in those whose mothers were HIV-negative, 62/1000py (95%CI 47-81) in those with HIV-positive mothers and 30/1000py (95%CI 22-40) in those with unknown HIV exposure.
Among the neonates, mortality was highest in the first week of life 784/1000py (95%CI 636-966) compared to 63/1000py (95%CI 43-93) at 7-28 days. The risk of death among neonates born to older mothers was 50% lower, RR 0.5 (95%CI 0.3-0.7) compared to teenage mothers and was higher in primiparous mothers than in those mothers who had given birth before: RR 2.6 (95%CI 1.8-3.8). In the older age groups, mortality rates were much higher in those with HIV-positive mothers, and highest in those whose HIV-positive mothers were already on ART at the time of child birth (though there were small numbers in these groups). Table 3 shows the causes of death by age group and child exposure to maternal HIV status at birth. Marked differences in leading causes of death between the age groups were observed. In age group 0-28 days the three leading causes of death were birth injury/asphyxia (35.1%) followed by neonatal sepsis (29.0%) and prematurity/low birth weight (18.3%). In age group 29-365 days the main causes of death were pneumonia (16.7%), HIV/AIDS (15.0%), diarrhoea (10.0%), malaria (9.2%), and other acute febrile diseases (17.5%). In the 1-4 year age group, malaria (23.7%), diarrhoea (17.0%), HIV/AIDS (7.6%) and other acute febrile diseases (19.5%) were most common.
Overall 26 mothers died during the study period. Of these, 9 (34.6%) died of AIDS, 5 (19.2%) of other communicable diseases, 6 (23.1%) of non-communicable diseases, 5 (19.2%) of maternal causes, and 1 died of external causes. Of the 8 mothers who died and whose children also died, 5 died of AIDS, 1 died of TB, and 2 died of maternal causes. Maternal death increased under-five mortality both one year before and after the mother had died (Table 2), but the numbers of maternal deaths were too few for us to investigate it further in the multivariate analysis.
The proportion of deaths attributable to HIV exposure among those with HIV-positive mothers (the attributable risk percent) or in the population (the population attributable fraction) was highest in the post-neonatal age group (Table 4) and lowest in neonates. In the post-neonatal period, maternal HIV-positivity accounted for more than one third of all deaths.
Table 5 shows results of multivariable models of the association between child mortality and child exposure to maternal HIV status at birth for each age group after adjusting for age within age groups, gender, maternal age, parity and source of drinking water as these are the only variables which confounded the association as described in the methods section above. After adjusting for potential confounding variables and taking into account the effect of child clustering by mother, the rate ratios of mortality in children born to HIV-positive mothers versus children born to HIV-negative mothers were higher than the crude rate ratios in all groups: 1.5 (95% CI 0.6-3.7) among neonates; 11.5 (95% 7.2-18.5) among post-neonatal infants, and 4.6 (95% CI 2.7-7.9) in 1-4 year olds.
Discussion
Overall, child mortality was higher in children born to HIV-positive mothers than those born to HIV-negative mothers, but this effect was only seen after the neonatal period. The main causes of death were different in the different periods. In the older children more than three quarters of the deaths were due to infections. Only a few children had their causes of death classified as AIDS, perhaps reflecting the challenge of assigning AIDS as a cause of death to under-five children using information from verbal autopsies, as noted by others.7 In the neonatal period one third of deaths were due to birth injury/asphyxia (in line with other studies),21 for which the HIV status of the mother might not play a role. However, half were due to infections or prematurity, so the relatively small effect of HIV exposure is surprising.
Both our neonatal and post-neonatal crude mortality rates were similar to those reported elsewhere in the sub-Saharan region.22-23 The crude death rate among children born to HIV-positive mothers was very high. It was 9 times higher than the rate in those born to HIV-negative mothers in the post-neonatal period and 4 times higher in children aged 1-4 years. But the difference by HIV status of the mother among neonates was negligible (rate ratio 1.1) and not statistically significant. In a previous study in Karonga district in 1998-2002 we had also found little effect in neonates, crude rate ratio 1.7 (0.7-4.4) compared to rate ratios of 3-5 in the older age groups.24 In a clinic based prospective study in Blantyre, Malawi, the greatest increase in mortality rates was also found in the post-neonatal period.13 In Zomba, Malawi, higher mortality for children of HIV-infected mothers was found in each period, including neonates.25 A study in Rwanda found no difference in child mortality by maternal HIV status at 6 months, but observed a significant difference in mortality at 9 months. The lack of an effect of HIV on mortality in the first few weeks of life, despite an increase in the stillbirth rate, has also been reported from South Africa.26 Most studies of the effect of maternal HIV status have not analysed the neonatal and post-neonatal periods separately.
In our study, maternal HIV-positivity accounted for most of the deaths of older children with HIV-positive mothers and one third of all deaths in the post-neonatal period (Table 4). It accounted for only a small proportion of neonatal deaths even among those with HIV-positive mothers, which is consistent with the reported common causes of neonatal death.
Children who lost their mothers were at greater risk of dying than those whose mothers were still alive. This accounted for a small number of deaths, but the effect was large in the post-neonatal period, with 3 babies dying before and another 3 dying after their mothers had died.
Interestingly, the children who were born to mothers who were on ART exhibited higher mortality rates than those born to mothers who were HIV-positive but not on ART. This is in contrast to other studies.27-30 This could be a chance finding because numbers were small as only a minority of mothers were already on full ART at the time of their child’s birth, but it may also reflect more advanced HIV disease in the mother which is associated with higher mortality in the children.8, 31-32
This is a large, population-based study with long follow-up and minimal loss to follow-up, detailed data on potential confounders and information on causes of death. However, the study was not designed to define the mechanisms underlying this excess mortality in the HIV exposed child. We had no information on the use of Nevirapine to prevent mother-to-child transmission and there was no HIV DNA testing for infants or routine HIV testing for children. We cannot therefore differentiate HIV-infected children from HIV-exposed and uninfected, who would be expected to have a much lower mortality rate33 although it may still be higher than that in HIV-unexposed children.34-35 Access and use of other interventions known to affect mortality such as, Vitamin A supplementation, midwife attended delivery, completeness of infant vaccine schedules, duration of breastfeeding, were not systematically assessed. Future work will need to focus on understanding modifiable factors and the relative contributions of HIV-infection in the child, inadequate provision of maternal care and biological factors predisposing to serious illness in exposed-uninfected children.
Conclusion
Maternal HIV-positive status at birth has a major impact on child mortality with the relative risk being felt differently across age groups. The new Malawi government policy on preventing mother-to-child transmission using the World Health Organisation Option B+ of “universal life-long ART for all HIV infected pregnant and breastfeeding women regardless of clinical or immunological stage”36 needs to be optimised to prevent more children from getting infected with HIV and maintaining the health of HIV-infected mothers. This can be expected to reduce post-neonatal and child mortality, but different interventions will be needed to reduce neonatal mortality.
Acknowledgements
Dr. Lumbani Munthali
Dr. Terence Tafatatha
Elenaus Mwaiyeghele
Fredrick Kalowebekamo
Lackson Kachiwanda
Vitumbiko Nyirenda
Taniel Njawala
Research was funded by: Wellcome Trust
Footnotes
Conflict of interest and source of funding: None of the authors declare any conflict of interest and source of funding: the study was funded by Wellcome Trust, grant # 079828/Z/06/Z and 096249/Z/11/A
“Data presented previously at the Global Health Metrics and Evaluation, 2013 Conference in Seattle, Washington, 17-19th June, 2013, and published as abstract in (The Lancet - 17 June 2013 (Vol. 381, Page S29 ) DOI: 10.1016/S0140-6736(13)61283-0)”
Facebook
Twitter
Google+