Neonatal mortality within 24 hours of birth in six low- and lower-middle-income countries

Abstract Objective To estimate neonatal mortality, particularly within 24 hours of birth, in six low- and lower-middle-income countries. Methods We analysed epidemiological data on a total of 149 570 live births collected between 2007 and 2013 in six prospective randomized trials and a cohort study from predominantly rural areas of Bangladesh, Ghana, India, Pakistan, the United Republic of Tanzania and Zambia. The neonatal mortality rate and mortality within 24 hours of birth were estimated for all countries and mortality within 6 hours was estimated for four countries with available data. The findings were compared with published model-based estimates of neonatal mortality. Findings Overall, the neonatal mortality rate observed at study sites in the six countries was 30.5 per 1000 live births (range: 13.6 in Zambia to 47.4 in Pakistan). Mortality within 24 hours was 14.1 per 1000 live births overall (range: 5.1 in Zambia to 20.1 in India) and 46.3% of all neonatal deaths occurred within 24 hours (range: 36.2% in Pakistan to 65.5% in the United Republic of Tanzania). Mortality in the first 6 hours was 8.3 per 1000 live births, i.e. 31.9% of neonatal mortality. Conclusion Neonatal mortality within 24 hours of birth in predominantly rural areas of six low- and lower-middle-income countries was higher than model-based estimates for these countries. A little under half of all neonatal deaths occurred within 24 hours of birth and around one third occurred within 6 hours. Implementation of high-quality, effective obstetric and early newborn care should be a priority in these settings.


Introduction
Neonatal mortality remains unacceptably high and the risk is greatest on the first day of life -these were the conclusions of the 14th annual State of the World's Mothers report, entitled Surviving the first day, 1 which presented estimates of the risk of maternal, neonatal and first-day deaths in 186 countries for 2011. Worldwide, it was estimated that more than 1 million babies died within 24 hours of birth in 2011, which represented more than one third of all neonatal deaths and 15% of all deaths in children younger than 5 years. The risk was highest in sub-Saharan Africa but the number of deaths was highest in South Asia.
Estimated risks for 2013, which have recently been published, were similar. 2 The estimates for 2011 and 2013 were both derived using a modelling approach based on data on live births and neonatal deaths from the United Nations Inter-agency Group for Child Mortality Estimation. Although data from vital registration systems were available for 109 countries, only those from 57 countries were judged to be of high quality -the proportion of deaths in the first day of life was calculated directly for these countries. For the other 129 countries without high-quality vital registration system data, the proportion of deaths in the first day of life was estimated from country-specific United Nations' data using a neonatal survival curve derived from Demographic and Health Survey data for 79 countries (median reporting date: 1999). Modelbased estimates provide essential information on countries lacking high-quality direct assessments: they illustrate the magnitude of the problem, enable cross-country comparisons to be made and help guide health policy and health service planning. However, the estimates must be accurate and upto-date. For many countries, epidemiological studies have collected prospective, population-based information on pregnancies, births and newborn deaths that could be used to replace model estimates with high-quality, up-to-date data. We obtained such information from community surveillance systems established as part of large, ongoing or recently completed studies in six low-and lower-middle-income countries in sub-Saharan Africa and South Asia: Bangladesh, Ghana, India, Pakistan, the United Republic of Tanzania and Zambia. The aims of this Objective To estimate neonatal mortality, particularly within 24 hours of birth, in six low-and lower-middle-income countries. Methods We analysed epidemiological data on a total of 149 570 live births collected between 2007 and 2013 in six prospective randomized trials and a cohort study from predominantly rural areas of Bangladesh, Ghana, India, Pakistan, the United Republic of Tanzania and Zambia. The neonatal mortality rate and mortality within 24 hours of birth were estimated for all countries and mortality within 6 hours was estimated for four countries with available data. The findings were compared with published model-based estimates of neonatal mortality. Findings Overall, the neonatal mortality rate observed at study sites in the six countries was 30.5 per 1000 live births (range: 13.6 in Zambia to 47.4 in Pakistan). Mortality within 24 hours was 14.1 per 1000 live births overall (range: 5.1 in Zambia to 20.1 in India) and 46.3% of all neonatal deaths occurred within 24 hours (range: 36.2% in Pakistan to 65.5% in the United Republic of Tanzania). Mortality in the first 6 hours was 8.3 per 1000 live births, i.e. 31.9% of neonatal mortality. Conclusion Neonatal mortality within 24 hours of birth in predominantly rural areas of six low-and lower-middle-income countries was higher than model-based estimates for these countries. A little under half of all neonatal deaths occurred within 24 hours of birth and around one third occurred within 6 hours. Implementation of high-quality, effective obstetric and early newborn care should be a priority in these settings.

Research
Neonatal mortality within 24 hours Abdullah H Baqui et al.
analysis were to estimate the neonatal mortality rate and the percentage of neonatal deaths occurring within the first 6 and 24 hours of life in these countries and to compare the results with modelbased estimates.

Methods
Our analysis involved epidemiological data from population-based, cohort or intervention studies carried out at seven sites in the six countries between 2007 and 2013 and coordinated by the World Health Organization (WHO; Table 1; available at: http://www.who.int/bulletin/volumes/94/10/15-160945) Intervention studies that recruited newborn babies were excluded as some deaths that occurred before enrolment could have been missed. Data were derived from five cluster randomized trials, one individually randomized trial and one cohort study. Three trials evaluated the effect of applying chlorhexidine to the umbilical cord on neonatal mortality, one examined low-dose vitamin A supplementation for women of reproductive age and three involved home visits by community health workers. Together the trials included over 2 million women in predominantly rural communities. All studies included a community-based surveillance system involving home visits by trained field workers who collected data on vital events, including pregnancies, births, deaths, illness and migration, every one to three months. For the analysis, data from both intervention and control arms were pooled if there was no evidence that the intervention affected neonatal mortality. Otherwise, only data from the control arm were included.
For Bangladesh, the analysis used data from the control arm of a cluster randomized trial evaluating the effect of different umbilical cord cleansing regimens on neonatal mortality and infection. 3,4 Between 2007 and 2009, community health workers made home visits every two months. The dates and times of births were recorded soon after birth and the dates and times of deaths were recorded during postnatal visits on days 1, 3, 6, 9, 15 and 28. For Ghana, data on all participants in two cluster randomized trials were used because the interventions had no significant effect on neonatal mortality. 5,6 The trials were conducted in seven predominantly rural districts in 2008: one examined the effect of low-dose vitamin A supplementation on maternal and infant mortality and one evaluated the effect of a home visit programme on neonatal mortality and newborn care practices. Trained field workers made home visits every four weeks, at which they recorded births and deaths. For each neonatal death, a verbal autopsy was performed by a trained supervisor, who determined whether or not the death occurred within the first 24 hours. For India, data from the control arm of a cluster randomized trial on the effect of an Integrated Management of Neonatal and Childhood Illness programme on neonatal and infant mortality were used. 7 The trial was conducted in rural and semi-urban areas of Haryana State between 2008 and 2010. Home visits took place monthly and all households with a live birth were visited on day 29 to document the vital status of the infant. As in Ghana, an independent verbal autopsy was performed for any death.
For Pakistan, data came from a Health and Demographic Surveillance system covering approximately 274 000 people in urban and peri-urban Karachi and from the control arm of cluster randomized trial in Sindh. In Karachi, data were collected by community health workers who visited households every three months in 2011. 8 In Sindh, the trial investigated the effect of interventions involving traditional birth attendants and Lady Health Workers on birth asphyxia, sepsis and the number of lowbirth-weight infants and, consequently, on neonatal mortality. For the United Republic of Tanzania, data from all participants in an individually randomized trial examining the effect of applying chlorhexidine to the umbilical cord stump on neonatal mortality were used because the intervention had no significant effect. The trial was conducted on Pemba Island and involved health-care workers in the local maternity ward, maternal and child health workers and traditional birth attendants. All births were registered centrally and households where a birth occurred were visited by a field supervisor; 90% of visits were made within 12 hours of delivery. The date and time of birth were confirmed with the birth attendant. For Zambia, data from the control arm of a cluster randomized trial on the application of chlorhexidine were used. The trial was conducted in the Southern Province between 2011 and 2013. 9 As over 95% of women attended antenatal clinics during pregnancy, trial enrolment took place at health centres and during community outreach programmes conducted by health workers. Data on births and deaths were collected through a community-based birth notification system and during home visits after delivery.

Statistical analysis
At all sites, a neonatal death was defined as a death within the first 28 days of life (i.e. day 0 to day 27). For Bangladesh, India, the United Republic of Tanzania and Zambia, we estimated mortality within the first 24 hours directly using the difference between the time and date of birth and the time and date of death. For Ghana, the occurrence of death within the first 24 hours was determined by verbal autopsy and, in Pakistan, deaths that occurred on the day of birth were taken as deaths that occurred within the first 24 hours. Mortality in the first 24 hours of life was expressed as deaths per 1000 live births and as a percentage of all neonatal deaths. In addition, mortality in the first six hours of life was calculated for the four sites where relevant data were available. For comparison, we used modelling estimates of neonatal mortality, mortality within the first 24 hours of birth and the percentage of all neonatal deaths occurring within 24 hours for each country derived by Oza et al. 2

Results
The total number of live births included in the analysis was 149 570: the number in each country ranged from 11 143 in Bangladesh to 44 450 in the United Republic of Tanzania ( Table 2). The pooled neonatal mortality rate across all studies was 30.5 per 1000 live births and country-specific neonatal mortality rates ranged from 13.6 per 1000 live births (95% confidence interval, CI: 12.0-15.3) in Zambia to 47.4 per 1000 (95% CI: 44.9-49.9) in Pakistan. Overall, mortality within the first 24 hours was 14.1 per 1000 live births -the lowest rate was 5.1 per 1000 live births in Zambia and the highest was 20.1 per 1000 live births in India. The percentage of neonatal deaths occurring within the first 24 hours was 46.3% overall: the figure ranged from 36.2% (95% CI: 33. 6 within the first six hours were available, the mortality rate was 8.3 per 1000 live births overall, which corresponded to 31.9% of the overall neonatal mortality rate at these sites. The proportion of neonatal deaths that occurred in the first 6 hours ranged from 17.5% in Zambia to 36.0% in the United Republic of Tanzania (Table 2). Fig. 1 shows a comparison of the neonatal mortality rates we observed in the six study countries with the corresponding rates derived in a modelling-based study by Oza et al. 2 According to the modelling study, 36% of neonatal deaths would be expected to occur in the first 24 hours of life in our six study countries, which is lower than the average of 46% we observed.

Discussion
Our main finding was that, overall, a little under half of all neonatal deaths in the study populations occurred within 24 hours of birth -the figure ranged from 36% to 66% across the study sites -and about one in three neonatal deaths occurred within six hours of birth. Our data confirm that neonatal mortality remains high in many low-and lowermiddle-income countries.
Our study has several strengths. The total study population was large and data on almost 150 000 live births and over 4500 neonatal deaths were included. Moreover, as the data came from large, high-quality and prospective surveillance systems that were established to capture population-based data on births and deaths, the numerators and denominators were accurate. In particular, our data were more likely to be complete than data from large countrywide surveys, such as Demographic and Health Surveys, which have, in the past, underestimated the true burden of rare events such as neonatal deaths and stillbirths. 10 In addition, nationally representative surveys suffer from recall bias, which may be particularly important for deaths occurring very close to birth. Many mothers may report neither the birth nor the death of such babies during surveys. The studies we analysed prospectively identified and followed up all births in the target populations and the possibility that stillbirths were misclassified as very early neonatal deaths was minimized in these studies by active surveillance and careful verbal autopsy.

Comparison with model-based estimates
Generally the neonatal mortality rates we observed were significantly different from the model-based estimates: the 95% CI for the observed neonatal mortality rate did not include the modelled estimate for any country except Ghana. Moreover, the observed proportion of neonatal deaths occurring in the first 24 hours was significantly higher than the modelled estimate of 36% for all countries except Pakistan and Zambia. In Pakistan, the observed proportion of deaths in the first 24 hours may have been underestimated because, in one trial, only deaths recorded on the date of birth were regarded as having occurred in the first 24 hours. This explanation does not apply to Zambia, however.
Our observation that a higher proportion of neonatal deaths occurred in the first 24 hours than predicted by modelling must be interpreted in the context of the overall neonatal mortality rate. The proportion of neonatal deaths occurring in the first day of life is generally higher when the neonatal mortality rate is low: in 2011, a proportion of 35% was reported in sub-Saharan Africa where the neonatal mortality rate was 34 per 1000 live births compared with a proportion of 67% for developed countries where the neonatal mortality rate was 3 per 1000 live births. 1 In our study, the observed neonatal mortality rate was higher than the modelled estimate for Bangladesh, Ghana, India and Pakistan but lower for the United Republic of Tanzania and Zambia. Although no special criteria were used to select the study populations, these populations did live in geographical areas with high mortality. Consequently, the neonatal mortality rates in these settings may have been higher than the national averages. Given higher neonatal mortality rates, the proportion of deaths occurring in the first 24 hours would have been expected to be lower, but we found the opposite. This observation has important implications for maternal and newborn health programmes in lowand lower-middle-income countries because most neonatal deaths globally occur in populations similar to our study populations, in which the neonatal mortality rate is likely to be near or above the national average.
The neonatal mortality rates observed in our study would have been expected to be lower than modelled estimates because the data were collected more recently: between 2007 and 2013 compared with a median survey year of 1999 for the data used in the model. However, improvements could have subsequently taken place in the study areas. In particular, there have been many advances in newborn care in recent years in settings where access to adequate health care is not universal. For example, in a large trial in Ghana, home visits by community health workers to promote essential newborn care practices have been shown to reduce neonatal mortality by 12% and have the potential to reduce childhood mortality by 4.9%. 6 In our study, the observed neonatal mortality rate was lower than the modelled estimate only in the United Republic of Tanzania and Zambia. We believe the higher rates we found in the other four countries were due to differences in methods used: our estimates were based on prospectively defined population groups in which all births were followed up whereas the model's estimates were based on cross-sectional surveys, which are more prone to recall bias and, conse-quently, to underestimation of the real neonatal mortality rate. Although the provision of intrapartum and postnatal care by skilled birth attendants is known to be crucial for neonatal survival, we observed no clear relationship between coverage by skilled birth attendants or antenatal care and neonatal mortality. This finding implies that neonatal mortality and the proportion of neonatal deaths in the first 24 hours may depend on other factors, such as: (i) the quality of the intrapartum and postnatal care; (ii) the characteristics of the infant, such as birth weight; or (iii) socioeconomic factors, such as maternal literacy.

Limitations
One limitation of our analysis is that our estimates of neonatal mortality were not nationally representative because the studies were conducted in predominantly rural areas. However, our findings are generalizable to similar, vulnerable populations in low-and lower-middleincome countries because the neonatal mortality rates we observed were not influenced by averaging across different population types. Another limitation is the variation between studies in the surveillance carried out to identify neonatal deaths. Although frequent home visits were made during the neonatal period at most sites, only one or two visits took place in Ghana and India. However, the dates and times of births and deaths were confirmed at all sites by verbal autopsies conducted by experienced field workers within three months of the death. These field workers spent one or two hours with families to determine the circumstances surrounding the death and its timing. Although there is a possibility that the time of death was recalled incorrectly, recall bias is substantially greater in surveys like Demographic and Health Surveys, where the interview may take place several months or even years after a neonatal death. Another limitation of our study is that the exact time of death was not available in Pakistan, where the date of birth was used to determine whether the death occurred within the first 24 hours of life. Finally, in Zambia, pregnant women enrolled at health-care facilities during antenatal visits and during community outreach programmes -they were not identified by routine community-based surveillance.
In conclusion, our findings suggest that maternal and child health programmes in low-and lower-middle-