Subnational estimates of maternal mortality in Nigeria: Analysis of female siblings’ survivorship histories.


 BACKGROUND High Maternal Mortality (MM) in Nigeria is further complicated by the lack of reliable estimates for subnational levels such as states and geopolitical regions. Disaggregating maternal mortality estimates by subnational levels is crucial to ensuring policy decisions and program implementation are adapted to areas with a high burden of mortality. This study involves a novel adaptation of small area estimation techniques to derive plausible estimates of levels and trends in Maternal Mortality rates and ratios for states and geopolitical regions in Nigeria. METHODS. Survivorship history data of 293,769 female siblings were provided by 114,154 women in the Nigeria Demographic and Health Surveys of 2008, 2013 and 2018. MM Rates and Ratios were estimated using the Empirical Bayesian technique for small area demographic estimates. The James-Stein estimator was used to shrink the estimates closer to the population mean values with 95% Confidence Interval (CI). RESULTS Levels of MMRatio were highest in the rural areas, States and regions in Northern Nigeria. MMRatio was consistently lower in the South West (2008=281; 2013=367; 2018=392) and higher among the Northern regions of the country, particularly the North-East (2008=654; 2013=612; 2018=901) for three consecutive surveys. Over the three surveys, mortality trends declined about 18% in the North West and 54.2% in the South East region. However, there was a 4.8% increase in MMRatio for South West between 2008 to 2018. CONCLUSIONS Nigeria has geopolitical and sub-national disparities that pose a burden to the country’s maternal health. Since several states in the Northern geopolitical zone still show high maternal mortality, targeted intervention at state levels should be explored to ensure that mothers who need help get it to ensure the sustainable development goals are met.


BACKGROUND
High Maternal Mortality (MM) in Nigeria is further complicated by the lack of reliable estimates for subnational levels such as states and geopolitical regions.Disaggregating maternal mortality estimates by subnational levels is crucial to ensuring policy decisions and program implementation are adapted to areas with a high burden of mortality.This study involves a novel adaptation of small area estimation techniques to derive plausible estimates of levels and trends in Maternal Mortality rates and ratios for states and geopolitical regions in Nigeria.

CONCLUSIONS
Nigeria has geopolitical and sub-national disparities that pose a burden to the country's maternal health.Since several states in the Northern geopolitical zone still show high maternal mortality, targeted intervention at state levels should be explored to ensure that mothers who need help get it to ensure the sustainable development goals are met.Background Elevated levels of maternal mortality are a challenge for population health and development.In 1987, the United Nations launched the Safe motherhood initiative (SMI) in Kenya.This initiative was established to reduce death during pregnancy and after childbirth.The SMI aimed to reduce the MM ratio by 50 percent by the year 2000 (1).
Several other initiatives such as the Millennium Development Goals (MDGs) and Sustainable Development Goals (SDGs) introduced various programmes targeted at reducing the global level of MM.Despite these efforts, evidence suggests only a modest reduction in maternal mortality in developing countries (2).According to the World Health Organization, 99% of all maternal mortality occurs in LMIC, and it is endemic in rural areas and poor communities (3).Globally, Nigeria and India rank top on the list of countries with the highest estimated number of maternal deaths according to WHO with an estimate of 67,000 and 35,000 maternal deaths (23% and 12% of global maternal deaths) respectively (2).Sub-Saharan Africa as a region has a high MMRatio (MMRatio = 542 in 2017).Nigeria with MMRatio = 917, is the most populous country in SSA and as such contributes largely to the burden of MM (2).
While several analyses of MM trends show that Nigeria is making progress in reducing the maternal mortality rate, the pace remains slow as a woman's chance of dying from pregnancy and childbirth is 1 in 13 and more startling is that most of these deaths are preventable (2).Several doubts have risen about the numbers that have been published as the rates of Maternal Mortality in Nigeria, considering the fluctuation and inconsistency of the figures and the uncertainty of their sources.The difficulty in measurement can be attributed rightly to the inadequate recording of adult deaths, misclassification of maternal death, and the relatively rare nature of maternal deaths (4-7).Nigeria, as a country, has an inefficient vital/civil registration system, a challenge several developing countries are battling (8,9).In the absence of a complete vital registration system, which should have been the accurate source of number and causes of deaths, these concerns about the estimates are not outrageous in themselves since estimates are generated by alternate methods based on several assumptions or from health facilities neglecting events that occurred out of the hospitals.Therefore, Nigeria does not only contribute enormously to the high maternal mortality rate in the world but also still has challenges in the measurement of the specific estimates.
Consequently, the various interventions and efforts to reduce maternal deaths and maternal mortality rates cannot be appreciated nor can impact be properly measured, if there are no adequate data and reliable estimates to measure the various performance indicators.Like most LMIC, there are relative inadequacies observed in the information on maternal mortality (MM) in Nigeria.Additionally, without valid estimates for the national and subnational subpopulations, the interventions cannot be targeted accurately to the groups of individuals who need them the most.This can be linked to the recent emphasis on a need to disaggregate data by variables such as socioeconomic status, geographical area, or even sex in the aim to reinforce data monitoring and accountability (10).
The question, therefore, remains, "what is the magnitude of maternal mortality and how is this burden distributed across different states, to ensure the government appropriates the interventions successfully?"There are no generally accepted consistent estimates of the maternal mortality rate in Nigeria.There seem to be differences in the estimates produced in various studies and used for various purposes (additional table 1).Not only are these figures displaying wide variation and disparity, but they concealed the differentials of these estimates within the different regions, states, and socio-economic groups in respective countries.
Several estimates that were provided in the past have been criticized for either being too low or too high (11).Besides, it has been argued that they do not reflect the impact of several interventions that have been implemented for maternal mortality reduction (12).
Another obvious inadequacy of the existing estimates for Nigeria is that they refer to the country as a whole: there are no differentials such as urban/rural, geopolitical zones, and administrative entities such as states that are necessary for disaggregated planning purposes.Meanwhile, States are semi-autonomous and empowered to design their policies and programmes.It is therefore essential to have subnational estimates of maternal mortality useful for state-level initiates on maternal health indices. .Therefore, this study involves a novel adaptation of small area estimation techniques to derive plausible estimates of Maternal Mortality rates and ratios for the thirty-six states, six geopolitical regions, rural and urban areas of Nigeria.

Data Source
This study is a demographic and statistical analysis of cross-sectional population-based data obtained from the Nigerian Demographic and Health Surveys of 2008, 2013, and 2018.
For this analysis, the maternal and adult mortality module is also known as the sibling survival module which was added to 2008, 2013, and 2018 Women's Questionnaire was used.The respondents were asked questions about their siblings born to the same biological mother.The name of each of the siblings is provided from the oldest to the youngest, with which the interview proceeds to find more details about each of the siblings.The current age of the siblings is required as well as the marital status, for living siblings.The age at death and year since death is asked for siblings that are reported to be dead.Female siblings who are above the age of 15 are further probed about.The interview asked if the sister died during pregnancy, childbirth, or during the postpartum period.Then MM rates and ratio were derived using the Empirical Bayesian Estimation of MM for states.This method was also adopted by Ahmed & Hill to generate similar estimates for MM in Bangladesh (13).Selected factors in line with the McCarthy and Maine analytical framework (14) were explored as covariates to get estimates that were being used in the comparison of MM levels across states in Nigeria.
In preparing the data for analysis, the period length is captured by computing reference period which is the seven-year period prior the survey, excluding the month of the interview i.e. 0 -6 years preceding the survey.The Individual sibling respondent dataset was then reconstructed into panel data (person-years) using the varstocases command in SPSS and each reported sibling was counted as an observation and is the unit of analysis from the siblings' history.This reconstructed dataset is labeled as the MM dataset.It has the records of all female siblings reported by the individual women.The data of female siblings who were dead from maternal causes were then used for further analysis.Female siblings who are reported to have died were assumed to be exposed to the risk of dying for 6 months in their year of death and this was considered in calculating the person-years of exposure.For entries with missing value on the survival of the siblings, it was excluded from the analysis.Age was adjusted for all the estimates generated and sampling weight was taken into consideration for all analyses.The dataset was then disaggregated to the various sub-population which include the 36 states and FCT.This was done using the IBM SPSS Syntax in Version 21.0.

Statistical Analysis
A direct estimator of MMRate was obtained based on sample weights of the information of maternal deaths from the NDHS.This method is insufficient to obtain the desired parameter in a small area because there might be small areas not represented adequately in the sample size or not large enough to provide a stable and precise estimate.
A synthetic estimate also called an indirect estimate was obtained using the equation: ε = error term X' = Vector of covariates, measured at aggregate/mean for every small area.X is a vector of auxiliary variables that are mortality predictors which would be measured as a mean of the values for the sub-national levels.So, the mixed model is optimally based on direct and indirect estimates of Y.This prediction is known as best linear unbiased prediction (BLUP) and is a weighted estimate of the direct and indirect estimators which "borrows strength/information" from related areas and groups.This information provided from other related areas increases the effectiveness of the sample size, and in return, the precision of the estimate derived.
However, the expected value of the   then ignores the error term It ignores the diversity (heterogeneity) of all the small areas based on the assumptions of the areas having similar characteristics; it then assumes that the MMRatio is the same.
One of the techniques the small area estimation makes use of is the Random effect model also known as the mixed model.This is different from the generalized linear models as it includes all models in the variance components procedure.MIXED model handles correlated data, unequal variances and complicated situations in which units are nested in a hierarchy, for example, data obtained from a sample of respondents from a sample of states and political regions in Nigeria, as in the NDHS data.
where   is the heterogeneity/diversity across the small areas.
~(0,   2 ) � ′  =Indirect estimator γ= Shrinkage factor (SF) for area j. Hence, The maternal death counts were treated as the response variable, and region of residence, wealth index, religion and level of education were the covariates in the model and an offset variable, the logarithm of the persons-year exposure.

Discussions
The study has successfully provided plausible estimates of MM, highlighting the critical areas where maternal mortality rates and ratios are highest in the major sub-populations in Nigeria.
Prior to this research, attempts have not been made at using the widely accepted Nigerian Demographic Health Survey datasets to generate disaggregated rates for MM in Nigeria.There were arguments also on the magnitude of mortality among women of reproductive age in Nigeria.
This study has ascertained the number of maternal deaths observed in each state in Nigeria.The Also, although the Northern region had a higher burden of MM, a few states contributed to the burden of MM reported in the various geo-political regions.further investigation has been made to ensure that each state in the geo-political region is accounted for, to reveal the magnitude of burden they contribute to each region.The observed differences in MM between the various states mirrors inequalities that has been observed in other developed countries (15).These states' estimates also differ greatly from hospital-based studies in the various states in the country, which are relatively high (additional table 1).This resonates with a previous study in Malawi (16).This highlights possible political will issue and administrative lag in commitment to the health services of individual states.This trickles to the allocation of resources from the central pool to address the healthcare needs of each state.If there are no small area sub-national estimates of mortality indices, in this case MM, and resources are being allocated to each state equally or based on other indicators other that the burden of mortality and monitored and evaluated healthcare needs, then, the real high risk areas will be neglected.This might in turn cause the heavy inequality in the MM experience of women in neighbouring states within the same geographical locations.In addition, the Northern region has states with high fertility in the country.This means that women of reproductive years are more exposed to the risk of child-bearing in these regions.It is also known that these regions are socially conservative and have practices of early girl-marriages most especially in their rural regions, which can be found largely in northern areas compared to the south (17).
In comparison with sub-national MM estimates, findings from this study suggests that facilitybased estimation of MMR, are not substantive representative of these states in which they were carried out.These studies might have over reported the phenomenon, in that it is concentrated for women that were able to access health care at the clinics where the study was carried out.This leaves out other deaths that occur at home, that could not reach the health care centres or hospitals, and in fact the deaths that were measured might just be emergencies that were rushed into the clinics.Hospital based MMR is rather influenced by a delay in the health seeking behaviour of the women, It can be concluded that facility-based estimates are unacceptably high.Also, worthy to be observed is that these model-based estimates slightly differs from the set of estimates presented by the Institute for Health Metrics and Evaluation (IHME) of the University of Washington in Seattle.Their regression model differed from the UN Interagency with the use of more AIDS or AIDS-related deaths in to the regression model used in obtaining the MMRatio.The IHME estimated maternal deaths to be 342,900 compared to the UN estimates of 358,000 maternal deaths.This was used to obtain IHME estimates of 251 per 100,000 live-birth (range 221-289) and UN estimates was 260 (range 200-370).According to Abouzahr (11) , these estimates differ in the statistical methods used in deriving the parameters and does not necessarily mean one is superior to the other.While the UN estimates used Gross National Income (GNI) as a covariate in their analysis, as well as general fertility rate and proportion of deliveries attended to by skilled birth attendants, in addition the IHME covariates included total fertility rate, HIV zeroprevalence, neonatal mortality, age-specific female education as well as age.Although still birth attendant was included in the IHME analysis, it was not an addition of the predictive validity of the estimates of MM.It is difficult to judge one method as superior to another as the statistical models are rather descriptive than explanatory in nature.Hence, it will suffice to say experts in various countries study county specific situation and data availability to solve issues of estimates for policy decision making.For this study, the Empirical Bayesian Method for small area estimates works perfectly in Nigerian.This approach centres the estimates around an average by borrowing information within the population to generate a refined estimates with assumptions suitable for small area estimations.This is a major strength for the small area estimation technique utilizing the empirical Bayesian method.Furthermore, the estimates from this method yielded a narrower 95% confidence intervals for generated estimates.
In order to tackle high MM in Nigeria, sub-national disparities need to be addressed.This can be urban-rural, geo-political region and even the various states' context.This is beside the concentrated effort made at the central government level.Socio-economic and health development imbalances impede the progress of a country's global or public health improvement.
If there are left behind groups in a population, achieving any of the sustainable goals will be sabotaged by huge spatial inequalities.The disaggregation of the data into the sub-population as adopted in this study has provided plausible estimates with which MM in Nigeria's sub-population can be described, monitored and curbed.At this stage, in Nigeria, level of MM produced in this study for each sub-population might not be precise estimates, but it is sufficient to raise the consciousness of the government and policy makers to the magnitude in various types of places of residence, geo-political zones and states.For instance, estimates bordering between 300-700 per 100,000 might be given same policy responses, however, sub-population with estimates higher than that are definitely red flagged areas.Evidence-based decisions clearly require reliable estimates, in the absence of which resources will be wasted undetected.This has provided researched evidence for a need to target intervention programmes to the high risks areas like the North Central, North West and some part of the South-South, where MM is highest and most likely to occur.
With increasing demographic transition and change in population dynamics, there is a need to delineate population data to accommodate the heterogeneity of various socio-demographic groups.In Africa, women of reproductive age differ by risks process, urbanization, and geopolitical regions, which provides a challenge for policy implementation.This study has provided estimates that allow for spatial mapping of small area MM experience in Nigeria.This helps for understanding geographical variation and allocating decentralized resources, and policies to curb MM in sub-national areas with high level of MM.This can also assist social demographers in assessing etiological hypotheses in researching the high-risk areas of MM in Nigeria per state.In many instances, maternal health policies are rather generic; they are extended to all women of reproductive ages and do not account for disparities among most vulnerable and underserved women.Consequently, since challenges and choices differ for women in various environments and socio-economic groups, pooling programs and intervention without adaptive solutions is not as effective.Despite several interventions, Maternal Mortality (MM) remains high in Nigeria.The focus ought then to shift from pushing out programmes and intervention arbitrarily to ensuring maternal health care are evidence-based, tailor-made and available for underserved population that contribute largely to maternal health inequities.It is widely accepted that actions that improve the maternal health of women of reproductive ages not only vary across the age groups but also from countries, communities, and other subpopulations as applies.This makes this study fulfil one of the basic tenants of public health in understanding spatial patterns of health-related problems (18), since public health interventions, even though will be a common thread, actions, programmes for each subpopulation should be guided by evidence drawn from sound scientificresearch (19).This has also in essence crossed the hurdles of unreliable national estimates due to unavailability of CVRS and the rareness of maternal deaths in a statistical sense (20).
Reliable sub-population data and estimates on mortality are essential for policy and for planning to monitor the progress and development of a country against set goals.In Nigeria, since Vital Statistics Registration System (CVRS) is unavailable, small area demographic estimation methods can be explored in the interim.This can be by disaggregating population-based data and exploring direct estimation or using model-based approaches (10,21).Within country comparison of demographic estimates, mortality will reveal the dimensions of inequalities in the population.
Whiles the availability of the NDHS has brought a rich dataset for demographers to understand the dynamics of population and estimates indices in Nigeria, strengthening the complete CVRS should be a key priority in the country.The registering of births and deaths should be an integral part of the nation's health surveillance culture.In the meantime, more investments should be put in place into the NDHS in enhancing the data quality.Small area datasets need to be collected in national surveys.It might be expensive to have a single survey capture all the information needed, however, data on both health and inequity might be gotten from different sources.For instance, if the data source captures studies for different purposes, it might decide to collect data not only at household level but also put into consideration disaggregation that allows for regional analysis and sub-national estimates which might include, race, ethnicity, economic status etc. Therefore it means sampling must always align with administrative stratification for uniformity.Also, since health intervention programmes are aimed to curb health menaces and also to reduce disparities, regional or state level monitoring of demographic indices will be a useful tool to provide benchmarking terms.This will ensure that there is appropriate resource allocation according to the magnitude of burden in each sub-national population.This is particularly more effective when the country's health system is decentralized and allows to capture the substantial differences that may occur in the various geographical areas.
It is no news that a population-wide intervention would cost more money and resources to implement, hence, focused sub-population-based interventions have been proven to bring about more reduction in MM (22).Building a sustainable evaluation capacity at the country and state levels will help in the allocation of scarce resources.Evidence-based intervention, programmes, and policies can be made to various states and geo-political zones.This enhances the cases of inclusiveness for rural residents and vulnerable people across the country.There is a need to improve and scale-up demographic estimates for mortality and fertility in different subpopulations in Nigeria exploring the robustness of the Bayesian method and more importantly to strengthen small area demographic estimates in Nigeria and Sub-Saharan Africa at large.The Bayesian method is a rich method that can utilize data from ranges of sources and measure uncertainty in resultant rates.It also has the capability of smoothing data across age, time and space as well as correct mortality data for its incompleteness.More investigation will be required, largely through qualitative researches and probably maternal surveillance audits and autopsies, to determine the factors contributing to a high level of maternity mortality (MM) in the high-risk zones in Nigeria.

Conclusion
In conclusion, our model-based estimates have provided disaggregation of population data in generating demographic estimates has also been introduced as a plausible means of handling the issues of health disparities across varying sociodemographic groups in the Nigerian population.
METHODS.Survivorship history data of 293,769 female siblings were provided by 114,154 women in the Nigeria Demographic and Health Surveys of 2008, 2013 and 2018.MM Rates and Ratios were estimated using the Empirical Bayesian technique for small area demographic estimates.The James-Stein estimator was used to shrink the estimates closer to the population mean values with 95% Confidence Interval (CI).RESULTS Levels of MMRatio were highest in the rural areas, States and regions in Northern Nigeria.MMRatio was consistently lower in the South West (2008=281; 2013=367; 2018=392) and higher among the Northern regions of the country, particularly the North-East (2008=654; 2013=612; 2018=901) for three consecutive surveys.Over the three surveys, mortality trends declined about 18% in the North West and 54.2% in the South East region.However, there was a 4.8% increase in MMRatio for South West between 2008 to 2018.
) dj = the number of deaths in each state Nj = the number of women in reproductive age in each state The mixed model combines the technique of the direct estimator and the indirect estimator to produce what is known as the BEST LINEAR UNBIASED PREDICTION.The Best Linear Unbiased Prediction estimators minimize the Mean Square Error among the other classes of linear unbiased estimators, and it generally does not depend on the normality of the random effects.

Figure 1
Figure 1 and 3 gives insight into how each state in the country fared compared to the

Figure 4 :
Figure 4 : Map showing Model-based sub-national maternal mortality ratio (MMRatio) estimates, findings show that MMR in Nigeria has not decreased significantly.It was in fact noticed that there was a slight spike in the estimates of MMR from the 2013 datasets as compared to the 2008 datasets and the percentage of maternal deaths increased across the three surveys.MM was relatively lower in the Southern part of Nigeria compared to the Northern regions.The South West experienced a slight increase in MMRatio of about 4.8% from 2008 to 2018.However, the Mortality trends declined about 18% in the North West and 54.2% in the South east from 2008 to 2018.

Table 1
shows the model-based estimates of MMRatio in the Northern states for 2008.Among all the Northern states, Nasarawa had highest MMRatio of 879 (95% CI: 718 -1075) maternal deaths per 100,000 live-births.Adamawa state had the highest MMR of 709 (95% CI: 621 -810) maternal deaths per 100,000 live-births among the North Eastern states and Kebbi state had the highest among the North Western states with MMR of 780 (95% CI; 633 -962).Table 2 shows the model-based estimates of the Southern states.Among the states in the Southern geopolitical zones, Lagos recorded the lowest MMR of 280 (95% CI: 172 -457) maternal deaths per 100,000 live-births and Bayelsa State had the highest MMRatio of 832 (95% CI: 671 -1033) maternal deaths per 100,000.Akwa Ibom State in the South South and Enugu State in the South East also had closely high MMRatio

Table 3 and
Table 4 show the model-based estimates of maternal mortality ratio for the Northern and Southern states for 2013 respectively.Katsina State in North Western part and Benue State in the North Central part of Nigeria had the highest MMRatio of 1621

Table 1 :
Model-Based Estimate of maternal mortality rates (MMRates) and

Table 2 :
Model-Based Estimate of maternal mortality rates (MMRates) and 249 maternal mortality ratio (MMRatio) in Southern states in Nigeria DHS, 2008

Table 3 :
Model-based Estimate of maternal mortality rates (MMRates) and maternal mortality

Table 4 :
Model-based Estimate of maternal mortality rates (MMRates) and maternal mortality

Table 5 :
Model-based Estimate of maternal mortality rates (MMRates) and maternal mortality

Table 6 :
Model-based Estimate of maternal mortality rates (MMRates) and maternal mortality For instance, in 2008, Taraba state in the North East and Kaduna and Zamfara states in the North West, contributed largely to the MMRatio of the Northern region compared to other states in the same region.Ebonyi state in the South East and Akwa Ibom State in the South South also had MMRatio that were as high as those observed in the Northern parts of the country.Similarly in 2013, Niger state in the North Central, Borno state in the North East and Kaduna and Kebbi states in the North West contributes largely to the high magnitude of MMR for the Northern states.Although the Southern states had lower level of MM compared to the states in the North, Ebonyi state in the South East and Akwa Ibom in the South South had relatively high MMRatio as well.This is one of the advantages of this study; This is a novel area in demographic research as attention becomes drawn to precision public health to enhance health outcomes through equitable, data-driven policies in population health.This same method can be applied to the under-five mortality and fertility patterns of the various states and geo-political zones in Nigeria.Small area estimation has shown promising possibilities of handling the data inadequacies in some demographic or geopolitical groups that might have insufficient sample sizes for direct estimations of demographic indicators.