Background
Adverse pregnancy outcomes (APOs) are a major cause of neonatal mortality and morbidity and affects millions of women of childbearing age worldwide and is a critical global health concern. While large strides have been made in tracking progress towards the attainment of global maternal health targets in Sub-Saharan Africa (SSA), the World Health Organization (WHO) estimates that approximately 810 women die each day as a result of preventable causes associated with pregnancy and childbirth [
1‐
4]. Despite the global priority and substantial investment in maternal and child health programs in recent decades [
5], the burden of APOs remains unacceptably high in SSA and Southern Asia, with both regions accounting for approximately 86% of the global maternal deaths [
1]. Recent estimates of APOs have been 546 maternal deaths per 100,000 live births, 25.9 neonatal deaths per 1000 live births and 13.9 stillbirths per 1000 total births [
6‐
8]. These sobering trends are mirrored in Kenya, which risks missing out on the 2030 neonatal mortality target of 12 deaths per 1000 live births given its high neonatal mortality rate of approximately 19.6 deaths per 1000 live births and 11.5 low birth weight babies per 100 live births reported in 2018 [
9]. To address this growing burden, national initiatives such as free maternity services, elimination of user fees for public primary health care and the beyond zero campaign have been established to improve pregnancy outcomes and increase institutional delivery rates [
10].
As a step towards renewing its focus towards maternal and child health programs and achieving global health targets, the United Nations Sustainable Development Goals (SDGs) set ambitious health targets for mothers, infants and children. Target 3.1 seeks to reduce maternal mortality ratio to less than 70 per 100,000 live births. In contrast, target 3.2 aims to end preventable neonatal mortality to 12 per 1000 live births and sustain universal healthcare by 2030 [
11]. Moreover, the WHO conceptualised two global initiatives namely; Ending Preventable Maternal Mortality (EPMM) [
12] and the Every Newborn Action Plan (ENAP) [
13]. These initiatives sought to catalyse global action into addressing disparities in maternal health outcomes among sub-populations. These ambitious targets were further entrenched by two of the six global nutritional targets for 2025 [
14], which sought to achieve; a 50% reduction of maternal anaemia and a 30% reduction of low birth weight (LBW) as vital components for improving maternal health.
The government of Kenya strives to achieve universal health coverage for its population and alleviate the burden imposed by adverse pregnancy outcomes in women of childbearing age. Mapping sub-county level estimates would provide a greater understanding of the variability in spatio-temporal patterns of adverse pregnancy outcomes, not possible by examination of direct national estimates or by examination of direct county-level estimates [
15]. This is crucial for; monitoring progress towards global maternal health priorities and orienting national policies to identify high burden sub-counties and high-risk populations [
11,
15,
16].
Sub-nationally, little is also known about the joint and shared risk estimates of adverse pregnancy outcomes and their association with related health outcomes across the continuum of maternal health care. This may be attributed to the complex interplay of risk factors that remains inadequately characterised in the spatial and temporal domain [
17‐
19]. This may be attributed to the paucity of data on individual pregnancy outcomes and risk factors sub-nationally. This becomes more apparent when data is stratified at policy-relevant spatial scales to accommodate different demographic dimensions. To stabilise estimates, strengthen inference occasioned by data scarcity, we borrow information across adjacent sub-counties, diseases/health outcomes and sub-populations with common risk factors. Previous studies have demonstrated the robustness of Bayesian techniques in jointly mapping two or more diseases/health outcomes with common risk factors by leveraging information among the diseases and across neighbouring geographical units [
20,
21]. To the best of our knowledge, little work on Bayesian space-time modelling has been done in maternal health research in Kenya. To quantify the underlying burden and increase epidemiologic interpretability of adverse pregnancy outcomes, we employed a Bayesian hierarchical shared component model that simultaneously accounts for both shared and specific risk components attributed to each pregnancy outcome.
Discussion
We employed a Bayesian hierarchical binomial model that simultaneously analysed the spatial pattern and temporal trend of adverse pregnancy outcomes (APOs) in Kenya between 2016 and 2019. Our modelling framework allowed us to incorporate information from multiple pregnancy outcomes that shared risk factors over the spatial extents and time periods. Information across the outcomes was specified via the spatial and temporal correlation structures, and our model was adjusted for both the covariate effect and an unstructured space-time interaction effect. To our knowledge, this analysis represents the first sub-national, high-resolution estimation of adverse pregnancy outcomes concomitantly over space and time in Kenya, which is a crucial input for strengthening national and county efforts towards the realisation of Universal Health Coverage (UHC) and achieving global health targets. Estimates of maternal health outcomes have historically relied on periodic household surveys such as the DHS and MIS. However, their sampling design limits their scope as they are best suited to yield estimates at the national or regional level. Compared to the unstable national-level estimates, our estimates provide greater descriptive granularity and give a baseline for bolstering maternal health programs in high-risk sub-counties.
The risk of preventable adverse pregnancy outcomes increased between 2016 and 2019 and was non-uniform across all the outcomes. Low birth weight (LBW) is considered a composite measure of foetal growth and an underlying determinant/contributor of adverse outcomes, and has been associated with an increased risk of poor physical and cognitive development in infants [
41,
42]. Our results suggest an elevated risk of LBW in sub-counties located along the Indian ocean coastline which may signal the role of malaria in pregnancy. On the other hand, the low risk observed in resource-constrained sub-counties located in the north-eastern parts of Kenya may signal limited/unreliable data leading to an underestimation of its burden. This heterogeneous burden of LBW calls into question the implementation of national policies particularly in resource constrained sub-counties and urban informal settlements. Results suggest the need for more targeted approaches; particularly in sub-counties with urban informal settings where LBW contributes to neonatal deaths but remain underestimated [
43,
44].
The overall pre-term birth rate was markedly lower than 9 per 100 live births previously estimated [
45,
46]. By 2019, the pre-term birth rate ranged between 0.12 to 13.0 per 100 live births, with elevated estimates being observed in a few sub-counties located in the south-western parts of Kenya and along the Indian ocean coastline. However, low rates were consistently observed in the north-eastern parts of Kenya which may be attributed to limited data quantities over the period. Thus, to better understand the epidemiology of pre-term birth in low-risk sub-counties the volume and quality of routinely collected data need to be strengthened. Conversely, the elevated burden observed in some sub-counties in the south-western parts, may be reflective of the unresolved burden of pre-term births in unintended adolescent pregnancies [
42,
47].
Our stillbirth estimate of 18.7 per 1000 live births was significantly lower than previous estimates for Kenya and SSA [
8,
46]. The rate declined by 6.7%, from 19.4 per 1000 live births in 2016 to 18.1 per 1000 live births in 2019. The slow decline of stillbirths primarily attributed to intrapartum deaths [
8], may be reflective on the inadequate global and national attention in mitigating the burden [
11]. Additionally, generated estimates need to be interpreted cautiously due to under-reporting attributed to recall bias, inconsistent definitions/misclassification of fresh versus macerated stillbirths and omission. This may limit comparison with other international studies using diverse classification criteria [
7,
11]. Improving the quality of care and birth and registry data will likely contribute to accurate estimates sub-nationally.
Neonatal death rate is an important index for tracking progress towards the attainment of SDG targets, that seeks to reduce neonatal mortality to at least 12 deaths per 1000 live births. Understanding its spatial extents and temporal trends is essential/critical due to its rising share in under-five mortality in Kenya [
25]. Our estimates was markedly lower compared to a facility-based study that estimated the neonatal death rate to be 16 per 1000 live births [
46]. Huge disparities in neonatal deaths persisted across sub-counties, with elevated risk observed in malaria-endemic regions located along the Indian Ocean coastline and Lake Victoria region. The highest risk was consistently observed in Kibra sub-county, with approximately 75 deaths in 1000 live births [
48]. It is important to note that Kibra inhabits the largest urban informal settlement in Africa [
44]. Thus, the elevated risk may have been compounded by poor sanitation practices and over-stretched facilities in the densely populated informal settlements [
49].
Our results suggest that disparities in the spatial extents of APOs may have been driven by demand-side barriers and supply-side challenges, with low-resource counties disproportionately vulnerable to adverse outcomes. Recent studies highlight a heterogenous timing and coverage of ANC4 across sub-counties in Kenya, with poorer/rural sub-counties in northern Kenya exhibiting low coverage levels [
50,
51]. The current guidance from the WHO suggesting at least four comprehensive and targeted visits [
16,
52]. However, a confidential inquiry into maternal deaths in Kenya indicated that only one in five women attending ANC had at least 4 ANC visits in 2014 [
53]. ANC attendance and coverage are essential as they are positively associated with increased use of birth attendants, reducing the risk of adverse pregnancy outcomes. Pregnant women are screened for anaemia, HIV, syphilis, urinalysis, and blood typing (ANC profile) during these visits. Additionally, they receive vital micronutrient supplementation essential to both the mother and her unborn child among other interventions such as educational information on safe motherhood. However, regional disparities may signify the inequitable access and the widespread unmet need of ANC services particularly in arid, resource-constrained, rural and pastoralist dominated sub-counties located eastern and northern parts of Kenya as previously documented [
11,
54‐
56].
Increasing concerns about the shortage and distribution of skilled healthcare workers sub nationally, adequacy of pre-existing healthcare infrastructure, stakeholder non-involvement in policy formulation has previously been raised [
10,
57]. These concerns have been exacerbated by the distance to health facility, timely access to quality maternal care, cost of transport, poverty, illiteracy, religious and social-cultural beliefs and tendencies. These factors tend to suppress the uptake of facility-based delivery with only the complicated cases being referred to the facility [
50,
51,
58‐
61]. This is more pronounced in the rural and pastoralist communities that prefer traditional birth attendants [
62,
63]. According to the Kenya Health Workforce Status Report in 2017, the nurse-to-population ratio of 0.1/10,000 in arid and semi-arid sub-counties in northern Kenya was markedly lower than 9.7/10,000 in sub-counties Nairobi [
64]. This might have contributed by high attrition rates of health staff in these sub-counties due to poor infrastructure and insecurity. This has led to poor utilisation of public health facilities, leading to increased home deliveries [
61]. Our results lend credence to efforts to strengthen the healthcare workforce through policy, planning, adequate financing, recruitment, training and deployment. This may help to improve sub-county health systems and help achieve global and national health targets such as the universal coverage of skilled birth attendants [
65].
In low-resource sub-counties in northern Kenya, macronutrient/maternal undernutrition - caused by inadequate intake of essential minerals and vitamins; contributes to anaemia in pregnancy, which might also precipitate APOs [
66]. Additionally, infectious diseases such as malaria and human immunodeficiency virus (HIV) have previously been associated with an increased risk of pre-term deliveries, stillbirths, and neonatal deaths. The burden is more pronounced in sub-counties located in western Kenya [
67‐
73]. Sexually transmitted infections such as chlamydia, gonorrhoea, trichomoniasis and syphilis have also been associated with an increased risk of stillbirths and low birth weight among pregnant women in rural settings [
74‐
76].
Implication of study
Despite the achievements in maternal and child health-related programs over the recent decades, preventable adverse pregnancy outcomes remain a key challenge to pregnant women and their infants. The study presents a unique dimension for prioritising high-risk areas and increasing the efficiency of control programmes to alleviate the burden imposed by preventable APOs. This will accelerate the attainment of maternal health targets and align sub-county estimates with global and national priorities. Crucially, our estimates can help to raise awareness of growing inequities in Kenya’s diverse and growing population inhabiting different regions with diverse levels of social and economic development.
The shared risk estimates underscore the need to establish integrated suites of interventions (policy/care package/services) at meaningful policy levels to reduce missed opportunities/comprehensively address healthcare needs of pregnant women at the sub-county level, especially through nutrition-specific and nutrition-sensitive interventions. Compared to the national level estimates, our joint estimates resonate with Kenya’s broader maternal health objectives, focusing on disease patterns and monitoring/accelerating progress in maternal health care equity as outlined in the SDGs. Our findings also inform the need of mapping the overlapping burden of infectious diseases and APOs in Kenya. Study approaches can also be applied to other health outcomes with shared risk factors.
As more data on maternal health outcomes becomes available, underestimation stemming from inaccurate reporting/documentation can be used to flag sub-counties with less reliable data on other maternal health indicators and help channel investments to sub-counties with poor surveillance systems, constrained diagnostic capacity and unmet reproductive health needs. Overall, the local burden of disease can be used to prioritise appropriate, cost-effective interventions targeting the poor, rural sub-counties and informal settlements in urban areas will likely address the wide disparities and accelerate progress in maternal and newborn health. There is also a pressing need to understand more about APO’s risk to urban slum dwellers in Kenya who are disproportionately bombarded with chronic illness, maternal illiteracy, and socio-economic instability.
Study strengths and limitations
At a much higher level of spatio-temporal precision, we demonstrate the importance of a functional health information system in providing routine estimates between periodic and expensive survey estimates. Facility-based data can help inform the local burden of disease. The continual adoption of standard definitions and consistent measurement procedures will be critical for meaningful comparisons towards the attainments of SDGs. The Bayesian shared component model adds versatility to substantive epidemiological questions. Its ability to borrow strength from outcomes and across neighbouring sub-counties makes it an important tool for monitoring the evolution of facility-based maternal health disparities across the continuum of care.
Our findings need to be interpreted cautiously due to several methodological challenges that might have confounded our estimates. First, the study presents facility-based estimates that may not be generalisable to the entire population, especially for women delivering outside the public health facilities. This may be due to the geographic underrepresentation of public antenatal clinics. Secondly, we cannot discount the potential incompleteness of our data, disproportionately available in malaria endemic sub-counties with more robust surveillance infrastructure and higher uptake of antenatal care services. Third, misclassification of pregnancy outcomes; e.g. neonatal deaths as stillbirths, pre-term birth as low birth weight (or vice versa), might have led to underestimating some adverse outcomes. This may in part explain previous concerns raised on the use of routine health metrics for international comparisons [
77]. Fourth, although we included covariates to stabilise our estimates, our study was ecological with data aggregated at the sub-county and yearly level; this might have led to residual confounding. Fifth, our study did not account for the adverse outcomes during the different pregnancy semesters, pregnancy history, underlying conditions and parity which has previously been associated with adverse outcomes.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.