In many sub-Saharan African countries, the burden of preventable maternal deaths is still unacceptably high. Recent evidence shows that sub-Saharan Africa (SSA) contributed to 70% of global maternal deaths in 2020 and that SSA's average Maternal Mortality Ratio (MMR) of 536 per 100,000 live births [Uncertainty Interval (UI) of 469–640] is more than twice the global average MMR of 223 per 100,000 live births (UI: 202–255) (1). As a result of this high burden of maternal deaths, the surviving children face adverse health and socioeconomic outcomes, often lasting more than one generation (2, 3). Therefore, eliminating preventable maternal deaths has the potential to safeguard the health and socioeconomic outcomes of families that would struggle to thrive without a mother.

To successfully implement a strategy for eliminating preventable maternal deaths, it is crucial to establish accurate metrics on the number of women dying during and after childbirth, including where and why they died (4). Global estimates of country-level maternal mortality ratios (MMRs) are typically calculated by international agencies such as the United Nations using multiple sources of data on maternal deaths that occur in both community and health facility settings (1, 5). While these country MMR estimates help inform the global agenda, tracking progress and comparison between countries, their utility at the sub-national level within individual countries may be limited (6). In addition, global estimates lack granularity regarding which deaths occur in the community vs. those at health facilities (7). The paucity of insight into sub-national data and where deaths occur may hinder the focused and contextualised approach to tackling the persistent problem of preventable maternal mortality in SSA (8).

To fill this void, models such as the Demographic and Health Survey (DHS) that use data from national surveys to estimate sub-national MMR estimates have been proposed (9, 10). Maternal mortality data reported directly from health facilities have occasionally been utilised to compute MMR estimates. However, there are concerns regarding selection bias compared to other data sources and that institutional data reflects only those women who can access health facility care (11, 12). Despite these concerns, institutional data are highly specific to local contextualised conditions and thus better-suited to inform local policymakers, clinicians and programs for strengthening health facilities and the health system (12). Institutional data, such as iMMR, can offer valuable insights into how delays in receiving appropriate care at a health facility (third delay) contribute to maternal deaths and inform quality improvement within health systems (13, 14).

Health systems in low- and middle-income countries have made significant progress in having women deliver under skilled birth attendants (15). For example, in Kenya, as per the 2022 Kenya Demographic and Health Survey (KDHS) report, nearly 90% of women deliver under skilled birth attendants (16). Delivery under the care of skilled birth attendants is associated with a significant reduction in maternal mortality (17). Countries with high maternal mortality rates often have low coverage of skilled birth attendance, whereas countries with low maternal mortality rates tend to have high coverage of skilled birth attendance (18). Despite this achievement in the proportions of women delivering under skilled birth attendants, some countries, such as Kenya, paradoxically continue to exhibit high maternal mortality ratios (19). One explanation is that skilled birth attendants require an enabling work environment for their presence at birth to translate into quality care and a reduction in preventable morbidity and mortality (19–21). An enabling work environment for skilled birth attendants has been described as one that provides necessary inputs such as supportive regulations, policies, infrastructure, communication, referral systems, logistics, and supplies (19, 22).

In Kenya specifically, evidence from a confidential enquiry into maternal deaths showed a need for quality improvement at the health facility level to avert preventable maternal deaths (23). Specific health facility-level recommendations included Regular and mandatory healthcare worker updates in emergency obstetric care, increased availability and safety of blood and blood products, improved monitoring during the antenatal, intrapartum and postpartum periods, especially for women with high-risk pregnancies, safer anaesthesia, audit and feedback processes and improved documentation (23, 24).Following documentation at the health facility level, health data on key processes and outcomes, including maternal deaths in Kenya, are reported through the District Health Information System Software-2 (DHIS-2) for aggregation nationally (25, 26). Since the enactment of a new constitution in 2010, healthcare delivery has devolved from the national government to 47 sub-national governments referred to as counties (27, 28). Health service delivery is organised into four levels: Community services, primary health services, county referral services and national referral services (28). The health facilities that serve these four levels are categorised into six categories: Level 1 (Community health facility), Level 2 (Clinics and dispensaries), Level 3 (Health centres and nursing and maternity homes), Level 4 (County Hospitals), Level 5 (County referral hospitals) and Level 6 (National referral, teaching and specialised hospitals) (29). All these levels have a role to play in the provision of maternal health care in Kenya.

We postulated that disaggregation of national MMR estimates into sub-national levels might reveal variability between regions and health facilities and facilitate the identification of the most appropriate approaches for improving maternal health outcomes and, by extension, tackling the causes of preventable maternal deaths.

While progress in health indices of most countries, including maternal mortality, is monitored at the national level, there is potential benefit in exploring variability within countries (sub-national level) for the purpose of ensuring equity in the distribution and use of health resources and tracking progress (8). Learning from sub-national variability may also provide unique behavioural and practice insights from those accessing and/or delivering maternal healthcare (30). Variability is defined by assessing the performance of an entity, such as a health facility, in relation to an expected performance or set target (31). The terms used to describe variability include high or low, positive or negative outlier or positive or negative deviance depending on the extent of deviation from the expected performance or set target (31–33). Unique behavioural and practice insights from positive outlier regions or health facilities may help the negative outliers to solve shared challenges with the overall effect of improving maternal health outcomes through a positive deviance approach (30, 34–36). Such a shared challenge may be maternal mortality, with positive outlier regions, counties and health facilities providing insights for improving the negative outlier ones.

This study explores Kenya's 2021 national health facility dataset on maternal deaths and live births for sub-national variability to describe the pattern and trend of variation in Institutional Maternal Mortality Ratios (iMMRs). We posed the following three research questions (RQ): (1) What is the pattern of sub-national variation in iMMR in Kenya? (2) What is the pattern of iMMR variation by health facility level? and (3) What is the pattern of month-to-month county iMMR variation?

The research protocol for this work was reviewed by the University of Birmingham (Reference ERN_22-0550) and the Kenyatta National Hospital—University of Nairobi Ethics and Research Committee (ERC) (Reference KNH-ERC/RR/712). This work was conducted under the National Commission for Science, Technology, and Innovation (NACOSTI) license number NACOSTI/P/22/20651.

In October 2022, we submitted a DHIS2 data access request to the Division of Research and Development team via the office of the Director General, Ministry of Health, Kenya. The data access was authorised under reference MOH/ADM/1/1/82 (263).

The specific datasets requested for this project included: Healthcare facility by name and level, the total number of maternal deaths per facility for 12 months by facility (the latest complete annual dataset) and the total number of live births per facility for the same period as the maternal deaths.

We analysed 1,162 maternal deaths, and 1,174,774 live births reported to the DHIS2 from facilities where maternal deaths occurred in 2021. The health facilities consisted of private, faith-based/mission, and public health facilities across all the 47 counties of Kenya. A total of 323 health facilities were distributed as 49 level 2, 53 level 3, 197 level 4, 19 level 5, and 5 level 6 health facilities.

We present the results of our analysis of sub-national variation in iMMR below in four parts: Regional, county, health facility, and month-to-month variation.

We aimed to explore Kenya's institutional maternal mortality ratios (iMMRs) for sub-national variability. We conducted our analysis at the regional, county and aggregated health facilities by level avoiding individual health facility comparisons. This approach in the analysis was because healthcare governance and implementation are devolved functions (37). We avoided individual health facility comparisons as a good practice to minimise the risk of breach of confidentiality, especially at health facilities that had few maternal deaths (4, 40). We felt that such an analysis would have required consent or ethical approval by each of the participating health facilities.

Our analysis of the 2021 maternal deaths and live births data for Kenya established the following key findings: (1) there is evidence of sub-national variability in institutional maternal mortality ratios at regional and county levels; (2) here is evidence of variation in the relative contribution to the overall iMMRs by the various levels of health facilities. However, the effect size was zero, implying limited practical application of this finding; (3) the monthly mean iMMR trend did not reveal any variation.

To contextualise, Kenya has had two Confidential Enquiry into Maternal Deaths (CEMD) reports on deaths between 2014 and 2016 (23, 24). The analysis in both reports hints about the presence of regional variation, with most maternal deaths occurring within the Rift Valley, Western and Coast regions (23). Regional administrative boundaries have been historical since the promulgation of Kenya's new constitution in 2010, when healthcare governance was devolved into the county-level (37). Therefore, a county-by-county analysis would have provided more insight into the pattern and causes of sub-national variability.

Despite this, there were commonalities in both CEMD reports regarding the most frequent causes of death and the association of maternal deaths with healthcare worker-related factors (23, 24). The leading causes of maternal deaths nationally were obstetric haemorrhage (39.7%), hypertensive disorders (15.3%), and non-obstetric complications (19.8%) (23, 24). The healthcare worker-related factors included delays in starting treatment (32.9%), inadequate monitoring (26.9%), inadequate clinical skills (28.1%), prolonged abnormal observation without action (23.6%), and incomplete initial assessment (22.7%) (7, 24). Therefore, based on the CEMD evidence, we postulate that a possible explanation for the sub-national variability in iMMR is differences in healthcare worker-related factors within the health facilities in the various counties.

The reasons for the regional and county level variability in iMMRs in Kenya could not be determined using the provided data. Existing literature acknowledges diversity in maternal health indices such as maternal deaths and cites various explanatory reasons for such variability. These reasons include sub-national variations in political commitment to reducing maternal mortality, health system status and resilience, inequitable access to healthcare, poverty, and education (42–45). Temporal variability in maternal mortality may also be influenced by seasonal factors such as weather and the burden of seasonal infectious diseases like malaria, political instability, conflict, and healthcare worker strikes (46–48). Factors within health facilities include variations in service delivery such as triage, monitoring, and referral; availability of life-saving medications, equipment and blood transfusion; and staffing-related issues such as absence, competence and supervision of juniors (20, 23, 24). There is a need to further examine and compare Kenya's counties to explain the spatial and temporal variability in iMMRs and their potential application to tackling the causes and factors contributing to maternal mortality within counties.

Regarding the observed variation between various levels of health facilities, we postulate that this may reflect the status referral system where complicated cases are referred upwards and often arrive late for the case of level 6 health facilities. On the other hand, level 2 facilities are dispensaries. Although the dispensaries are closest to the community, they may be ill-equipped to handle even basic obstetric emergencies. The maternal healthcare referral system may cushion level 4 and 5 health facilities because they can refer complex cases upwards to level 6 facilities. Our meta-analysis revealed no effect size, implying that the variations in iMMR at the health facility level are likely to be of limited practical application.

A recent study that analysed pregnancy outcomes in health facility deliveries in Kenya and Uganda found that more pre-discharge deaths occurred after a maternal referral and following a caesarean section (49). Referred mothers are especially at risk because of sub-optimal care in ill-equipped lower-level health facilities and bottlenecks within the referral pathway (50).

Finally, the lack of month-to-month variation was unexpected, given evidence from Mozambique and Burkina Faso linking seasonal iMMR fluctuations to income declines, healthcare access barriers, and malaria-related maternal deaths (46, 51, 52). We did not find published work examining seasonality and maternal mortality in Kenya. We postulate that our finding of the absence of month-to-month variation may be due to the low number of maternal deaths per month or the limited one-year study period.

There is evidence of sub-national variability in Kenya's iMMRs at regional and county levels and between various health facility levels in the referral system. This variability presents an opportunity to explore the reasons for any observed positive and negative variability. Understanding these reasons is crucial for developing alternative and targeted strategies for improving maternal health outcomes. If positively deviant behaviours and practices are identified, they could form the basis for adopting asset-based approaches such as the positive deviance approach to improve maternal healthcare delivery processes and outcomes and reduce preventable maternal deaths.