Spatial access to comprehensive emergency obstetric and neonatal care and its relationship to mortality at the regional level in Sub-Saharan Africa and at a national level in Kenya

Background: Geographic access to hospitals is a key determinant of health outcomes and is critical in achieving Universal Health Coverage. Hospitals are the entry points for provision of comprehensive emergency obstetric and neonatal care. In sub-Saharan Africa where maternal and newborn deaths are disproportionately higher than anywhere else, defining those who are marginalised from these services is critical in in efforts to reduce these deaths. This thesis aimed to define geographic access to Comprehensive Emergency Obstetric and Neonatal Care (CEmONC) at a regional level in sub-Saharan Africa (SSA) and at a national level in Kenya and relate these access quotients to maternal and neonatal mortality. Methods: A spatial database of public hospitals covering 48 countries and islands in SSA, was assembled using a range of sources from both national and international organizations. Based on a review of minimum essential services at first level referral hospitals, the assembled hospitals were assumed to provide CEmONC. These were then used in a cost distance algorithm that adjusted for proximity to roads, to estimate the proportion of women of childbearing age (WoCBA) living within two hours of the nearest public hospital. The accessibility algorithm accounted for complexity in transportation derived from a review of transport speeds to hospitals and different modes of transportation. The derived access quotients were then used to define the relationship between access to hospitals and modelled estimates of both neonatal mortality rate per 1,000 livebirths (NMR) and maternal mortality rate per 100,000 livebirths (MMR) while controlling for other confounding factors. Confounders chosen were physician workforce density, poverty, adolescent-specific fertility, risk of catastrophic expenditure for surgery, the proportion of urban population and fragility. A country case study was then chosen to undertake a more exhaustive analysis of geographic accessibility and its impact on the two outcomes. Using Kenya as an example, service availability assessment was carried out using different datasets to map hospitals that provide services for caesarean section (CS) and very low birthweight (VLBW) newborns. These were chosen as tracer services for provision of CEmONC. A cost distance algorithm that adjusted for the proximity to roads, road condition, land use, elevation and rainfall patterns was developed to define geographic accessibility to both services in Kenya. Lastly, an assessment of the relationship between access to CS and VLBW hospitals in Kenya with maternal and neonatal mortality respectively was undertaken. Relationships were evaluated using maternal mortality data from the 2009 census while the Equitable Strategies Save Lives Tool (EQUIST) tool was used to generate both maternal and neonatal mortality. Results: 4908 public hospitals were mapped for the 48 countries. Accessibility results showed that 704 million (71%) people and 164 million (72%) WoCBA were living within 2 hours of the nearest public hospital across, which varied from 22.8 in South Sudan to 97.4 in Zanzibar. Only seventeen countries had more than 80% of their respective populations within 2 hours of a public hospital. In the continental level regression model, a 1% increase in WoCBA within 2 hours was associated with a reduction of MMR by -2.75 (p=0.039) but was not significantly associated with a reduction in NMR, with a coefficient of -0.01 (p=0.717). In the Kenya accessibility analysis, 228 and 293 hospitals were determined to provide VLBW and CS services respectively, out of a total 431 possible hospitals. Overall, 82% and 80% of the births needing CS and VLBW services respectively occurred within 2 hours of the nearest hospital. Access to CS and VLBW services was heterogenous and varied from 25.7 and 21.8% in Turkana to 100% in Nairobi and Vihiga. Regression models that accounted for confounders showed a 1% increase in access to CS hospitals was associated with are reduction of MMR by -11.92 (p=0.018) using census MMR and -3.81 (p=0.000) using the EQUIST MMR. Finally, using NMR output from the EQUIST model, increasing access to VLBW hospitals was associated with a reduction of NMR by -0.24 (p=0.050). Thus, using the census outcome as an example suggests that MMR in high burden counties of Mandera, Marsabit, Turkana and Wajir would reduce by 457, 452, 647 and 540 per 100,000 livebirths respectively by ensuring all their livebirths use hospitals within 2 hours. On the other hand, Samburu, Turkana and West Pokot counties can reduce their NMRs by approximately 12 deaths per 1,000 livebirths by ensuring all births are within two hours of a VLBW hospital. Conclusion: Understanding the distribution of hospitals and geographic access to these services is critical in estimating the underserved. At the continental level, geographic access is significantly associated with variation in MMR but not NMR. Using Kenya as an example shows that accessibility models can be improved by evaluating the services available in hospitals in addition to using improved data on covariates, thus allowing for better assessment of how access relates to mortality. As such, Kenya can reduce MMR by 240.7 per 100,000 livebirths and NMR by 5.2 per 1,000 livebirths by ensuring universal access to CS and VLBW hospitals. This highlights the importance of implementing interventions that bridge geographic accessibility gaps, especially in the marginalized populations across Africa. However, there are still challenges with availability and quality of hospital services data in addition to reliable data on outcomes such as mortality and ultimately, improving the quantification of impact geographic access to hospitals on health outcomes will be dependent on improving data collection tools.