Local use of geographic information systems to improve data utilisation and health services: mapping caesarean section coverage in rural Rwanda


Corresponding Author Leanna Sudhof, Women and Infants Hospital, Providence, RI, USA. Tel.: (401) 274-1122; Fax: (401) 453-7599; E-mail: LSudhof@wihri.org



To show the utility of combining routinely collected data with geographic location using a Geographic Information System (GIS) in order to facilitate a data-driven approach to identifying potential gaps in access to emergency obstetric care within a rural Rwandan health district.


Total expected births in 2009 at sub-district levels were estimated using community health worker collected population data. Clinical data were extracted from birth registries at eight health centres (HCs) and the district hospital (DH). C-section rates as a proportion of total expected births were mapped by cell. Peri-partum foetal mortality rates per facility-based births, as well as the rate of uterine rupture as an indication for C-section, were compared between areas of low and high C-section rates.


The lowest C-section rates were found in the more remote part of the hospital catchment area. The sector with significantly lower C-section rates had significantly higher facility-based peri-partum foetal mortality and incidence of uterine rupture than the sector with the highest C-section rates (P < 0.034).


This simple approach for geographic monitoring and evaluation leveraging existing health service and GIS data facilitated evidence-based decision making and represents a feasible approach to further strengthen local data-driven decisions for resource allocation and quality improvement.



Montrer l'utilité de combiner les données recueillies en routine et la situation géographique à l'aide d'un Système d'Information Géographique (SIG), afin de faciliter une approche guidée par les données, pour déceler les éventuelles lacunes dans l'accès aux soins obstétricaux d'urgence dans un district rural de santé rwandais.


Le total des naissances attendues pour 2009 à l’échelle des sous-districts a été estimé à l'aide des données démographiques recueillies par les agents de santé communautaires. Les données cliniques ont été extraites des registres des naissances de 8 centres de santé et de l'hôpital du district. Les taux de césariennes par rapport au total des naissances attendues ont été cartographiés par secteurs. Les taux de mortalité fœtale péri-partum par naissances dans un service ainsi que le taux de rupture utérine nécessitant une césarienne, ont été comparés entre les zones à taux élevé et bas de césarienne.


Les plus bas taux de césariennes ont été trouvés dans la partie la plus éloignée de la zone desservie par l'hôpital. Le secteur ayant des taux significativement plus faibles de césariennes avait une mortalité fœtale péri-partum et une incidence de rupture utérine significativement plus élevées dans un service que le secteur avec les taux de césarienne plus élevés (P < 0,034).


Cette approche simple pour le suivi et l’évaluation géographique s'appuyant sur les services de santé existants et des données SIG a facilité la prise de décision fondée sur des preuves et représente une approche possible pour renforcer les décisions locales guidées par les données pour l'allocation des ressources et l'amélioration de la qualité.



Demostrar la utilidad de combinar los datos recogidos de forma rutinaria con la localización geográfica obtenida utilizando un Sistema de Información Geográfico (SIG), para facilitar un sistema basado en datos que identifique potenciales brechas en el acceso a los cuidados obstétricos de emergencia en un distrito sanitario de una zona rural de Ruanda.


Se calculó el número total de nacimientos esperados a nivel subdistrital en el 2009 utilizando datos poblacionales recogidos por trabajadores sanitarios comunitarios. Los datos clínicos se tomaron de los registros de nacimientos de 8 centros sanitarios y el hospital distrital. Se realizó un mapa por célula de las tasas de cesáreas como proporción del total de nacimientos esperados. Se compararon las tasas de mortalidad fetal durante el periparto de los nacimientos ocurridos en centros sanitarios, al igual que la tasa de rupturas uterinas como indicación de cesárea, de las áreas con tasas bajas y con tasas altas de cesáreas.


Las tasas de cesáreas más bajas se encontraron en las partes más remotas del área de influencia del hospital. El sector con tasas de cesáreas significativamente más bajas tenía una mortalidad fetal ocurrida durante el periparto y en centros sanitarios significativamente más alta, así como una incidencia de rupturas uterinas significativamente más alta que el sector con las tasas de cesáreas más altas (P < 0.034).


Esta aproximación simple de monitorización geográfica y evaluación, que maximiza los datos existentes en servicios sanitarios y del SIG, facilita la toma de decisiones basada en evidencias y representa un enfoque factible, que fortalece la toma de decisiones apoyada en datos para la distribución de recursos y una mejora de la calidad.


Most maternal morbidity and mortality are preventable, and yet in 2008, nearly 350 000 women died in pregnancy or childbirth worldwide. Although progress has been made in reducing maternal mortality, the world will not achieve the target of Millennium Development Goal (MDG) 5: a 75% reduction in Maternal Mortality Ratio (MMR) by 2015 (Hogan et al. 2010). Caesarean section rates are a commonly used indicator to monitor access to, and use of, emergency obstetric care, one of the critical components in reducing maternal mortality (WHO et al. 2009). WHO estimates that population rates of Caesarean sections (C-sections) between 5 and 15% reflect appropriate access and utilisation (2009). Populations with lower rates potentially represent compromised access and utilisation, raising the risk of preventable maternal death.

In Rwanda, where 85% of the population live in rural areas, significant progress has been made towards achieving MDG 5 (Hogan et al. 2010). The MMR has decreased from an estimated 1300 deaths per 100 000 live births during the period 2000–2004, to 487 for the period 2004–2010 (Hill et al. 2007; National Institute of Statistics of Rwanda 2011). Despite this progress, the Rwandan Health Sector Strategic Plan-II noted that MDG 5 was ‘proving the most difficult to achieve’ (2009), and in response, the Rwandan Ministry of Health (MOH) has prioritised maternal health.

One priority for countries in sub-Saharan Africa working towards maternal mortality reduction is improving the use of existing data to improve health systems (Gething et al. 2007). Local decision makers often lack the tools to efficiently and effectively use data to identify gaps and ensure evidence-based decision making and equitable allocation of limited resources. One tool that has been increasingly used in sub-Saharan African settings to address this challenge is Geographical Information System(s) (GIS). This work has included the documentation of the negative impact of distance on service utilisation and identification of areas with low access to services (Noor et al. 2003; Heard et al. 2004; Feikin et al. 2009; Cooke et al. 2010). Tanser et al. (2001) analysed the geographic variability in usage rates of clinics in rural South Africa to identify areas of under-performance. GIS was also used for program monitoring and visualising progress in coverage of insecticide-treated nets in malaria-endemic regions between 2000 and 2007 (Noor et al. 2009).

Most GIS use reported in the literature was by groups with good data management infrastructure and expertise in the context of research institutions or large initiatives. In this article, we describe the implementation of a relatively low-cost, community-based approach to integrate GIS analysis and data use into district-level monitoring and evaluation in the catchment area of a single district hospital (DH) in southern Kayonza District. Local decision makers then used the geographic disaggregation and visualisation of locally available data to identify potential gaps in access to and utilisation of emergency obstetric care, providing guidance for evidence-based health resource allocation to improve equity at a sub-district level.


Study setting

Since 2005, Partners In Health (PIH) has collaborated with the Rwandan Ministry of Health to strengthen the health system in three rural districts. Joint interventions at the DH, health centre and community level have been employed to improve financial and social access to health care, including community health worker initiatives to accompany pregnant women in obtaining prenatal care and incentivised grants to the health centres (HCs) to minimise barriers to care as well as improve the quality of care. In 2009, PIH introduced GIS mapping and analysis as an adjunct to existing monitoring and evaluation efforts in the uniformly hilly southern region of Kayonza District to examine geographic access.

The health system in the southern part of Kayonza District includes one DH and eight HCs serving 7.5 administrative sectors, which make up the catchment area of the DH (see Figure 1). The names of the DH, HCs and sectors were replaced with letter designations, with the HC and the sector they serve given the same letter. Sectors included a population of 15 000–25 000 people, covering an average area of 48 km2. Each sector is divided by the Rwandan government into four or five administrative cells, each containing 5–16 villages. Each HC serves one sector, with the exception of the HC in Sector C, which serves a single cell in the DH catchment. Clinically, HCs are staffed by nurses alone, and standard obstetrical services include normal labour and vaginal delivery. Complicated labour and high-risk cases are referred to the DH, and C-sections are only performed at the DH. Operative vaginal deliveries (forceps and vacuum) are not performed at the HCs or the DH. One HC in the catchment, HC (H), does not provide childbirth services because women deliver at the adjoining DH.

Figure 1.

Map of study area.

Geographic data collection

To support integration of GIS into routine monitoring and evaluation, PIH employed a Rwandan Bachelor-level GIS studies graduate to provide training and support data use through analysis, interpretation and feedback of results to programs. The GIS team also included a local high school graduate program associate who acted as coordinator of data collection activities, which included mapping at the village level with community health workers. District-wide geographic data encoding sector boundaries and roads were obtained from the Centre for GIS at the National University of Rwanda, and the National Institute for Statistics in Rwanda provided geographic data encoding cell boundaries. Sector and district maps were created with ArcGIS 9.3.1 software, which was obtained through an academic partnership with Harvard University. All analysis was carried out by the local GIS team.

Data collection for facility-based delivery and caesarean sections

C-section and facility delivery data were manually collected over two months by a medical student and data officer. Maternity data in paper-based labour registries at the eight HCs and the DH for the 4583 women who delivered between 1 January 2009 and 31 December 2009 were extracted and entered in a secure Excel database. The 104 patients from outside of the DH catchment were excluded. Data extracted included patient name (for identification of duplicated entries), age, address, admission date, discharge date, delivery date, APGAR score, referral status and reason for referral. Indications for pre-partum referral of labouring patients documented in HC registries and indications for C-section documented at the DH were categorised according to first recorded diagnosis: failure to progress, cephalopelvic disproportion, foetal distress, repeat Caesarean, malpresentation, uterine rupture, failed induction, placental abnormality and pre-eclampsia. C-sections with documented reasons that are not recognised indications for C-section were coded as elective (‘voluntary’, ‘tubal ligation’ or ‘normal labour’).

Estimating total births

Total expected births were calculated using total population at the cell and sector levels as estimated below and published crude national birth rates. We used the estimate of Rwanda birth rate of 38.1 births per 1000 population from the US Census Bureau 2009 estimate, which was the most recent and also consistent with the trend in birth rates reported in the 2005 and 2007 Rwanda DHS (US Census Bureau 2009, Institut National de la Statistique du Rwanda (INSR), 2006, Ministry of Health (Rwanda), Macro International, Inc 2007–2008, 2010). The last Census in Rwanda was conducted in 2002, with results only made available at the provincial level to the authors (Minnesota Population Center 2011). The rapid increase in Rwanda's population since 2002 made the applicability of that data to estimate births in 2009 a concern (World Bank 2009). We thus turned to Rwanda's robust CHW system, which first began collecting and reporting population data on households in mid-2009. The CHWs (2–5 per village) submitted monthly reports, which included the number of inhabitants. HCs routinely collected monthly CHW reports and aggregated the data at the sector level. We used CHW-level paper reports for up to 4 months (August, September, and December 2009 and January 2010) and aggregated the total population numbers in these reports by village by month, before calculating the median monthly village population for each village. The village-level estimates were then aggregated to give cell and sector population number estimates. These population estimates, along with the estimated national crude birth rate, were used to calculate expected births by cell and sector. To test the variability in the CHW population data, we redid the calculations using the lowest and highest observations for each village, and the effect on the cell- and sector-level C-section rates was marginal, with a less than 5% difference for all but two cells, where the difference was less than 7%. Once the DHS 2010 was available, we again tested the reliability of the CHW data by calculating expected births using AfriPop 2010 estimates of women of child-bearing age, and population distribution and rural age-specific fertility rates averaged across the 2007–2008 and 2010 DHSs. A paired t-test showed no statistical difference between the two methods (P = 0.5098).

Outcome indicators

The main outcome indicator, C-section rate, was measured using two different denominators: estimated total births in the HC catchment areas (population-based C-section coverage rate) and total health facility births (health facility-based C-section rate) (Table 1). Other indicators included the hospital-based delivery rate and facility-based peri-partum foetal mortality rate. Both of these rates were calculated as a proportion of total documented births by patient cell or sector of residence. Health facility refers to both the HCs and the DH.

Table 1. Definitions of outcome indicators
Population-based C-section coverage rate (per 100 estimated total births)Total C-sections at the DH for women residing in a cell/Expected total births for that cell. This rate is a UN indicator for access to emergency obstetric care (WHO et al. 2009).
C-section rate in health-facility based deliveries (per 100 health-facility based deliveries)Total C-sections at the DH in women residing in a sector/Total documented health-facility based deliveries at any health facility in S. Kayonza for women residing in that sector. This rate shows the percentage of women presenting for health-facility based deliveries who received a C-section.
DH delivery rate (per 100 health-facility based deliveries)Total deliveries that occurred at the DH for women residing in a sector/Total documented health-facility based deliveries at any health facility in S. Kayonza for women residing in that sector. This measured the proportion of documented health-facility based deliveries that occurred at the DH, as a measure of health facility to health facility access.
Facility-based peri-partum fetal mortality rate (per 100 health-facility based deliveries)Total neonates that were documented to be dead at delivery or to have died in the first ten minutes of life whose mothers lived in a cell/Total documented health-facility based deliveries at any health facility in S. Kayonza for women residing in that cell.


Maps were created in ArcGIS 9.3.1, linking C-section coverage rates and facility-based peri-partum foetal mortality rates to cells, the government-determined sub-sector administrative boundaries. Distances by road between each HC and the DH were calculated using the network analysis extension. Ambulance travel time was estimated using the calculated distances and an assigned average speed on the two different types of road, paved road and major unpaved road. Health facility and population-based C-section rates, facility-based peri-partum foetal mortality rates and the rates of C-section for uterine rupture were compared across sectors. Chi-squared test or Fisher exact test were used to compare the sectors with the lowest and highest C-section rates to the other sectors.

Workshops led by the GIS team were held with Ministry of Health and PIH staff at the district, hospital and local levels to discuss how GIS data can be used in program monitoring and evaluation as well as quality improvement activities.

Ethical approval

This project was reviewed by the institutional review boards of Partners Health care in Boston, USA and the Rwanda National Ethics Committee. Patient data were aggregated at the village or cell level and were anonymised for analysis and reporting.


HC catchments and distances to DH

Documentation of patient addresses at the HCs allowed linkage of over 97% of women presenting for delivery at a health facility to their cell of residence, although one HC (G) had lower rates of documentation (Table 2). Most women presenting in labour resided within the HC's catchment, with the exception of HC (C), where about 50% of presenting patients lived in neighbouring sectors. The average age of women that delivered in the DH catchment ranged between 25.8 and 27.7 by sector. As shown in Figure 2, distances by road between the HCs and the DH ranged from 12 km to 32 km, and the estimated travel times varied from a half hour to two hours.

Table 2. Geographic characteristics of women registered in maternity ward registries in S. Kayonza
Health facility of presentation, represented by name of sectorTotal patients in labor registeredat each health facilitya% Patients linked to cell of residenceTotal patients presenting from assigned HC catchment
  1. a

    Including only patients originating from within DH catchment.

  2. b

    Patients identified as having been referred from one of the seven HCs were excluded.

Sector A429428 (99.8)421 (98.1)
Sector B797795 (99.7)777 (97.5)
Sector C236235 (99.6)131 (55.5)
Sector D384384 (100.0)381 (99.2)
Sector E347344 (99.1)331 (95.4)
Sector F527525 (99.6)520 (98.7)
Sector G419369 (88.1)412 (98.3)
Sector H (HC+DH)b13401302 (97.2)NA
Figure 2.

Travel time and distance by road from each HC to the DH.

Based on the CHW data, cell populations ranged from 2 300 to 8 200 people, and sector populations ranged from 14 000 to 31 000. The calculated expected births in 2009 ranged from 88 to 313 births by cell, and from 551 to 1183 births by sector.

The number of patients presenting in labour from within the DH catchment ranged from 236 at HC (C) to 1340 at the DH (Table 2). The DH delivery rate was significantly lower for women residing in Sector E (P < 0.016) and significantly higher for women residing in Sector B (P < 0.001) than for women living in other sectors (Table 3).

Table 3. Distribution by sector of residence of health-facility (HF) based deliveries and C-sections in S. Kayonza
Sector of residenceEstimatedtotal populationEstimated total birthsTotal documented HF-based deliveriesDeliveries at the DH (% of HF-based deliveries)Total C-sections (% of HF-based deliveries)bC-section coverage rates (per 100 estimated total births)Range inby-cell C-section coverage rateC-sections for Uterine Rupture (% C-sections)
  1. a

    Only one cell in Sector C.

  2. b

    Sectors E and H significantly different from the other sectors (P < 0.001).

Sector A31 0381183544149 (27.4)59 (10.8)4.993.0–8.71 (1.6)
Sector B24 953951961330 (34.3)112 (11.7)11.7810.1–16.70
Sector Ca443716914724 (16.3)13 (8.8)7.697.70
Sector D17 408663502144 (28.7)49 (9.8)7.394.9–10.22 (3.8)
Sector Eb14 45955135177 (21.9)24 (6.8)4.363.0–7.03 (12.5)
Sector F26 213999712188 (26.4)91 (12.8)9.118.5–9.63 (3.3)
Sector G14 998571468118 (25.2)56 (12.0)9.818.8–13.02 (3.6)
Sector Hb23 642901790NA145 (18.4)16.0911.7–29.62 (1.4)

Population-based C-section coverage rates by cell

The map of population-based C-section coverage rates (Figure 3) varied from 3.0 to 29.6%. The highest rates were in the area directly surrounding the hospital, while the lowest rates were in the eastern, more remote part of the DH catchment area. For the lowest C-section rate (3.0%), the 95% confidence interval extended from 1.5 to 5.6%, barely overlapping the 5% minimum WHO-recommended C-section rate. The 95% confidence interval for the highest C-section rate (29.6%) in the DH Sector – 22.1 to 38.2% – was above the WHO-recommended threshold of 15%.

Figure 3.

C-section coverage rate (per 100 expected births) by cell of residence with road network.

Health facility-based C-section rates

In comparing the rates of C-sections in all health facility-based deliveries, Sector E had a significantly lower rate than other sectors (P < 0.001), and Sector H had a significantly higher rate than other sectors (P < 0.001). Figure 4 shows the inverse relationship between rate of C-sections in health facility-based deliveries and travel time between the HC and DH.

Figure 4.

C-section rates in facility-based deliveries vs. HC to DH travel time.

Facility-based birth outcomes

Facility-based peri-partum foetal mortality rates ranged between 0.4 and 4.6%. Figure 5 suggests the co-existence of low C-section coverage rates and higher facility-based peri-partum foetal mortality rates in the eastern DH catchment. Using sectors of residence rather than cell of residence due to the low incidence of immediate peri-partum foetal mortality documented in the 2009 birth registries, women living in the catchment area of HC (E) had a higher facility-based peri-partum foetal mortality rate compared with women living in other sectors in the DH catchment (P < 0.034).

Figure 5.

Overlay of facility-based peri-partum foetal mortality rates (per 100 facility-based births) and population-based C-section coverage rates (per 100 estimated births) by cell of residence.

Uterine rupture as an indication for C-section indicates delay in accessing emergency obstetrical care. Differences in uterine rupture as the indications for C-sections based on sector of residence were seen between Sector E (3 of 24 C-sections) and Sector H, where all patients automatically deliver at the DH, obviating physical access barriers to emergency obstetrical care in HC-to-DH referrals (P = 0.02).


We found that a locally implemented, low-resource GIS program successfully integrated routinely available service delivery data with available geographic data. Feedback of results to local decision makers using maps of C-section data allowed visualisation of C-section rates by cell of patient residence, linking physical access, and in particular accessibility by ambulance (road density and total distance from the DH), with high (>15%) and low (<5%) population-based C-section coverage rates. This addition of the geographic analysis provided program managers with the ability to rapidly identify distance as a modifiable factor associated with lower access.

The use of GIS in the context of program monitoring also allowed for the geographic overlay of different data types. Facility-based peri-partum foetal mortality rates and rates of C-section for uterine rupture added additional evidence suggesting that utilisation of C-sections for women living in the part of the sub-district furthest from the DH may have been too low. In developing countries, uterine rupture is usually the result of prolonged dystocia and can be a marker for delay of accessing emergency obstetric care. Other studies have linked decreased physical access to health facilities with adverse outcomes in maternal health and other areas of health care (O'Meara et al. 2009; Pirkle et al. 2010).

The local integration of GIS into the review of available data also provided district program managers with potential evidence for physical access as a cause of C-section rate variability, in particular in Sector E, where lower road accessibility and higher calculated travel times overlapped with the lowest C-section rates. Based on these findings, program managers were concerned that there may be an unmet need for access to C-sections in that area. This information was presented in a workshop to key DH and program decision makers. As a result, programmatic decisions based on the available data were made to address the decreased physical access, and an additional ambulance was placed in the more remote HC. In contrast, the unexpectedly high rate of C-sections for women residing in cells adjacent to the paved road and DH suggests that this may be an over-utilised mode of delivery for women with easy access, as reflected in the higher number of cases coming from these cells where the indication for C-section was coded as elective (data not presented here).

The value of GIS data to drive local decision making and resource allocation has also been seen in rural Indonesia, which documented the use of spatial analytical skills six months following the training of health officers (Fisher & Myers 2011). Health officers implemented the open-source mapping system and used the results for auditing health infrastructure in two districts, directing new staff placement based on the location of existing midwives, prioritising clinic upgrades and road repairs based on priority areas, and coordinating transportation to health facilities for term pregnant women. In Rwanda, integrating GIS into decision making is already serving as a platform for further strengthening data use in the health sector. For example, in the Eastern Province, the use of geographic data to evaluate and plan programs has been used to identify areas with particularly high proportions of underweight HIV patients, allowing for focusing of CHW efforts in providing nutritional supplements and education in those areas most in need (Munyaneza et al. 2011).


Similar to many settings in sub-Saharan Africa, data on actual total births were outdated and did not reflect the population movement that occurred in the intervening 7 years. While this may have altered the calculated population-based rates, we constructed the estimate that took advantage of the CHW village-level monitoring and combined these locally derived population estimates with the national crude birth rate.

Ascertainment of facility births was limited to health facilities within the DH catchment area. If patients living at the periphery of the catchment area presented to other health facilities, the estimated C-section rate may be artificially low. While this may be a consideration in Sector A, where a health centre outside the DH catchment is located nearby, the area with the lowest C-section rate (Sector E) is remote with no access to alternative facilities.

Neonatal mortality, defined as the rate of infants that die in the first 28 days of life, is a more valid measure for adverse birth outcomes than immediate perinatal mortality (WHO et al. 2009), but this data were unavailable. We were also unable to measure maternal mortality as an outcome at the sector level due to the low incidence over the one year of the study.

The driving times from the HCs to DH calculated using standard speeds on different types of road and distances by road were not confirmed with actual travel time data under varying conditions. Given the terrain and lack of paved roads to the more remote HCs, the travel times were likely underestimated for routes to those HCs, and the association observed in this study may be more pronounced than presented here. Measuring actual ambulance travel times between health facilities and travel times from villages to the HCs would be important in follow-up studies.

In addition to the geographic barriers identified in the study, other factors which we were not able to capture may have affected C-section rates. Studies have shown patient factors such as perceived severity of illness, socio-economic status and perceived quality of care to be significant determinants of the accessing of health care in rural sub-Saharan African settings (Heard et al. 2004; Kiwanuka et al. 2008; Feikin et al. 2009; Van Hemelrijck et al. 2009). These data were not available as a part of the routinely available health facility data. Patient characteristics such as parity, age at first birth and birth interval are important determinants of C-section rates. While we were not able to include as cofactors, these are unlikely to vary geographically across sectors within a single district, and therefore, were unlikely to have had a significant effect on the geographic relationships found in this study.


In conclusion, our use of GIS to combine geographic and routine clinical data along with the involvement of local leaders resulted in the integration of GIS as a tool for program monitoring, data-driven decision making at the district level and measurement of impact of these decisions in the future. We believe that this low-resourced, simple geographic monitoring and evaluation system can increase the effective use of routinely collected data and help countries improve healthcare access and outcomes at the local level. Although the use of GIS to assess access to health services in East Africa is not new, novel aspects of our approach include the community involvement from initial conception, the focus on sub-district-level outcomes and the low technical expertise required to train staff and perform the data collection and analysis. Plans are already underway to replicate in other health service areas and other districts supported by PIH, with the goal of strengthening the capacity of district health officials and their partners to use this approach to identify and guide responses to gaps in the delivery of a specific health service in the communities.


The authors thank Deo Rutamu and Paulin Basinga for their help in acquiring the Rwanda-specific GIS data; Didi Bertrand Farmer, Denise Uwera and Elias Ngizwenayo for their support in facilitating this work; and Issa Kamatari for assistance with data collection. Many thanks also go to Bethany Hedt, Ann Miller and Mary Kay Smith-Fawzi for their help with design- and analysis-related questions and to Dana Thomson for her help with the population estimates. Most of all, we are grateful to the community health workers and maternity staff in Kayonza District for their help with data collection and their dedication to reducing maternal mortality for the women in their communities.