Use and Quality of Cesarean Delivery in a Low and Middle Income Setting: National Results from Two Emergency Obstetric and Newborn Care Assessments

Background: Cesarean delivery (CD) rates have reached epidemic levels in many high and middle income countries while increasingly, low income countries are challenged both by high urban CD rates and high unmet need in rural areas. The managing authority of health care institutions often plays a role in these disparities. This paper shows changes between 2008 and 2016 in CD rates and the capacity of the Ethiopian health system to deliver quality CD services, highlighting the role of the management sector. Methods: We compare results from two national cross-sectional emergency obstetric and newborn care assessments using descriptive statistics. The sample includes 111 hospitals in 2008 and 316 hospitals in 2016, and 275 CD case reviews in 2008 and 568 in 2016. Our primary outcome measures include population- and institutional-based CD rates; hospital readiness to perform CD; quality of clinical management; and the relative size of Robson classication groups Results: The national population-based rate increased (< 1% to 2.7%) as did all regional rates. Rates ranged from 24% in urban settings to less than 1% in several rural regions. The institutional rate was 54% in the private for-prot sector in 2016, up from 46% in 2008. Hospital readiness to perform CDs increased in public and for-prot hospitals. Only half of the women whose cases were reviewed received uterotonics after delivery of the baby, but use of prophylactic antibiotics was high. Partograph use increased from 9% to 42% in public hospitals, but was negligible or declined elsewhere. In 2016, a third of case reviews from the public sector were among low-risk nulliparous women (Robson group 1). Conclusions: Between 2008 and 2016, government increased the availability of CD services, improved public hospital readiness and some aspects of clinical quality. Strategies tailored to further reduce the high unmet need for CD and what appears to be an increasing number of unnecessary cesareans are discussed. Adherence to best practices and universal coverage of water and electricity will improve the quality of hospital services while the use of the Robson classication system may serve as a useful quality improvement tool.

ranged from 24% in urban settings to less than 1% in several rural regions. The institutional rate was 54% in the private for-pro t sector in 2016, up from 46% in 2008. Hospital readiness to perform CDs increased in public and for-pro t hospitals. Only half of the women whose cases were reviewed received uterotonics after delivery of the baby, but use of prophylactic antibiotics was high. Partograph use increased from 9% to 42% in public hospitals, but was negligible or declined elsewhere. In 2016, a third of case reviews from the public sector were among low-risk nulliparous women (Robson group 1).
Conclusions: Between 2008 and 2016, government increased the availability of CD services, improved public hospital readiness and some aspects of clinical quality. Strategies tailored to further reduce the high unmet need for CD and what appears to be an increasing number of unnecessary cesareans are discussed. Adherence to best practices and universal coverage of water and electricity will improve the quality of hospital services while the use of the Robson classi cation system may serve as a useful quality improvement tool.

Background
Increasingly, low and middle income countries face the two extremes of unmet need for and unnecessary cesareans, or "too little too late and too much too soon" (1). Just what the number of cesareans should be has evolved (2,3). The World Health Organization (WHO) began producing global guidance on this topic as early as 1985 (4). In 2015, backed by evidence that rates greater than 10% were not associated with a reduction in maternal and newborn mortality, the WHO stressed that all women who need a CD should receive one, while cautioning against pursuing speci c targets (5,6). Women who have a cesarean undergo perioperative risks such as blood loss, anesthetic accidents, wound infection and iatrogenic stula, and put future pregnancies at risk due to uterine scaring (7)(8)(9). A prospective study in Norway found that 1 in 5 women experienced at least short-term complications as a result of cesarean surgery (10). Nonclinical consequences as well as the health of newborns and children are of equal concern (11).
The slow rise has been attributed to weak health systems and the lack of resources to ensure quality of care. Recent global and regional studies on surgery have highlighted extensive unmet need for surgery.
They also documented safety concerns when undertaken too late or in environments where de cits in surgical personnel and resources are common, such as oxygen, medical equipment and electricity (11,12).
Ethiopia is a country with high maternal and newborn mortality, high prevalence of stula and historically low cesarean and institutional delivery rates. According to the 2016 Demographic and Health Survey, the CD rate was 2% based on births during the ve years prior to the survey (13). Earlier trend data from the capital indicated rising cesarean rates that were positively correlated with income and education, as in many other country settings (14,15). The current Health Sector Transformation Plan set a populationbased cesarean target of 8% for 2020 as a step towards addressing the unmet need for life-saving surgery (16). This paper presents a broad overview of recent changes in the service coverage and quality of CD, with a focus on differences across public and private sectors. We describe changes in national and regional CD rates; institutional rates; the readiness of hospitals to provide obstetric surgery; and the quality of CD services.

Study design and data collection
This secondary data analysis draws on two Emergency Obstetric and Newborn Care (EmONC) assessments from 2008 and 2016 (17,18). Both were national cross-sectional censuses of public and private health facilities.
The data collection instruments for 2008 were adapted from a set of standard modules previously used in many countries (19). The 2016 assessment used the same core tools administered in 2008 that underwent a global revision with local adaptation in 2016. The six modules relevant to this study were: M1 -basic infrastructure; M2 -human resources; M3 -inventory of drugs, equipment and supplies; M4summary service statistics; M5 -performance of signal functions; and M8 -case review for CDs. All facilities received modules 1-5; only facilities that performed CDs received M8. The latter was modi ed in 2016 to include information not in 2008 such as a woman's characteristics needed to classify her into one of the Robson 10 groups, type of anesthesia and professional cadre who performed the operation.
The rst assessment was launched October 1, 2008 and completed by January 15, 2009; the 2016 survey commenced in mid-May and was completed by mid-December 2016. In 2008, a private company conducted the assessment and prepared the databases (20) and in 2016, this responsibility shifted to the Ethiopian Public Health Institute. Details about the data collector training and survey execution can be found in the nal reports (17,18).

Study population and setting
If the Ethiopian Food, Medicine and Health Control Authority approved a facility as a site to deliver routine and/or operative childbirth services, it was eligible for the assessment. Between 2008 and 2016 the Ministry of Health led a massive infrastructure expansion, adding about 3000 health centers and 200 hospitals (21,22). In 2008, 751 health facilities with childbirth services were visited, including 111 hospitals. In 2016, 3,804 were visited, 316 of which were hospitals. This paper focuses on hospitals designated to provide comprehensive emergency obstetric care.
Although we call both assessments a census, in 2008 15 facilities were not visited because they did not appear on the master list of licensed facilities, 12 of which were in the capital of Addis Ababa. In 2016, 11 facilities were not visited due to civil unrest but few if any of these service sites were hospitals. Finally, two hospitals refused to participate in the 2016 assessment.
In each hospital, a subset of women who delivered by cesarean had their records reviewed. In 2008 data collectors identi ed three women per hospital and in 2016 only two cases per hospital were selected because of the increase in the number of hospitals. The selection criteria remained the same: 1) cases occurred in the previous 12 months, and 2) they were the last women who had had a cesarean but were no longer under postoperative care, regardless of survival.

Processes and comparisons
This paper features three units of analysis: aggregated hospital service statistics, hospitals and individual women who delivered by cesarean. Aggregated service statistics (M4) covered 12 consecutive months prior to the assessment (July 2007-June 2008 and January-December 2015). Service data included the number of deliveries by mode of delivery; these data were used to estimate CD rates. Modules 1, 2, 3 and 5 provided information to assess hospital readiness to perform CD while the CD chart review (M8) was the information source for service quality and record-keeping at the individual level.

Variables and index creation
For both population-based and institutional CD rates, the numerator was the sum of all CDs performed at each hospital. The denominator for the population-based rate was the number of expected births in each region, calculated from population gures from the Planning and Programming Department of the Federal Ministry of Health and the crude birth rate established by the 2016 Demographic and Health Survey (13,23). The denominator for the institutional CD rate was the sum of all births at each hospital that had provided cesarean services in the three months prior to the assessment. To assess health system readiness to provide CD services, we created a binary summary score (yes or no), based on an algorithm de ned by the availability of at least one health professional able to perform the operation and another to provide anesthesia, plus readiness items that had to be functioning and included EITHER an anesthesia machine + (halothane or ketamine) OR regional anesthesia (lignocaine/ lidocaine 4% or bupivacaine) AND an oxygen cylinder with manometer and owmeter (low ow) tubes and connectors, an operating table and a functioning adjustable light (20). Although not included in the algorithm, interruptions in water and electricity in the operation theaters were assessed.
Quality of clinical management was measured by use of a partograph, administration of prophylactic antibiotics and uterotonics, time interval from decision to incision, type of anesthesia, clinician who performed the CD and maternal and newborn outcomes.
A nal analysis of clinical management was based on the Robson 10-group classi cation scheme, designed to determine institutional cesarean rates for clinically relevant and mutually exclusive groups (24). The classi cation system depends on six characteristics of women that are easily captured: parity (nulliparous/multiparous), number of fetuses (singleton/multiple), onset of labor (spontaneous or induced/CD before labor started), previous CD (yes/no), fetal lie (cephalic/ transverse/breech) and gestational age (<37 weeks/ >37 weeks). Because the 10 groups (see Figure 1) tend to have different cesarean rates, the classi cation scheme is used to inform where changes in clinical management should be made. One of the goals of its usage is to reduce cesareans among nulliparous women (groups 1 and 2), who are known to be vulnerable to unnecessary CDs, and who are often the biggest group (25). Overuse of CD among these women sets up a domino effect that contributes to repeated cesareans.
Our aim in using the Robson classi cation was to determine the distribution of cases according to the 10 groups to show their relative size, if and how the group sizes varied across managing authority, and the extent to which the group sizes aligned with other studies.
Our key stratifying variable -hospital managing authority -was de ned with three categories: public or government, private for-pro t and private not-for-pro t management by non-governmental organizations and/or religious missions.
To produce descriptive statistics (frequencies, percentages, means and medians) we used SPSS version 24. Since our data sources were censuses and not random samples, nor did they represent some theoretical population, we performed no statistical tests.

Ethical considerations
This paper utilizes secondary data; permission to use the data was granted by the Ethiopian Public Health Institute and the Family Health Division at the Federal Ministry of Health.

Coverage: Cesarean delivery rates
The national population-based cesarean rate in 2008 was 0.6% and 2.7% in 2016 (Table 1). The highly urbanized regions of Harari, Addis Ababa and Dire Dawa exhibited the highest rates in 2016: 17%, 24%, and 10%, respectively. Elsewhere, rates increased but remained very low.
At the national level, 72% and 74% of the cesareans were performed in the public sector in 2008 and 2016, respectively. Half of the cesareans performed in Addis Ababa took place in the private sectors. The CD rate in Addis Ababa was likely underestimated in 2008 due to the 12 private hospitals not eligible for the assessment. Notes: SNNP = Southern Nations, Nationalities, Peoples' region The institutional CD rate is the percentage of institutional births delivered by cesarean and is in uenced by patient mix and provider practice ( Table 2). Patterns of institutional CD rates in 2008 and 2016 were similar: the private for-pro t sector had the highest rates (46% and 54%, respectively) and the public sector had the lowest (15% and 19%, respectively). The greatest percentage increase occurred among public hospitals (23%).  (Table 3). Among hospitals that had performed CDs, the availability of surgeons, obstetricians and medical doctors declined between 2008 and 2016 when emergency surgical o cers (ESOs) began to play a major role, especially in the public sector. The private for-pro t hospitals relied almost entirely on obstetricians, while the not-for-pro t hospitals increased their reliance on both obstetricians and ESOs. An increase from 95-100% of hospitals reported at least one professional on staff to administer anesthesia.
The availability of drugs and equipment did not change dramatically between the two assessments. In 2016, public sector hospitals were at some disadvantage compared to private hospitals. For example, in 2016 85% of public hospitals had vaporizers compared to 96-100% among private hospitals.
According to the readiness algorithm, hospital readiness increased in the public sector from 70-85% and among for-pro t facilities from 87-97%, but only 67-68% of not-for-pro t facilities were staffed and equipped to provide CDs.
Although not included in the readiness algorithm, we also examined whether hospitals experienced interruptions in electricity and running water. In 2008 virtually all hospitals that regularly performed cesareans had functioning water and electricity in the operation theater or in the hospital itself on the day of the assessment. In 2016 the percentage dropped to closer to 90% for both electricity and water.  3 In 2008, the question asked about lidocaine 1 or 2%. 4 De ned as [at least 1 professional to conduct operation + 1 for anesthesia] AND [EITHER an anesthesia machine + halothane or ketamine OR regional anesthesia (lignocaine/lidocaine 4% or bupivacaine)] AND [an oxygen cylinder with manometer, owmeter, tubes and connectors + an operating table + overhead light]. 5 In 2008, the question referred to functioning power in the facility on day of interview, not the OT.
Case reviews: quality of clinical care and record-keeping In 2008, 95 hospitals provided a total of 275 case reviews while in 2016, 288 hospitals provided 568 cases. The average age of the women was 26 years in both assessments. However, women whose cesareans were performed in the private for-pro t hospitals were on average 28 years of age. In 2016 nearly half (46%) of the women attending public hospitals were nulliparous, while only 25-28% of women attending private hospitals were nulliparous.
In both EmONC assessments, more than 75% of women had an emergency cesarean (Table 4).
Emergency cesareans were least frequent in the private for-pro t settings (47% in 2008 and 53% in 2016); emergency cesareans in public and not-for-pro t hospitals ranged from 83-85% in 2008 to 90-91% in 2016. Proportionally, about three times as many cases with a previous cesarean or uterine scar were seen in the private hospitals compared to public hospitals. This pattern repeated itself in 2016. Indications for CD did not change dramatically although the proportion of CPD/prolonged labor cases increased (34-45%) while breech cases decreased (14-3%).  4 Meconium stained amniotic uid, post-term, intrauterine fetal death, HIV + mother and premature rupture of membranes.
Information related to the quality of clinical care is found in Table 5. In 2016, 52% of the women in the public hospitals were administered prophylactic uterotonics after the baby was delivered. The percentage was 46% among women in for-pro t hospitals and 41% in the private not-for-pro t sector. The use of prophylactic antibiotics was higher overall, with an increase from 87% in 2008 to 94% in 2016. Increases occurred only in public and private not-for-pro t hospitals.
Among women whose cesareans were an emergency, de ned by having gone into labor, partograph use increased from 9-42% in public hospitals, was negligible in the private for-pro t hospitals and declined from 47-27% in the private not-for-pro t sector.
We analyzed the time from decision to surgery for women with emergency cesareans, despite a high rate Questions regarding the type of anesthesia administered and the cadre of the surgeon were asked only in 2016. Sixty percent of the women in public hospitals received a spinal while 24% were administered general anesthesia. In the private not-for-pro t sector 81% received a spinal and 9% general while 49% of women received a spinal in the private for-pro t hospitals. In public hospitals, 25% of women had an obstetrician in attendance, 38% had a general surgeon and 29% an emergency surgical o cer. In the private for-pro t hospitals, 88% of women were attended by an obstetrician and in the not-for-pro t hospitals the percentage was 69%. Fetal outcomes improved over time with live births increasing from 79-95%. The improved fetal outcomes were most evident in government hospitals where perinatal deaths among the reviewed cases declined from 18-3%. Two maternal deaths were documented in 2008 and one in 2016 (data not shown).  1 In 2008, the timing of antibiotics was not speci ed. In 2016, timing re ects antibiotics given either pre-or post-operatively. If no documentation of antibiotics was recorded, we assumed the woman did not receive them. 2 Asked of emergency cesarean deliveries only.
Robson classi cation Figure 1 shows the relative contribution of each Robson group by managing authority. In the public sector, Robson group 1 accounted for 33% of the cesareans reviewed, followed by group 8 (12%). In the for-pro t hospitals, group 5 dominated (28%), followed by group 2 (17%). This aligns with the ndings on indications (Table 4), in which "previous cesarean" was the second most frequent indication in for-pro t hospitals. Forty-two percent of the reviewed cesareans from the not-for-pro t hospitals were women belonging equally to groups 1 and 5. As stated earlier, about a third of the cases suffered missing information and could not be grouped.

Discussion
Nationally, Ethiopia exhibits a large unmet need for cesarean delivery. When disaggregated by region, CD rates ranged from 24% in Addis Ababa to under 1% in pastoralist regions. Where rates were highest, the private sectors contributed between 25 and 50% of all cesareans. Although the public sector provides the bulk of obstetric services, by 2016 more than half (54%) of all deliveries in private for-pro t hospitals were cesareans. These two extremes -widespread underutilization of life-saving surgical care and the overutilization of a costly procedure not without risk -are faced by other low and middle income settings, and is an example of how the obstetric transition fails to follow a linear trajectory (26). Cesarean delivery rates, like the maternal mortality ratio, are also indicators for social and wealth disparities. Chart reviews revealed how patient pro les and provider practices varied by managing authority. The forpro t and not-for-pro t hospitals disproportionately attracted women with a previous CD and nonemergency cases. The case reviews also pointed to mixed results regarding the practice of evidencebased interventions. The use of partographs to monitor labor more than tripled in the public sector while they were hardly used or declined in private hospitals. Uterotonics administered after the baby was delivered and prophylactic antibiotic use were also more evident in public hospitals than elsewhere. These three interventions help prevent serious complications and all women should receive them. Thus, it is encouraging that contrary to other experiences (27), the public sector performed as well if not better than private hospitals.
The inclusion of the Robson classi cation parameters in 2016 enabled us to see clearly the perpetuation of "once a cesarean, always a cesarean," especially among the women from the private hospitals and the contribution of group 5 to these two groups of hospitals.
Typically, groups 1, 2, 3 and 5 contribute heavily to the overall CD rate, while groups 6-10 account for a smaller proportion of all cesareans. According to a WHO multi-country study from 21 countries and 287 hospitals, groups 6-10 accounted for only 20% of the cesareans (28). In our case, groups 6-10 accounted for 38% of the cases. This might re ect Ethiopia's overall low cesarean rate and that higher risk women contributed more to the pool of cesareans than in the WHO multicenter study. As the overall cesarean rate increases, the contribution of groups 6-10 relative to other groups may decline.
Nevertheless, the largest Robson group in the public sector was group 1, a group vulnerable to unnecessary cesarean deliveries.

Strengths and Limitations
A strength of this paper is the richness of its data sources: two national health facility censuses that enabled comparisons between public and private sectors. Analyses drew on data from interviews with health workers, observation of infrastructure, 12 months of service statistics, and individual level data from women who had undergone a cesarean, resulting in a multifaceted overview of how the use and quality of cesarean delivery changed and suggesting some steps going forward.
Although each assessment was designed as a census of facilities, we know that about 15 facilities where cesarean services were provided were not eligible and therefore not visited in 2008 and at least two in 2016. These omissions affected 2008 more than 2016 and probably caused us to underestimate the 2008 population-based CD rate for Addis Ababa. However, we do not believe that their inclusion would modify our conclusions.
We recognize that observation generally produces more accurate results than reporting, especially if recall is required. Hospital readiness depended heavily on responses from staff rather than observation, for example, whether an oxygen cylinder was both available and functional. Furthermore, readiness results will change as does the de nition of the algorithm, and we encourage strengthening this de nition.
Data quality of the primary sources -admission, operating theater and discharge logbooks -is a wellrecognized limitation when working with health facility data. We are not clear why an increase in missing information occurred between the two assessments, especially among the case reviews. The inability to classify a third of the cesareans reviews into one of the 10 Robson groups points to omissions in recordkeeping, even though the six parameters used to group women are considered standard data points. The group that suffered the least amount of missing information (20 cases) was group 8, de ned only by singleton or multiple gestation. Group 5 suffered the most from missing information: 132 of 568 cases lacked at least one of the ve variables that de ned group 5, with "previous cesarean" the most frequently missing.
The case reviews were systematically selected but were technically a convenience sample, and do not represent all cesareans, especially those conducted in the private sectors where missing data tended to be higher. Similar studies have struggled with the use of non-randomly selected chart reviews but also recognized their value when presenting an overview of a single service delivery intervention (29,30).

Conclusions
During the 8-year interval between assessments, the government tripled the number of hospitals, raised the CD delivery rate in underserved regions, and improved both the readiness to perform obstetric operations and the quality of clinical care. The data also point to dangerously low CD rates in underserved rural areas. More equitable access to cesarean services can be achieved through strengthened referral systems, hospital maternity waiting homes and the continued expansion of facilities that ll geospatial gaps (31,32). Meanwhile, high institutional CD rates in the private sector suggest that not all cesareans may be medically indicated. Federal normative bodies as well as professional societies should investigate local conditions to identify what is driving the demand: nancial incentives on the supply side, professional inexperience with non-routine vaginal births, or other forces.
The use of the Robson classi cation would be a useful tool for future quality improvement efforts -to ensure that the right women are receiving this procedure while others are protected from unnecessary procedures. This paper is the result of secondary data analysis of two national assessments. We received permission to access and use the data from the Ethiopian Public Health Institute and the Family Health Division of the Federal Ministry of Health.

Consent to publish
Not applicable.

Availability of data and materials
The datasets generated and analyzed for this study are available from the Ethiopian Public Health Institute Director on reasonable request.

Competing interests
None of the authors has any competing interests.

Funding
The Bill and Melinda Gates Foundation provided initial support to all of the authors. But the Foundation did not in uence the writing process or substantive content.
Authors' contribution MGB took the lead on conceptualizing the paper and writing the rst draft. TGZ, AH, ALR and PB reviewed and revised. TGZ and PB did the statistical analyses.
All authors have read and approved the manuscript.