Objective To examine the changing temporal association between caesarean birth and neonatal death within the context of Ethiopia from 2000 to 2016. Design Secondary analysis of Ethiopian Demographic and Health Surveys. Setting All administrative regions of Ethiopia with surveys conducted in 2000, 2005, 2011 and 2016. Participants Women aged 15-49 years with a live birth during the 5 years preceding the survey. Main outcome measures We analysed the association between caesarean birth and neonatal death using log-Poisson regression models for each survey adjusted for potential confounders. We then applied the Three Delays Model’ to 2016 survey to provide an interpretation of the association between caesarean birth and neonatal death in Ethiopia. Results The adjusted prevalence ratios (aPR) for neonatal death among neonates born via caesarean section versus vaginal birth increased over time, from 0.95 (95% CI: 0.29 to 3.19) in 2000 to 2.81 (95% CI: 1.11 to 7.13) in 2016. The association between caesarean birth and neonatal death was stronger among rural women (aPR (95% CI) 3.43 (1.22 to 9.67)) and among women from the lowest quintile of household wealth (aPR (95% CI) 7.01 (0.92 to 53.36)) in 2016. Aggregate-level analysis revealed that an increased caesarean section rates were correlated with a decreased proportion of neonatal deaths. Conclusions A naïve interpretation of the changing temporal association between caesarean birth and neonatal death from 2000 to 2016 is that caesarean section is increasingly associated with neonatal death. However, the changing temporal association reflects improvements in health service coverage and secular shifts in the characteristics of Ethiopian women undergoing caesarean section after complicated labour or severe foetal compromise.
We used data from the Ethiopian DHS completed in 2000, 2005, 2011 and 2016. The Ethiopian DHS are nationally representative cross-sectional surveys conducted in nine regional states (Tigray, Affar, Amhara, Oromia, Somali, Benishangul-Gumuz, SNNPR, Gambela and Harari) and two city administrations (Addis Ababa and Dire Dawa). Each of the surveys involved a two-stage, stratified, clustered sampling design. The survey datasets are deidentified and made freely available online. Permission to use these data was granted by the DHS Program. The details about the methodology and standards for protecting the privacy of study participants in all DHS can be accessed online (http://www.dhsprogram.com/What-We-Do/methodology.cfm). The DHS questionnaire asks women about pregnancy, antenatal and delivery care for live births they have reported in the past 5 years. The data on caesarean section and other variables in the DHS were collected based on mothers’ self-report. For example, the self-reported data on caesarean section were collected by asking mothers a question that reads, ‘Was (NAME) delivered by caesarean section, that is, did they cut your belly open to take the baby out?’ in the 2016 survey. Stanton and colleagues25 in their study demonstrated that the DHS caesarean section rates, compared with facility-based records of caesarean section rates, are reliable for national and global monitoring in developing countries. For this study, the exposure group were infants delivered by caesarean section and unexposed group comprised infants born vaginally. Neonatal death includes infants who were born alive in the 5 years before each survey, but died within the first 28 days of life. The outcome variable, neonatal death, was measured from two variables (whether the child is alive and age at death (in days)). The following potential confounders were identified based on a priori subject-matter and expert knowledge. They included place of delivery (public, private, non-governmental organisation and home), type of residence (urban/rural), sex of child (male/female), size of baby at birth (very large, larger than average, average, smaller than average, very small and do not know), mother’s age at birth (in years), mother’s education (no education, primary, secondary and higher), birth order (1, 2–3 and 4+) and household wealth quintile (poorest, poorer, middle, richer and richest). The size of baby at birth was assessed based on mother’s perception (estimate) of baby size at birth. It has previously been shown that, in the absence of complete enumeration of birth weight, mother’s perception of baby size at birth can be used as a proxy to birth weight in nationally representative surveys.26 Mother’s age at birth was calculated as a difference (in years) between infant’s date of birth and mother’s date of birth. The DHS computes the wealth index for each survey based on household assets using principal components analyses27 and categorises households into wealth quintiles. These asset-based measures represent the wealth distribution relative to other households within the country. They are widely used and are consistent with comparisons to household expenditures and the measurement of inequalities in child mortality, education and healthcare use in low-income and middle-income countries.28 Missing information is uncommon in DHS because the data are collected by a trained interviewers at a face-to-face interview. All analyses (ie, Ethiopian DHS 2000, 2005, 2011 and 2016) were weighted to be nationally representative. As women may have had more than one births within the 5-year survey periods, we also accounted for both clustering of caesarean deliveries within women as well as the complex survey design during the data analyses using the unit of analysis (ie, children) study number and sample weights. We then conducted both individual-level and aggregate-level analyses. Our 2016 data analysis was also supplemented by an application of the ‘Three Delays Model’ to interpret the association between caesarean birth and neonatal death both empirically and theoretically. All analyses were conducted using STATA/SE V.15.1 (Stata Corporation). Associations between caesarean birth and neonatal death at individual-level were analysed using log-Poisson regression models using data from Ethiopian DHS conducted in 2000, 2005, 2011 and 2016. We calculated unadjusted and adjusted prevalence ratios (aPR) and their 95% CIs for each survey. We have then compared the strength of association between caesarean birth and neonatal death across all surveys analysed. After noting the increasing association between caesarean birth and neonatal death over time, we conducted a series of analyses to explore what was during the change. We used the 2016 data because the association was more pronounced. We first restricted the analysis to participants living in regions with the highest caesarean section rates to examine whether the increased access to caesarean section affected the proportion of neonatal deaths. We then estimated the effect of caesarean birth on neonatal death in regions with low caesarean section rate (ranged: 0.4%–5.3%) or where access to caesarean section is limited, by excluding births in relatively high caesarean section rate regions—Addis Ababa (21.4%) and Harari (9.0%).29 Both low-level and high-level of caesarean use has risks exceeding the risks of spontaneous vaginal deliveries.15 30 It was demonstrated that low levels of caesarean are related to lack of access and can contribute to maternal and newborn deaths.21 31 Given the very large rural–urban differences in caesarean section rates in Ethiopia,29 32 we also conducted similar analyses separately for rural women. In addition, we evaluated the association by restricting the analyses to births from the lowest quintile of household wealth, births from the highest quintile of household wealth, and births in public health facilities separately. These alternative analyses were exploratory in nature and helped us understand contextual factors leading to inequalities in caesarean use that may occur not only due to inadequate access among the poorest women, but also due to overuse among the richest population subgroups.33 34 The subgroup analyses allowed us to explain how contextual factors such as unequal access, infrastructural and workforce constraints could play role in the association between caesarean section and neonatal death because these factors will result in delay in accessing emergency caesarean section, which is usually accessible at specialised health facilities. The 2016 DHS included an additional question regarding ‘timing of decision to conduct caesarean section (ie, whether it was before or after the onset of labour pains)’. We used this variable as a proxy to the types of caesarean birth (indicative of intrapartum or prelabour caesarean section) and conducted analysis to examine the association between types of caesarean section and neonatal death. As this was confined only to 2016 data, we have provided the results in online supplementary table A1. bmjopen-2018-027235supp001.pdf Data on the caesarean section rates and proportion of neonatal deaths were disaggregated by urban–rural areas for each of the nine regional states and two city administrations in Ethiopia for each of the surveys completed in 2000, 2005, 2011 and 2016. However, the urban–rural stratification for Addis Ababa is only available for the 2005 survey. These results in a total of 85 data points (observations). In order to assess the correlation between caesarean section and neonatal death at the aggregate level, we conducted simple linear regression for overall surveys together and for individual surveys separately. The ‘Three Delays Model’ is a conceptual framework developed by Thaddeus and Maine to examine factors contributing to maternal mortality with specific focus on those that affect the ‘interval between the onset of obstetric complication and its outcome’.24 The ‘Three Delays Model’ summarises the various factors that affect this interval into three phases of delay—delay in deciding to seek care (phase I delay); delay in identifying and reaching medical facility (phase II delay); and delay in receiving adequate and appropriate treatment (phase III delay). Some of the key factors that shape the model include status of women; distance from health facility; availability and cost of transportation; condition of roads; distribution of health facilities; shortage of supplies, equipment and skilled birth attendants and adequacy of referral system.24 The pictorial presentation of the ‘Three Delays Model’ is provided in online supplementary figures A1–A4. As maternal and neonatal mortality share many risk factors, we adopted the ‘Three Delays Model’ as a framework to help interpret the association between caesarean birth and neonatal mortality within the context of Ethiopia using the 2016 survey because factors contributing to the ‘three delays’ aggravate the underlying medical indications for caesarean intervention that make neonatal death difficult to prevent. The 2016 survey was selected for interpretation of the association between caesarean birth and neonatal death using the ‘Three Delays Model’ because the association was more pronounced in the 2016 data. Previous studies conducted in India,35 Tanzania36 and Uganda37 have applied the ‘Three Delays Model’ to their analyses of perinatal deaths. We have identified some contributing factors underlying the ‘Three Delays Model’ from the 2016 survey. For example, information regarding problems faced by women of reproductive age (15–49 years) in accessing healthcare to obtain medical advice or treatment for themselves when they are sick were gathered. It consisted of four questions: distance to health facility (big problem/not big problem); getting money for treatment (big problem/not big problem); getting permission to go for treatment (big problem/not big problem) and not wanting to go alone (big problem/not big problem). Furthermore, data on skilled assistance during delivery, and women’s socioeconomic and demographic status are also available in the DHS. This information can particularly be important to understand and address the barriers that women face in seeking care during pregnancy and delivery.32 We have, therefore, analysed the 2016 data to describe these factors empirically in the context of Ethiopia. This research was done without patient involvement in setting the research question or the outcome measures, and in the design and implementation of the study. No patients were asked to advise on interpretation or writing up of results. There are no plans to disseminate the results of this research to study participants or the relevant patient community.