Does facility birth reduce maternal and perinatal mortality in Brong Ahafo, Ghana? A secondary analysis using data on 119 244 pregnancies from two cluster-randomised controlled trials

Background: Maternal and perinatal mortality are still unacceptably high in many countries despite steep increases in facility birth. The evidence that childbirth in facilities reduces mortality is weak, mainly because of the scarcity of robust study designs and data. We aimed to assess this link by quantifying the influence of major determinants of facility birth (cluster-level facility birth, wealth, education, and distance to childbirth care) on several mortality outcomes, while also considering quality of care. Methods: Our study is a secondary analysis of surveillance data on 119 244 pregnancies from two large population-based cluster-randomised controlled trials in Brong Ahafo, Ghana. In addition, we specifically collected data to assess quality of care at all 64 childbirth facilities in the study area. Outcomes were direct maternal mortality, perinatal mortality, first-day and early neonatal mortality, and antepartum and intrapartum stillbirth. We calculated cluster-level facility birth as the percentage of facility births in a woman’s village over the preceding 2 years, and we computed distances from women’s regular residence to health facilities in a geospatial database. Associations between determinants of facility birth and mortality outcomes were assessed in crude and multivariable multilevel logistic regression models. We stratified perinatal mortality effects by three policy periods, using April 1, 2005, and July 1, 2008, as cutoff points, when delivery-fee exemption and free health insurance were introduced in Ghana. These policies increased facility birth and potentially reduced quality of care. Findings: Higher proportions of facility births in a cluster were not linked to reductions in any of the mortality outcomes. In women who were wealthier, facility births were much more common than in those who were poorer, but mortality was not lower among them or their babies. Women with higher education had lower mortality risks than less-educated women, except first-day and early neonatal mortality. A substantially higher proportion of women living in areas closer to childbirth facilities had facility births and caesarean sections than women living further from childbirth facilities, but mortality risks were not lower despite this increased service use. Among women who lived in areas closer to facilities offering comprehensive emergency obstetric care (CEmOC), emergency newborn care, or high-quality routine care, or to facilities that had providers with satisfactory competence, we found a lower risk of intrapartum stillbirth (14·2 per 1000 deliveries at >20 km from a CEmOC facility vs 10·4 per 1000 deliveries at ≤1 km; odds ratio [OR] 1·13, 95% CI 1·06–1·21) and of composite mortality outcomes than among women living in areas where these services were further away. Protective effects of facility birth were restricted to the two earlier policy periods (from June 1, 2003, to June 30, 2008), whereas there was evidence for higher perinatal mortality with increasing wealth (OR 1·09, 1·03–1·14) and lower perinatal mortality with increasing distance from childbirth facilities (OR 0·93, 0·89–0·98) after free health insurance was introduced in July 1, 2008. Interpretation: Facility birth does not necessarily convey a survival benefit for women or babies and should only be recommended in facilities capable of providing emergency obstetric and newborn care and capable of safe-guarding uncomplicated births. Funding: The Baden-Württemberg Foundation, the Daimler and Benz Foundation, the European Social Fund and Ministry of Science, Research, and the Arts Baden-Württemberg, WHO, US Agency for International Development, Save the Children, the Bill & Melinda Gates Foundation, and the UK Department for International Development.

Our study is a secondary analysis of data from two cluster-randomised controlled trials, ObaapaVitA35 and Newhints,36 for which data were continuously collected between 2000 and 2009 in seven contiguous districts of the Brong Ahafo region in Ghana. ObaapaVitA35 tested the effect of low-dose vitamin A supplementation on mortality of women of reproductive age (enrolled at age 15–45 years) and of their babies, and collected data from Dec 11, 2000, enrolling women in a staggered way across districts, until Oct 31, 2008. Newhints36 tested the effect of home visits by community-based surveillance volunteers on neonatal mortality, and collected data from Nov 1, 2008, to Dec 31, 2009. Neither study showed a significant effect on mortality.35, 36 The surveillance system established for the trials included home visits every 4 weeks to women of reproductive age to identify and register pregnancies, births, and deaths. Data were collected on place of delivery, caesarean section, pregnancy-related mortality, stillbirth, and neonatal mortality, as well as sociodemographic characteristics. Data collection is described in the key trial publications.35, 36 We harmonised and jointly analysed data from the ObaapaVitA35 and Newhints36 trials. The unit of all analyses was the delivery episode (including deaths in women who had not delivered), which meant a woman could contribute several delivery episodes over time and that twin or triplet births were considered as one episode. A delivery episode was considered to result in stillbirth or early neonatal death if at least one baby fulfilled the criteria for this outcome, so in a few cases, a delivery episode was counted as having resulted in two different outcomes (eg, if twins died at different timepoints). Births in hospitals, health centres, clinics, or maternity homes were considered to be facility births. The mortality outcomes we considered were: stillbirth (born dead after at least 6 months of gestation), separated into antepartum and intrapartum stillbirth (further details available in the study by Ha and colleagues);37 early neonatal death (death of a liveborn infant within the first 7 days of delivery), with the subgroup first-day neonatal death (death of a liveborn infant within 24 h of delivery); perinatal death (stillbirth or early neonatal death); and direct maternal death (death from obstetric complications or interventions during pregnancy or within 42 days thereof). Livebirths with incomplete follow-up for the first 7 days were excluded from the analyses of early neonatal and perinatal mortality. Cluster-level facility birth was calculated as the percentage of facility births in a village or suburb. We used cluster-level facility birth in the preceding 2 calendar years as a predictor for the index birth. Unlike using births in the same year, this strategy avoids confounding by complications at the cluster level. Some delivery episodes from a few very small villages were excluded from this analysis because they had no births recorded in the preceding 2 years, leading to missing values in cluster-level facility birth. The same is true for births before 2003, when no childbirth records of the previous 2 calendar years were available. To measure wealth, we calculated household asset quintiles using principal component analysis of household assets according to DHS methodology.38 Mother’s education was coded in four levels: none; primary school; middle school or junior secondary school; and technical, commercial, or senior secondary school, or post-middle college, or post-secondary or higher education. We used global positioning system coordinates of health facilities and village centroids to calculate distances from the woman’s regular place of residence to the closest health facility and to the closest high-quality health facility, considering several quality dimensions. Straight-line distances to a comprehensive emergency obstetric care (CEmOC) facility ranged from less than 1 km to 84 km.39 Women in three of the larger towns (Nkoranza, Techiman, and Kintampo) were assigned the centroid of the respective suburbs as their place of residence. Road network data were used to calculate road distance and travel-time measures for sensitivity analyses.39 For the purpose of this analysis, we visited all 86 health facilities in the study area in 2010 to assess quality of obstetric and newborn care. Of the 64 facilities offering childbirth care, 24 were classified as capable of providing high-quality routine care, 12 as capable of providing emergency obstetric care (EmOC), of which eight were capable of providing CEmOC, and five were capable of providing emergency newborn care (EmNC).40 Detailed information on methods and findings of this comprehensive health facility assessment have been published elsewhere.40, 41 Briefly, we used information on key signal functions, availability of drugs, equipment, and trained health professionals to create quality scores of different dimensions of care, including routine childbirth care, CEmOC, and EmNC. Furthermore, we used clinical vignettes to assess health professional competence, interviewing the most experienced provider, present at the day of visit, who manages childbirth and newborn infants at the facility. Two vignette cases tested ability to diagnose and manage conditions that threatened the lives of both mother and baby—pre-eclampsia and severe antepartum haemorrhage. On average, providers mentioned 11 of 20 necessary actions correctly, with the number of correct answers ranging from one to 15.41 The four quality-of-care variables used in this analysis were distance to the closest health facility offering CEmOC, distance to the closest facility offering EmNC, distance to the closest facility offering high-quality routine childbirth care, and distance to the closest facility with staff who achieved a vignette score of at least 12 of 20. To assess the effect of Ghana’s 2005 policy on free childbirth care and its 2008 policy on free national health insurance for pregnant women, we studied the association between facility birth and mortality during three time periods, defined in previous analyses.33 The first period reflected the time before the policy change, starting June 1, 2003 (because variables that were needed to adjust for confounding were consistently collected from this date) and finishing March 31, 2005. The second period started on April 1, 2005, when the nationwide delivery fee exemption policy was introduced, and finished June 30, 2008. The third period started on July 1, 2008, when free national health insurance was introduced for pregnant women; this period ended with the end of Newhints surveillance on Dec 31, 2009.36 Although data on stillbirth, early neonatal mortality, first-day neonatal mortality, and perinatal mortality were available for the full sample (2000–09), data on antepartum stillbirth and intrapartum stillbirth were available only from June, 2003, to October, 2008, and data on direct maternal mortality only until the end of the ObaapaVitA trial35 in October, 2008. The total numbers of pregnancies, deliveries, and deaths in adjusted and unadjusted analyses are shown in a flowchart (appendix, p 2). For presentation in figure 1, we categorised continuous exposure variables into a small number of groups, so that the proportion of facility births, caesarean sections, and all types of mortality risks could be plotted by category. We then assessed associations in crude and multivariable two-level logistic regression models, with village of residence at level two, thus taking the similarities of births from the same village into account. Health service use and mortality outcomes by cluster-level facility birth, wealth, education, and distance to facilities offering various levels of care Facility birth and caesarean section (right axis), and mortality (left axis) by cluster-level facility birth (A), household wealth (B), mother’s education (C), distance to closest childbirth facility of any level (D), distance to closest facility providing CEmOC (E), distance to closest facility providing EmNC (F), distance to closest facility offering high-level routine care (G), and distance to closest facility with satisfactory provider competence (H). CEmOC=comprehensive emergency obstetric care. EmNC=emergency newborn care. We analysed the effect of the proportion of cluster-level facility birth in the preceding 2 years as a continuous variable. The effects of household wealth were estimated per wealth quintile and those of mother’s education were estimated per highest education level reached. To establish the functional shape of the association between distance and outcomes, we used fractional polynomials of first degree, assuming a monotone dose-response relationship.42 Across associations we found that transformations with slopes that flatten for larger distances, such as the logarithm or the square root of distance, were better than linear or quadratic slopes. Thus, all distance variables were log-transformed for the analyses. Multivariable analyses were adjusted for year of birth, multiple birth, mother’s age, parity, religion, ethnicity, occupation, education, wealth, and distance to closest CEmOC (in the models with wealth and education as main exposures) and restricted to births after June 1, 2003, because different data collection procedures before that date led to more missing values for adjustment variables. We then dropped observations with missing values in any of the adjustment variables, which amounted to about 1% of the sample after June 1, 2003. The direct maternal mortality outcome was rare, with only 200 deaths during the entire observation period, and we wished to use all pregnancies from the year 2000 onwards, despite missing data on household wealth, education, occupation, and multiple birth for many women who died before 2003. We used multiple imputation (mi command in Stata) with 20 imputations for these four variables in an imputation model that included year of birth, mother’s age, parity, religion, ethnicity, and the respective main exposure and the outcome variables. Thus, the regression models for direct maternal mortality were adjusted for year of birth, mother’s age, parity, religion, ethnicity, occupation (partly imputed), education (partly imputed), wealth (partly imputed), multiple birth (partly imputed), and distance to the closest CEmOC (in the models with wealth and education as main exposures). This and all other analyses were done with Stata IC 14 software.43 For completeness and comparability to other studies, we also examined the association between individual-level facility birth and mortality outcomes in adjusted analyses (appendix, p 6). We also did four sensitivity analyses that are described and summarised in the appendix: crude analyses in the restricted sample from June, 2003 (appendix, p 14), using road distance and travel time (appendix, p 18), restricting the sample to women with good pregnancy surveillance (appendix, p 20) and using a three-level random-effects model (appendix, p 23). Results were very similar to the main results presented herein. We obtained ethical approval from the London School of Hygiene and Tropical Medicine in London, UK, and from the Kintampo Health Research Centre in Kintampo, Ghana. All participants of the ObaapaVitA35 and Newhints36 trials provided written informed consent on recruitment. Health workers provided written informed consent for the health facility assessment before the start of data collection. The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of this report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

• gabrysch, sabine
• nesbitt, robin c.
• schoeps, anja
• hurt, lisa
• soremekun, seyi
• edmond, karen
• manu, alexander
• lohela, terhi j.
• danso, samuel
• tomlin, keith
• kirkwood, betty r.
• campbell, oona m.r.