Facility delivery is an important aspect of the strategy to reduce maternal and newborn mortality. Geographic access to care is a strong determinant of facility delivery, but few studies have simultaneously considered the influence of facility quality, with inconsistent findings. In rural Brong Ahafo region in Ghana, we combined surveillance data on 11,274 deliveries with quality of care data from all 64 delivery facilities in the study area. We used multivariable multilevel logistic regression to assess the influence of distance and several quality dimensions on place of delivery. Women lived a median of 3.3 km from the closest delivery facility, and 58% delivered in a facility. The probability of facility delivery ranged from 68% among women living 1 km from their closest facility to 22% among those living 25 km away, adjusted for confounders. Measured quality of care at the closest facility was not associated with use, except that facility delivery was lower when the closest facility provided substandard care on the EmOC dimension. These results do not imply, however, that we should increase geographic accessibility of care without improving facility quality. While this may be successful in increasing facility deliveries, such care cannot be expected to reduce maternal and neonatal mortality.
This study is a secondary data analysis of the Newhints cluster-randomized trial from November 2008 to December 2009 on the impact of home visits by community-based surveillance volunteers on neonatal mortality17. In 2009, the study area comprised seven districts in the Brong Ahafo region of Ghana, a predominantly rural area with approximately 600,000 residents, of which over 100,000 are women of reproductive age17,18. Newhints surveillance entailed monthly home visits by resident field workers to all women of reproductive age and women were enrolled in the trial once they became pregnant. Information was collected on socioeconomic characteristics, obstetric history and pregnancy outcome for all women. The neonatal mortality rate in the study area was 31 per 1000 live births18, and the national maternal mortality ratio was estimated at 350 per 100,000 live births in 200819. In the same year, Ghanaian national health insurance was made free for all pregnant women, covering all costs associated with pregnancy and delivery, although informal costs may persist20. In 2010, we conducted a health facility assessment (HFA) at all 86 health facilities in the study area. We interviewed the staff member most qualified in maternity care about provision of essential interventions (signal functions) and on staffing, checked competence using clinical vignettes and verified availability of drugs and equipment (tracer items). Quality of routine delivery care, emergency obstetric care (EmOC) and emergency newborn care (EmNC) were each categorized into five levels, combining reported performance of signal functions, tracer items, and minimum requirements on numbers of skilled staff employed at the facility7. The health facility assessment identified 64 facilities offering delivery care in the study area: 11 hospitals, 11 maternity homes, 34 health centers and 8 clinics7. The facility type “clinics” comprises clinics, ‘community-based health planning and service’ (CHPS) compounds21 and health posts. More than half of the 64 facilities (53%, n = 34) were found to provide “good quality” routine care (defined as facilities classified as high or highest on the quality assessment), while less than 20% (n = 12) provided basic or comprehensive emergency obstetric care (BEmOC or CEmOC), and only 8% (n = 5) provided basic or comprehensive emergency newborn care (BEmNC or CEmNC, Table 1). *Good routine care = high or highest on the quality of care categorization, i.e. providing ≥8/12 signal functions; ≥3 health professionals conducting deliveries (≥1 of whom midwives) employed at the facility. **EmOC = basic or comprehensive emergency obstetric care i.e. providing all 6 basic signal functions (or all except assisted vaginal delivery); ≥3 health professionals (≥1 of whom is a midwife) managing obstetric complications and ≥1 present during the visit. ***EmNC = basic or comprehensive emergency newborn care i.e. providing ≥5 signal functions (or all except dexamethasone to mother for premature labour); ≥3 health professionals managing sick newborns and ≥1 present during the visit. See ref. 7 for details on the quality classification. A geospatial database of the study area was created in ArcMap (ESRI California) mapping all health facilities, roads and villages where pregnant women lived. Distances between village centroids and health facilities were calculated using several different methods, including straight-line distance, road network distance and raster least cost paths, which incorporated topographical barriers22. Straight-line distance from the woman’s village to the closest delivery facility proved to be an adequate proxy for potential spatial access to delivery care in this context22; therefore, the average village-level distance to the closest health facility was used for all women in the same village in this analysis. Analyses were conducted in Stata version 12.0, using multilevel logistic regression (xtmelogit command), with the lowest level of analysis being the delivery, counting multiple births (twins and triplets) as one delivery, and random intercepts at the village level. Health facility catchment area was considered as an alternative second level, but results were similar and village level accounted for more of the variation in facility use between women. The exposures of interest were straight-line distance to closest facility, facility type and quality of care; the outcome of interest was delivery in a health facility of any type (hospital, clinic, health center, or maternity home). We included the following potential confounders in multivariable models: age, religion, marital status, parity, ethnicity, occupation, wealth quintile, education, multiple birth, previous stillbirth, health insurance and Newhints intervention vs. control group. Distance to the closest facility was modelled both as a categorical and as a continuous variable using a square-root transformation to approximate a linear association with the log-odds of facility delivery (linearity was checked using lowess plots and fractional polynomials). Health facility type and quality of care were evaluated by adding a categorical variable for the type or quality of care offered at the closest facility to the models including distance as a continuous variable. We also calculated marginal predicted probabilities of facility delivery using the margins and associated marginsplot post-estimation commands in Stata. Distance and quality were also modelled in several alternative ways (see supplementary files): distance to the closest facility of a certain quality level (using categorization as described above) and the highest quality facility within a certain distance. Furthermore, two alternative quality measures were evaluated: a simple score counting one point per signal function, per doctor conducting cesarean sections (up to 3) and per health professional (up to 3) at each facility (total maximum 32 points), and health worker competence as evaluated with two clinical vignettes (total maximum 20 points, for details see23). This study uses data collected for the Newhints trial, which was approved by the ethics committees of the Ghana Health Service, Kintampo Health Research Center and the London School of Hygiene and Tropical Medicine (LSHTM) (trial registration number http://ClinicalTrials.gov: {“type”:”clinical-trial”,”attrs”:{“text”:”NCT00623337″,”term_id”:”NCT00623337″}}NCT00623337)18. The additional analyses were approved by these committees.