Original research The influence of distance and quality on utilisation of birthing services at health facilities in Eastern Region, Ghana Winfred Dotse-G borgbortsi 1,2 , Duah Dwomoh 3 , Victor Alegana,1,2,4,5 Allan Hill,6 Andrew J Tatem,1,2 Jim Wright1 To cite: Dotse-G borgbortsi W, AbsTrACT Dwomoh D, Alegana V, et al. Objectives Skilled birth attendance is the single most Key questions The influence of distance important intervention to reduce maternal mortality. and quality on utilisation of However, studies have not used routinely collected health What is already known? birthing services at health service birth data at named health facilities to understand ► Skilled attendance at birth is a key intervention in facilities in Eastern Region, reducing maternal mortality. Ghana. BMJ Global Health the influence of distance and quality of care on childbirth service utilisation. Thus, this paper aims to quantify ► Skilled attendants generally supervise births at 2020;4:e002020. doi:10.1136/ the influence of distance and quality of healthcare on health facilities in Ghana.bmjgh-2019-002020 utilisation of birthing services using routine health data in ► Cross- sectional studies using household surveys Eastern Region, Ghana. have found both distance and quality of care affect Handling editor Sanni Yaya Methods We used a spatial interaction model (a model skilled attendance at birth. ► Additional material is that predicts movement from one place to another) published online only. To view What are the new findings?drawing on routine birth data, emergency obstetric care please visit the journal online ► Distance travelled had a more profound influence on (http://d x. doi.o rg/ 10.1 136/ surveys, gridded estimates of number of pregnancies and health facility births than quality of care. bmjgh-2 019- 002020). health facility location. We compared travel distances by ► Most women travelled beyond their nearest health sociodemographic characteristics and mapped movement facility to give birth, sometimes crossing regional patterns. boundaries to do so. Received 24 September 2019 results A kilometre increase in distance significantly Revised 20 December 2019 reduced the prevalence rate of the number of women What do the new findings imply? Accepted 9 January 2020 giving birth in health facilities by 6.7%. Although quality ► Access to skilled birth should be improved by placing care increased the number of women giving birth in health facilities closer to communities and improving health facilities, its association was insignificant. Women the quality of care at existing health facilities. travelled further than expected to give birth at facilities, on ► We can further understand maternal health utilisa- average journeying 4.7 km beyond the nearest facility with tion through routine health management information a recorded birth. Women in rural areas travelled 4 km more systems data analysis. than urban women to reach a hospital. We also observed that 56% of women bypassed the nearest hospital to their community. intervention for the reduction of maternal Conclusion This analysis provides substantial mortality.4 When births occur outside health opportunities for health planners and managers to understand further patterns of skilled birth service facilities in low and middle-i ncome countries, utilisation, and demonstrates the value of routine health they are less likely to have a skilled attendant data. Also, it provides evidence- based information for present. 5 Previously, a skilled attendant was improving maternal health service provision by targeting defined as a qualified health professional specific communities and health facilities. such as a midwife, doctor or nurse specially trained with skills to manage pregnancy, supervise births and care immediately after birth.6 An updated definition of a skilled InTrOduCTIOn birth attendant has a broader team including © Author(s) (or their employer(s)) 2020. Re- use Although the maternal mortality ratio fell by nurses, midwives, anaesthetists and special- permitted under CC BY. 37% globally between 2000 and 2015,1 sub- ised doctors such as obstetricians and paedia- Published by BMJ. Saharan Africa still records 546 deaths per tricians.7 Instead of emphasising job roles, the For numbered affiliations see 100 000 live births.2 While the global maternal updated definition of a skilled birth attendant end of article. mortality ratio decreased by 1.5% per year focuses on actual professional competencies Correspondence to from 1990 to 2015, in sub- Saharan Africa it alongside relevant training of personnel, 3 Winfred Dotse- Gborgbortsi; declined at half this rate. Skilled attendance together with an enabling working environ- w wdg1n15@ soton.a c.u k at childbirth has been identified as the key ment. Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 1 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health When a woman goes into labour or when obstetric facilities is low, the Eastern Region was chosen because its complications occur, reaching the nearest well-e quipped healthcare facilities are largely publicly funded. Finally, health facility in the shortest possible time is essential for the study focused on childbirth in hospitals only, because the prevention of maternal mortality. The decision to use of the lack of electronic birth records in primary health- facility- based maternal health services is affected by the care facilities. geographical distribution of health facilities and by the distance to such facilities.8 There is evidence to support a data link between distance to facility and health outcomes.9 In The study used four secondary data sets, namely HMIS addition, the effects of facility services and infrastructure records of hospital-b ased births; locations of health facil- on attendance are examined here. Quality is assessed ities reporting through the HMIS; a nationally represent- using measures of human resources, infrastructure, 10 ative sample survey of emergency obstetric and newborn equipment, medicines and supplies. care (EmONC) services at health facilities; and gridded Few studies have assessed the influence of both distance estimated number of pregnancies. and the quality of healthcare on the way skilled childbirth The HMIS data were extracted from Ghana’s District attendance services are used in low and middle- income 18 countries.11 12 Health Information Management Systems 2 (DHIMS2) These studies have used data from specific 11 database. DHIMS2 is a database for storing routine research projects such as clinical trials or Demographic and Health Surveys (DHS),13 14 rather than being based health data from patient interactions at health facilities on routinely collected health management informa- in Ghana. They comprised the individual-l evel records of tion systems (HMIS) data. Studies using the DHS have 26 563 routine births at 19 public hospitals in the Eastern examined attendance at any health facility by distance to Region, Ghana, for 2016. In addition to the facility of the nearest facility, but have not examined attendance birth, DHIMS2 recorded mother’s place of residence, at specific, named facilities nor the impact of quality of age, occupation, educational status, health insurance, maternal care services provided on health facility use.15 parity, type of birth and birth outcome. Alongside these The DHS and similar surveys (eg, multiple indicator individual- level data, DHIMS2 records separately the cluster survey) do not record facility names and such monthly aggregate number of women giving birth in analyses are further complicated by the random displace- health facilities also used here. ment of household cluster locations for data protection Healthcare quality at these facilities was assessed based purposes.16 on a nationally representative sample survey of EmONC Therefore, the aim of this study was to use routinely services at 124 health facilities in Eastern Region 19 collected childbirth data from an HMIS for hospitals (including 22 hospitals) in 2010. In the Eastern Region, in Eastern Region, Ghana, to determine the influence health facilities that recorded at least five births each of distance and the quality of healthcare on the use of month in 2009 were included in the EmONC survey. birthing services. In addition, we analysed the charac- EmONC data on just two hospitals (Seventh-d ay Adven- teristics of women who bypassed their nearest health tist and St Joseph) were not available. The EmONC assess- facility in order to attend an alternative centre. The study ment collected data on infrastructure, human resources, thus seeks to develop methods for making fuller use of drugs, equipment, supplies, signal functions and other routinely collected childbirth data for policy and inter- essential maternal and newborn health variables to quan- ventions. To our knowledge, the study also produced the tify indicators that reflect the state of EmONC as recom- 20 first visualisation of childbirth- related patient flows to mended by the WHO. facilities at a subnational level in Africa. The spatial distribution of potential demand for obstetric care was quantified using a gridded (100×100 m) map layer of estimated pregnancy for 2015.21 22 The MeTHOds WorldPop programme generated these estimates by A cross- sectional study design was used to examine the redistributing census- derived number of pregnancies for number of women using public secondary healthcare electoral areas down to grid cells based on land cover and facilities among women expected to give birth in the settlement extents derived from satellite imagery. Eastern Region of Ghana from January to December A 5 km buffer zone was created around the Eastern 2016. Less than 1% of births occurring in Eastern Region to partially account for usage of the region’s Region’s health facilities do not have a skilled attendant, healthcare facilities by residents of neighbouring while conversely, skilled attendance of home births in the regions. Following preliminary exploration of HMIS region is very rare. Our study therefore uses birthing at a data, the Ga East and West districts from the Greater facility as a proxy for skilled birth attendance.17 Accra region (figure 1) were added to the model due to the large patient inflows from these districts. The selection of sites demand for services from communities in the buffer The Eastern Region was selected because there were zone was included but not their health facilities, since we spatial data on location of health facilities and commu- had no access to the data on births in the neighbouring nities. In addition, since HMIS coverage of private health regions. 2 Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health Figure 1 Flows from communities to health facilities; expected flow of women determined by assigning each woman to the nearest health facility that reported a birth; flow line width and colour intensity depicts frequency of flows whereas length shows distance. CHPS, community health planning and services. Rural/urban analysis of results was achieved by over- more sophisticated distance metrics, such as drive times, laying a map layer of mothers’ places of residence on an in settings without significant physical barriers to move- urban extent map layer derived from satellite imagery.23 ment,24 straight- line distance was calculated between each mother’s geocoded community of residence and the Locating mothers’ places of residence Geocoding, a technique for converting place names to hospital where the childbirth took place. For comparison map coordinates, was used to locate mothers’ places of with these observed distances travelled to give birth, two residence for subsequent mapping and analysis. Mothers’ expected distances travelled were calculated, assuming places of residence in HMIS records were geocoded by that facility choice was affected only by geography. The manually matching settlement names where women lived first expected distance was calculated to the nearest with locational data sets from the Centre for Geograph- hospital from each mother’s community of residence, ical Information Systems and Remote Sensing-U niversity while the second was calculated to the nearest primary of Ghana, Google Maps and a Global Positioning System or secondary healthcare facility that had recorded a survey of settlements conducted by the Ghana Health birth. For the expected travel to any facility (primary or Services. There were no differences in sociodemographic secondary) providing birthing services, 323 health facil- characteristics between women whose place of residence ities comprising 134 (41%) community health planning was successfully geocoded and those without location and services, 26 (8%) clinics, 110 (34%) health centres, information (see online supplementary appendix A). 37 (11%) hospitals and 16 (5%) maternity homes were Individual-level analysis of facility attendance for childbirth used as destination health facilities for the second set of The completeness of individual-l evel records at hospi- expected distances, since they recorded births. Then, a tals was assessed by comparing individual and aggre- non- parametric Kruskal-W allis test was used to compare gate numbers of women in DHIMS2. Since straight-l ine the differences in distance (observed minus expected) by distance has been shown to be highly correlated with sociodemographic characteristics. Dotse- Gborgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 3 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health Population-level analysis and flow visualisation of facility Table 1 Results from the spatial interaction model attendance for childbirth predicting births in 19 hospitals in Eastern Region, Ghana, To visualise the mothers’ travel from place of residence in 2016 to destination health facility, three maps were developed, Prevalence ratio (95% two illustrating expected flows of mothers to facilities Predictors CI) P value and the other showing observed flows. The first expected travel was to any health facility providing birthing services, Distance (km) 0.935 (0.930, 0.940) <0.001 and the second one to hospitals only. Patient flow lines Per cent quality score 1.01 (0.999, 1.02) 0.069 linking communities and facilities were generated in Number of estimated 1.002 (1.001, 1.002) <0.001 QGIS (V.3.4) and mapped in ArcGIS (V.10.5). Flows less pregnancies than five women were excluded to improve clarity.25 Per cent completeness 1.02 (1.016, 1.024) <0.001 of birth data spatial interaction modelling Number of inpatient 1.002 (1.001, 1.004) 0.001 A spatial interaction model predicts the flow of people beds between an origin and a destination based on destina- Model inflation distance 1.08 (1.074, 1.087) <0.001 tion attractiveness, demand for services at the origin and distance between origin and destination. This study imple- mented a spatial interaction model to predict recorded Patient and public involvement patient flows from communities to facilities, since such This research was done without patient involvement. models quantify how distance affects healthcare facility Patients were not invited to comment on the study attendance and provide goodness-o f- fit diagnostics of design and were not consulted to develop patient- model performance.26–28 The number of women travel- relevant outcomes or interpret the results. Patients were ling from a given origin community to a given hospital not invited to contribute to the writing or editing of this was modelled as a function of the number of pregnan- document for readability or accuracy. cies in the community the woman came from, healthcare quality at the destination hospital, the number of beds resuLTs as a proxy for hospital size and the distance between the Individual-level analysis of facility attendance for childbirth community and hospital, expressed as an exponential Figure 2 shows the proportion of individual records docu- distance decay function. mented via the HMIS for 35 secondary care facilities, A summary quality of healthcare index was created compared with the number of women documented in from the EmONC facility survey data. This bespoke aggregate routine reports. The 26 563 individual records quality index was based on six categories of variables received were incomplete, comprising only 63.4% of representing physical size, availability of medicines, aggregate reported number of women giving birth at medical and non- medical supplies, infection preven- hospitals in 2016. Individual-l evel records were almost tion and EmONC signal functions. The categories were incomplete or totally non- existent in 16 out of 35 hospital selected based on Hulton and colleagues’ framework for facilities, whereas individual records exceeded aggregate assessing quality of maternal care using two dimensions, number of women in five hospitals (above dotted line in namely experience and provision of care.29 The catego- figure 2). Most of the health facilities without individual ries were range standardised and weighted to achieve a records were either private health facilities or faith- based summary index. Higher weights were assigned to those hospitals and returned fewer aggregate birth counts. The categories (physical size and non-m edical supplies) more 16 hospitals with incomplete records plus two hospitals apparent to women giving birth. The variables in each (Koforidua Seventh-d ay Adventist and St Joseph) lacking category and weights are provided in online supplemen- EmONC service quality data were excluded from subse- tary appendix B. quent population analysis, leaving 17 hospitals (see To estimate the demand for maternal healthcare, esti- online supplementary appendix C for flow diagram on mated pregnancies from the WorldPop gridded map excluded data). layer were summed for zones representing women’s Although there were non-i dentifiable, ambiguous or communities of residence. To control for varying incom- unavailable locations for some patients, the places of pleteness of individual- level records per facility, the ratio residence of 23 246 (87.5%) women were successfully of individual- level to aggregate reported births (from geocoded to communities of origin and after further HMIS) was also included in the model. cleaning, 21 856 (82.3%) were included in the individual- Four different models (Poisson, negative binomial, level analysis. Table 2 summarises the characteristics of zero- inflated Poisson and zero-i nflated negative binomial these 21 856 women who gave birth in health facilities. models) were fitted to quantify the effect of the predictors More women (81.9%) were between 20 and 40 years and on number of facility births (table 1). The performance resided mainly in rural areas. The majority were educated of these models was evaluated based on the Akaike infor- up to junior high school level (56.2%) and were either mation criterion. trading or farming (42.6%), whereas 5.6% were still in 4 Dotse- Gborgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health 2000 women. Although Nsawam Hospital had a low bed complement (proxy for physical size), it had the largest number of women giving birth. There was more varia- bility among comprehensive facilities compared with the partially designated hospitals. There is a clear difference in utilisation patterns, as shown in figures 1, 4 and 5. Women would have trav- elled shorter distances if they had attended the nearest health facility or nearest hospital. The observed journeys show that women travelled farther than expected to use a birthing service. As expected, most of the larger flows were closer to the hospitals. The Regional, Nsawam and Holy Family Hospitals located at Koforidua, Nsawam and Nkawkaw, respectively, stand out as the most used, irre- spective of distanceThere were large flows from Greater Figure 2 Completeness of health management information systems (HMIS) individual birth records for 35 hospitals Accra region to Nsawam Hospital in the Eastern Region. in Eastern Region, Ghana, during 2016. Hospitals below The results from the zero- inflated negative binomial the dotted line returned less individual records compared model showed that, for each kilometre increase in distance with aggregates and more than 100% individual records in from a health facility, the prevalence ratio of the number hospitals above the line. of women giving birth in a health facility between origin and destination pairs decreases by 6.7% (95% CI 6 to 7; school. Almost all women had health insurance (99%), p<0.001). The inflate coefficient for distance suggests but there were 100 more uninsured women in rural dwell- that for each unit increase in distance from facility, the ings relative to urban settings. Most women gave birth by prevalence rate of an inflated zero increases by 8% (95% spontaneous vaginal births and were discharged without CI 7.4 to 8.4; p<0.001). For each additional expected complications. Generally, women travelled 5.73 km to pregnancy, the prevalence ratio of the number of women a hospital, but rural women travelled farther (7.53 km) giving birth in a health facility increases by 0.2% (95% than urban women (1.04 km). CI 0.1 to 0.2; p<0.001). Unexpectedly, quality of care was The majority of women bypassed the nearest hospital marginally insignificant (p=0.069). The effect of other (56%) or the nearest health facility (76%). More than facility birth predictors can be found in table 1. Other half of the women bypassed both the nearest hospital model results from initial exploration are presented as and primary health facility (54.2%). Bypassing the online supplementary appendix D. nearest primary health facility was more prevalent than The effect of distance on skilled birth service utilisation among hospitals. Table 3 reflects the way mothers bypass was assessed using the modelled distance decay curve in secondary and primary health facilities providing birthing figure 6. The exponential effect of distance on skilled services by comparing the observed travel to expected births is seen as the number of women decreased rapidly distances at the health facility tiers. The median expected within the first 10 km and a less steep slope thereafter. travel distances to secondary and primary facilities were There was a more rapid decrease in distance for urban 3.3 km (IQR: 7.1) and 1 km (IQR: 2.7), respectively, but women compared with their rural counterparts. The the observed was 5.7 km (IQR: 10.6). This shows that slopes for both groups were similar between 5 and 12 km, women were travelling longer distances and bypassing then after that urban flows decreased marginally faster. closer health facilities. Older women travelled longer The rug plot indicates fewer travels beyond 25 km. distances than expected to hospitals (p<0.001). Educa- tion, occupation and parity had significant differences among the various groups of women in travel to both dIsCussIOn tiers of health facilities (p<0.001). In contrast, there were This paper presents the first geovisualisation of skilled no significant differences for health insurance, outcome birth interaction between named hospitals, and commu- of birth and whether a woman had a caesarean section nities, using routine HMIS data. The flow visualisation or not. reveals the previously unknown true catchment extent of some hospitals and provides significant opportunities Population-level analysis and flow visualisation of facility to improve accessibility to quality maternal healthcare attendance for childbirth services. Likewise, health managers can plan and imple- Figure 3 highlights the relationship between births and ment maternal health interventions to meet observed healthcare quality. The derived quality index ranged demand beyond administrative boundaries. As HMIS from 43% to 87%. The Eastern regional hospital had the data availability, completeness and quality improve, data highest quality. Most hospitals with higher number of sets such as the one used in this study should enable women had comprehensive EmONC services, whereas all regularly updated analyses of geographic patterns of hospitals with partial EmONC services recorded less than birthing service utilisation. Further investment in health Dotse- Gborgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 5 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health Table 2 Rural/urban characteristics of women and summary of distance travelled (n=21 856) Characteristics Rural (%) Urban (%) Total (%) Age (years) (n=21 856) 10–20 2710 (16.5) 841 (15.6) 3551 (16.2) 20–30 8653 (52.6) 2901 (53.6) 11 554 (52.9) 30–40 4784 (29.1) 1550 (28.7) 6334 (29.0) 40–50 301 (1.8) 116 (2.1) 417 (1.9) Education (n=21 289) N o formal education 2332 (14.6) 655 (12.3) 2987 (14.0) Primary 1118 (7.0) 937 (17.6) 2055 (9.7) Junior high 9018 (56.5) 2954 (55.5) 11 972 (56.2) Senior high 2495 (15.6) 452 (8.5) 2947 (13.8) Tertiary 1008 (6.3) 320 (6.0) 1328 (6.2) Occupation (n=21 856) Employed 3878 (23.6) 1124 (20.8) 5002 (22.9) T rader/farmer 6815 (41.4) 2496 (46.2) 9311 (42.6) Student 954 (5.8) 270 (5.0) 1224 (5.6) U nemployed 1958 (11.9) 584 (10.8) 2542 (11.6) U nspecified 231 (1.4) 138 (2.6) 369 (1.7) Others 2612 (15.9) 796 (14.7) 3408 (15.6) Health insurance (n=21 856) Non-i nsured 158 (1.0) 51 (0.9) 209 (1.0) Insured 16 290 (99.0) 5357 (99.1) 21 647 (99.0) Outcome of birth (n=21 856) Absconded 5 (0.0) 5 (0.1) 10 (0.0) Died 4 (0.0) 0 (0.0) 4 (0.0) Discharged 16 424 (99.9) 5401 (99.9) 21 825 (99.9) Transferred 13 (0.1) 2 (0.0) 15 (0.1) Unspecified 2 (0.0) 0 (0.0) 2 (0.0) Type of birth (n=21 856) N ormal births and normal births with episiotomy 12 625 (76.8) 4051 (74.9) 16 676 (76.3) C aesarean section and other surgical procedures 3823 (23.2) 1357 (25.1) 5180 (23.7) Parity (n=21 119) Never given birth 3817 (24.0) 1185 (22.6) 5002 (23.7) 1–3 10 081 (63.5) 3499 (66.7) 13 580 (64.3) 4 or more 1978 (12.5) 559 (10.7) 2537 (12.0) Distance travelled (km) Mean (SE) 8.41 (0.05) 4.44 (0.09) 7.47 (0.05) Median (IQR) 7.53 (9.21) 1.04 (4.49) 5.73 (10.59) information management to address challenges such countries.11–14 31 Distance significantly reduced the prev- as incompleteness of records could support precise alence of mothers giving birth in a hospital by 6.7% per decision- making30 through greater reliance on routinely km increase, an effect comparable to the 4.4% found collected data and a lower need for makeshift, externally in rural India.31 Our finding is however lower than funded projects. the 29% reduction in facility- based births reported in The relationship found between distance and health Zambia12 when distance to the facility doubled. The facility births within secondary care supports existing odds of delivery decreased by 61% per km increase literature on the negative influence of distance on in the log of distance, which is broadly comparable skilled childbirth attendance in low and middle-i ncome to a nationwide study in Ghana that reported 54%.14 6 Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health Table 3 Differences in distance travelled calculated by subtracting expected distance to primary and secondary health facilities from the observed distances (n=21 856) E2: Expected Test statistic Test statistic E1: Expected distance distance to nearest (for difference in (for difference in O: Observed distance to nearest secondary facility recording O- E1 by patient O-E 2 by patient travelled (km) care (km) births (km) characteristic) characteristic) Median IQR Median IQR Median IQR Kruskal- Wallis Kruskal-W allis Age (years) <0.001 0.66 10–20 6.31 10.11 4.04 7.85 0.99 2.64 20–30 5.46 10.31 3.16 6.90 0.97 2.71 3 0–40 5.77 11.06 3.16 6.66 0.94 2.84 40–50 6.22 10.49 3.79 7.93 0.93 2.77 Education <0.001 <0.001 N o formal education 5.59 6.59 3.95 6.25 0.84 0.98 P rimary 3.19 9.23 1.84 8.23 0.84 0.85 J unior high 6.22 11.02 3.16 6.91 1.07 3.29 S enior high 5.54 8.50 3.16 5.97 1.20 2.84 Tertiary 4.41 9.18 1.83 5.84 0.72 1.50 Occupation <0.001 <0.001 Employed 5.72 11.15 2.12 5.61 1.05 3.05 Trader/farmer 5.62 11.06 3.16 6.95 0.98 2.94 Student 5.54 8.50 3.63 6.70 1.00 2.31 Unemployed 6.22 10.42 3.63 7.00 1.09 2.84 Unspecified 8.79 15.07 6.02 11.93 0.60 2.91 Others 5.50 7.22 3.97 6.92 0.84 0.96 Health insurance 0.83 0.40 Non-i nsured 4.82 12.43 3.16 8.38 0.98 2.78 I nsured 5.73 10.53 3.34 7.08 0.97 2.73 Outcome of birth 0.04 0.46 Absconded 2.95 3.97 0.84 3.00 0.84 0.86 Died 1.70 19.93 1.69 17.56 0.50 1.52 Discharged 5.73 10.53 3.34 7.11 0.97 2.73 Transferred 7.98 12.38 5.45 9.38 0.64 0.82 Unspecified 14.98 1.82 14.98 1.82 1.51 1.06 Type of birth 0.84 0.18 N ormal births and 5.75 10.01 3.39 7.08 0.98 2.85 normal births with episiotomy Caesarean section 5.62 11.15 3.16 7.13 0.84 2.30 and other surgical procedures Parity <0.001 <0.001 N ever given birth 6.02 9.72 3.21 7.13 0.89 2.66 1 –3 5.45 10.31 3.16 6.67 0.96 2.64 4 or more 6.49 10.68 3.75 7.37 1.05 2.99 However, it was higher than the 38% found in the Brong origin–destination pairs in a spatial interaction model, Ahafo region of Ghana.11 All of these other studies used so the different methods applied could account for survey data, analysed individual events and controlled some of the apparent differences in observed distance for their individual determinants using logistic models decay relative to other studies. Furthermore, the effect that specified childbirth outside a health facility as the of distance on rural/urban women shown in figure 6 reference category. Some studies also log transformed did not show much difference within the first 10 km as their distance measures. In contrast, this study used expected because the spatial data used to classify rural/ Dotse- Gborgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 7 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health place of birth included variables such as availability of water, privacy, electricity, waiting area for family members and functioning patient toilets in addition to other measures such as drugs, level of care, human resource and physical size in the summary index.12 The difference between expected and observed flows could be explained by the influence of distance and quality of care, which makes women prone to bypassing the nearest hospital to use another. Further explora- tion revealed that 76% of the women bypassed the nearest health facility (primary or secondary) providing birthing services, similar to the 75.4% found bypassing the nearest facility in Tanzania, which was explained Figure 3 Number of women giving birth in health facilities by quality of care.32 However, rates of bypassing hospi- in 2016 versus quality index, available beds and emergency tals only were approximately 20% lower. Other reasons obstetric and newborn care (EmONC) status for 17 hospitals in Eastern Region, Ghana (excludes Seventh- day Adventist that could account for bypassing of health facilities are and St Joseph). previous experience in a facility, which influences quality, access to services and mistrust.33 34 Bypassing is due over- urban settlements captured most periurban areas as whelmingly to patient choice, since only 18 cases were rural. transferred between hospitals by health staff. Further Quality of care did not significantly increase the number studies are needed to find the reasons for bypassing in of women giving birth in health facilities, comparable to the region. related studies in other sub- Saharan African countries.11 13 There was a high inter-r egional flow of patients from Since our choice of indicators for the healthcare quality the Ga East and West municipalities in Greater Accra of care in this study shares similarities with a study in rural region to use birthing services at Nsawam Hospital in Zambia, in which quality of care was significant,12 it was Eastern Region. Although the nearest hospitals to these surprising we did not observe such an association. Our areas of Greater Accra in terms of straight- line distance range of indicators for the quality of care index reflected lay within the same region, an examination of drive times characteristics more apparent to women, a feature that within Google Maps suggested that drive times from is essential for observing an association between quality these neighbourhoods to Nsawam Hospital and hospitals of care and increased skilled births.13 The study in rural within Greater Accra were similar. This inter- regional flow Zambia on the influence of distance and level of care on may thus reflect patients avoiding the congested journey Figure 4 Flows from communities to health facilities; expected flow of women determined by assigning each woman to the nearest hospital; flow line width and colour intensity depicts frequency of flows whereas length shows distance. CHPS, community health planning and services. 8 Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health Figure 5 Flows from communities to health facilities; observed flow of women to hospitals; flow line width and colour intensity depicts frequency of flows whereas length shows distance. CHPS, community health planning and services. to attend Accra’s hospitals. More generally, such inter- from both regions. Also, health managers and planners regional flows may affect estimates of regional skilled need to account for the true catchment of health facili- birth attendance rates derived from regional population ties as opposed to planning within administrative bound- counts and HMIS aggregate number of women. The aries not designed for healthcare delivery. attending (denominator) population served by Eastern In applying study quality criteria used in a recent Region’s hospitals would be underestimated, leading systematic review of healthcare attendance literature,35 to an overestimate of skilled birth attendance rates in our study based on HMIS records has several advantages Eastern, while the opposite effect would occur in Greater over the largely community-b ased approaches reported Accra. To appropriately plan for such inter- regional flows, to literature included in the review. First, in contrast there should be a dialogue between health managers to our study’s use of clinical records, community- based approaches generally rely on self- reported facility atten- dance by mothers, with potential for social desirability36 and recall biases.35 Second, in many studies included in previous systematic reviews, distance or travel time to facility is similarly self-r eported, rather than calculated between origin and facility. Finally, in contrast to many of these previous studies, our analysis is based around clearly defined journey endpoints, the locations of secondary care facilities. Studies of completeness of HMIS data in Ghana suggest the proportion of missing data is declining over time, following continued investment and data auditing.37 Thus, while we included a covariate to account for the proportion of missing data in our spatial interaction model, in future years, greater facility reporting may reduce the need for such measures. Figure 6 Effect of distance to facility on the number of women using birthing services based on a zero- inflated The major limitation of this study is potential selection negative binomial regression model. The short lines on the bias arising from gaps in HMIS data. Primary healthcare axes show the distribution of origin–destination points along facilities were excluded from our analysis due to incom- the lines. pleteness of HMIS data, although they report a significant Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 9 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health number of births38 and compete with hospitals. The utilisation in this study could be used to inform the routine birth data were incomplete for some hospitals design of more realistic catchments for hospitals irre- and non- existent for others, which leaves gaps in the visu- spective of district boundaries. Calculation of skilled alisation and analysis. Since more private and faith- based birth rates per facility and district should also consider hospitals lacked individual-l evel HMIS data, such facil- the actual rather than assumed catchment population. ities are under-r epresented in our analysis. This means Given that women cross regional boundaries to attend that our analysis of bypassing is based on a subgroup of facilities for childbirth, there should be inter-r egional hospitals where public facilities predominate and may collaboration to better serve women on the borders of not reflect birthing uptake at all facilities. However, the Eastern and Greater Accra regions. In future, continued 2017 Ghana Maternal Health Survey reported only 6.3% investment in HMIS and the development of an address of births in private health facilities in the Eastern Region, referencing system in Ghana should reduce the main so this is likely to be a small proportion of births.39 Among limitations affecting our analysis, namely issues arising those secondary care facilities that did hold individual- from gaps in healthcare records and incomplete or level records of births, the majority had fewer indi- inaccurate geocoding of patients’ place of residence. vidual records than the aggregate total annual number of women, suggesting record incompleteness. We were Author affiliations unable to assess if systematic differences existed between 1School of Geography and Environmental Science, University of Southampton, women with and without digital records, but anecdotal Southampton, UK 2 evidence suggests data entry backlogs meant births later WorldPop Research Group, School of Geography and Environmental Science, in the year were less likely to be digitised. University of Southampton, Southampton, UK3Department of Biostatistics, School of Public Health, College of Health Sciences, Incomplete geocoding of women’s places of residence University of Ghana, Legon, Accra, Ghana may also potentially lead to a biased subset of birthing 4Population Health Unit, Kenya Medical Research Institute - Wellcome Trust records being included in our analysis. However, there Research Programme P.O. Box 43640-00100, Nairobi, Kenya was no systematic difference in the demographics of 5Faculty of Science and Technology, Lancaster University, Lancaster, UK 6 women at places that were geocoded and those that Social Statistics and Demography, University of Southampton, Southampton, UK were not (online supplementary appendix A). Further- Acknowledgements To Zoe Matthews for her insights and valuable comments more, given the launch of a Ghanaian National Digital on maternal health utilisation and quality of care. To Andy Newing for sharing Address System and associated postal codes in 2017,40 resources and advice on spatial interaction modelling. To the Ghana Health Service in future, loss of spatial data during geocoding should for making their data available for analysis. reduce as this system becomes more widely used. Contributors WDG and JW conceived the study and it was further developed by Furthermore, the survey data used to estimate the AH. AJT and VA contributed to the methods and provided review comments for the initial draft. WDG and DD analysed the data. WDG and JW wrote the initial draft and healthcare quality index were collected in 2010, leaving all authors contributed to subsequent drafts. a 6- year gap between the birth data and the quality Funding Wellcome Trust (grant number: 204613/Z/16/Z) and UK Department index. Due to the aggregate nature of the variables for International Development (DFID). VA is funded through a Wellcome Trust included in a classic spatial interaction model, the Fellowship (number 211208). Part of this work was done during WDG’s time as study could not account for individual confounders a Commonwealth Scholar and currently supported by the Economic and Social Research Council through the South Coast Doctoral Training Partnership (grant such as financial accessibility, education and perceived number ES/P000673/1). need for birthing services; or investigate the individual Map disclaimer The depiction of boundaries on the map(s) in this article do not factors relating to bypassing of health facilities. While imply the expression of any opinion whatsoever on the part of BMJ (or any member there are more realistic measures of proximity, a simple of its group) concerning the legal status of any country, territory, jurisdiction or area Euclidean distance measure was adopted.24 However, or of its authorities. The map(s) are provided without any warranty of any kind, either a sample of origin–destination pairs was used to express or implied. compare travel time, mechanised network distance and Competing interests WDG was employed by the Ghana Health Service from Euclidean distance to evaluate this choice of distance February 2008 to August 2015. He conducted this study during a study leave. All other authors have no competing interests. metric, and this showed a high correlation between all Patient consent for publication Not required. three measures. Finally, we assumed women giving birth in hospitals should have a skilled attendant at birth. ethics approval Ethical approval for this study was received from the University of Southampton (ERGO ID: 26328). Although this might not always be the case in some primary care facilities, this assumption seems plausible Provenance and peer review Not commissioned; externally peer reviewed. given the staff complements in the region’s secondary data availability statement Data are available in a public, open access repository. Data may be obtained from a third party and are not publicly available. care facilities. Data on the number of estimated pregnancies are available from the WorldPop at https://www.w orldpop. org/. Data on women delivering at health facilities in Ghana can be obtained from the Ghana Health Service. COnCLusIOn Open access This is an open access article distributed in accordance with the This study has demonstrated the utility of routine HMIS Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits data in research to improve service provision. It shows others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, that quality of care and distance are important influ- and indication of whether changes were made. See: https:// creativecommons. org/ encers of choice of hospital for childbirth. Patterns of licenses/b y/ 4.0 /. 10 Dotse- Gborgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright. BMJ Global Health OrCId ids 20 Maine D. Monitoring emergency obstetric care: a Handbook. World Winfred Dotse-G borgbortsi http://o rcid. org/0 000-0 001-7 627-1 809 Health Organization, 2009. Duah Dwomoh http://o rcid. org/0 000-0 002-2 726- 9929 21 Tatem AJ, Campbell J, Guerra- Arias M, et al. Mapping for maternal and newborn health: the distributions of women of childbearing age, pregnancies and births. Int J Health Geogr 2014;13:2. 22 James WHM, Tejedor- Garavito N, Hanspal SE, et al. Gridded birth and pregnancy datasets for Africa, Latin America and the Caribbean. Scientific Data 2018;5:180090 https://www. nature. com/ articles/ REFEREnCEs sdata201890# supplementary-i nformation 1 United Nations. Progress towards the sustainable development 23 Schneider A, Friedl MA, Potere D. Mapping global urban areas using goals: report of the Secretary- General 2016. MODIS 500- m data: New methods and datasets based on ‘urban 2 Alkema L, Chou D, Hogan D, et al. Global, regional, and national ecoregions’. Remote Sens Environ 2010;114:1733–46 https://d oi. levels and trends in maternal mortality between 1990 and 2015, with org/ scenario-b ased projections to 2030: a systematic analysis by the 24 Nesbitt RC, Gabrysch S, Laub A, et al. Methods to measure potential un maternal mortality estimation Inter- Agency group. The Lancet spatial access to delivery care in low- and middle- income countries: 2016;387:462–74. a case study in rural Ghana. Int J Health Geogr 2014;13:25. 3 Kassebaum NJ, Barber RM, Bhutta ZA, et al. Global, regional, 25 Rae A. From spatial interaction data to spatial interaction and national levels of maternal mortality, 1990–2015: a systematic information? Geovisualisation and spatial structures of migration analysis for the global burden of disease study 2015. The Lancet from the 2001 UK census. Comput Environ Urban Syst 2016;388:1775–812. 2009;33:161–78. 4 Campbell OMR, Graham WJ. Strategies for reducing 26 Dennett A. Estimating flows between geographical locations:‘get me maternal mortality: getting on with what works. The Lancet started in’spatial interaction modelling. UCL working paper series 2006;368:1284–99. 2012;181:1–24. 5 Houweling TAJ, Ronsmans C, Campbell OMR, et al. Huge poor-r ich 27 Flowerdew R, Aitkin M. A method of fitting the gravity model based inequalities in maternity care: an international comparative study of on the Poisson DISTRIBUTION*. J Reg Sci 1982;22:191–202. maternity and child care in developing countries. Bull World Health 28 Garcia AJ, Pindolia DK, Lopiano KK, et al. Modeling internal Organ 2007;85:745–54. migration flows in sub- Saharan Africa using census microdata. 6 WHO. Making pregnancy safer: the critical role of the skilled Migration Studies 2015;3:89–110. attendant: a joint statement by who, ICM and FIGO. making 29 Hulton LA, Matthews Z, Stones RW. Applying a framework for pregnancy safer: the critical role of the skilled attendant. WHO, ICM assessing the quality of maternal health services in urban India. Soc and FIGO, 2004. Sci Med 2007;64:2083–95. 7 WHO. Defining competent maternal and newborn health 30 Mutale W, Chintu N, Amoroso C, et al. Improving health information professionals. World Health Organisation, 2018. systems for decision making across five sub- Saharan African 8 Gabrysch S, Campbell OMR. Still too far to walk: literature review of countries: implementation strategies from the African health the determinants of delivery service use. BMC Pregnancy Childbirth Initiative. BMC Health Serv Res 2013;13:S9. 2009;9:34. 31 Kumar S, Dansereau EA, Murray CJL. Does distance matter for 9 Okwaraji YB, Edmond KM. Proximity to health services and child institutional delivery in rural India? Appl Econ 2014;46:4091–103. survival in low- and middle- income countries: a systematic review 32 Kanté AM, Exavery A, Phillips JF, et al. Why women bypass front- and meta- analysis. BMJ Open 2012;2:e001196. line health facility services in pursuit of obstetric care provided 10 Hulton L, Matthews Z, Bandali S, et al. Accountability for quality of elsewhere: a case study in three rural districts of Tanzania. Trop Med care: monitoring all aspects of quality across a framework adapted Int Health 2016;21:504–14. for action. Int J Gynaecol Obstet 2016;132:110–6. 33 Kruk ME, Mbaruku G, McCord CW, et al. Bypassing primary care 11 Nesbitt RC, Lohela TJ, Soremekun S, et al. The influence of distance facilities for childbirth: a population- based study in rural Tanzania. and quality of care on place of delivery in rural Ghana. Sci Rep Health Policy Plan 2009;24:279–88. 2016;6. 34 Tappis H, Koblinsky M, Doocy S, et al. Bypassing primary care 12 Gabrysch S, Cousens S, Cox J, et al. The influence of distance facilities for childbirth: findings from a multilevel analysis of skilled and level of care on delivery place in rural Zambia: a study of birth attendance determinants in Afghanistan. J Midwifery Womens linked national data in a geographic information system. PLoS Med Health 2016;61:185–95. 2011;8:e1000394. 35 Wong KLM, Benova L, Campbell OMR. A look back on how far 13 Lohela TJ, Campbell OMR, Gabrysch S. Distance to care, facility to walk: systematic review and meta- analysis of physical access delivery and early neonatal mortality in Malawi and Zambia. PLoS to skilled care for childbirth in sub-S aharan Africa. PLoS One One 2012;7:e52110. 2017;12:e0184432. 14 Masters SH, Burstein R, Amofah G, et al. Travel time to maternity 36 Nyongesa C, Xu X, Hall JJ, et al. Factors influencing choice of skilled care and its effect on utilization in rural Ghana: a multilevel analysis. birth attendance at ANC: evidence from the Kenya demographic Soc Sci Med 2013;93:147–54. health survey. BMC Pregnancy Childbirth 2018;18:88. 15 MEASURE DHS. Methodology - Collecting Geographic Data. 37 Adaletey DL, Jolliffe B, Braa J, et al. Peer- performance review as a Available: https:// dhsprogram. com/ What- We- Do/G PS- Data- strategy for strengthening health in-f ormation systems: a case study Collection. cfm from Ghana. J Health Inform Afr 2014;2. 16 International I. Demographic and health survey sampling and 38 Sakeah E, McCloskey L, Bernstein J, et al. Is there any role for household listing manual. Maryland, USA: ICF International community involvement in the community- based health planning Calverton, 2012. and services skilled delivery program in rural Ghana? BMC Health 17 GSS, GHS, ICF. Ghana demographic and health survey 2014. Serv Res 2014;14:340. Rockville M, USA: USA: GSS, GHS, and ICF International, 2015. 39 GSS, GHS, ICF. Ghana maternal health survey 2017. Accra Ghana: 18 Ghana Health Service. Available: https://d hims.c himgh.o rg/d hims/ GSS, GHS, ICF, 2017. 19 Ghana Health Service. National assessment for emergency obstetric 40 Ghana Post. Unique digital address for all locations across Ghana. and newborn care, Ghana 2011. Available: https://www.g hanapostgps. com/ Dotse-G borgbortsi W, et al. BMJ Global Health 2020;4:e002020. doi:10.1136/bmjgh-2019-002020 11 BMJ Glob Health: first published as 10.1136/bmjgh-2019-002020 on 10 February 2020. Downloaded from http://gh.bmj.com/ on June 24, 2020 by guest. Protected by copyright.