Geographical accessibility in assessing bypassing behaviour for inpatient neonatal care, Bungoma County-Kenya

Background: Neonatal mortality rate in Kenya continues to be unacceptably high. In reducing newborn deaths, inequality in access to care and quality care have been identified as current barriers. Contributing to these barriers are the bypassing behaviour and geographical access which leads to delay in seeking newborn care. This study (i) measured geographical accessibility of inpatient newborn care, and (ii), characterized bypassing behaviour using the geographical accessibility of the inpatient newborn care seekers. Methods: Geographical accessibility to the inpatient newborn units was modelled based on travel time to the units across Bungoma County. Data was then collected from 8 inpatient newborn units and 395 mothers whose newborns were admitted in the units were interviewed. Their spatial residence locations were geo-referenced and were used against the modelled travel time to define bypassing behaviour. Results: Approximately 90% of the sick newborn population have access to nearest newborn units (< 2 h). However, 36% of the mothers bypassed their nearest inpatient newborn facility, with lack of diagnostic services (28%) and distrust of health personnel (37%) being the major determinants for bypassing. Approximately 75% of the care seekers preferred to use the higher tier facilities for both maternal and neonatal care in comparison to sub-county facilities which mostly were bypassed and remained underutilised. Conclusion: Our findings suggest that though majority of the population have access to care, sub-county inpatient newborn facilities have high risk of being bypassed. There is need to improve quality of care in maternal care, to reduce bypassing behaviour and improving neonatal outcome. Bungoma County (coordinates 0.4213°N to 1.1477° N along the latitude and 34.3627° E to 35.0677° E along the longitude) is located in the western region of Kenya, bordering Uganda and covering an area of approximately 2069 km2. The County is administratively divided into nine sub-counties namely Bumula, Kanduyi, Sirisia, Kabuchai, Kimilili, Tongaren, Webuye East, Webuye West and Mt. Elgon. The sub-counties are divided further into forty five county assembly wards, with the county headquarters based in Bungoma town, Kanduyi sub-county (Fig. 1). Bungoma sub counties and the eight newborn units used in the study Bungoma County had an estimated population of 1.7 million in 2017 based on the 2009 National Population Census [30]. The county was purposely selected because of its high NMR of 32 per 1000 live births which is 45% above the national NMR of 22 per 1000 live births [5]. The county is served by 184 health facilities: 12 hospitals, 17 health centres, 102 dispensaries, and 52 clinics [31]. The hospitals are categorised into level IV and level V categories, with level IV facilities being sub-county hospitals which offer basic services and level V facilities representing county level facilities which offer comprehensive services. The health centres, dispensaries and clinics are categorized into lower levels in the health system hierarchy; level III, II and I respectively [32]. In an effort to reduce the high neonatal mortality in the county, eight public hospitals were selected to handle inpatient newborn incidences, the facilities included two county level facilities; Bungoma and Webuye hospitals and six sub-county level hospitals namely; Kimilili, Chwele, Bumula, Mt. Elgon, Naitiri and Sirisia (Fig. ​(Fig.1).1). In these facilities, newborn units (NBUs) were recently renovated, equipped with diagnostic equipment, and their health workers trained on newborn care [33]. Newborn units geographic data: The geographic coordinates of the eight selected inpatient newborn units were obtained from Google Earth [34] and existing public hospitals database [35]. The road network of Bungoma county was assembled from OpenStreetMaps (OSM, 2018), then classified into primary, secondary, county and rural roads [36]. Each road segment was assigned a specific travel time based on the speed limit dependent on the road class, from a study over western Kenya [37] and Kenya National Highway Authority [38]. Land cover map was obtained from Kenya Ministry of Environment [39] at 30 m spatial resolution and was used to demarcate the land use and landcover classification. The land cover had five major classes; forestland, grassland, cropland, wetland and bare land. A digital elevation model (DEM) was obtained from the Shuttle Radar Topography Mission (SRTM) [40] at a spatial resolution of 30 m. Lastly, a gridded surface of live births population of Kenya at 1 km spatial resolution from WorldPop database [41] was resampled to 30 m spatial resolution and used for this analysis. This gridded surface was developed from an integration of land cover data, census data and household survey data using dasymetric techniques. A detailed description of this live births layer is provided in Tatem et al. [42] and WorldPop database [41]. To estimate the inpatient newborn population, the live births population was multiplied by an estimated constant rate of burden of newborn (183/1000) who require inpatient from prior studies [43]. A cross sectional inpatient newborn unit survey was conducted in the eight selected inpatient newborn facilities between May 2018 and August 2018.The purpose of the survey was to identify the care seekers hospital of admission, care seekers awareness on the nearest inpatient newborn units and the bypassing determinants. It is approximated that there are 29,645 births annually in Bungoma County [5], however, only 21% of the newborn are born with complications requiring inpatient newborn services [33], giving a population of 6268 neonates. To estimate the sample population needed for the study, a statistical method for calculating sample population was used (Eqn1 and Eq. 2), having an allowable margin of error of 5% at a confidence interval of 95%. Where Za/2 is the critical value of a normal distribution at a/2 (for a confidence level of 95% critical value is 1.96), MOE is the margin of error, p is sample proportion (50% being a default value) and N is the sample population. The sample size was estimated to be 363. All mothers whose newborn were admitted in the selected NBUs during the survey period were approached for participation until the sample size was reached. The primary question in the survey was “Is this the nearest inpatient neonatal facility to your home?” Other independent variables used in the study were selected based on existing literature regarding their influence in bypassing behaviour on childbirth and newborn care. They included maternal age, maternal education level, occupation, and marital status, mode of transport and household assets. The name of the nearest school and village of residence of the care seeker were also required. See Additional file 1: appendix 1 for the sample questionnaire. We collected the data from the care seekers (mothers) using a structured interviewer questionnaire. The questionnaire was designed in the English language but administered to the respondent either in English or translated to Kiswahili based on care seeker preference. Care seekers (mothers), in the newborn wards, whose newborns had been admitted in the inpatient NBU during this period were eligible to participate in the survey. The ethical approval was granted by the Ethical Review Committee of Mount Kenya University. Subsequent permissions were sought from the Bungoma County Ministry of Health. A written consent was obtained from the care seekers involved in the study after they were informed on the objectives of the study. For mothers who were underage (< 18 years), assent was sought from their guardians to participate in the survey. Participation was on a voluntary basis, and participants were informed of their right to withdraw their participation at any time if and when they desired. A number of techniques have been developed to measure access to healthcare namely: gravity model [44], population provider ratio, travel time model [45] and network analysis [46]. Compared to the other methods, travel time model has been credited to be the most efficient in SSA [47, 48] and also recommended by WHO as it represents the near real world reality in accessing care [49]. Travel time model was selected as it captures and integrate corrections due to different land cover surface and terrain [28, 29, 50], it reflects most probable decision care seekers make, is intuitive and comparable across different countries [29, 51]. A geographical accessibility surface was generated using AccessMod version 5 software [52] this surface reflected the least accumulative distance or path to the nearest inpatient newborn unit [53]. In creating this surface, an initial surface impedance was created by combining data on the road network, land cover data and elevation. These surfaces were rasterized and merged into a single raster layer where travel speed was assigned to each cell at a spatial resolution of 30 m. The travel speeds for each surface and the mode of transport was assigned based on prior studies [37]. Forested areas and wetlands were assigned higher impedance values with low travel speed of 0.01 km/hr. as they acted as barriers, while low impedance values were assigned to the road networks which have high travel speed as shown in Table 1. Travel speed assigned to different land surfaces and their respective mode of transportation The DEM was used to generate slope which was used to calculate different walking and cycling speeds for each degree rise based on Toblers’s equation (Eq. 3) [54]. Where W, is the speed calculated and S is the slope in degrees. The resulting impedance surface was used in a cost distance analysis to create an overall travel time surface to the newborn facilities. This analysis required two inputs; the selected NBU locations and the impedance surface. The cost distance tool calculated the cumulative geographic “cost”, in units of time, required to transverse from each cell in Bungoma County to the grid cell containing the selected inpatient NBU. The population of sick newborns who require inpatient care was then estimated based on aggregated inpatient newborn population at sub-county level by various travel time 30 min, 1 h and 2 h of an inpatient newborn unit using zonal statistics tool of ArcGIS (ESRI Inc.). The nearest school and village names were used to georeferenced the residence location of the care seekers. The georeferenced data were overlaid on the modelled geographical accessibility surface and travel time from each resident home to the nearest hospital was calculated. If the travel time to the nearest facility was less than the travel time to the facility where the newborn was admitted, the care seeker was classified as a bypasser. The field data were analysed using IBM SPSS Statistics Version 20 (SPSS Inc., Chicago, IL, USA). The characteristics of the mothers were analysed and compared between the bypassers and non-bypassers. Frequencies and percentages were used in analysing categorical data, while the mean and standard deviations were used for continuous data. We conducted a Principal Component Analysis (PCA) to generate wealth index quintiles of the care seekers using their household characteristics namely: source of water, type of sanitation used, source of fuel and source of lighting. Then a bivariate logistic regression analysis was carried out to assess the association between bypassing status and bypassing variables, and the results reported using odd ratios (OR), confidence interval (95% CI) and p-values (α = 0.05).