Menu
Menu
Menu
Menu
Background: There is substantial evidence that immunization is one of the most significant and cost-effective pillars of preventive and promotive health interventions. Effective childhood immunization coverage is thus essential in stemming persistent childhood illnesses. The third dose of pentavalent vaccine for children is an important indicator for assessing performance of the immunisation programme because it mirrors the completeness of a child’s immunisation schedule. Spatial access to an immunizing health facility, especially in sub-Sahara African (SSA) countries, is a significant determinant of Pentavalent 3 vaccination coverage, as the vaccine is mainly administered during routine immunisation schedules at health facilities. Rural areas and densely populated informal settlements are most affected by poor access to healthcare services. We therefore sought to determine vaccination coverage of Pentavalent 3, estimate the travel time to health facilities offering immunisation services, and explore its effect on immunisation coverage in one of the predominantly rural counties on the coast of Kenya. Methods: We used longitudinal survey data from the health demographic surveillance system implemented in Kaloleni and Rabai Sub-counties in Kenya. To compute the geographical accessibility, we used coordinates of health facilities offering immunisation services, information on land cover, digital elevation models, and road networks of the study area. We then fitted a hierarchical Bayesian multivariable model to explore the effect of travel time on pentavalent vaccine coverage adjusting for confounding factors identified a priori. Results: Overall coverage of pentavalent vaccine was at 77.3%. The median travel time to a health facility was 41 min (IQR = 18–65) and a total of 1266 (28.5%) children lived more than one-hour of travel-time to a health facility. Geographical access to health facilities significantly affected pentavalent vaccination coverage, with travel times of more than one hour being significantly associated with reduced odds of vaccination (AOR = 0.84 (95% CI 0.74 – 0.94). Conclusion: Increased travel time significantly affects immunization in this rural community. Improving road networks, establishing new health centres and/or stepping up health outreach activities that include vaccinations in hard-to-reach areas within the county could improve immunisation coverage. These data may be useful in guiding the local department of health on appropriate location of planned immunization centres.
We utilized longitudinal survey data from the Kaloleni-Rabai Community Health Demographic Surveillance System (KRHDSS) in the coast of Kenya. This system is nested on the local community health infrastructure and that regularly captures demographic and health information and vital status and migration in a local community. It tracks a cohort of more than 92,000 population in over 18,000 households and covers 113 villages in this area. Households of interest in our study were those with children aged between 14 weeks and 11 months. The 113 villages are distributed among 10 Community Health Units (CHU), which are the lowest-level tier in the Kenyan health system structure interfacing the health system on one hand and the households on the other. Three CHUs (Buni, Vishakani, Mwele-Kisurutini) were considered peri-urban as they were adjacent to urban areas or encompassed parts of local rural towns within the sub-counties of interest. The cohort has been followed up semi-annually since 2017 and by 2019, six rounds of data collection had been completed. Longitudinally linked individual level information (using unique identification numbers) was collected during each round. We accessed the data in December 2019 and identified 4,442 eligible children aged 14 weeks to 11 months from the cohort for the purposes of this study. This age-bracket represents the optimal times to assess coverage of pentavalent vaccination as recommended within the Kenya’s Ministry of Health community health data collection guidelines [21], as it is during this period that the three doses of pentavalent vaccine are considered complete. New individuals can enter this cohort by either birth or in-migration, while cohort members can exit by either out-migration or death. A detailed profile of this cohort has been presented elsewhere [22]. For each round of data collection, a trained community health volunteer (CHV) visited the longitudinally tracked households and interviewed the mother or caretaker of the child who provided the following data: vaccination data (based on child’s vaccination card or on maternal recall if card is unavailable), demographic information, reproductive, maternal and child-health data, child orphan status, school attendance among children, social determinants of disease (e.g. insecticide-treated bed-net use, Water, Sanitation and Hygiene (WaSH) practises, access to HIV testing etc.), child nutritional data (MUAC measurements),vital events (births, migration, and deaths) and pentavalent immunization data for all children 14 weeks – 11 months of age. Global positioning system (GPS) coordinates of the households were also collected. A preconfigured open data kit (ODK) installed in electronic tablets was used for data collection, and upon completion of the interview, data were reviewed for completeness and synced to a central server. Further data screening was performed by a data manager for any errors (omissions and inconsistencies) and the feedback sent to CHV for verification. The whole process of data collection was supervised and coordinated by KRHDSS field officers and the local public health personnel. We assembled information on coordinates of health facilities, land cover, digital elevation model, road network, and barriers within Kaloleni and Rabai sub-counties in Kilifi County (Fig. 1) to compute travel time which is a marker of geographical accessibilities of health facilities offering immunization services. Map of Kaloleni-Rabai Subcounties where the Community Health Demographic Surveillance System is implemented. Source of map: generated by the author using open-source software QGIS v3.12 We obtained a list of all facilities that offer immunization services within the study area from the Kenya master health facility list [23] and the Kenya health information system [24]. We merged facilities from these two sources, eliminated duplicates and obtained their GPS coordinates, which we validated against the recently geocoded master database of all health facilities in sub-Saharan Africa [25]. We also included health facilities from the sub-counties neighbouring the study area, with the assumption that the nearest health facility might be in a neighbouring sub-county especially for household along the borders of the study area as shown in Fig. 2. Further, we ensured that the resultant heath facilities were within the settlement and not on waterbodies by checking their coordinates using Google Earth. Map showing the distribution of households with children aged < 11 months in the study area. Source of map: generated by the author using open-source software QGIS v3.12 Data for road networks were assembled from OpenStreetMaps (OSM) and Google Map Maker (GMM). Duplicates and short sections of roads disconnected from the main network were removed. As done elsewhere [26, 27], we classified roads into 4 categories: primary (class A & B) roads that mainly connect international borders, secondary (class C & D) roads that feed into primary roads or connected to major towns, county (class E) roads that feed into secondary roads and connect smaller towns or market centers, and rural (class U) roads that connect rural areas. These roads were assigned different speeds depending on the probable mode of transport as follows: primary and secondary roads whose modes of transport were vehicular were assigned speeds of 80 km/h and 50 km/h, respectively. County roads with bicycling as a mode of transport were assigned 11 km/h, while rural roads were assigned 5 km/h based on similar studies in Kenya [27, 28]. We obtained data for the land cover and digital elevation model (DEM) at a spatial resolution of 30 m from the Regional Centre for Mapping of Resources for Development (RCMRD) [29]. This is the centre responsible for disseminating open geospatial datasets for Eastern and Southern Africa. Land cover for the study area consisted of 9 categories, which we assigned walking speed based on previous studies [27, 28, 30]; tree cover (4 km/h), shrub cover (5 km/h), grassland (5 km/h), cropland (2 km/h), aquatic vegetation (0.01 km/h), sparse vegetation (2 km/h), bare areas (5 km/h), built-up areas (5 km/h), and open water (0.01 km/h). Walking and bicycling speeds were further adjusted accordingly based on the topography derived from the DEM. This correction used Tobler's equation [31] that linked walking and bicycling speeds with the slope of the terrain. Land covers and the DEM showing different elevations of the study area are provided in supplementary file 1. Methods for estimating geographical accessibility have been developed over time, namely, the travel time model [26], network analysis [32], and gravity model [33]. In this study, we used the travel time model because it has been recommended by the WHO as a suitable method of modeling healthcare accessibility [34] and because it takes into consideration other key aspects of accessing care, such as terrain and land cover surfaces [35]. We used AccessMod (version 5) [36] to model geographical accessibility. The software uses the Manhattan distance method to cumulatively determine the time needed to cross contiguous cells using the least cost path from settlement to immunizing health facilities. Therefore, to estimate travel time, we first generated a travel impedance raster surface by merging land cover, elevation, and road network. To each contiguous cell of the resultant raster layer, we assigned travel speeds accordingly as described earlier. Lastly, we combined the location of the immunizing health facilities to the rasterized layer and estimated the time in minutes needed to travel to the nearest facility at 30 m spatial resolution. For further analyses, we extracted the travel time for each household’s geographical coordinates from the generated raster file. The obtained travel time was then assigned to children within a given household. Maps of travel time to the nearest immunizing health facility and the average time per household were plotted in QGIS (version 3.12). In our analyses, we included other factors likely to influence the association between travel time and uptake of pentavalent vaccination, either as confounders or effect modifiers. These included i) location of the household of interest (rural or peri-urban), surrogates of contact with health facilities for services other than for pentavalent immunization, and iii) individual characteristics (e.g. whether the index child was an orphan based on our previous findings from the area [37] and uptake of health behaviors such as use of insecticide treated bed nets, and positive WaSH practices). We used a Bayesian hierarchical logistic regression model to explore the effect of geographical accessibility on pentavalent coverage on the population of 4,442 children aged 14 weeks to 11 months in the cohort. Community Health Units (CHUs) and round of data collection were used as random effects. To stabilize computations, we used weakly informative priors that also served to bind the estimates within the acceptable ranges [38]. We specified four chains each with 5000 iterations, half of which were used to warm the sample and were discarded before estimations were made. The convergence of the model was determined by examining trace plots of the model. We adjusted for confounding due to sociodemographic and other factors described above. In keeping with previous studies investigating the effect of travel time [28], we grouped travel time into two groups: less than 1-h and more than 1-h travel to a health facility. To compare differences between two groups, we used an independent t-test statistical technique, and the results were interpreted using a p-value at the significance level of α = 0.05. The results from the multivariable model were reported as odds ratios (ORs) and 95% credible intervals. Significance of odds ratios was assumed if the 95% credible intervals excluded one. All analyses were performed using R Version 3.4.3.
N/A
