Geospatial analysis and decision support for Health Services Planning in Uganda

Shuaib Lwasa, PhD, MA, MSc, BA International Potato Center/Urban Harvest s.lwasa@cgiar.org; lwasa_s@arts.mak.ac.ug Tel: 256 41 4287571, 256 772461727 Abstract The role of geospatial technologies in planning and management of location-based services has been underscored by the several studies related social services provision. As the growth and expansion of geospatial techniques continues to surge, the utilization of spatial information in decision making for social services has intensified among different users including governments, corporations, research institutions and private sector with the adoption of geospatial tools to support decisions making processes. In Uganda the utilization of geo-spatial techniques in provision of health services has gained momentum after a complete survey of health units and development of a national geodatabase. Planning for the provision of health infrastructure services requires quality information to rationalize the location and allocation of services in relation to population. Health service planners are always faced with a question of where services are located in relation to need and how such distribution would affect the allocation of resources to meet the requirements of the population. Because resources are scarce, prioritization is indispensable and such analysis of needs as well as gaps becomes important in planning. This paper analyzes access to health facilities using the population gridding approach with geospatial techniques in relation to location of health infrastructure facilities. The paper also illustrates the invaluable potential of geospatial analysis in supporting decision making. 1.0 Introduction Health infrastructure in Uganda refers to the physical structure and supporting equipment established for provision of health services. It usually involves a structure with facilities for different health service needs, equipment such as cold-chain facilities for storage, management and use in the provision of health services to the population. Health systems in Uganda are increasingly facing challenges of ensuring health care provision to the growing population that is also disproportionately spatially distributed. The challenges are due to a number of factors including; population growth, uneven population distribution related to natural resource bases, transportation networks which have made some areas remote, human resource and the financial requirements for managing and provision of health services. Health service provision is one of the many basic locationbased social services to be provided in-tandem with the spatial distribution of the population. Planning for the provision of health infrastructure therefore requires quality information on location of services, capacity of facilities and catchment population. Although a countrys population requires health infrastructure and health care within a given proximity, because of resource constraints, prioritization is indispensable. Prioritization of health service location would therefore consider several factors but importantly are needs and gaps. According to the National Health Policy (MoH 2002), health infrastructures are to be established in various administrative units within 2 km reach by population in rural as well as urban areas. The Uganda Bureau of Statistics (UBOS 2002) conducted a socio-economic survey on health service access and the result indicates health facility access is 69.6% by the rural population, 95.8% by the urban population and national average of 73.2% within 5 km radius. This was considered as an

improvement from a staggering proportion of the population within 10 km in 1991. This paper analyzes access to health facilities using the population gridding approach and combining it with location of health infrastructure facilities. The analysis is intended to enhance the understanding of location-based service analysis and provision of information for planning of health services. 1.1. Study Issue

Location-based services such as health present a research and policy challenge in dealing with the spatial distribution and relation of such services to other variables including population. Geospatial tools have been applied in the analysis of information much needed in decision making and data analysis can range from visualization, exploration through spatial statistics to spatial econometrics (Bivand 1998; Jeong and Gluck 2002; Davis 2003). With the underlying spatial relationships notwithstanding, exploration and visualization of data is very significant for decision making because it clarifies in relatively clear ways where need is or where hotspots and gaps exist. Such information provides a basis for evidence-based planning and management of services. Spatial observations upon which analysis can be undertaken include fields or surfaces, point patterns and lattice observations where attribute values relate to a grid that could be an administrative unit (Bivand 1998). The research question that this study attempts to address is that given the spatial distribution of population and the location of services, how can the relationships between the services, capacity and catchment distance on one hand and population distribution on another, enhance information for decision making regarding the planning of health services? This question is premised on the fact that current services planning mechanisms are largely based on econometric models and could be enhanced if coupled with spatial analytical techniques that would explore the historical explanations of locating services but also provide pointers for policy on what actions can be taken where gaps exist. The study employed a triangulation of spatial techniques including gridding the population and spatial aggregation of health services based on the populated areas and spatial statistics to generate information for decision making. 1.2. Objectives and purpose of the study

The objective of this study was to spatially analyze health facility access by relating location and population. Specifically the study objectives included; 1.3. To apply location-based analyzes of access to health facilities in Uganda To generate spatial information for visualization and support planning and delivery of health infrastructure. To demonstrate the use of geospatial information and techniques in provision of information required for planning and health service delivery Methodology

Geographic Information Science (GIS) techniques are utilized to analyze the spatial distribution and relationships based on health service locations and population. Because population data is administrative based, this study used the Gridding approach to spatiallize population in the country. This technique is more robust than administrativebased summarizes which are not spatially populated. Data was collected from secondary data sources of Population and housing census, spatial datasets from National Forestry Authority (NFA) and the health infrastructure database created by Global Positioning System data capture accomplished between 2000 2003 and continuously updated. The population of Uganda was gridded at a 5 sq km resolution (map 1) spatiallizing national 2

population in areas excluding water bodies (Uwe D. et. Al, 2001) using population density. Grids were generated from county administrative data level since there has been little change in counties than district or lower level administrative units. Additional data utilized in the analysis in terms of grades and type of ownership, catchment distance, bed capacity was collected during the survey. Geostatistical tools of Geoda 9.5i (Anselin 2004) and ArcGIS are utilized to spatially explore the data, visualize and statistically analyze relationships between the key variables of population, number of health facilities, average catchment distance and bed capacity. Data are explored in terms of its randomness locations within the grids. Bivariate statistical analyses were conducted for relationships between variable generating significance maps which are visualized and enhancing understanding access to health by distance and the spatial distribution of the facilities. 2.0 Exploration of Population data Population were spatiallized using the gridding approach (Diechmann and Balk 2001; Davis 2003). This was to establish where people are concentrated using a fairly accurate method. The gridded dataset is utilized to visualize the spatial distribution of population in Uganda before relating it to services provision for decision making. Decision making requires robust data on location of services and population which provides the demand side of services. Exploration of population distribution in Uganda indicates concentration of population in areas around the major water bodies in Uganda of Lake Victoria and Lake Kioga, mountainous areas although there are a few outliers especially in the areas of conflict lying to the north eastern and north of the country. But the exploration also reveals distribution in areas with population less than the mean of the grids as shown in map 2. The implication of this analysis is that location of services needs to be sensitive and responsive to the distribution. A further analysis of the data indicates that 251 grids have population which is lying in the upper outlier implying high concentration of population in relatively small areas which are areas of conflict. This distribution has an implication on services location and planning and is later in the paper utilized to analyze the relationship between population and services to generate information useful for decision making.

Map 1

Map 2

3.0 Health Infrastructure Development in Uganda The Ministry of Health in Uganda has a program for health infrastructure construction and equipping of the health centers across the country of various grades from HC II, HC III, HC IV and Hospitals1. This was in response to the need for improving access to the basic need of health recognizing the inadequacy of health care facilities(UNDP 2005). Health facilities in Uganda are graded based on health services offered and administrative or political unit serviced in addition to the quality of staff associated with each grade. Health Centers II (HC II) are managed by nursing officers and services include; treatment of non-complicated ailments that require clinical services such as Malaria, bed rests for some cases of patients and act as points for near-neighborhood advise on health needs. HC III are managed by a Clinical Officer and services include; maternal health care, a labour ward, non-complicated surgery services, general treatment and services a Sub county. While the HC IV is managed by a qualified Doctor, has a labor ward, surgery theatre, offers general health care services besides supervising all the lower level health centers, provision of outreach programs and guidance. The Hospital on the other hand offers all health care services, can have several qualified Doctors as well as lower level health workers, manages outreach programs and some serve as regional referrals for complicated ailments. These health services tier have been linked to the administrative units and levels in which management responsibility is also shared between local governments (which are districts, Subcounties or parishes) and the Central Government with the lead institution being the Ministry of Health. An implicit assumption in health service provision is that the nearer to the health facility, the more access to health services. According to (UBOS 2005) access to health facilities and therefore services has improved from 49% population coverage in 1992 to 69.9% in 2005, of the population, living within five kilometers of a health service unit but this is based on social surveys conducted during the census which may not accurately provide for the distance factor. The challenge is that distance units have differing interpretations by the many rural communities in Uganda. Thus self-reported data on distances may not be very well represented by data from surveys. Distances to health facilities influence the level of access to services since the transportation of the sick people and time taken can have a bearing on response in case of a health problem. Rural communities are particularly affected because there are still marked variations in access to health facilities both within and between districts. Beyond physical access, many of these health facilities do not provide the full range of essential primary health care services. There is a total of 2314 health facilities country wide for all grades and their distribution by district also shown in map 1 as dots. The distribution of health centers by grade is analyzed in the table below. Health center II distribution appears satisfactory but HC III, IV and hospitals are still requiring especially when compared with the population and services offered.

HC II, Health Center Two serves an administrative unit which is a Parish HC III, Health Center Three serves an administrative unit which is a Subcounty HC IV, Health Center Four also Health Sub district headquarter, serves Constituency Hospital, serves as district or can be a regional referral health unit

HC II
Mean Standard Error Median Mode Standard Deviation Sum

HC III

HC IV

HOSPITAL

Grand Total

23.05454545 Mean 13.89286 Mean 3.117647 Mean 2.224489796 Mean 41.32143 2.732078818 Standard Error 1.73349tandard Error S 0.268105 Standard Error 0.33025617 Standard Error 4.60125 20Median 12Median 3 Median 2 Median 34 20Mode 15Mode 3 Mode 1 Mode 31 20.2616388 Standard Deviation12.97225 Standard Deviation 1.914649 Standard Deviation 2.31179319 Standard Deviation 34.4326 1268Sum 778Sum 159Sum 109Sum 2314

Table 1 showing descriptive statistics of health facilities by grade

3.1.

Decision making in health services planning

Decision making is a process of solving a problem which is said to exist if 'some one is in doubt as to which choice is best to remove his dissatisfaction with his present state'. Such a person or in this context government can identify three aspects related to the choices; first one or more outcomes that he desires; and secondly two or more unequally efficient or effective courses of action and thirdly an environment containing factors that affect the outcomes. In decision making there is an ideal of behaving objectively and rationally in which optimal courses of action are found and relevant information for the decision assumed to be readily available. In practical terms decision makers usually do not have all the relevant information when making decisions because of the time and cost constraints for gathering such information. A decision maker will stop gathering for more information once some information is available on the basis of which a decision can be reached. Rarely does it occur that more information either gathered or analyzed from existing data will lead to yet better decisions. Thus decisions tend to focus on procedures that lead to a solution that may not necessarily be optimal. It turns out that the process usually followed is the procedural rationality in which a course of action, which decision maker thinks of as good enough is taken. Searching for a satisfying alternative rather than an optimal alternative. Information search about a problem and its evaluation in decision making is very critical and when spatial data is considered, the available tools for evaluation provide a wide range of possibilities for manipulation to ensure that existing information can be utilized for better decisions(ILRI and CBS 2002). At national level part of the decision making process is how spatial information can be captured and what methods would be used to evaluate such information. Decision makers usually operate within a tight time frame with inadequate resources and information. They are buffeted by special-interests, bureaucratic imperatives, and political forces whose vision extends no further than the next election cycle. The current health services planning system utilizes population data tied to administrative units with different spatial attributes and thus do not spatialize population for right allocation.

Evaluate outcome

Crisis/ problem

Spatial data
Policy Action Assessment/ Collect Information

The Place of spatial data in Decision making process

Having discussed the importance of decision making and the process itself, it is prudent to highlight its application in health services planning. Like other location-based services, health services are very critically tied to space in which population exist. It is a challenge to provide health services acceptable when population is unevenly distributed. Therefore all possible courses of action in health services location need to be explored but the value of geospatial tools in analyzing and visualizing service gaps and needs in unchallengeable. Therefore visualizing service gaps, needs is a crucial step toward allocation for service delivery. 3.2. Analysis of access to Health Facilities basing on Distance

Physical access to health facilities is critical in health services planning. This is because the distance to a health facility is a significant factor for morbidity. Although literature indicates improvement in health service access (NEMA 2000/01)and there is continued investment in health infrastructure especially in rural areas by the Government through the Ministry of Health, there are still questions regarding health service access in relation to population distribution. Access level in literature was analyzed on a basis of a survey conducted country-wide on services provision in the country but does not adequately portray the access levels accurately. Given the shortcomings of self-reported data and the many interpretations of distance unit in the various languages of the country, the reported statistics needed to be verified with more robust, spatial analysis techniques to identify health facility gaps and need. Thus analysis of coverage as presented in this paper has been enhanced with spatially analyzed access which integrates grided population with location of health facilities.

Based on the population by grid and applying the spatial analytical technique of proximity analysis with a tight distance algorithm, the analysis reveals that an estimated 6.6 million people which is 27.3% of the total is within the 5 km radius to a health facility. This result is based on a tight spatial analysis for all grids (cells) which are completely contained in the 5 km zone. This excludes grids which may be touched by the 5 km zone. A further analysis of access within the grids revealed that there are differences in access as determined by the location of the health facility within a grid. A centrality assessment returned 60.4% proportion of health 4 18 774 cmmore50 00 4.2 km facilities at BT than -50 1755.95 1713 Tm /F2 from the centroid of the grid which denotes a random location of health facilities within the grids. It should be noted that

facilities especially the lower health centers. In terms of functionality, some NGO operated health centers suffer from intermittent operations due to drug unavailability and inadequacy of personnel.

Considering regional access, as shown in map 5 and the grids highlighted in yellow, part Map 3 Map 4 of the central region around Lake Victoria, eastern and south western regions have physical access to health facilities that is better than central, north western and northern regions. The clustering of health facilities around Lake Victoria is possibly due to the high population and connectivity of the region while in northwestern region, the presence of refugees and international organizations such as World Food Program, UN Human Center for Refugees explains the concentration of health facilities in some of the districts.

4.0 Relating population to health services location Map 5

Map 6 7

As noted earlier, the research question that this paper attempts to address is whether the spatial distribution of population and the location of services relate with key variables including bed capacity and catchment distance in trying to generate information vital for decision making regarding the planning of health services. Access of health facilities to the population was also recognized in the national population policy as a key ingredient of sustainable development (1995). Utilizing spatial analysis techniques the key variables of population by grid and number of facilities in a grid were plotted to establish any relationships. The result of the plot as indicated in figure 1 indicated that an increase in population relates with an increase in number of facilities for the grids. A further analysis of the relationship using Morans I statistic indicates a weak but positive relationship between population and number of facilities of 0.1932 with high significance indicated in map 7. The interpretation of this statistic is that population in as much it is an important variable in planning for health facilities, is statistically not significant in the historical location of facilities. On the other hand it reveals the need for consideration of population in locating health services.

Figure 2 Figure 1 The Local index Spatial Autocorrelation as shown in map 7 also indicates the spatial significance of the relationship between population and number of facilities. The results were computed with a high statistical significance of 0.0001 in areas of high population and high number of health facilities. Two important scenarios are discerned from the significance map that is areas with high population and high number of health facilities which according to the map are categorized as high-high and areas with high population but low number of health facilities. Where as the former might depict over allocation, the later reveals under allocation of health facilities which is a gap that needs to be filled by health services planning but utilizing fairly accurate information. In terms of decision making, visualizing the maps such as map 7, communicates to decision makers where improvement is required more readily than if such information was communicated using tables and graphs.

A spatial regression statistical model using Least squares also reveals a positive but weak causal-effect relationship between population and number of facilities. The regression model was applied to estimate the prediction of number of facilities by population. The model results indicate R = 0.067750 at a significance of 0.0000 with a low spatial dependence. The importance of this statistic in decision making is emphasizes the need for consideration of population distribution in health services planning in order to improve access. The number of facilities variable was also tested against average catchment distance. Data on average cathcment distance was taken from recordings about the origin of patients treated by distance of most of the health services. To smoothen the data and remove errors of estimate the range distance was computed for each health facility. Data was then linked to the health facility and spatially joined to be identified with the grid and estimated population of the grid. The model results indicated a weak prediction of average catchment distance of R = 0.000424 at 0.0000 by number of facilities in a grid. The assumption when testing this relationship was that the average catchment distance would increase with increased number of health facilities since services would be available to the population not only within the administrative unit serviced by neighboring units as well. The decision-making implication of this analysis is that even if health facilities are located in given administrative areas, they will not necessarily guarantee access to services to neighboring administrative units. In addition catchment distance was analyzed in relation to bed capacity and the assumption was that with average catchment distance, bed capacity would be predicted. Although results indicated R = 0.092575 at a significance of 0.06466, there is a high degree of dependence between these two variables. These statistical results can be looked at on one hand as satisfying the research needs but have a potential in health services planning because of the revelation of underlying relationships. Bed capacity analysis is crucial in treating complicated ailments such as malaria during which patients may need resting or admission. Although data on admissions would be most appropriate in this analysis, average catchment distance gives an indication of service reach in space and with the population distribution can be utilized in planning for bed capacity. The following section attests to the implications of the spatial analysis in planning for health services in Uganda. 5.0 Implications for health services planning As indicated in the spatial data exploration as well as statistical analyzes and coupled with the importance of public decision making for improvement of health services, this paper discusses the results in the context of planning. Population distribution is a very critical factor in planning location of health services just like other location-based services. Need for health facilities is obvious where population is high and current access distance is higher than the national target of 5 km but where it is above 10 km then such need is very critical. In map 6, the populated grids where need is high show a dispersed population with a facility more than 10 Km and a grid population of > 1500. The total population in greatest need is 2.8 million which is 11.5% of the national population as shown on map 5 in yellow cells. In eastern Uganda the conflict districts, the presence of 9

Internally Displaced Peoples camps creates a great need for health facilities while in central northern, the need is also clearly displayed on the map. In the context of prioritization, health facilities planning needs to consider where the greatest need exists and this is improved with utilization of geospatial techniques in the analysis and visualization of data. Information provided by these techniques can enahance better decision making for the health sector in Uganda. The statistical results show that the location of health facilities was not based on spatial distribution of population. This is because there are areas with high-high and high-low population in relation to number of health facilities. The implication is that spatializing population can greatly improve health services planning because it would be based on needs. The statistical significance maps also show where gaps exist especially grids with high population and low health facilities as well as low population and high health facilities further underscoring the population factor in location of health facilities. Thus some key highlights of the implication are as below; Health facilities have not been located based on spatial distribution of population. This is understandable because allocation is based on population aggregation by administrative units but can be improved with spatial distribution of population. Health services are accessible in space based on distance as recognized by planning authority but the 5 km distance threshold has implication to inadequacy to mortality as the case in remote areas and this is coupled with inadequate ambulance services. More distance tight analyzes are necessary in planning for health facilities. Spatiallizing population by grids although has weaknesses including boundaries provides accurate mechanisms for distributing population and a potential for adequately estimating the needs.

2.0 Conclusion In conclusion, this study demonstrates how geospatial tools and information can aid planning, decision-making and delivery of health services. By combining population and location of facilities in an innovative way using grids, location-based services can be analyzed for gaps and planned better. It is also recognized that there are several means of analyzing access distance is just one of them, while at the same time there are shortcomings of the gridding approach especially the estimation of population which is based on density. Despite these short comings, the geospatial approach can provide valuable information for planning. References (1995). National Population Policy for Sustainable Development. Kampala, Government of Uganda. Anselin, L. (2004). GeoDaTM 0.9.5-i Release Notes, Center for Spatially Integrated Social Science. Bivand, R. (1998). A review of spatial techniques for location studies: 25. Davis, B. (2003). Choosing a method for poverty mapping, FAO.

10

Diechmann, U. and D. Balk (2001). Transforming Population data for Interdisciplinary Usages: From census to grid. New York. ILRI and CBS (2002). Mapping Poverty in Kenya and Uganda'. Nairobi, ILRI. Jeong, W. and M. Gluck (2002). Multimodal bivariate thematic maps with auditory and haptic display. Proceedings of the 2002 International Conference on Auditory Display, Kyoto, Japan, July 2-5. MoH (2002). National Health Policy. Kampala. NEMA (2000/01). State of the Environment Report for Uganda. Kampala. UBOS (2002). Uganda Population and Housing Census. Kampala, Bureau of Statistics. UBOS (2005). Statistical Abstract. Kampala, Bureau of Statistics. UNDP (2005). Human Development Report 2005.

Copyright Notice As a condition for inclusion in the GSDI 10 Conference Proceedings, this work is licensed under the Creative Commons Attribution 3.0 United States License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/us/. Any use of the work in contradiction of the Creative Commons License requires express permission by the authors of the article.

11