Determining health-care facility catchment areas in Uganda using data on malaria-related visits
Kate Zinszer a, Katia Charland a, Ruth Kigozi b, Grant Dorsey c, Moses R Kamya d & David L Buckeridge a
a. Surveillance Lab, Clinical and Health Informatics Research Group, McGill University, 1040 Pine Avenue West, Montreal, Quebec, Canada H3A 1A3.
b. Uganda Malaria Surveillance Project, Kampala, Uganda.
c. Department of Infectious Diseases, University of California, San Francisco, United States of America.
d. Department of Medicine, Makerere University College of Health Sciences, Kampala, Uganda.
Correspondence to Kate Zinszer (e-mail: firstname.lastname@example.org).
(Submitted: 27 May 2013 – Revised version received: 08 October 2013 – Accepted: 21 October 2013 – Published online: 10 January 2014.)
Bulletin of the World Health Organization 2014;92:178-186. doi: http://dx.doi.org/10.2471/BLT.13.125260
Knowledge of a health-care facility’s catchment area is important for assessing health service utilization, for calculating population-based rates of disease and for performing other important analyses. Different approaches to defining catchment areas have been developed, mostly in the field of health service research.1–5 One simple way of establishing the boundaries of a catchment area is to use distance from the facility – either the straight-line distance, the distance patients have to travel or the distance travelled by patients in a given time.6,7 Under this approach it is assumed that people will visit the closest facility, which implies that distance is the overriding factor influencing attendance. However, distance is only one of many factors that influence the choice of health-care facility; others are the services available and the perceived quality of care.4,6
Another approach, termed the patient-flow method, is based on the proportion of patients visiting or admitted to a health-care facility who come from a particular administrative area, such as a census tract or a postal code area: if the proportion exceeds a set minimum, that administrative area is included in the facility’s catchment area.5,8 With this approach, the catchment area is not limited by the distance between a patient’s residence and the facility. However, an arbitrary threshold is usually imposed on the minimum proportion of patients who must come from a particular area for it to be included in the catchment area. For example, postal code areas that account for less than 1% of admissions to a facility may be excluded from the catchment area.9 Consequently, some individuals who live in an area not considered part of a facility’s catchment area may regularly attend the facility. The likelihood that these minority “users” would be regarded as living outside the catchment area increases with the size of the administrative area. Another limitation is that an area may be excluded from the catchment area even though a large proportion of its population, or even the entire population, uses the facility because the proportion of patients attending the facility from that area does not exceed the minimum.3 In this case, the chance of exclusion increases as the area’s population decreases.
Here we propose a new method for defining the catchment area of a health-care facility that builds and improves on the patient flow approach: the catchment area is defined using a statistical measure – the cumulative case ratio, which is the ratio of the observed to the expected utilization of the health-care facility for a particular condition by patients in an administrative area. We illustrate our method by using data on the utilization of malaria-related services to define the catchment areas of six health-care facilities in Uganda. Then, for each facility, we compare the cumulative rate of confirmed malaria cases in the catchment area derived using this approach with the rate in areas derived using the straight line or road network distance from the facility.
In this analysis we used data on outpatients attending health-care facilities for suspected malaria collected by the Uganda Malaria Surveillance Project, in which a sentinel-site approach to monitoring the malaria burden in the country was adopted. The surveillance programme was implemented in a staggered fashion: it started in July 2006 and the final site opened in August 2008. We selected six sites to represent the diversity of malaria epidemiology in Uganda. They were all government, level-IV health centres that provided care free of charge, including diagnostic testing and medications, as has been described previously,10 and all had the laboratory equipment and trained staff needed for malaria diagnosis, case management and data collection. The data collected for each patient presenting to outpatient clinics included the patients’ demographic characteristics and parish of residence, the results of malaria diagnostic tests, the diagnosis and the treatments prescribed. The parish is the second smallest administrative unit in Uganda and each parish contains 5000 to 6000 inhabitants. A standardized case report form was used and data were entered electronically at each site by a data officer, who was supported by the Uganda Malaria Surveillance Project.
We determined which parishes should be included in the catchment area of each facility using three different parameters: the straight-line distance from the facility, the road network distance from the facility and the cumulative case ratio for malaria-related visits. To derive catchment areas based on straight-line distance, we used distances of 5, 10, 20 and 30 km. These distances were selected because the 2009 Uganda Malaria Indicator Survey11 found that 96% of respondents lived within 9 km of a health-care facility and because the 2009 Uganda National Household Survey12 reported that the average distance of a household from a government hospital was 20 km. The catchment area included all parishes that fell within circles centred on the facility with radii of 5, 10, 20 and 30 km, respectively. To derive catchment areas based on the road network distance, we used road distances of 5, 10, 20 and 30 km along the road networks surrounding each facility. The catchment area included all parishes located within a road distance of 5, 10, 20 or 30 km, respectively, from the facility. In addition, parishes were included if they were located less than 2 km from the nearest road. A parish that did not lie entirely within the distance circle or within the road network distance was included in the catchment area only if over 50% of its surface area lay within the relevant limit. Otherwise, it was excluded.
The cumulative case ratio was defined as the ratio of the observed to the expected number of malaria-related visits to a facility from a parish. Malaria-related visits included all visits between 1 January 2010 and 31 December 2012 by patients who had suspected or confirmed malaria or who tested negative for the disease. We used malaria-related visits because we wanted to include all users of malaria-related services, not only confirmed cases. The expected number of malaria-related visits to a facility from a particular parish was calculated by multiplying the parish’s population by the cumulative case rate for that facility. A parish was included in the catchment area if the upper limit of the 95% confidence interval for the cumulative case ratio for that parish was 1 or greater because a ratio less than 1 indicated that the parish contributed significantly fewer malaria-related visits than expected for its population.
Catchment areas were derived for each of the six sentinel sites using the three parameters and the cumulative case rate for each catchment area, however derived, over a 1-year period was calculated. The numerator was the total number of malaria cases confirmed between 1 January 2010 and 31 December 2012 from all parishes included in the catchment area. The denominator was the total population of all parishes included in the catchment area, which was derived using population estimates from the 2002 Uganda Population and Housing Census. During this period, an average of 98% of all patients with malaria symptoms were tested for malaria: the proportion ranged from 97% to 100% over the six sites.
Catchment areas were plotted on mapping files obtained from the Uganda Bureau of Statistics and the geographical coordinates of all parishes were recorded using zone 35 north of the Universal Transverse Mercator coordinate system. All analyses were performed using R software v2.14.0 and ArcGIS 10 (esri, Redlands, USA).
Fig. 1 displays the cumulative case rate per 1000 population in each parish for all malaria cases confirmed during 2012 at one of the six Uganda Malaria Surveillance Project health-care facilities. The figure also shows the locations of the six facilities and the variation in disease burden and geographical spread. No catchment area definition was applied. It can be seen from the figure that the parishes with the highest cumulative case rates either contained a facility or was adjacent to one. Over 40% of parishes had a cumulative case rate of 1 per 1000 or less in 2012. Fig. 2 shows the catchment area of the Nagongera health-care facility, as determined using the three parameters: straight-line distance, road network distance and cumulative case ratio for malaria-related visits. The largest geographical area was obtained using a straight-line distance of 30 km, whereas the smallest was obtained using the cumulative case ratio. In addition, use of the cumulative case ratio led to the inclusion of noncontiguous parishes. Figures illustrating the corresponding catchment areas for the other five health-care facilities are shown in Appendix A (available at: http://surveillance.mcgill.ca/users/kzinszer/WHObulletin/index.php).
Fig. 1. Cumulative rate of confirmed malaria cases in parishes containing patients who attended six health-care facilities, Uganda, 2012
Fig. 2. The Nagongera health-care facility’s catchment area as determined using the three parameters: straight-line distance,a road network distanceb and cumulative case ratio for malaria-related visits,c Uganda, 2012
As shown in Table 1, the cumulative rate of confirmed malaria cases varied considerably for most sites according to the way in which the catchment area was defined. In particular, the rate decreased with increasing distance from the facility for both straight line and road network distances and, generally, was highest when the catchment area was defined using a distance of 5 km. The largest rates were observed for the catchment area of the Kamwezi health-care facility: 234.9 per 1000 when defined using a straight-line distance of 5 km, 193.1 per 1000 when defined using the cumulative case ratio and 156.1 per 1000 when defined using a road network distance of 5 km. Although the denominator in the cumulative case rate calculation became much larger as distance increased, there was no corresponding increase in the numerator. Hence, generally, the cumulative case rate decreased as the size of the catchment area increased. Table 2 shows the number of parishes included in each catchment area. When catchment areas were defined using a straight-line distance of 30 km, they included an average of 105 parishes; in contrast, when defined using the cumulative case ratio, they included an average of 10 parishes. The catchment area of the Aduku health-care facility did not contain any parishes when defined using a road network distance of 5 km (i.e. less than 50% of each parish’s area lay within the defined distance).
Table 1. Cumulative rate of confirmed malaria cases at six health-care facilities in Uganda,a by catchment area definition, 2012
Table 2. Parishes, population and malaria cases in health-care facility catchment areas, by catchment area definition, Uganda, 2012
The differences observed between different estimates of the cumulative rate of confirmed malaria cases in catchment areas generally occurred because the numerator and denominator in the case rate calculation increased differentially with the distance used to define the catchment area. For example, the catchment area and its population were largest when a straight-line distance of 30 km was used; consequently, the denominator was also large. When a straight-line distance of 5 km was used, the numerator was only slightly smaller but the denominator was much smaller. Clearly the distance between a patient’s residence and the health-care facility was important but doubling the distance did not double the number of cases. Defining a catchment area according to distance from the facility has the advantage of simplicity but this approach does not take into account where patients actually live. Moreover, although distance is important, it is not the only factor influencing a patient’s choice. Use of the cumulative case ratio is not affected by distance since it uses patients’ actual addresses. As Fig. 2 demonstrates, this can result in catchment areas made up of noncontiguous parishes.
The main limitation of our approach follows from the assumption that the reason the number of malaria-related visits from a particular parish was lower than expected was primarily because utilization of the health-care facility by the parish’s population was low. However, lower than expected utilization could have been due to a low incidence of symptoms characteristic of malaria in the parish. If the purpose of defining a catchment area is to estimate a facility’s utilization rate, this assumption may not be of concern since areas of low utilization are excluded. However, if the purpose is to obtain a population-based estimate of disease burden, the assumption may introduce an error and additional data on the background level of the disease in question would be required to assess the potential level of that error. There are also limitations inherent in using health-care utilization data to calculate population-based estimates of disease burden. Although attendance at a health-care facility is influenced by several factors associated with the individual patient, it is also affected by the characteristics of the facility, such as its capacity, which will limit the number of cases that can be seen. In addition, we had no information on “competing” facilities, whose presence may have influenced attendance at the six facilities we studied. Consequently, we may have underestimated the true cumulative case rate of malaria in the catchment areas since our analysis included only cases at Uganda Malaria Surveillance Project facilities.
Accurate identification of a facility’s catchment area is important for: understanding the population served; for planning and evaluating service delivery, including the accessibility of services; and for deriving population-based health indicators, such as disease burden. Conversely, an erroneous view of the catchment area can lead to inefficient and inadequate services, misspecification of the catchment population and potentially flawed decision-making on other facilities, such as deciding where to locate a new facility. Despite its limitations, the cumulative case ratio approach is more likely to produce an accurate estimate of the true catchment area than an approach using the straight line or road network distance because it is based on where patients actually live. In contrast, the distance approaches can lead to the inclusion of parishes where there are no cases of malaria.
The first consideration in using the cumulative case ratio approach to estimate a facility’s catchment area is to choose the basic administrative unit. It is best to use the smallest unit possible. The geographical location of patients must be recorded at the facility level and population data with the same geographical resolution must be obtained. Another consideration is the time period during which the catchment area is defined since catchment areas are affected by factors that vary over time, such as changes in the capacity of or the services provided by the facility or the opening or closing of other health-care facilities nearby. After exploring this issue, our findings (not reported) suggest that data from the most recent 2 or 3 years are sufficient for establishing current catchment areas. The use of a longer period would provide an insight into the stability of the catchment area over time. Catchment areas should be reassessed periodically. A final consideration is whether the catchment area should be based on all admissions to the facility or on the utilization of a particular service.
Our analysis demonstrates how population-based measures of disease burden, such as the malaria case rate, are dependent on the method used to define the catchment areas of health-care facilities. The cumulative case ratio approach to defining catchment areas we propose identified administrative units in which the utilization of a health-care facility was substantially lower than expected, thus enabling those units to be excluded from the facility’s catchment area. Our approach is simple and reproducible and is based on using a statistical measure to decide which administrative units should be included in catchment areas.
Research discussed in this publication has been financed with funds provided by the UK Department For International Financing (DFID) to the International Initiative for Impact Evaluation, Inc. (3ie) and managed through the Global Development Network (GDN), as well as support from the World Bank. The findings, interpretations, and conclusions expressed here are those of the authors and do not necessarily reflect the views of any of the funders, the Board of Executive Directors of the World Bank or the governments they represent.
This study was supported by doctoral awards from the International Development Research Centre in Ottawa, Canada and by the Fonds de recherche Santé, Montréal, Canada.
- Alexandrescu R, O’Brien SJ, Lyons RA, Lecky FE, Trauma Audit and Research Network. A proposed approach in defining population-based rates of major injury from a trauma registry dataset: delineation of hospital catchment areas (I). BMC Health Serv Res 2008; 8: 80 http://dx.doi.org/10.1186/1472-6963-8-80 pmid: 18402693.
- Jones S, Wardlaw J, Crouch S, Carolan M. Modelling catchment areas for secondary care providers: a case study. Health Care Manag Sci 2011; 14: 253-61 http://dx.doi.org/10.1007/s10729-011-9154-y pmid: 21455707.
- Gilmour SJ. Identification of hospital catchment areas using clustering: an example from the NHS. Health Serv Res 2010; 45: 497-513 http://dx.doi.org/10.1111/j.1475-6773.2009.01069.x pmid: 20050933.
- Diesfeld HJ. The definition of the hospital catchment area and its population as a denominator for the evaluation of hospital returns in developing countries. Int J Epidemiol 1973; 2: 47-53 http://dx.doi.org/10.1093/ije/2.1.47 pmid: 4774785.
- Baker LC. Measuring competition in health care markets. Health Serv Res 2001; 36: 223-51 pmid: 11327175.
- Schuurman N, Fiedler RS, Grzybowski SC, Grund D. Defining rational hospital catchments for non-urban areas based on travel-time. Int J Health Geogr 2006; 5: 43 http://dx.doi.org/10.1186/1476-072X-5-43 pmid: 17018146.
- Boscoe FP, Henry KA, Zdeb MS. A nationwide comparison of driving distance versus straight-line distance to hospitals. Prof Geogr 2012; 64: 188-96 http://dx.doi.org/10.1080/00330124.2011.583586.
- Shortt NK, Moore A, Coombes M, Wymer C. Defining regions for locality health care planning: a multidimensional approach. Soc Sci Med 2005; 60: 2715-27 http://dx.doi.org/10.1016/j.socscimed.2004.11.016 pmid: 15820582.
- Garnick DW, Luft HS, Robinson JC, Tetreault J. Appropriate measures of hospital market areas. Health Serv Res 1987; 22(1): 69-89.
- Sserwanga A, Harris JC, Kigozi R, Menon M, Bukirwa H, Gasasira A, et al., et al. Improved malaria case management through the implementation of a health facility-based sentinel site surveillance system in Uganda. PLoS One 2011; 6: e16316 http://dx.doi.org/10.1371/journal.pone.0016316 pmid: 21283815.
- Uganda malaria indicator survey 2009. Kampala: Uganda Bureau of Statistics & Calverton: Macro International, Inc.; 2010.
- Uganda national household survey 2009/2010. Kampala: Uganda Bureau of Statistics; 2010.