Medical geography uses the concepts and techniques of geography to investigate health-related issues. This report aims to make the health services research community more aware of methods and approaches offered by the discipline of geography to the field of health services research. With the computer programs and applications now available, it is possible to create maps and spatial interpretations that are not complex and do not require special data or systems. Data that apply to standard geographic areas can be mapped quickly and included in almost any type of printed report or visual medium.
This report leads the reader through several commonly used methods for producing maps and visual displays that can be used for policy analysis, research, and/or planning purposes. It incorporates demonstrations and interpretations of applications using real-world examples. The applications reflect both the disease ecology and spatial analysis aspects of medical geography.
Print copies of Using Geographic Methods to Understand Health Issues (AHCPR Pub. No. 97-N013) are available from the AHCPR Publications Clearinghouse (800-358-9295).
Overview
Methodology
Rural-Urban Continuum
Demographics and Access
Health Professions and Regionalization
Health Conditions
Conclusion
For Further Study
Technical Notes
Selected Software Manufacturers
Authors, Grant Information, and Citation
Maps are the basic tools that geographers use to present information. Computers have revolutionized mapmaking and placed geography at the forefront of research. Geographic Methods for Health Services Research, the book on which this is based, is the result of a multidisciplinary effort funded by the U.S. Agency for Health Care Policy and Research (AHCPR) to make the health services research community more aware of methods and approaches offered by the discipline of geography to the field of health services research. In a nontechnical format, the authors introduce tools and techniques of medical geography that apply to health services.
Medical geography uses the concepts and techniques of geography to investigate health-related issues. Although some date the field back to the time of Hippocrates almost 2,500 years ago, the term medical geography was first introduced by 18th and 19th century physicians who attempted to understand the relationship between the occurrence of disease and environmental conditions. The most widely used example is John Snow's hand-drawn dot map of cases of cholera (a water-borne disease) concentrated around the Broad Street well. The map, by some accounts, led authorities to remove the pump handle during the 1854 epidemic in London.
This use of medical geography can be thought of as disease ecology, investigating the causes of diseases, and it has played an important role in many epidemiologic and demographic studies. Medical geography also employs spatial analysis, the study of the distribution of discrete variables or indicators.
The three major types of spatial analysis employed to understand geographical variation in disease frequency and/or health status are:
Another focus of medical geography relates to the spatial analysis of population-based measures of access, health care delivery, and resource allocation. These issues are particularly relevant to policymakers, health services researchers, and planners. Examples of these types of studies include:
Effective cartography and map interpretation have been important elements in the evolution of modern medical geography from both the disease ecology and spatial analysis perspectives. The widespread adoption of maps and other geographic displays as analytic and decision-support tools has been fueled recently by the increased availability of software and databases for personal computers. These graphical displays can be powerful tools that reveal underlying relationships or processes. Maps and cartographic analyses are appearing more often in health services research journals and monographs, and their utility and ease of use are demonstrated here.
Atlases and collections of maps relevant to health are now available as reference sources and as guides for policymaking. There are thousands of data sets that can be used with hundreds of commercially available mapping software programs.
This publication leads the reader through several commonly used methods for the production of maps and visual displays that can be used for policy analysis, research, and/or planning purposes. By bridging the gap between the development and use of the tools and techniques of geography for health policy analysis, health services research, and health care planning, this effort incorporates both demonstrations and interpretations of applications using real-world examples. These applications reflect both the disease ecology and spatial analysis aspects of the field of medical geography, depicting the utility of the analytic and summary capabilities available to researchers, planners, and analysts.
Mapping is part of the larger field of cartography (defined as the art, science, and technology of making maps and studying them) and has enjoyed a resurgence in interest as computers have made it possible to create maps and spatial interpretations. These computer programs and applications are not complex, nor do they require special data or systems to produce excellent and informative maps. Given the widespread availability of personal computers and the broad dissemination of data in electronic formats, there is no reason why data that apply to standard geographic areas cannot be mapped quickly and included in almost any type of printed report or visual medium.
Most maps depend on defined boundaries to make them recognizable and meaningful. These boundaries include physical features, such as coastlines or rivers, or political boundaries, such as State and county lines. Although drawing boundaries may seem simple and fundamental to the lay user of maps, it comprises one of the more complex tasks involved in computer mapping. Boundary definitions make up most of the data involved in the creation of a computer-drawn map, and the quality of any map depends on the quality of the boundary data.
Boundary files usually come packaged with a mapping program, but they may be bought separately when special boundaries are of interest or to update changes in boundaries such as ZIP Code areas. The U.S. Bureau of the Census produces and updates a complex boundary set called the TIGER (Topologically Integrated Geographic Encoding and Referencing) file system. TIGER incorporates most physical features of interest, including most streets and roads and a wide array of political boundaries down to the smallest census divisions. The TIGER files are often used as the source of less complex sets of boundaries.
The most common form of cartography involves creating maps that represent spatial relationships as they appear from above the earth. However, certain distortions of the physical relationship of places may more accurately represent the variables of interest and their relationship to human perception. When a map is distorted to compensate for a factor such as population density or travel time, the map is more accurately called a cartogram. (Cartograms are presentations of statistical data in diagrammatic form that resembles or relates to a map.) Many personal computer packages have the capacity to produce cartograms that adjust the size of a place to a particular factor, such as population. These distorted images require careful explanation for accurate interpretation because maps often reflect physical features.
One of the most vexing problems in health care delivery in the United States is how to provide an adequate and equitable level of health care services to rural populations. Rurality is not a fixed and simple concept; some would argue that rural and urban are the two ends of a continuum of patterns of human settlement, with intermediate points along that continuum defined by population density, distance from services, and the type of economic activity prevalent in a place. To represent that continuum, one can use choroplethic maps that divide political jurisdictions by degree of rurality. (In choroplethic maps, appropriate area shadings such as value, texture, and intensity are assigned to regions based on information collected prior to mapping.)
Figure 1 illustrates one such system that classifies counties using a system developed by the U.S. Department of Agriculture. The rural-urban continuum code is based on both county population size and whether a county is a core metropolitan county, some other type of metropolitan county, adjacent to or separated from a metropolitan county. The map shows that the State of Minnesota contains a mix of densely and more sparsely populated areas, with the largest concentration of population centered around Minneapolis/St. Paul in the southeast quadrant of the State and a smaller concentration surrounding Duluth in the northeast quadrant. The variation in the size and population density of the counties illustrates the expected pattern of settlement from the metropolitan centers outward.
Populations included in a county's borders may reflect a range of age, gender, and race distributions. The size of the population that is very young or very old has important implications for the delivery of health services. A county with a large pediatric population will require different specialists and services than will a county that is made up mostly of older people. To characterize the age distribution of a county, one can use the population pyramid.
The population pyramid shown in Figure 2 breaks population into 5-year age increments and divides it by gender, then graphs each increment using horizontal bars. This pyramid, which represents Brown County in Minnesota, shows 5-to-14-year-olds and people in their mid-twenties through thirties predominating, indicating that this county probably has a high proportion of young families raising children. Population pyramids can also be used to examine other demographic variables of interest, such as race.
Often, a problem or condition is widespread, and its most important characteristic is the degree to which it is distributed across regions or concentrated in a single place. Over the past 15 years, the number of rural hospitals that have had to close their doors has been recognized as a significant problem.
Figure 3 provides a picture of the regional nature of the phenomenon using a dot map. In a dot map, point symbols are used to show the spatial distribution of some feature and give readers an idea of its magnitude. The relative concentration of closures in Texas, Appalachia, and along the Mississippi River is apparent. One broader interpretation that can be derived is that hospital closures during this time period were primarily concentrated in the East and Midwest. The dot map, unless combined with some other shading, does not convey information about the denominator (in this case, the total number of hospitals), and these maps must be interpreted carefully.
From a geographical perspective, multiple dimensions of access to care can be investigated among vulnerable subpopulations residing in a given area. For example, groups can be examined based on rural or urban residence, low income/Medicaid coverage, race, ethnicity, age, gender, or transportation availability. These analyses can be compared against some societal norm or objective. However, typically internal comparisons are made that highlight inequalities or patterns that can be visually assessed through map displays. Adjusting for population density is particularly important in conducting these types of cartographic analyses.
Examples of these types of cartographic adjustments included in this section are demographic base maps (Figure 4), isoline maps of population potential (Figure 6), and two maps that relate access and demographics using choroplethic and vector mapping (Figures 5 and 7).
A demographic base map alters the representation of actual spatial units (e.g., counties) so the size of the unit is proportional to the numbers of people or resources living/located within its boundaries. This is an important adjustment when the areal size of selected spatial units varies disproportionately with the settlement patterns (or population density and dispersion) of these areas.
Figure 4 depicts an actual areal map of central North Carolina in the upper right corner, with the demographic base map reflecting the proportional differences in the residential location of the population aged 75 and over for an 11-county Health Service Area. Counties with proportionately smaller land areas (such as Durham County) can become more prominent after adjusting for population density.
Hospital planners are often interested in evaluating service areas and geographic sources of admission for patients. This information can prove useful in planning community outreach or alliances of institutions, as well as conducting competitor analyses. One way to analyze these patterns is to calculate a commitment index, which analyzes the proportion of a hospital's admissions deriving from each discrete region within its service area. Commitment indexes are typically calculated at the county or ZIP Code level using hospital discharge information.
Figure 5 is a choroplethic map that displays the ZIP Code origins for patients admitted during a set period. Shading is used to identify a primary service area from which a set proportion of the patients come—in this case, those ZIP Code areas that contribute the first 60 percent of the hospital's patients. To make this map, values were calculated using a spreadsheet, and the ZIP Code proportions were imported to a mapping software program.
Population potential is a measure of distribution that describes the nearness or accessibility of people to a certain point or place. The general formula for population potential essentially calculates the degree of interaction or the population potential for a point (typically representing an areal centroid or geometric center). This point is equal to the population of the area divided by the distance between the area's center and another point at the center of an area summed over all points in a larger region. Population potential values can be used to measure a population's exposure or potential for the spread of a disease. Population potential computations can be created on a spreadsheet and then imported to a mapping program.
In Figure 6, the population 75 years of age and over in a region is displayed as an isoline map, reflecting equal levels of weighted distance from county centers compared with the region's population. This map could be incorporated into a larger needs assessment in planning for health care resource needs for an elderly population.
A vector map provides an opportunity to depict the direction of movement of a population across a defined area. Vector maps use directional lines or arrows to reflect flow from an origin point to some destination across space. Variations in the width of these arrows can be used to indicate differences in the magnitude of the flow; color or shading schemes can be used to show differences in flow among subpopulations.
Figure 7 is a vector map that shows both the percentage of hospital cancer discharges treated in a given county (depicted as proportional circles) and the percentage of cross-county outmigration of cancer patients for inpatient treatment (depicted as vectors or arrows). The upper portion of the map display depicts cross-county migration to Mecklenburg County, North Carolina, for inpatient cancer care; the bottom portion shows cross-county migration for inpatient cancer care across the remaining counties in the defined area. Percentages are calculated using inpatient claims data from a uniform discharge database; in this case, data from two States are used. Providers and policy analysts can use vector maps like this to plan for regionalized delivery of oncology services.
Assessment by geographers of provider and service availability in an area can make an important contribution to studies of health care access. Geographic analysis techniques can be used to identify inequities in the distribution or to plan for a reasonable mechanism for allocation and/or regionalization of health care resources. These map displays demonstrate three alternatives: the use of choroplethic maps to depict service and provider availability in North Carolina (Figures 8 and 9), the use of map boundaries to reflect geopolitical borders for planning and resource allocation purposes (Figures 10 and 11), and the potential use of geographic techniques in regionalizing health care services across an area (Figure 12).
The spatial distribution of oncologists across the United States, as in North Carolina, is very uneven, with several prominent clusters. This is a possible concern for health planners and policymakers at different levels—State, regional, and local. Maps can provide a visual summary that identifies areas of underservice, which can then be addressed through provider networking and/or policy actions that encourage providers to locate in and serve a particular area.
Figure 8 shows the availability of cancer care at the county level, based on a classification scheme that integrates provider and facility resources for oncology services. In this choroplethic map, more darkly shaded areas reflect a greater intensity of service availability. Superimposed on shaded areas are the numbers of practicing oncologists located in each county. This map illustrates that while some counties appear to be well served, others have poor potential access to oncologists.
In contrast to Figure 8, Figure 9 displays the absolute number of oncologists at the county level. Although this choroplethic map provides useful information about the distribution of providers across the State, it does not fully depict geographic access to cancer care services.
Regionalization of health care resources has not been a leading feature in the organization of health care in the United States. However, in several cases, planning and resource allocation occur in a regional structure. The Department of Veterans Affairs (VA, formerly the Veterans Administration) system of health care delivery is the clearest example of a national system implemented and administered through a regional model.
Figure 10 shows VA medical service areas in North Carolina. Boundaries for these service areas were determined more by expected patterns of use than by established boundaries. This type of map, together with a patient origin map, such as Figure 5, as an overlay, could be used to suggest periodic boundary redefinitions to compensate for shifts in utilization patterns and population migration.
Similar to Figure 10, Figure 11 depicts service areas for the Area Health Education Centers (AHEC) Program in North Carolina. The AHEC Program was established in 1971 to encourage the establishment and support of regionalized training programs to alleviate health professional shortages in rural areas. The AHECs link health care providers to central health sciences centers. In the North Carolina AHEC Program, staff from the central medical schools visit regional AHEC hospitals to provide both training and health care. The AHEC regions are strongly hierarchical and are focused on university medical centers. Program boundaries follow county lines and, in almost all cases, State borders.
Many States constructed Perinatal Care Regions in the 1970s and 1980s to reduce perinatal and maternal mortality and to improve the health prospects of perinatal survivors. The delivery system that evolved was strongly hierarchical, with more difficult cases routinely referred to higher order centers in a three-tiered system. Given a need to minimize distance to these services, a variety of methods can be used to construct regions. One simple technique is the use of Thiessen polygons.
Thiessen polygons are used in Figure 12 to define Level III Obstetrical Delivery Regions in North Carolina. (Level III is the highest service level and includes a neonatal intensive care unit.) Boundary location is a major problem with regional delimitation, and this technique provides a method of roughly anticipating regional boundaries with very little data. It is particularly appropriate when considering regions surrounding a number of nodal centers, Level III nurseries in this example. The 11 hospitals with Level III nurseries are located in 10 communities.
Polygon construction demands some decisionmaking that is best informed by analysts familiar with the travel and practice patterns in the area. It begins with the construction of lines connecting each adjacent or nearby pair of nodal centers. Once the join lines are drawn, perpendicular lines bisecting each join line at its midpoint are added. These perpendiculars are extended until they meet another perpendicular line. The perpendiculars form the Thiessen polygon regional boundaries and thus map the regional pattern. It is often necessary to construct a regional boundary system several times to achieve an acceptable regionalization. Some join lines may be inappropriate because of topographical barriers, cultural or linguistic patterns, or a misreading of the equivalence of the nodes. In constructing regions, it is particularly important to consider features of the physical geography, concerns of spatial contiguity, existing political borders, settlement patterns of the population, existing patterns of utilization, and relevance and logic.
Maps are uniquely able to display incidence data for contagious diseases because physical proximity and human interaction are key elements in their spread. Likewise, many chronic diseases are restricted to certain populations or caused or exacerbated by environmental factors that have a particular physical or geographic distribution. Mapping can graphically track the spread of a disease or be used to search for causal agents in the environment and help understand the origins of a disease or condition.
Acquired immunodeficiency syndrome (AIDS) has, over a brief span of time, become the deadliest contagious disease in the Nation. AIDS began as a primarily urban disease but soon began to make inroads in almost every community in the Nation. Figure 13 depicts the relative incidence of AIDS cases in counties in North Carolina using location quotients to identify high-incidence areas. The location quotient is a ratio of proportions.
The quotient for each county is shown in the top map of Figure 13. To arrive at this quotient, first, the number of cases in the county is divided by the county population. Second, that figure is divided by the number of cases in North Carolina divided by the total population of the State. A location quotient greater than one indicates that a county has more than its share of cases based on its population; less than one indicates a lower share. The bottom map of Figure 13 indicates the probability that the observed quotient is higher or lower than expected due only to chance. It shows that the higher rates in the two pairs of adjacent counties in the center of the State are probably not due to random distribution of cases.
Tuberculosis has reappeared as an important threat to the health of both rural and urban populations. Its incidence and prevalence show marked geographic characteristics.
Figure 14 illustrates a regional pattern of incidence for 1 year using simple population rates to create a choroplethic map. In this map, the divisions of the rates at certain cut points dramatize the regional nature of the disease. The cut points for the shading were arbitrarily chosen based on the distribution of the rates in steps of 10 cases per 100,000. Such thresholds may relate to specific policies or targets set by policymakers, but they are essentially arbitrary.
The modes of transmission of tuberculosis and AIDS are relatively well understood compared with the etiology of cancer. Occasionally, several cancer cases appear in the same general area, giving rise to fears that there is a proximate cause for this "cluster" of cases. Geographic methods are used to determine if the observed cluster is a result of coincidence or random clustering or if there is a probability that there is some specific cause.
Figure 15 illustrates the arbitrary division of a two-county region into synthetic regions to identify high- and low-incidence areas for pancreatic cancer. The synthetic regions described in this map by the dotted lines are based on perceptions of natural areas in the counties and were drawn without prior knowledge of the cancer case locations. They may also be drawn randomly to focus attention on specific parts of an established area, such as a county. The degree of cluster in these synthetic areas can be assessed statistically. In this example, surveillance for any change in rates continued. Five cases occurred in 1990, two in Alleghany and three in Ashe County. Their occurrence had no significant impact on the regional crude rates.
Certain factors in the environment may be overlaid on an incidence map to visually test for relationships between disease and the factor of interest.
Figure 16 includes the locations of power transmission lines (greater than 100 kilovolt), the residences of persons dying of brain cancer during the period 1980-89, and the residences of an equal number of people whose deaths during the same period were caused by a disease that had a similar population rate—in this case, septicemia. The mapping is one part of a test of clustering that also involves analysis of incidence over time and statistical analysis of the relationships between the variables involved. In this instance, the process concluded that the rates were not due to clustering around power transmission lines because their incidence was very much the same as the incidence for a disease that has no apparent relationship to electromagnetic field strength.
This general introduction to the use of geography and cartography in health services research—especially in the field of rural health services—was designed to illustrate how geographic methods can be applied to a wide range of data sets using a variety of commercially available mapping software programs (see Technical Notes).
Most of the methods and techniques introduced here do not demand sophisticated computer software or hardware, nor do they require a substantial background in either geography or statistics. The health care issues selected are mostly common ones dealt with by providers and planners across the rural-urban continuum. We are confident that the user can identify other appropriate applications.