A Review of Spatial Methods in Epidemiology, 2000–2010

Abstract

Understanding the impact of place on health is a key element of epidemiologic investigation, and numerous tools are being employed for analysis of spatial health-related data. This review documents the huge growth in spatial epidemiology, summarizes the tools that have been employed, and provides in-depth discussion of several methods. Relevant research articles for 2000–2010 from seven epidemiology journals were included if the study utilized a spatial analysis method in primary analysis (n = 207). Results summarized frequency of spatial methods and substantive focus; graphs explored trends over time. The most common spatial methods were distance calculations, spatial aggregation, clustering, spatial smoothing and interpolation, and spatial regression. Proximity measures were predominant and were applied primarily to air quality and climate science and resource access studies. The review concludes by noting emerging areas that are likely to be important to future spatial analysis in public health.

INTRODUCTION

Place has long been recognized as an essential component in epidemiological science (118). Over the past 20 years, interest in place has experienced resurgence, facilitated by improved accessibility of computer-based geographic information systems (GIS). Personal computing, improvements in processing speed, and user-friendly applications have placed spatial analysis within reach of a large number of researchers (38, 86). In the late 1990s and early 2000s, at least three books (26, 35, 134) and a number of reviews (23, 36, 45, 87, 103, 105) were published on spatial methods in health research. These publications have advanced the application of spatial methods to public health data and likely contributed to the growth of the field.

This article reviews the past ten years of literature on spatial analysis methods from major epidemiology journals and assesses trends in the research; it discusses the methods applied and research areas that are emerging. Previous reviews covered the history of GIS and functional capabilities for health research (23), spatial analysis approaches, and tools, along with examples from public health research (23, 36, 45, 87, 103, 105). Most reviews discussed visualization or mapping and emphasized the importance of choice of spatial scale, the mismatch between the most relevant areas for disease risk and the data actually available, the modifiable area unit problem, ecologic fallacies, and incomplete data and measurement error. Given the breadth of previous reviews, the primary focus of this review is spatial analysis.

When examining place-based effects as determinants of risks and rates of diseases, epidemiologists have relied primarily on pre-existing discrete georeferenced data to specify exposures or outcomes; for example, census data are frequently utilized for area-level exposure. Although these studies are useful in their own right, their methods remain essentially aspatial. Many studies have examined characteristics that are spatially defined, and many incorporate multilevel data structures; yet most do not consider local relationships between spatially contiguous entities or incorporate spatial autocorrelation into statistical modeling. The interest of this review is not just in discrete spatial geographic units per se, but in the utilization of methods employing spatial aggregation, proximity, clustering, or spatial regression.

METHODS

Journal Selection

This article reviews literature from the seven English-language journals listed in Table 1. We selected journals targeted to epidemiologists, along with one leading public health journal frequently read and cited by epidemiologists (American Journal of Public Health). Because this review aims to inform trends in the general field of epidemiology, journals with a primary focus on spatial analysis/health geographics were not included. For example, Health & Place, International Journal of Health Geographics, Environmental Health Perspectives, and Environmental Research all emphasize geography but do not target epidemiology as their primary discipline and so were excluded. Journals that were not published for the full 2000–2011 period of the review were also excluded (Epidemiologic Perspectives and Innovation and Spatial and Spatio-Temporal Epidemiology), as were journals emphasizing reviews and commentaries (Annual Review of Public Health and Epidemiologic Reviews).

Table 1.

Articles published January 1, 2000 through February 28, 2011, in seven target journals and number (and percent) of articles that utilized a spatial analysis method (see search criteria in text)

Journal name	Stage 1 All citations	Stage 2 After applying broad search terms		Stage 3 Selected for inclusion
	N	N	%	N	%
American Journal of Epidemiology	3,239	1,225	38%	52	2%
American Journal of Public Health	3,442	1,293	38%	30	1%
Annals of Epidemiology	1,228	415	34%	21	2%
Epidemiology	1,177	419	36%	44	4%
European Journal of Epidemiology	1,290	570	44%	11	1%
International Journal of Epidemiology	1,466	712	49%	15	1%
Journal of Epidemiology and Community   Health	1,921	1,007	52%	34	2%
Total	13,763	5,641	41%	207	1%

Search Method

The literature review consisted of three stages (Table 1). In stage 1, all abstracts appearing between 1/1/2000–2/28/2011 were extracted from PubMed into a citation database. In stage 2, citations were limited to those that began with “spa,” “geo,” “neigh,” “area,” “dist,” “proxim,” “map,” “census,” or “place” anywhere in the citation. Approximately 40% of stage 1 citations met stage 2 criteria. Relevant articles lacked key word uniformity, necessitating manual review. In stage 3, articles utilizing a spatial analysis method were identified. Articles were excluded if spatial analysis (a) was a small subcomponent of the study [e.g., see Silhol et al. (116)]; or (b) was not described in the body of the full article [e.g., see Stark Casagrande et al. (120)]; or (c) if geographic data came from predefined discrete spatial data that were not spatially altered by the researchers and no explicitly spatial methods were used in the analysis of the article’s primary findings [for example of an article excluded, see Turrell et al. (130)]. Area-level studies using census data were not included unless the investigators employed a spatial analytic tool (for example, block group smoothing or calculation of distance statistics) or accounted for spatial autocorrelation in regression.

Abstraction Method

Stage 3 abstracts were reviewed and coded using a term list derived from the frequency of words in the abstracts (summative content analysis of words in the abstracts) and phrases based on expert knowledge of spatial methodology (directed content analysis) (55) (see Supplemental Table 1 for the list of topics. Follow the Supplemental Material link from the Annual Reviews home page at http://www.annualreviews.org). Each abstract received multiple spatial analysis codes and substantive codes. The results summarized the frequency of spatial methods classifications, and the graphs explore trends in classes over time.

SUBSTANTIVE TOPICS

Table 1 shows the distribution of articles published by each journal and the proportion selected for the review. Approximately 14,000 articles were published in the 7 selected journals. Among these, 5,641 were reviewed manually for inclusion. One percent (n = 207) of articles utilized a spatial analysis method, and the proportion rose to 4% for articles published in Epidemiology. Selected for inclusion were 193 spatial methods citations (shown in Figures 1 and 2) plus 14 articles that address georeferencing issues (see Supplemental Table 2 for complete list of articles). Figure 1 illustrates the steep increase in the number of articles using spatial methods over the past 10 years. The most common spatial methods or spatial tools employed were calculations of distances (proximity calculation), estimation of summary measures across prespecified geographic areas (aggregation methods), tools used to assess various forms of clustering, spatial smoothing and interpolation methods, and spatial regression. The largest increase during the period examined was in proximity and cluster analyses methods.

Figure 1.

Number of articies using spatial methods, total by year and total for first part of decade and second part of decade (2000–2011).

Figure 2.

Percent of total articles using spatial methods by substantive topic and publication during the first (2000–2005) and second parts of decade (2006–2010) “Resources” category includes health care, food and physical activity environments, facility location analyses etc.; “Other environmental” includes toxic waste, pesticides, electromagnetic fields, etc.

During the decade under review, some shifts occurred in substantive areas utilizing spatial methods. Cancer and air pollution epidemiology have long applied sophisticated spatial methods, and these fields continue to dominate the spatial literature (16, 49, 53, 58, 132). The declines evident in Figure 2 may have resulted from investigators who submitted their work to alternate journals. By contrast, from the first to second half of the decade, studies in resource availability rose fourfold (8% to 32%; Figure 2), spatial methods applied to disparities research more than doubled (6% to 17%), and studies of health-related behaviors increased the use of spatial methods. The spatial patterning of resources by race and/or socioeconomic status was a popular area of study; most studies focused on differential access to health care, to food or alcohol stores, and to health-promoting features of the built environments (8, 12, 60, 79, 88, 98, 117, 139). Spatial methods facilitate this work; however, estimating the contribution of local resources to health disparities requires cross-classification of resources within sociodemographic groups. Because of residential segregation, there is usually very little cross-classification; thus when adjusted for confounders, this work has found generally weak or nondetectable spatial effects for disparities in health conditions/outcomes (48, 76, 92).

Georeferencing

A number of articles focused on georeferencing (the procedure for assigning spatial coordinates by matching addresses to a reference map or “geocoding”). Of 14 georeferencing articles, most were concerned with positional accuracy and repeatability of geocoding procedures. Geocoding accuracy was generally insensitive to urbanized areas at a census tract resolution or larger. However, if accuracy of street address was critical, then geocoding method discrepancies became much larger (75, 112, 135). Several articles discussed novel ways to improve accuracy of geocodes by using telephone exchanges in combination with other address data (111, 133). The articles cautioned that spatial alignment of overlays (94) and the quality of the reference map (41, 140) have received insufficient attention. The articles caution that the ease of using GIS software masks complexities in geocoding and, thus, novice users need to be careful that they are applying best practices.

Spatial Methods

The most common spatial methods or spatial tools employed in the papers reviewed were calculations of distances (proximity calculation), estimation of summary measures across prespecified geographic areas (aggregation methods), assessment of various forms of clustering, spatial smoothing and interpolation methods, and spatial regression.

Spatial Proximity

Measuring distance between an index location (often a residence or workplace) and resources or adverse exposures is relatively simple, and the measurement is frequently used to estimate environmental exposure. In the reviewed articles, air pollution and climate researchers often calculate distance to the nearest major roadway [for example, Allen et al. (2, 54) and Houston et al. (2, 54)], and behavioral researchers assess proximity to resources such as fast-food restaurants and grocery stores (28, 139) (see Supplemental Table 1). The majority of studies incorporating proximity use simple straight line/Euclidean distances (n = 51, or 80% of proximity articles). Straight-line distance can be a poor proxy for estimating access if there are no direct roads or other means of traveling to a particular destination. For this reason, four articles use network street distance or another network path such as path distance to a park (84) or a pharmacy (137). When travel time is a strong impediment, proximity measures have adjusted distance for travel mode [e.g., walking or taking public transportation (125) or road speed limits, type of road surface, or topography (97)].

Aggregation Methods

Aggregating spatial features within a given area is another method frequently used to characterize exposure. Often, a simple summary total or a simple average is computed within a predefined buffer size or an administratively defined unit such as a census tract. Examples are simple summaries of vehicle counts to determine traffic exposure (10, 42, 134) and number of street intersections or other features of the built environment to assess environmental suitability for walking (18, 74, 76). Aggregation can be combined with proximity analysis via distance-weighted averaging such as kernel densities [used in nine of the reviewed studies (see examples in 29, 79, 89, 126)]. Kernel densities assign greater weight to nearby features and thus are often used in resource access studies (10, 134).

A challenge in using aggregation is accounting for variability in the underlying population distribution. For example, when studying resource availability, it may be important to account for population density if the presence of a larger population implies more competition for resources; two approaches can address this. First, the buffer size can vary depending on population density [referred to as adaptive buffers, which are utilized in several studies (12, 21)]. Second, researchers can use a fixed buffer size, with densities weighted by population. For an example of the latter, one article uses kernel densities for physical activity resources weighted by population density (29) because facilities are a fixed quantity (e.g., a gym with 10 treadmills); thus, when demand is high and the resource is in use it is no longer available.

Although buffer size aggregation is a relatively simple tool, complexities can arise. For example, correction factors need to be applied when deriving buffers for partially measured regions (29). In addition, there is often little prior knowledge regarding the relevant buffer size thus necessitating extensive sensitivity analysis for alternate buffer sizes (29, 37).

Cluster Detection Techniques

Used in 36 studies, spatial cluster methods are the most common tool for assessing nonrandom spatial patterns. Many statistical methods have been developed to determine if disease clusters are of sufficient geographic size and concentration to have not occurred by chance (65, 69). Global clustering tests evaluate without pinpointing the specific locations of clusters, whereas local clustering tests specific small-scale clusters and focused clustering assesses clustering around a prefixed point source such as a nuclear installation. These tests have different substantive interpretations and detection methods; the presence of one does not imply the presence of the other (57). Overall, in this literature a total of 35 studies use spatial cluster detection methods. Such methods have had to overcome limitations to statistical power, including the availability of few cases, high variability in the background population density, multiple testing, and size and shape of cluster windows.

Twelve studies tested for global clustering to determine the existence of clustering in a study area without pinpointing specific locations. Tests most frequently used in the selected literature are Diggle and Chetwynd’s bivariate K-function (9, 13, 31), Mantel-Bailar’s test (27), and the Potthoff-Whittinghill method (PW) (1, 63, 85, 100, 109). Global tests used less frequently in this literature are Moran’s I statistic (91), and Cuzick and Edwards’s nearest neighbors and Tango’s maximized excess events tests (MEET) (68, 124). Tango’s MEET generally has the best statistical power, adjusts for multiple testing, and has the added value of being able to evaluate spatial autocorrelation and spatial heterogeneity (119).

Local cluster detection, also called hot spot analysis, can be assessed using Anselin’s local indicator of spatial association (LISA) (4, 47) and the Besag-Newell test (14, 96). Kulldorff’s spatial scan statistic tends to be the preferred local cluster detection method because it can adjust for multiple testing and heterogeneous background population densities, along with other confounding variables, is applicable to both point and aggregated data, and has been adapted to detect noncircular clusters (67, 70, 71). Implemented in SaTScan, the spatial scan statistic is the most common cluster detection method used in the studies identified in this review (at least 11 studies), including a recent extension for cluster detection of survival data (48, 58). Finally, Diggle’s test (30) and Stone’s conditional test (122) evaluate focal clustering and determine whether risk declines from prespecified point sources (85, 96).

Although most investigations of disease clusters are purely spatial in nature, a number of studies explore space-time clustering of cancer and infectious diseases. Seven studies apply a global space-time Knox technique (49, 53, 64, 80–82, 101, 128) and two studies apply Diggle’s global space-time K-function (32, 53, 82). The K-function is a preferred method because it corrects for edge effects and allows for a range of spatial and temporal scales. Kuldorff’s spatial scan was extended for local space-time clustering and was employed to investigate space-time of clustering of childhood mortality in rural Burkina Faso (108) and for sentinel cluster detection of influenza outbreaks in urban areas in southern Japan (95). Other space-time cluster detection tests include the dynamic continuous-area space-time (DYCAST) system, which was used to identify and prospectively monitor high-risk areas for West Nile virus (128); generalized Bayesian maximum entropy (GBME), which is useful when there is considerable data uncertainty (138); and spatial velocity techniques, used to explore the speed of epidemic waves in long-term temporal data such as influenza (24).

Spatial Interpolation and Smoothing Methods

Interpolation and smoothing methods are utilized by epidemiologists in many contexts to improve estimation. Applied to spatial epidemiology, these tools can be used to derive a spatial surface from sampled data points (filling in where data are unobserved) or to smooth across polygons (aggregate data) to create more robust estimates (87). Both approaches use nearby observations or spatially contiguous entities to fill in or otherwise improve spatial estimation.

Spatial interpolation methods include fitting spatial coordinates as penalized splines (8, 131) and as weighted averages for local populations (115). Both methods have low computational requirements and are advantageous when the large-scale trend is strong. Penalized thin-plate splines have the added advantage of being distribution-free and are thereby useful when spatial continuity is complex and prone to mis-specification. Local interpolation methods are preferred in the presence of a sufficient number of observations and small-scale spatial variability. Kriging is one such local interpolation procedure that uses a weighted linear combination of nearby observations to obtain an exact best linear unbiased predictor (46, 59). Four articles used model-based geostatistical kriging to predict disease rates at unmeasured locations and/or produce smoothed rates (43, 44, 99, 107).

Smoothing methods are used frequently to improve the accuracy of death or disease rates for small areas with few observations. However, most smoothing has been aspatial even when applied to geographic data. For example, empirical Bayes estimation has been used to weight small area estimates that have high random variation toward an observed global average derived from all areas. Most public health researchers have weighted toward the global average, which is potentially problematic because it can oversmooth the data and remove informative local variability. Spatial empirical Bayes estimation incorporates information from local, spatially contiguous areas; it goes beyond simply utilizing the overall global mean. Five studies included in this review derive estimates using spatial empirical Bayes; examples include smoothing estimates of deaths due to stomach cancer and stroke across 54 counties in England and Wales (77) and cryptosporidiosis incidence across 163 statistical local areas in Brisbane, Australia, during an 8-year period (56).

Multivariable Spatial Regression

Standard statistical regression models, which assume independence of the observations, are not appropriate for analyzing spatially dependent data (10). Spatial modeling requires iterative assessment of the strength of spatial autocorrelation in raw and adjusted data. If data are spatially autocorrelated and covariate information does not fully account for that pattern, then incorporating spatial dependencies into the modeling is likely a necessity. Although differences between spatial and aspatial models can be sizable (20, 21), there may be low sensitivity to the choice of spatial model. For example, among the reviewed articles, a number of studies evaluate sensitivity to various spatial regression approaches and find negligible differences (8, 15, 102). Nevertheless, notable differences among spatial regression models include their computational complexity, their ability to capture spatial heterogeneity, and their ability to quantify the uncertainty associated with parameter estimates.

Spatial regression, both frequentist and Bayesian, was used in 45 studies to address spatial autocorrelation and/or spatial heterogeneity. This review does not provide details on these complex methods, and the reader is advised to consult standard references for the field (3, 35, 72, 134).

Spatial autoregressive models

Simultaneous autoregressive (SAR) models are frequentist approaches designed to address spatial autocorrelation. They incorporate spatial autocorrelation using neighborhood matrices that specify relationships between neighboring data points. The form of the SAR model depends on whether the spatial autocorrelation processes are thought to derive from omitted variables (motivating a spatially lagged error term) or from the dependent variable (motivating a spatially lagged term for the dependent variable); a mixed SAR model assumes both processes (3). SAR models are commonly used in spatial econometrics, and only three reviewed articles use SAR models (8, 37, 50). A more recently developed SAR local regression technique, geographically weighted regression (GWR) (17, 40), explicitly allows for heterogeneity in the relation between predictor and response variables over space; thus, it can be used to diagnose spatial heterogeneity. Using this tool, one study identified considerable heterogenity in local effect estimates predicting childhood obesity. The study identifies microarea factors that are likely to be most important for preventing childhood obesity, thus this tool can be used to design appropriate microarea interventions (34).

Bayesian regression models

Bayesian regression models provide an alternative to SAR models. One of these models, the Besag York and Molliè (BYM) model, can be used to estimate the effects of potential risk factors related to a disease by including fixed covariates along with the random effects. At least 11 reviewed articles employ the BYM model to estimate local regression coefficients of covariates at the aggregated level (15, 33, 52, 61, 62, 66, 102, 106, 114, 123, 129). Although most of the reviewed articles model a single disease, at least four articles use the BYM model with shared component models or multivariate CAR models for joint spatial modeling of bivariate and multivariate diseases with common risk factors (39, 51, 78, 127). Shared component models were employed to investigate the similarities of the spatial patterns of childhood acute lymphoblastic leukemia and diabetes mellitus type 1 in the United Kingdom (39), incidence of ischemic stroke and acute myocardial infarction in Finland (51), and spatial-temporal similarities of childhood leukemia and diabetes in the United Kingdom (78). In addition, two reviewed articles extend the spatial BYM model for spatial-temporal analysis (63, 113). At least six articles also apply Bayesian geostatistical models for spatial regression of point-referenced data where the spatial correlation random effects are modeled using a multivariate normal distribution, with the covariance matrix defined as a function of the distance between locations (19–21, 43, 110). For example, Chaix and colleagues use Bayesian geostatistical logistics to study neighborhood effects on health care utilization (20) in France and mental disorders in Malmo, Sweden (21).

Spatial regression has been adapted to many complex data structures and methodologies, including Bayesian maximum entropy methods for spatiotemporal analysis (138) and spatial classification and regression tree models (56). However, there are simple methods to incorporate spatial parameters that may be suitable for many research questions and data types. Sometimes referred to as trend surface modeling, coordinates can be entered as polynomials (22) or flexibly modeled using smoothed terms [e.g., LOESS (136)], or a deterministic distribution-free approach can be implemented using thin-plate splines (8, 121). Spatially implicit modeling is another approach that can be applied on its own or in tandem with spatial covariance modeling. At its most basic, a spatially implicit model includes spatially distributed predictors and assumes their sufficiency to account for the spatial autocorrelation in model residuals. At least 11 reviewed studies rely on this technique for some stage of their analyses (see examples in 16, 93, 104). Spatially colocated covariate data, required for this approach, can be critical to any modeling effort. Covariate data can stabilize nonstationarity in the mean, and importantly, examining spatial variation before and after adjustment for covariates can identify which factors determine—at least in part—the observed spatial patterns (6, 8). Whereas detection of spatial patterns can lead us to closer investigation of the underlying phenomenon, covariate data are needed for real insight into which factors are driving observable patterns.

SUMMARY

The past ten years saw a steep growth in the use of spatial methods in the scientific literature, and much of the growth was in spatial applications to resource availability and disparities research. Overall, proximity measures were the predominant spatial method. Other frequently utilized methods were tools to aggregate spatial data, with spatial smoothing and interpolation methods employed less frequently. Use of cluster detection tools continues to expand, and local clustering methods and public health studies that incorporate space-time clustering and spatial regression have been growing increasingly sophisticated.

Despite recent growth, viewed as a percent of all articles published in major epidemiology journals, use of any spatial method was small (1%, as defined by this review). This may be explained by several factors: (a) Research questions that involve spatial relationships have remained a minority in the discipline; (b) expertise in spatial methods tends to lie in the disciplines of geography and statistics, thus necessitating interdisciplinary collaboration; and (c) spatial epidemiologists can publish their work in many excellent journals not including the seven journals selected for this review. The seven selected journals are well read and well cited by epidemiologists and are able to illustrate trends in the discipline. A review of all public health journals was beyond the scope of this article. Researchers applying spatial methods are advised to consult comprehensive texts (5, 10, 25, 35, 69, 83, 134) because this review was also not intended as a definitive review of best practices, although clearly many best practices have been employed. Rather, the review was intended to summarize the most prevalent methods and substantive applications and to set the stage for discussing where the field may be headed.

FUTURE DIRECTIONS

Employing Spatial Methods to Understand Better How Place-Based Features Affect Health

This review excluded numerous studies that did not execute a spatial analysis—defined as calculating a spatial metric, examining local relationships between spatially contiguous entities, or utilizing a spatial statistic—even though the study investigated spatial exposures or outcomes. Despite an exponential growth of interest in place, most research in public health, with few exceptions, remains aspatial (20). Even among articles included in the review, spatial interpolation or smoothing methods to estimate geographically varying exposures (8) and the use of spatial regression alone or in combination with multilevel models (21, 123) have been underutilized. These techniques should be applied more widely because they are likely to improve specificity of exposure and disease relationships, reduce measurement error, and deepen our understanding of the relationships between place and health.

Integration of Spatial Methods, Digital Media, and Social Networks to Enable Space-Time Research

Reviews have noted the need for more space-time health-related research, as well as a lack of data to support these investigations (83). A significant boost to availability of data has come from the ubiquity of GPS receivers, including wearable GPS devices and cellular phones. These devices facilitate active and passive data gathering to support space-time research. In addition, location-stamped social network Web communications (from Web sites such as Foursquare and Facebook Places) continue to grow in popularity, and these data sources have not yet been utilized in spatial epidemiology. In general, the integration of spatial epidemiology with social network analyses has been underexplored. Only two papers reviewed here examined the joint effects of social and geographic networks on disease or utilized GPS devices in the research (11, 90). Integration of spatially referenced data and social network methods into spatial epidemiology presents methodological challenges (73), and results can be difficult to interpret (11). However, assuming that progress is made to address those challenges, this area will likely be of growing interest to public health researchers.

Spatial Simulation

Interest is increasing in the use of simulation models to better understand and capture the bidirectional relationships between people and places over time (7). Simulation models can be used to create synthetic populations in space and can include individuals who interact with each other in spatially constrained ways that are in turn affected by spatially patterned environments. Computer modeling can be used to program space-time interactions in response to spatial dimensions and environmental features. One of the reviewed articles illustrates this potential: Investigators derived a synthetic population with microarea data and then used geographically weighted regression to suggest how antiobesity policies could be tailored to the specific needs of each microarea (34). A number of challenges and limitations exist for merging spatial methods and simulation frameworks (7). Nevertheless, we anticipate that, increasingly, these two broad methodologies will work in tandem to yield additional insights into the spatial determinants of health and the interventions that may be most useful.

CONCLUSION

To our knowledge, this article is the first review to derive an inventory of spatial epidemiology during the past decade, to document huge growth in this field systematically, and to summarize the tools that have been employed. Space and place have been key dimensions of epidemiology and public health for decades, yet advanced spatial methods have been relatively slow to trickle into epidemiology, likely owing to disciplinary focus and training. Today there are many rich tools to employ for a more sophisticated treatment of space and to understand better the interacting contributions to health of individual characteristics and spatially varying place-based factors. Such factors are potentially amenable to policy and thus have particular importance to public health. Increased use of spatial methods is likely to continue, in tandem with interest in the relationship between place and health, as spatial software tools become more accessible and geographically referenced data become more available.