Positional accuracy of geocoded addresses in epidemiologic research

Using imputation to provide location information for nongeocoded addresses

Background

The importance of geography as a source of variation in health research continues to receive sustained attention in the literature. The inclusion of geographic information in such research often begins by adding data to a map which is predicated by some knowledge of location. A precise level of spatial information is conventionally achieved through geocoding, the geographic information system (GIS) process of translating mailing address information to coordinates on a map. The geocoding process is not without its limitations, though, since there is always a percentage of addresses which cannot be converted successfully (nongeocodable). This raises concerns regarding bias since traditionally the practice has been to exclude nongeocoded data records from analysis.

Methodology/Principal Findings

In this manuscript we develop and evaluate a set of imputation strategies for dealing with missing spatial information from nongeocoded addresses. The strategies are developed assuming a known zip code with increasing use of collateral information, namely the spatial distribution of the population at risk. Strategies are evaluated using prostate cancer data obtained from the Maryland Cancer Registry. We consider total case enumerations at the Census county, tract, and block group level as the outcome of interest when applying and evaluating the methods. Multiple imputation is used to provide estimated total case counts based on complete data (geocodes plus imputed nongeocodes) with a measure of uncertainty. Results indicate that the imputation strategy based on using available population-based age, gender, and race information performed the best overall at the county, tract, and block group levels.

Conclusions/Significance

The procedure allows for the potentially biased and likely under reported outcome, case enumerations based on only the geocoded records, to be presented with a statistically adjusted count (imputed count) with a measure of uncertainty that are based on all the case data, the geocodes and imputed nongeocodes. Similar strategies can be applied in other analysis settings.

Spatial autocorrelation among automated geocoding errors and its effects on testing for disease clustering

Automated geocoding of patient addresses is an important data assimilation component of many spatial epidemiologic studies. Inevitably, the geocoding process results in positional errors. Positional errors incurred by automated geocoding tend to reduce the power of tests for disease clustering and otherwise affect spatial analytic methods. However, there are reasons to believe that the errors may often be positively spatially correlated and that this may mitigate their deleterious effects on spatial analyses. In this article, we demonstrate explicitly that the positional errors associated with automated geocoding of a data set of more than 6000 addresses in Carroll County, Iowa are spatially autocorrelated. Furthermore, through two simulation studies of disease processes, including one in which the disease process is overlain upon the Carroll County addresses, we show that spatial autocorrelation among geocoding errors maintains the power of two tests for disease clustering at a level higher than that which would occur if the errors were independent. Implications of these results for cluster detection, privacy protection, and measurement error modeling of geographic health data are discussed.

Methods for retrospective geocoding in population studies: the Jackson Heart Study

The increasing use of geographic information systems (GIS) in epidemiological population studies requires careful attention to the methods employed in accomplishing geocoding and creating a GIS. Studies have provided limited details,hampering the ability to assess validity of spatial data. The purpose of this paper is to describe the multiphase geocoding methods used to retrospectively create a GIS in the Jackson Heart Study (JHS). We used baseline data from 5,302 participants enrolled in the JHS between 2000 and 2004 in a multiphase process to accomplish geocoding2 years after participant enrollment. After initial deletion of ungeocodable addresses(n=52), 96% were geocoded using ArcGIS. An interactive method using data abstraction from participant records, use of additional maps and street reference files,and verification of existence of address, yielded successful geocoding of all but 13 addresses. Overall, nearly 99% (n=5,237) of the JHS cohort was geocoded retrospectively using the multiple strategies for improving and locating geocodable addresses. Geocoding validation procedures revealed highly accurate and reliable geographic data. Using the methods and protocol developed provided a reliable spatial database that can be used for further investigation of spatial epidemiology. Baseline results were used to describe participants by select geographic indicators, including residence in urban or rural areas, as well as to validate the effectiveness of the study's sampling plan. Further, our results indicate that retrospectively developing a reliable GIS for a large, epidemiological study is feasible. This paper describes some of the challenges in retrospectively creating a GIS and provides practical tips that enhanced the success.

Local indicators of geocoding accuracy (LIGA): theory and application

BACKGROUND

Although sources of positional error in geographic locations (e.g. geocoding error) used for describing and modeling spatial patterns are widely acknowledged, research on how such error impacts the statistical results has been limited. In this paper we explore techniques for quantifying the perturbability of spatial weights to different specifications of positional error.

RESULTS

We find that a family of curves describes the relationship between perturbability and positional error, and use these curves to evaluate sensitivity of alternative spatial weight specifications to positional error both globally (when all locations are considered simultaneously) and locally (to identify those locations that would benefit most from increased geocoding accuracy). We evaluate the approach in simulation studies, and demonstrate it using a case-control study of bladder cancer in south-eastern Michigan.

CONCLUSION

Three results are significant. First, the shape of the probability distributions of positional error (e.g. circular, elliptical, cross) has little impact on the perturbability of spatial weights, which instead depends on the mean positional error. Second, our methodology allows researchers to evaluate the sensitivity of spatial statistics to positional accuracy for specific geographies. This has substantial practical implications since it makes possible routine sensitivity analysis of spatial statistics to positional error arising in geocoded street addresses, global positioning systems, LIDAR and other geographic data. Third, those locations with high perturbability (most sensitive to positional error) and high leverage (that contribute the most to the spatial weight being considered) will benefit the most from increased positional accuracy. These are rapidly identified using a new visualization tool we call the LIGA scatterplot.Herein lies a paradox for spatial analysis: For a given level of positional error increasing sample density to more accurately follow the underlying population distribution increases perturbability and introduces error into the spatial weights matrix. In some studies positional error may not impact the statistical results, and in others it might invalidate the results. We therefore must understand the relationships between positional accuracy and the perturbability of the spatial weights in order to have confidence in a study's results.

Evaluating geographic imputation approaches for zip code level data: an application to a study of pediatric diabetes

BACKGROUND

There is increasing interest in the study of place effects on health, facilitated in part by geographic information systems. Incomplete or missing address information reduces geocoding success. Several geographic imputation methods have been suggested to overcome this limitation. Accuracy evaluation of these methods can be focused at the level of individuals and at higher group-levels (e.g., spatial distribution).

METHODS

We evaluated the accuracy of eight geo-imputation methods for address allocation from ZIP codes to census tracts at the individual and group level. The spatial apportioning approaches underlying the imputation methods included four fixed (deterministic) and four random (stochastic) allocation methods using land area, total population, population under age 20, and race/ethnicity as weighting factors. Data included more than 2,000 geocoded cases of diabetes mellitus among youth aged 0-19 in four U.S. regions. The imputed distribution of cases across tracts was compared to the true distribution using a chi-squared statistic.

RESULTS

At the individual level, population-weighted (total or under age 20) fixed allocation showed the greatest level of accuracy, with correct census tract assignments averaging 30.01% across all regions, followed by the race/ethnicity-weighted random method (23.83%). The true distribution of cases across census tracts was that 58.2% of tracts exhibited no cases, 26.2% had one case, 9.5% had two cases, and less than 3% had three or more. This distribution was best captured by random allocation methods, with no significant differences (p-value > 0.90). However, significant differences in distributions based on fixed allocation methods were found (p-value < 0.0003).

CONCLUSION

Fixed imputation methods seemed to yield greatest accuracy at the individual level, suggesting use for studies on area-level environmental exposures. Fixed methods result in artificial clusters in single census tracts. For studies focusing on spatial distribution of disease, random methods seemed superior, as they most closely replicated the true spatial distribution. When selecting an imputation approach, researchers should consider carefully the study aims.

Access to mammography screening in a large urban population: a multi-level analysis

Objective

To understand area-based sociodemographics, physician and medical practice characteristics, and community indicators associated with mammography use in Los Angeles County. An earlier multi-level analysis by Gumpertz et al. found that distance to the nearest mammography facility helped explain the higher proportion of Latinas diagnosed with late stage breast cancer compared with non-Latina Whites in Los Angeles County. Our study examined whether Latinas also have lower rates of mammography use.

Methods

We used a multi-level spatial modeling approach to examine individual and community level associations with mammography use among a diverse group of women aged 40–84 years in Los Angeles County. To build our multi-level spatial data set, we integrated five data sources: (1) 2001 California Health Interview Survey (CHIS) data, (2) 2001 Food and Drug Administration (FDA) certified mammography facility data, (3) 2003 LA Transit Authority data, (4) 2000 US Decennial Census data, and (5) 2001 Community Tracking Study (CTS) Physician’s Survey data.

Results

Our study confirmed for Los Angeles County many associations for mammography use found in other locations. An unexpected finding was that women with limited English proficiency (predominantly Latina) were significantly more likely to have had a recent mammogram than English-proficient women. We also found that, after controlling for other factors, mammography use was higher in neighborhoods with a greater density of mammography facilities.

Conclusion

Women with limited English proficiency were especially likely to report recent mammography in Los Angeles. This unexpected finding suggests that the intensive Spanish-language outreach program conducted by the Every Woman Counts (EWC) Program in low-income Latina communities in Los Angeles has been effective. Our study highlights the success of this targeted community-based outreach conducted between 1999 and 2001. These are the same populations that Gumpertz et al. identified as needing intervention. It would be useful to conduct another study of late-stage diagnosis in Los Angeles County to ascertain whether increased rates of mammography have also led to less late-stage diagnosis among Latinas in the neighborhoods where they are concentrated in Los Angeles.

BME estimation of residential exposure to ambient PM10 and ozone at multiple time scales

BACKGROUND

Long-term human exposure to ambient pollutants can be an important contributing or etiologic factor of many chronic diseases. Spatiotemporal estimation (mapping) of long-term exposure at residential areas based on field observations recorded in the U.S. Environmental Protection Agency's Air Quality System often suffer from missing data issues due to the scarce monitoring network across space and the inconsistent recording periods at different monitors.

OBJECTIVE

We developed and compared two upscaling methods: UM1 (data aggregation followed by exposure estimation) and UM2 (exposure estimation followed by data aggregation) for the long-term PM10 (particulate matter with aerodynamic diameter < or = 10 microm) and ozone exposure estimations and applied them in multiple time scales to estimate PM and ozone exposures for the residential areas of the Health Effects of Air Pollution on Lupus (HEAPL) study.

METHOD

We used Bayesian maximum entropy (BME) analysis for the two upscaling methods. We performed spatiotemporal cross-validations at multiple time scales by UM1 and UM2 to assess the estimation accuracy across space and time.

RESULTS

Compared with the kriging method, the integration of soft information by the BME method can effectively increase the estimation accuracy for both pollutants. The spatiotemporal distributions of estimation errors from UM1 and UM2 were similar. The cross-validation results indicated that UM2 is generally better than UM1 in exposure estimations at multiple time scales in terms of predictive accuracy and lack of bias. For yearly PM10 estimations, both approaches have comparable performance, but the implementation of UM1 is associated with much lower computation burden.

CONCLUSION

BME-based upscaling methods UM1 and UM2 can assimilate core and site-specific knowledge bases of different formats for long-term exposure estimation. This study shows that UM1 can perform reasonably well when the aggregation process does not alter the spatiotemporal structure of the original data set; otherwise, UM2 is preferable.

Development of a wearable global positioning system for place and health research

Background

An increasing number of studies suggest that characteristics of context, or the attributes of the places within which we live, work and socialize, are associated with variations in health-related behaviours and outcomes. The challenge for health research is to ensure that these places are accurately represented spatially, and to identify those aspects of context that are related to variations in health and amenable to modification. This study focuses on the design of a wearable global positioning system (GPS) data logger for the purpose of objectively measuring the temporal and spatial features of human activities. Person-specific GPS data provides a useful source of information to operationalize the concept of place.

Results

We designed and tested a lightweight, wearable GPS receiver, capable of logging location information for up to 70 hours continuously before recharging. The device is accurate to within 7 m in typical urban environments and performs well across a range of static and dynamic conditions.

Discussion

Rather than rely on static areal units as proxies for places, wearable GPS devices can be used to derive a more complete picture of the different places that influence an individual's wellbeing. The measures are objective and are less subject to biases associated with recall of location or misclassification of contextual attributes. This is important for two reasons. First, it brings a dynamic perspective to place and health research. The influence of place on health is dynamic in that certain places are more or less relevant to wellbeing as determined by the length of time in any location and by the frequency of activity in the location. Second, GPS data can be used to assess whether the characteristics of places at specific times are useful to explaining variations in health and wellbeing.

Methodologic issues and approaches to spatial epidemiology

Spatial epidemiology is increasingly being used to assess health risks associated with environmental hazards. Risk patterns tend to have both a temporal and a spatial component; thus, spatial epidemiology must combine methods from epidemiology, statistics, and geographic information science. Recent statistical advances in spatial epidemiology include the use of smoothing in risk maps to create an interpretable risk surface, the extension of spatial models to incorporate the time dimension, and the combination of individual- and area-level information. Advances in geographic information systems and the growing availability of modeling packages have led to an improvement in exposure assessment. Techniques drawn from geographic information science are being developed to enable the visualization of uncertainty and ensure more meaningful inferences are made from data. When public health concerns related to the environment arise, it is essential to address such anxieties appropriately and in a timely manner. Tools designed to facilitate the investigation process are being developed, although the availability of complete and clean health data, and appropriate exposure data often remain limiting factors.

Children at risk: measuring racial/ethnic disparities in potential exposure to air pollution at school and home

STUDY OBJECTIVE

This paper addresses the environmental justice implications of children's health by exploring racial/ethnic disparities in potential exposure to air pollution, based on both school and home locations of children and three different types of pollution sources, in Orange County, Florida, USA.

METHODS

Using geocoded school and residence locations of 151 709 children enrolled in the public school system, distribution functions of proximity to the nearest source are generated for each type of air pollution source in order to compare the exposure potential of white, Hispanic, and black children. Discrete buffer distances are utilised to provide quantitative comparisons for statistical testing.

MAIN RESULTS

At any given distance from each type of pollution source, the cumulative proportion of Hispanic or black children significantly exceeds the corresponding proportion of white children, for both school and home locations. Regardless of race, however, a larger proportion of children are potentially exposed to air pollution at home than at school.

CONCLUSIONS

This study addresses the growing need to consider both daytime and nighttime activity patterns in the assessment of children's exposure to environmental hazards and related health risks. The results indicate a consistent pattern of racial inequity in the spatial distribution of all types of air pollution sources examined, with black children facing the highest relative levels of potential exposure at both school and home locations.

Geocoding Accuracy Considerations in Determining Residency Restrictions for Sex Offenders

Geocoding is commonly employed to determine the location of addresses for use in spatial analysis, including the establishment of residency restriction zones for sex offenders. Street geocoding has known limitations in terms of positional accuracy, which may adversely affect spatial analytic methods. A case study on sex offenders in Orange County, Florida, is used to characterize the positional accuracy of street geocoding and its impact on spatial analysis. Positional accuracy of street geocoded locations of sex offenders' residences, schools, and day care facilities was determined using Geographic Information Systems (GIS) by measuring the distance to the correct property boundaries. Results show that positional errors in street geocoding are substantial and may bias conclusions drawn from proximity analysis. Findings strongly suggest that street geocoding is not appropriate for assessing residency restriction violations for sex offenders. These findings have important implications for criminal justice policies related to residency restrictions for sex offenders.

A GIS-based method for household recruitment in a prospective pesticide exposure study

BACKGROUND

Recent advances in GIS technology and remote sensing have provided new opportunities to collect ecologic data on agricultural pesticide exposure. Many pesticide studies have used historical or records-based data on crops and their associated pesticide applications to estimate exposure by measuring residential proximity to agricultural fields. Very few of these studies collected environmental and biological samples from study participants. One of the reasons for this is the cost of identifying participants who reside near study fields and analyzing samples obtained from them. In this paper, we present a cost-effective, GIS-based method for crop field selection and household recruitment in a prospective pesticide exposure study in a remote location. For the most part, our multi-phased approach was carried out in a research facility, but involved two brief episodes of fieldwork for ground truthing purposes. This method was developed for a larger study designed to examine the validity of indirect pesticide exposure estimates by comparing measured exposures in household dust, water and urine with records-based estimates that use crop location, residential proximity and pesticide application data. The study focused on the pesticide atrazine, a broadleaf herbicide used in corn production and one of the most widely-used pesticides in the U.S.

RESULTS

We successfully used a combination of remotely-sensed data, GIS-based methods and fieldwork to select study fields and recruit participants in Illinois, a state with high corn production and heavy atrazine use. Our several-step process consisted of the identification of potential study fields and residential areas using aerial photography; verification of crop patterns and land use via site visits; development of a GIS-based algorithm to define recruitment areas around crop fields; acquisition of geocoded household-level data within each recruitment area from a commercial vendor; and confirmation of final participant household locations via ground truthing. The use of these procedures resulted in a sufficient sample of participants from 14 recruitment areas in seven Illinois counties.

CONCLUSION

One of the challenges in pesticide research is the identification and recruitment of study participants, which is time consuming and costly, especially when the study site is in a remote location. We have demonstrated how GIS-based processes can be used to recruit participants, increase efficiency and enhance accuracy. The method that we used ultimately made it possible to collect biological samples from a specific demographic group within strictly defined exposure areas, with little advance knowledge of the location or population.

Geocoding accuracy and the recovery of relationships between environmental exposures and health

Background

This research develops methods for determining the effect of geocoding quality on relationships between environmental exposures and health. The likelihood of detecting an existing relationship – statistical power – between measures of environmental exposures and health depends not only on the strength of the relationship but also on the level of positional accuracy and completeness of the geocodes from which the measures of environmental exposure are made. This paper summarizes the results of simulation studies conducted to examine the impact of inaccuracies of geocoded addresses generated by three types of geocoding processes: a) addresses located on orthophoto maps, b) addresses matched to TIGER files (U.S Census or their derivative street files); and, c) addresses from E-911 geocodes (developed by local authorities for emergency dispatch purposes).

Results

The simulated odds of disease using exposures modelled from the highest quality geocodes could be sufficiently recovered using other, more commonly used, geocoding processes such as TIGER and E-911; however, the strength of the odds relationship between disease exposures modelled at geocodes generally declined with decreasing geocoding accuracy.

Conclusion

Although these specific results cannot be generalized to new situations, the methods used to determine the sensitivity of results can be used in new situations. Estimated measures of positional accuracy must be used in the interpretation of results of analyses that investigate relationships between health outcomes and exposures measured at residential locations. Analyses similar to those employed in this paper can be used to validate interpretation of results from empirical analyses that use geocoded locations with estimated measures of positional accuracy.

Validation of a GIS facilities database: quantification and implications of error

PURPOSE

To validate a commercial database of community-level physical activity facilities that can be used in future research examining associations between access to physical activity facilities and individual-level physical activity and obesity.

METHODS

Physical activity facility characteristics and locations obtained from a commercial database were compared to a field census conducted in 80 census block groups within two U.S. communities. Agreement statistics, agreement of administratively defined neighborhoods, and distance between locations were used to quantify count, attribute, and positional error.

RESULTS

There was moderate agreement (concordance: nonurban: 0.39; urban: 0.46) of presence of any physical activity facility and poor to moderate agreement (kappa range: 0.14 to 0.76) of physical activity facility type. The mean Euclidean distance between commercial database versus field census locations was 757 and 35 m in the nonurban and urban communities, respectively. However, 94% and 100% of nonurban and urban physical activity facilities, respectively, fell into the same 5-digit ZIP code, dropping to 92% and 98% in the same block group and 71% along the same street.

CONCLUSIONS

Our findings suggest that the commercial database of physical activity facilities may contain appreciable error, but patterns of error suggest that built environment-health associations are likely biased downward.

Validation of the geographic position of EPER-Spain industries

Background

The European Pollutant Emission Register in Spain (EPER-Spain) is a public inventory of pollutant industries created by decision of the European Union. The location of these industries is geocoded and the first published data correspond to 2001. Publication of these data will allow for quantification of the effect of proximity to one or more such plant on cancer and all-cause mortality observed in nearby towns. However, as errors have been detected in the geocoding of many of the pollutant foci shown in the EPER, it was decided that a validation study should be conducted into the accuracy of these co-ordinates. EPER-Spain geographic co-ordinates were drawn from the European Environment Agency (EEA) server and the Spanish Ministry of the Environment (MOE). The Farm Plot Geographic Information System (Sistema de Información Geográfica de Parcelas Agrícolas) (SIGPAC) enables orthophotos (digitalized aerial images) of any territorial point across Spain to be obtained. Through a search of co-ordinates in the SIGPAC, all the industrial foci (except farms) were located. The quality criteria used to ascertain possible errors in industrial location were high, medium and low quality, where industries were situated at a distance of less than 500 metres, more than 500 metres but less than 1 kilometre, and more than 1 kilometre from their real locations, respectively.

Results

Insofar as initial registry quality was concerned, 84% of industrial complexes were inaccurately positioned (low quality) according to EEA data versus 60% for Spanish MOE data. The distribution of the distances between the original and corrected co-ordinates for each of the industries on the registry revealed that the median error was 2.55 kilometres for Spain overall (according to EEA data). The Autonomous Regions that displayed most errors in industrial geocoding were Murcia, Canary Islands, Andalusia and Madrid. Correct co-ordinates were successfully allocated to 100% of EPER-Spain industries.

Conclusion

Knowing the exact location of pollutant foci is vital to obtain reliable and valid conclusions in any study where distance to the focus is a decisive factor, as in the case of the consequences of industrial pollution on the health of neighbouring populations.

Error and bias in determining exposure potential of children at school locations using proximity-based GIS techniques

BACKGROUND

The widespread availability of powerful tools in commercial geographic information system (GIS) software has made address geocoding a widely employed technique in spatial epidemiologic studies.

OBJECTIVE

The objective of this study was to determine the effect of the positional error in geocoding on the analysis of exposure to traffic-related air pollution of children at school locations.

METHODS

For a case study of Orange County, Florida, we determined the positional error of geocoding of school locations through comparisons with a parcel database and digital orthophotography. We used four different geocoding techniques for comparison to establish the repeatability of geocoding, and an analysis of proximity to major roads to determine bias and error in environmental exposure assessment.

RESULTS

RESULTS INDICATE THAT THE POSITIONAL ERROR IN GEOCODING OF SCHOOLS IS VERY SUBSTANTIAL: We found that the 95% root mean square error was 196 m using street centerlines, 306 m using TIGER roads, and 210 and 235 m for two commercial geocoding firms. We found bias and error in proximity analysis to major roads to be unacceptably large at distances of < 500 m. Bias and error are introduced by lack of positional accuracy and lack of repeatability of geocoding of school locations.

CONCLUSIONS

These results suggest that typical geocoding is insufficient for fine-scale analysis of school locations and more accurate alternatives need to be considered.

Estimating the Intensity of a Spatial Point Process from Locations Coarsened by Incomplete Geocoding

The estimation of spatial intensity is an important inference problem in spatial epidemiologic studies. A standard data assimilation component of these studies is the assignment of a geocode, that is, point-level spatial coordinates, to the address of each subject in the study population. Unfortunately, when geocoding is performed by the standard automated method of street-segment matching to a georeferenced road file and subsequent interpolation, it is rarely completely successful. Typically, 10-30% of the addresses in the study population, and even higher percentages in particular subgroups, fail to geocode, potentially leading to a selection bias, called geographic bias, and an inefficient analysis. Missing-data methods could be considered for analyzing such data; however, because there is almost always some geographic information coarser than a point (e.g., a Zip code) observed for the addresses that fail to geocode, a coarsened-data analysis is more appropriate. This article develops methodology for estimating spatial intensity from coarsened geocoded data. Both nonparametric (kernel smoothing) and likelihood-based estimation procedures are considered. Substantial improvements in the estimation quality of coarsened-data analyses relative to analyses of only the observations that geocode are demonstrated via simulation and an example from a rural health study in Iowa.

Fetal deaths and proximity to hazardous waste sites in Washington State

BACKGROUND

The in utero period is one of increased susceptibility to environmental effects. The effects of prenatal exposure to environmental toxicants on various adverse pregnancy outcomes, including fetal death, are not well understood.

OBJECTIVE

We examined the risk of fetal death in relation to maternal residential proximity to hazardous waste sites.

METHODS

We conducted a population-based case-control study using Washington State vital records for 1987-2001. Cases were women with fetal deaths at > or = 20 weeks (n = 7,054). Ten controls per case were randomly selected from live births. Locations of 939 hazardous waste sites were identified from the Department of Ecology registry. We measured distance from maternal residence at delivery to the nearest hazardous waste site, and calculated odds ratios (ORs) and 95% confidence intervals (CIs).

RESULTS

The risk of fetal death for women residing < or = 0.5 miles, relative to > 5 miles, from a hazardous waste site was not increased (adjusted OR = 1.06; 95% CI, 0.90-1.25). No associations were observed for any proximity categories < or = 5 miles from sites with contaminated air, soil, water, solvents, or metals; however, fetal death risk increased among women residing < or = 1 mile from pesticide-containing sites (OR = 1.28; 95% CI, 1.13-1.46).

CONCLUSION

These results do not suggest that fetal death is associated with residential proximity to hazardous waste sites overall; however, close proximity to pesticide-containing sites may increase the risk of fetal death.

Positional accuracy and geographic bias of four methods of geocoding in epidemiologic research

PURPOSE

We examined the geographic bias of four methods of geocoding addresses using ArcGIS, commercial firm, SAS/GIS, and aerial photography. We compared "point-in-polygon" (ArcGIS, commercial firm, and aerial photography) and the "look-up table" method (SAS/GIS) to allocate addresses to census geography, particularly as it relates to census-based poverty rates.

METHODS

We randomly selected 299 addresses of children treated for asthma at an urban emergency department (1999-2001). The coordinates of the building address side door were obtained by constant offset based on ArcGIS and a commercial firm and true ground location based on aerial photography.

RESULTS

Coordinates were available for 261 addresses across all methods. For 24% to 30% of geocoded road/door coordinates the positional error was 51 meters or greater, which was similar across geocoding methods. The mean bearing was -26.8 degrees for the vector of coordinates based on aerial photography and ArcGIS and 8.5 degrees for the vector based on aerial photography and the commercial firm (p < 0.0001). ArcGIS and the commercial firm performed very well relative to SAS/GIS in terms of allocation to census geography. For 20%, the door location based on aerial photography was assigned to a different block group compared to SAS/GIS. The block group poverty rate varied at least two standard deviations for 6% to 7% of addresses.

CONCLUSION

We found important differences in distance and bearing between geocoding relative to aerial photography. Allocation of locations based on aerial photography to census-based geographic areas could lead to substantial errors.

Comparing a single-stage geocoding method to a multi-stage geocoding method: how much and where do they disagree?

Background

Geocoding methods vary among spatial epidemiology studies. Errors in the geocoding process and differential match rates may reduce study validity. We compared two geocoding methods using 8,157 Washington State addresses. The multi-stage geocoding method implemented by the state health department used a sequence of local and national reference files. The single-stage method used a single national reference file. For each address geocoded by both methods, we measured the distance between the locations assigned by each method. Area-level characteristics were collected from census data, and modeled as predictors of the discordance between geocoded address coordinates.

Results

The multi-stage method had a higher match rate than the single-stage method: 99% versus 95%. Of 7,686 addresses were geocoded by both methods, 96% were geocoded to the same census tract by both methods and 98% were geocoded to locations within 1 km of each other by the two methods. The distance between geocoded coordinates for the same address was higher in sparsely populated and low poverty areas, and counties with local reference files.

Conclusion

The multi-stage geocoding method had a higher match rate than the single-stage method. An examination of differences in the location assigned to the same address suggested that study results may be most sensitive to the choice of geocoding method in sparsely populated or low-poverty areas.

Influence of geocoding quality on environmental exposure assessment of children living near high traffic roads

Background

The widespread availability of powerful geocoding tools in commercial GIS software and the interest in spatial analysis at the individual level have made address geocoding a widely employed technique in epidemiological studies. This study determined the effect of the positional error in street geocoding on the analysis of traffic-related air pollution on children.

Methods

For a case-study of a large sample of school children in Orange County, Florida (n = 104,865) the positional error of street geocoding was determined through comparison with a parcel database. The effect of this error was evaluated by analyzing the proximity of street and parcel geocoded locations to road segments with high traffic volume and determining the accuracy of the classification using the results of street geocoding. Of the original sample of 163,886 addresses 36% were not used in the final analysis because they could not be reliably geocoded using either street or parcel geocoding. The estimates of positional error can therefore be considered conservative underestimates.

Results

Street geocoding was found to have a median error of 41 meters, a 90^th percentile of 100 meters, a 95^th percentile of 137 meters and a 99^th percentile of 273 meters. These positional errors were found to be non-random in nature and introduced substantial bias and error in the estimates of potential exposure to traffic-related air pollution. Street geocoding was found to consistently over-estimate the number of potentially exposed children at small distances up to 250 meters. False positives and negatives were also found to be very common at these small distances.

Conclusion

Results of the case-study presented here strongly suggest that typical street geocoding is insufficient for fine-scale analysis and more accurate alternatives need to be considered.

Risk of malformations associated with residential proximity to hazardous waste sites in Washington State

Hazardous waste sites often contain substances harmful to fetal development. Using linked birth-hospital discharge and hazardous sites data for Washington State, we evaluated the association between malformation occurrence and maternal residential proximity to hazardous waste sites. Cases (N=63,006) were infants born 1987-2001 with malformations. Controls (N=315,030) were randomly selected infants without malformations born during these years. Distance between maternal residence and nearest hazardous waste site was measured using geographic information systems (GIS) software. Odds ratio (OR) estimates of the relative risk of malformation at varying distances were calculated. Relative to living >5 miles from a site, living < or = 5 miles was associated with increased risk of any malformations in offspring (for >2- < or = 5 miles: OR 1.15: 95% Confidence Interval (CI): 1.10, 1.21; for >1- < or = 2 miles: OR 1.26, 95% CI: 1.20, 1.32; for >0.5- < or = 1 miles: OR 1.28, 95% CI: 1.22, 1.35; for < or = 0.5 miles: OR 1.33, 95% CI: 1.27, 1.40.) Risk estimates varied by urban vs. rural maternal residence and by specific malformation type. Hazardous waste sites are often located within populated areas. Thus, the possibility of increased malformation occurrence among those in close proximity deserves closer scrutiny.

The accuracy of address coding and the effects of coding errors

Studies that estimate the influence of characteristics of place on health often use geocoded addresses to identify location of study subjects. This study uses housing built before 1990 selected for the 1995-2001 National Health Interview Survey (N=252,421) to develop a standard against which geocodes obtained from an address-coding program are compared. The results show that geocoding is generally accurate and is more successful in urban areas. Blockgroups with missing codes are more rural and somewhat poorer than blockgroups with correct codes. The effect of incorrect codes on statistical analyses depends on the proportion rural in a study population.

Modeling the probability distribution of positional errors incurred by residential address geocoding

Background

The assignment of a point-level geocode to subjects' residences is an important data assimilation component of many geographic public health studies. Often, these assignments are made by a method known as automated geocoding, which attempts to match each subject's address to an address-ranged street segment georeferenced within a streetline database and then interpolate the position of the address along that segment. Unfortunately, this process results in positional errors. Our study sought to model the probability distribution of positional errors associated with automated geocoding and E911 geocoding.

Results

Positional errors were determined for 1423 rural addresses in Carroll County, Iowa as the vector difference between each 100%-matched automated geocode and its true location as determined by orthophoto and parcel information. Errors were also determined for 1449 60%-matched geocodes and 2354 E911 geocodes. Huge (> 15 km) outliers occurred among the 60%-matched geocoding errors; outliers occurred for the other two types of geocoding errors also but were much smaller. E911 geocoding was more accurate (median error length = 44 m) than 100%-matched automated geocoding (median error length = 168 m). The empirical distributions of positional errors associated with 100%-matched automated geocoding and E911 geocoding exhibited a distinctive Greek-cross shape and had many other interesting features that were not capable of being fitted adequately by a single bivariate normal or t distribution. However, mixtures of t distributions with two or three components fit the errors very well.

Conclusion

Mixtures of bivariate t distributions with few components appear to be flexible enough to fit many positional error datasets associated with geocoding, yet parsimonious enough to be feasible for nascent applications of measurement-error methodology to spatial epidemiology.

Evaluating common de-identification heuristics for personal health information

BACKGROUND

With the growing adoption of electronic medical records, there are increasing demands for the use of this electronic clinical data in observational research. A frequent ethics board requirement for such secondary use of personal health information in observational research is that the data be de-identified. De-identification heuristics are provided in the Health Insurance Portability and Accountability Act Privacy Rule, funding agency and professional association privacy guidelines, and common practice.

OBJECTIVE

The aim of the study was to evaluate whether the re-identification risks due to record linkage are sufficiently low when following common de-identification heuristics and whether the risk is stable across sample sizes and data sets.

METHODS

Two methods were followed to construct identification data sets. Re-identification attacks were simulated on these. For each data set we varied the sample size down to 30 individuals, and for each sample size evaluated the risk of re-identification for all combinations of quasi-identifiers. The combinations of quasi-identifiers that were low risk more than 50% of the time were considered stable.

RESULTS

The identification data sets we were able to construct were the list of all physicians and the list of all lawyers registered in Ontario, using 1% sampling fractions. The quasi-identifiers of region, gender, and year of birth were found to be low risk more than 50% of the time across both data sets. The combination of gender and region was also found to be low risk more than 50% of the time. We were not able to create an identification data set for the whole population.

CONCLUSIONS

Existing Canadian federal and provincial privacy laws help explain why it is difficult to create an identification data set for the whole population. That such examples of high re-identification risk exist for mainstream professions makes a strong case for not disclosing the high-risk variables and their combinations identified here. For professional subpopulations with published membership lists, many variables often needed by researchers would have to be excluded or generalized to ensure consistently low re-identification risk. Data custodians and researchers need to consider other statistical disclosure techniques for protecting privacy.

Spatial confidentiality and GIS: re-engineering mortality locations from published maps about Hurricane Katrina

Background

Geographic Information Systems (GIS) can provide valuable insight into patterns of human activity. Online spatial display applications, such as Google Earth, can democratise this information by disseminating it to the general public. Although this is a generally positive advance for society, there is a legitimate concern involving the disclosure of confidential information through spatial display. Although guidelines exist for aggregated data, little has been written concerning the display of point level information. The concern is that a map containing points representing cases of cancer or an infectious disease, could be re-engineered back to identify an actual residence. This risk is investigated using point mortality locations from Hurricane Katrina re-engineered from a map published in the Baton Rouge Advocate newspaper, and a field team validating these residences using search and rescue building markings.

Results

We show that the residence of an individual, visualized as a generalized point covering approximately one and half city blocks on a map, can be re-engineered back to identify the actual house location, or at least a close neighbour, even if the map contains little spatial reference information. The degree of re-engineering success is also shown to depend on the urban characteristic of the neighborhood.

Conclusion

The results in this paper suggest a need to re-evaluate current guidelines for the display of point (address level) data. Examples of other point maps displaying health data extracted from the academic literature are presented where a similar re-engineering approach might cause concern with respect to violating confidentiality. More research is also needed into the role urban structure plays in the accuracy of re-engineering. We suggest that health and spatial scientists should be proactive and suggest a series of point level spatial confidentiality guidelines before governmental decisions are made which may be reactionary toward the threat of revealing confidential information, thereby imposing draconian limits on research using a GIS.

Match rate and positional accuracy of two geocoding methods for epidemiologic research

PURPOSE

This study compares the match rate and positional accuracy of two geocoding methods: the popular geocoding tool in ArcGIS 9.1 and the Centrus GeoCoder for ArcGIS.

METHODS

We first geocoded 11,016 Texas addresses in a case-control study using both methods and obtained the match rate of each method. We then randomly selected 200 addresses from those geocoded by using both methods and obtained geographic coordinates of the 200 addresses by using a global positioning system (GPS) device. Of the 200 addresses, 110 were case maternal residence addresses and 90 were control maternal residence addresses. These GPS-surveyed coordinates were used as the "true" coordinates to calculate positional errors of geocoded locations. We used Wilcoxon signed rank test to evaluate whether differences in positional errors from the two methods were statistically significantly different from zero. In addition, we calculated the sensitivity and specificity of the two methods for classifying maternal addresses within 1500 m of toxic release inventory facilities when distance is used as a proxy of exposure.

RESULTS

The match rate of the Centrus GeoCoder was more than 10% greater than that of the geocoding tool in ArcGIS 9.1. Positional errors with the Centrus GeoCoder were less than those of the geocoding tool in ArcGIS 9.1, and this difference was statistically significant. Sensitivity and specificity of the two methods are similar.

CONCLUSIONS

Centrus GeoCoder for ArcGIS for geocoding gives greater match rates than the geocoding tool in ArcGIS 9.1. Although the Centrus GeoCoder has better positional accuracy, both methods give similar results in classifying maternal addresses within 1500 m of toxic release inventory facilities when distance is used as a proxy of exposure.

Accuracy of commercial geocoding: assessment and implications

BACKGROUND

Published studies of geocoding accuracy often focus on a single geographic area, address source or vendor, do not adjust accuracy measures for address characteristics, and do not examine effects of inaccuracy on exposure measures. We addressed these issues in a Women's Health Initiative ancillary study, the Environmental Epidemiology of Arrhythmogenesis in WHI.

RESULTS

Addresses in 49 U.S. states (n = 3,615) with established coordinates were geocoded by four vendors (A-D). There were important differences among vendors in address match rate (98%; 82%; 81%; 30%), concordance between established and vendor-assigned census tracts (85%; 88%; 87%; 98%) and distance between established and vendor-assigned coordinates (mean rho [meters]: 1809; 748; 704; 228). Mean rho was lowest among street-matched, complete, zip-coded, unedited and urban addresses, and addresses with North American Datum of 1983 or World Geodetic System of 1984 coordinates. In mixed models restricted to vendors with minimally acceptable match rates (A-C) and adjusted for address characteristics, within-address correlation, and among-vendor heteroscedasticity of rho, differences in mean rho were small for street-type matches (280; 268; 275), i.e. likely to bias results relying on them about equally for most applications. In contrast, differences between centroid-type matches were substantial in some vendor contrasts, but not others (5497; 4303; 4210) p(interaction) < 10(-4), i.e. more likely to bias results differently in many applications. The adjusted odds of an address match was higher for vendor A versus C (odds ratio = 66, 95% confidence interval: 47, 93), but not B versus C (OR = 1.1, 95% CI: 0.9, 1.3). That of census tract concordance was no higher for vendor A versus C (OR = 1.0, 95% CI: 0.9, 1.2) or B versus C (OR = 1.1, 95% CI: 0.9, 1.3). Misclassification of a related exposure measure--distance to the nearest highway--increased with mean rho and in the absence of confounding, non-differential misclassification of this distance biased its hypothetical association with coronary heart disease mortality toward the null.

CONCLUSION

Geocoding error depends on measures used to evaluate it, address characteristics and vendor. Vendor selection presents a trade-off between potential for missing data and error in estimating spatially defined attributes. Informed selection is needed to control the trade-off and adjust analyses for its effects.

Comparison of residential geocoding methods in population-based study of air quality and birth defects

Our population-based case-control study of air quality and birth defects in Texas relied on the geocoding of maternal residence from vital records for the assignment of air pollution exposures during early pregnancy. We attempted to geocode the maternal addresses for 5,338 birth defect cases and 4,574 frequency-matched controls using an automated procedure with standard matching criteria in ArcGIS 8.2 and 8.3. Initially, we matched 7,266 observations (73%). To increase the proportion of successful matches, we used an interactive procedure for the 2,646 addresses that were initially not geocoded by the software. This yielded an additional 985 matches (37%). Using the same 2,646 initially unmatched addresses, we compared the results of this interactive procedure to those of an automated procedure using lower standards. The automated procedure with lower standards yielded more matches (n=1,559, 59%) but with questionable accuracy. We included the interactively geocoded observations in our final data set. Their inclusion did not affect the estimates of air pollution exposure but increased our statistical power to detect associations between air quality and risk of selected birth defects. The geocoded and not geocoded populations differed in the distribution of Latino ethnicity (51% vs 59%) and ethnicity was independently associated with air pollution exposures (P<0.05). Geocoding status also appeared to modify the association between ethnicity and risk of birth defects; Latina women appeared to have a slightly lower risk of birth defects than non-Latina women in the geocoded population and to have a slightly higher risk in the not geocoded population. Incomplete geocoding may have resulted in a selection bias because of the under-representation of Latinas in our study population.

Location of residence associated with the likelihood of patient visit to the preoperative assessment clinic

BACKGROUND

Outpatient preoperative assessment clinics were developed to provide an efficient assessment of surgical patients prior to surgery, and have demonstrated benefits to patients and the health care system. However, the centralization of preoperative assessment clinics may introduce geographical barriers to utilization that are dependent on where a patient lives with respect to the location of the preoperative assessment clinic.

METHODS

The association between geographical distance from a patient's place of residence to the preoperative assessment clinic, and the likelihood of a patient visit to the clinic prior to surgery, was assessed for all patients undergoing surgery at a tertiary health care centre in a major Canadian city. The odds of attending the preoperative clinic were adjusted for patient characteristics and clinical factors.

RESULTS

Patients were less likely to visit the preoperative assessment clinic prior to surgery as distance from the patient's place of residence to the clinic increased (adjusted OR = 0.52, 95% CI 0.44-0.63 for distances between 50-100 km, and OR = 0.26, 95% CI 0.21-0.31 for distances greater than 250 km). This 'distance decay' effect was remarkable for all surgical specialties.

CONCLUSION

The present study demonstrates that the likelihood of a patient visiting the preoperative assessment clinic appears to depend on the geographical location of patients' residences. Patients who live closest to the clinic tend to be seen more often than patients who live in rural and remote areas. This observation may have implications for achieving the goals of equitable access, and optimal patient care and resource utilization in a single universal insurer health care system.

Geocoding in cancer research: a review

There is now widespread agreement that geographic identifiers (geocodes) should be assigned to cancer records, but little agreement on their form and how they should be assigned, reported, and used. This paper reviews geocoding practice in relation to major purposes and discusses methods to improve the accuracy of geocoded cancer data. Differences in geocoding methods and materials introduce errors of commission and omission into geocoded data. A common source of error comes from the practice of using digital boundary files of dubious quality to place addresses into areas of interest. Geocoded data are linked to demographic, environmental, and health services data, and each data type has unique accuracy considerations. In health services applications, the accuracy of distances computed from geocodes can differ markedly. Privacy and confidentiality issues are important in the use and release of geocoded cancer data. When masking methods are used for disclosure limitation purposes, statistical methods must be adjusted for the locational uncertainty of geocoded data. We conclude that selection of one particular type of geographic area as the geocode may unnecessarily constrain future work. Therefore, the longitude and latitude of each case is the superior basic geocode; all other geocodes of interest can be constructed from this basic identifier.

Influence of geographic location in modeling blood pesticide levels in a community surrounding a U.S. Environmental protection agency superfund site

In this study we evaluated residential location as a potential determinant for exposure to organochlorine compounds. We investigated the geographic distribution characteristics of organochlorine levels in approximately 1,374 blood samples collected in 1974 from residents of a community with a potential organochlorine source. Street addresses of Washington County, Maryland, residents were obtained and geocoded in a geographic information system. We used multivariate linear regression models to characterize the blood organochlorine levels of these residents that had been analyzed as part of previous studies using both environmental- and individual-level covariates. This was done to evaluate if the geographic distribution of blood levels in participants was related to the environmental source in the community. Model inference was based on generalized least squares to account for residual spatial variation. A significant inverse relationship was found between blood dieldrin levels and residential distance from the potential source. For every mile of distance from the source, blood dieldrin levels decreased 1.6 ng/g in study participants (p-value = 0.042), adjusting for age, sex, education level, smoking status, and drinking water source. 1,1-Dichloro-2,2-bis(p-chlorophenyl)ethylene (DDE) levels in the blood did not change significantly based on residential distance from the source, taking the same covariates into account. However, these results are limited by the inability to account for several potential confounders. This study demonstrates that spatially distributed covariates may play an important role in individual exposure patterns. Spatial information may enable researchers to detect a potential exposure pattern that may not be revealed with only nonspatial variables.

Geographic bias related to geocoding in epidemiologic studies

BACKGROUND

This article describes geographic bias in GIS analyses with unrepresentative data owing to missing geocodes, using as an example a spatial analysis of prostate cancer incidence among whites and African Americans in Virginia, 1990-1999. Statistical tests for clustering were performed and such clusters mapped. The patterns of missing census tract identifiers for the cases were examined by generalized linear regression models.

RESULTS

The county of residency for all cases was known, and 26,338 (74%) of these cases were geocoded successfully to census tracts. Cluster maps showed patterns that appeared markedly different, depending upon whether one used all cases or those geocoded to the census tract. Multivariate regression analysis showed that, in the most rural counties (where the missing data were concentrated), the percent of a county's population over age 64 and with less than a high school education were both independently associated with a higher percent of missing geocodes.

CONCLUSION

We found statistically significant pattern differences resulting from spatially non-random differences in geocoding completeness across Virginia. Appropriate interpretation of maps, therefore, requires an understanding of this phenomenon, which we call "cartographic confounding."

Methodological aspects of a GIS-based environmental health inspection program used in the Athens 2004 Olympic and Para Olympic Games

Background

The use of geographical information system (GIS) technologies in public health surveillance is gradually gaining momentum around the world and many applications have already been reported in the literature. In this study, GIS technology was used to help county departments of Public Health to implement environmental health surveillance for the Athens 2004 Olympic and Para Olympic Games.

Methods

In order to assess the workload in each Olympic county, 19 registry forms and 17 standardized inspection forms were developed to register and inspect environmental health items requiring inspection (Hotels, restaurants, swimming pools, water supply system etc), respectively. Furthermore, related databases were created using Epi Info 2002 and a geographical information system (GIS) were used to implement an integrated Environmental Health inspection program. The project was conducted in Athens by the Olympic Planning Unit (OPU) of the National School of Public Health, in close cooperation with the Ministry of Health and Social Solidarity and the corresponding departments of Public Health in all municipalities that were scheduled to host events during the Athens 2004 Olympic and Para Olympic games.

Results

A total of 44,741 premises of environmental health interest were geocoded into GIS databases and several electronic maps were developed. Using such maps in association with specific criteria, we first identified the maximum workload required to execute environmental health inspections in all premises within the eleven Olympic County Departments of Public Health. Six different scenarios were created for each county, based on devised algorithms in order to design the most effective and realistic inspection program using the available inspectors from each municipality. Furthermore, GIS applications were used to organize the daily inspection program for the Olympic games, provide coloured displays of the inspection results and link those results with the public health surveillance of specific cases or outbreak investigation.

Conclusion

Our computerised program exhibited significant efficiency in facilitating the prudent use of public health resources in implementing environmental health inspections in densely populated urban areas as well as in rural counties. Furthermore, the application of simple algorithms in integrating human and other resources provided tailored and cost-effective applications to different public health agencies.

Comparison of spatial scan statistic and spatial filtering in estimating low birth weight clusters

Background

The purpose of this study is to examine the spatial and population (e.g., socio-economic) characteristics of low birthweight using two different cluster estimation techniques. We compared the results of Kulldorff's Spatial Scan Statistic with the results of Rushton's Spatial filtering technique across increasing sizes of spatial filters (circle). We were able to demonstrate that varying approaches exist to explore spatial variation in patterns of low birth weight.

Results

Spatial filtering results did not show any particular area that was not statistically significant based on SaTScan. The high rates, which remain as the filter size increases to 0.4, 0.5 to 0.6 miles, respectively, indicate that these differences are less likely due to chance. The maternal characteristics of births within clusters differed considerably between the two methods. Progressively larger Spatial filters removed local spatial variability, which eventually produced an approximate uniform pattern of low birth weight.

Conclusion

SaTScan and Spatial filtering cluster estimation methods produced noticeably different results from the same individual level birth data. SaTScan clusters are likely to differ from Spatial filtering clusters in terms of population characteristics and geographic area within clusters. Using the two methods in conjunction could provide more detail about the population and spatial features contained with each type of cluster.

Assessing spatio-temporal variability of risk surfaces using residential history data in a case control study of breast cancer

BACKGROUND: Most analyses of spatial clustering of disease have been based on either residence at the time of diagnosis or current residence. An underlying assumption in these analyses is that residence can be used as a proxy for environmental exposure. However, exposures earlier in life and not just those in the most recent period may be of significance. In breast cancer, there is accumulating evidence that early life exposures may contribute to risk. We explored spatio-temporal patterns of risk surfaces using data on lifetime residential history in a case control study of breast cancer, and identified elevated areas of risk and areas potentially having more exposure opportunities, defined as risk surfaces in this study. This approach may be more relevant in understanding the environmental etiology of breast cancer, since lifetime cumulative exposures or exposures at critical times may be more strongly associated with risk for breast cancer than exposures from the recent period. RESULTS: A GIS-based exploratory spatial analysis was applied, and spatio-temporal variability of those risk surfaces was evaluated using the standardized difference in density surfaces between cases and controls. The significance of the resulting risk surfaces was tested and reported as p-values. These surfaces were compared for premenopausal and postmenopausal women, and were obtained for each decade, from the 1940s to 1990s. We found strong evidence of clustering of lifetime residence for premenopausal women (for cases relative to controls), and a less strong suggestion of such clustering for postmenopausal women, and identified a substantial degree of temporal variability of the risk surfaces. CONCLUSION: We were able to pinpoint geographic areas with higher risk through exploratory spatial analyses, and to assess temporal variability of the risk surfaces, thus providing a working hypothesis on breast cancer and environmental exposures. Geographic areas with higher case densities need further epidemiologic investigation for potential relationships between lifetime environmental exposures and breast cancer risk. Examination of lifetime residential history provided additional information on geographic areas associated with higher risk; limiting exploration of chronic disease clustering to current residence may neglect important relationships between location and disease.

Lumping or splitting: seeking the preferred areal unit for health geography studies

Background

Findings are compared on geographic variation of incident and late-stage cancers across Connecticut using different areal units for analysis.

Results

Few differences in results were found for analyses across areal units. Global clustering of incident prostate and breast cancer cases was apparent regardless of the level of geography used. The test for local clustering found approximately the same locales, populations at risk and estimated effects. However, some discrepancies were uncovered.

Conclusion

In the absence of conditions calling for surveillance of small area cancer clusters ('hot spots'), the rationale for accepting the burdens of preparing data at levels of geography finer than the census tract may not be compelling.

Breast Cancer Risk and Exposure in Early Life to Polycyclic Aromatic Hydrocarbons Using Total Suspended Particulates as a Proxy Measure

Polycyclic aromatic hydrocarbons (PAH) are ubiquitous in the environment. We hypothesized that early life exposure to PAHs may have particular importance in the etiology of breast cancer. We conducted a population-based, case-control study of ambient exposure to PAHs in early life in relation to the risk of breast cancer. Total suspended particulates (TSP), a measure of ambient air pollution, was used as a proxy for PAHs exposure. Cases (n = 1,166) were women with histologically confirmed, primary, incident breast cancer. Controls (n = 2,105) were frequency matched by age, race, and county of residence to cases. Annual average TSP concentrations (1959-1997) by location were obtained from the New York State Department of Environmental Conservation for Erie and Niagara Counties. Based on the monitor readings, prediction maps of TSP concentrations were generated with ArcGIS 8.0 (ESRI, Inc., Redlands, CA) using inverse distance squared weighted interpolation. Unconditional logistic regression was used to estimate odds ratios and 95% confidence intervals. In postmenopausal women, exposure to high concentrations of TSP (&gt;140 μg/m3) at birth was associated with an adjusted odds ratio of 2.42 (95% confidence interval, 0.97-6.09) compared with exposure to low concentrations (&lt;84 μg/m3). However, in premenopausal women, where exposures were generally lower, the results were inconsistent with our hypothesis and in some instances were suggestive of a reduction in the risk of breast cancer. Our study suggests that exposure in early life to high levels of PAHs may increase the risk of postmenopausal breast cancer; however, other confounders related to geography cannot be ruled out.

Improving geocoding practices: evaluation of geocoding tools

This study examined the sources of error involved in geocoding, by systematically evaluating the strengths and weaknesses of three widely used tools for geocoding. We tested them against a random sample of addresses from a state administrative address master file and found considerable variation in identification of census block geocodes of addresses. This high variation was mainly attributable to differences in preprocessing of addresses before geocoding and the reference street data used for geocoding. Preprocessing includes not only parsing and standardizing, but also correcting addresses against the US Postal Service Zip+4 Database, the master mailing address database maintained and updated regularly by USPS.

Current practices in spatial analysis of cancer data: data characteristics and data sources for geographic studies of cancer

The use of spatially referenced data in cancer studies is gaining in prominence, fueled by the development and availability of spatial analytic tools and the broadening recognition of the linkages between geography and health. We provide an overview of some of the unique characteristics of spatial data, followed by an account of the major types and sources of data used in the spatial analysis of cancer, including data from cancer registries, population data, health surveys, environmental data, and remote sensing data. We cite numerous examples of recent studies that have used these data, with a focus on etiological research.

Historical measures of social context in life course studies: retrospective linkage of addresses to decennial censuses

BACKGROUND: There is evidence of a contribution of early life socioeconomic exposures to the risk of chronic diseases in adulthood. However, extant studies investigating the impact of the neighborhood social environment on health tend to characterize only the current social environment. This in part may be due to complexities involved in obtaining and geocoding historical addresses. The Life Course Socioeconomic Status, Social Context, and Cardiovascular Disease Study collected information on childhood (1930-1950) and early adulthood (1960-1980) place of residence from 12,681 black and white middle-aged and older men and women from four U.S. communities to link participants with census-based socioeconomic indicators over the life course. RESULTS: Most (99%) participants were linked to 1930-50 county level socioeconomic census data (the smallest level of aggregation universally available during this time period) corresponding to childhood place of residence. Linkage did not vary by race, gender, birth cohort, or level of educational attainment. A commercial geocoding vendor processed participants' self-reported street addresses for ages 30, 40, and 50. For 1970 and 1980 censuses, spatial coordinates were overlaid onto shape files containing census tract boundaries; for 1960 no shape files existed and comparability files were used. Several methods were tested for accuracy and to increase linkage. Successful linkage to historical census tracts varied by census (66% for 1960, 76% for 1970, 85% for 1980). This compares to linkage rates of 94% for current addresses provided by participants over the course of the ARIC examinations. CONCLUSION: There are complexities and limitations in characterizing the past social context. However, our results suggest that it is feasible to characterize the earlier social environment with known levels of measurement error and that such an approach should be considered in future studies.

Using geographic information systems for exposure assessment in environmental epidemiology studies

Geographic information systems (GIS) are being used with increasing frequency in environmental epidemiology studies. Reported applications include locating the study population by geocoding addresses (assigning mapping coordinates), using proximity analysis of contaminant source as a surrogate for exposure, and integrating environmental monitoring data into the analysis of the health outcomes. Although most of these studies have been ecologic in design, some have used GIS in estimating environmental levels of a contaminant at the individual level and to design exposure metrics for use in epidemiologic studies. In this article we discuss fundamentals of three scientific disciplines instrumental to using GIS in exposure assessment for epidemiologic studies: geospatial science, environmental science, and epidemiology. We also explore how a GIS can be used to accomplish several steps in the exposure assessment process. These steps include defining the study population, identifying source and potential routes of exposure, estimating environmental levels of target contaminants, and estimating personal exposures. We present and discuss examples for the first three steps. We discuss potential use of GIS and global positioning systems (GPS) in the last step. On the basis of our findings, we conclude that the use of GIS in exposure assessment for environmental epidemiology studies is not only feasible but can enhance the understanding of the association between contaminants in our environment and disease.

Accuracy of city postal code coordinates as a proxy for location of residence

BACKGROUND: Health studies sometimes rely on postal code location as a proxy for the location of residence. This study compares the postal code location to that of the street address using a database from the Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease (APPROACHCopyright ). Cardiac catheterization cases in an urban Canadian City were used for calendar year 1999. We determined location in meters for both the address (using the City of Calgary Street Network File in ArcView 3.2) and postal code location (using Statistic Canada's Postal Code Conversion File). RESULTS: The distance between the two estimates of location for each case were measured and it was found that 87.9% of the postal code locations were within 200 meters of the true address location (straight line distances) and 96.5% were within 500 meters of the address location (straight line distances). CONCLUSIONS: We conclude that postal code locations are a reasonably accurate proxy for address location. However, there may be research questions for which a more accurate description of location is required.

Positional error in automated geocoding of residential addresses

BACKGROUND: Public health applications using geographic information system (GIS) technology are steadily increasing. Many of these rely on the ability to locate where people live with respect to areas of exposure from environmental contaminants. Automated geocoding is a method used to assign geographic coordinates to an individual based on their street address. This method often relies on street centerline files as a geographic reference. Such a process introduces positional error in the geocoded point. Our study evaluated the positional error caused during automated geocoding of residential addresses and how this error varies between population densities. We also evaluated an alternative method of geocoding using residential property parcel data. RESULTS: Positional error was determined for 3,000 residential addresses using the distance between each geocoded point and its true location as determined with aerial imagery. Error was found to increase as population density decreased. In rural areas of an upstate New York study area, 95 percent of the addresses geocoded to within 2,872 m of their true location. Suburban areas revealed less error where 95 percent of the addresses geocoded to within 421 m. Urban areas demonstrated the least error where 95 percent of the addresses geocoded to within 152 m of their true location. As an alternative to using street centerline files for geocoding, we used residential property parcel points to locate the addresses. In the rural areas, 95 percent of the parcel points were within 195 m of the true location. In suburban areas, this distance was 39 m while in urban areas 95 percent of the parcel points were within 21 m of the true location. CONCLUSION: Researchers need to determine if the level of error caused by a chosen method of geocoding may affect the results of their project. As an alternative method, property data can be used for geocoding addresses if the error caused by traditional methods is found to be unacceptable.