Using imputation to provide location information for nongeocoded addresses

The quality of social determinants data in the electronic health record: a systematic review

OBJECTIVE

The aim of this study was to collect and synthesize evidence regarding data quality problems encountered when working with variables related to social determinants of health (SDoH).

MATERIALS AND METHODS

We conducted a systematic review of the literature on social determinants research and data quality and then iteratively identified themes in the literature using a content analysis process.

RESULTS

The most commonly represented quality issue associated with SDoH data is plausibility (n = 31, 41%). Factors related to race and ethnicity have the largest body of literature (n = 40, 53%). The first theme, noted in 62% (n = 47) of articles, is that bias or validity issues often result from data quality problems. The most frequently identified validity issue is misclassification bias (n = 23, 30%). The second theme is that many of the articles suggest methods for mitigating the issues resulting from poor social determinants data quality. We grouped these into 5 suggestions: avoid complete case analysis, impute data, rely on multiple sources, use validated software tools, and select addresses thoughtfully.

DISCUSSION

The type of data quality problem varies depending on the variable, and each problem is associated with particular forms of analytical error. Problems encountered with the quality of SDoH data are rarely distributed randomly. Data from Hispanic patients are more prone to issues with plausibility and misclassification than data from other racial/ethnic groups.

CONCLUSION

Consideration of data quality and evidence-based quality improvement methods may help prevent bias and improve the validity of research conducted with SDoH data.

Uncertainty in geospatial health: challenges and opportunities ahead

PURPOSE

Uncertainty is not always well captured, understood, or modeled properly, and can bias the robustness of complex relationships, such as the association between the environment and public health through exposure, estimates of geographic accessibility and cluster detection, to name a few.

METHODS

We review current challenges and future opportunities as geospatial data and analyses are applied to the field of public health. We are particularly interested in the sources of uncertainty in geospatial data and how this uncertainty may propagate in spatial analysis.

RESULTS

We present opportunities to reduce the magnitude and impact of uncertainty. Specifically, we focus on (1) the use of multiple reference data sources to reduce geocoding errors, (2) the validity of online geocoders and how confidentiality (e.g., HIPAA) may be breached, (3) use of multiple reference data sources to reduce geocoding errors, (4) the impact of geoimputation techniques on travel estimates, (5) residential mobility and how it affects accessibility metrics and clustering, and (6) modeling errors in the American Community Survey. Our paper discusses how to communicate spatial and spatiotemporal uncertainty, and high-performance computing to conduct large amounts of simulations to ultimately increase statistical robustness for studies in public health.

CONCLUSIONS

Our paper contributes to recent efforts to fill in knowledge gaps at the intersection of spatial uncertainty and public health.

Impact of geo-imputation on epidemiologic associations in a study of outdoor air pollution and respiratory hospitalization

Imputation of missing spatial attributes in health records may facilitate linkages to geo-referenced environmental exposures, but few studies have assessed geo-imputation impacts on epidemiologic inference. We imputed patient Census tracts in a case-crossover analysis of fine particulate matter (PM_2.5) and respiratory hospitalizations in New York State (2000-2005). We observed non-significantly higher PM_2.5 exposures, high accuracy of binary exposure assignment (89 to 99%), and marginally different hazard ratios (HRs) (-0.2 to 0.7%). HR differences were greater in urban versus rural areas. Given its efficiency and nominal influence on accuracy of exposure classification and measures of association, geo-imputation is a candidate method to address missing spatial attributes for health studies.

The association of neighborhood-level social class and tobacco consumption with adverse lung cancer characteristics in Maryland

INTRODUCTION

Although both active tobacco use and passive tobacco exposure are well-established as being risk factors for lung cancer, it is challenging to measure tobacco-related exposures at the population level, while considering other factors (gender, race, socioeconomic status) that may modify the relationship between tobacco and lung cancer. Moreover, research to date has focused primarily on relationships between tobacco and endpoints of lung cancer incidence or mortality. Tobacco's role in disease progression, through association with important disease characteristics such as tumor histological type and grade, and stage of disease at diagnosis, has been less well examined.

METHODS

This research examines associations between area-level tobacco use and social class, as well as individual gender, race and age, and three adverse disease characteristics (tumor type, grade and stage) among incident cases of lung cancer reported to the Maryland Cancer Registry in 2000. Cases were geocoded by residential address. Multi-level logistic regression models included Census block group-level estimates of per capita tobacco spending, from Consumer Expenditure Survey data, and a 4-item social class index, from Census estimates of rates of high school graduation, employment, white collar occupation, and per capita income.

RESULTS

Analyses of 3223 cases found no significant differences by race, however, results differed by gender. Lower block-group social class and higher tobacco spending were associated with squamous and small cell histological types and poorly differentiated or undifferentiated tumor grade. However, for later stage at diagnosis (SEER stages 2-7), both higher social class and greater tobacco spending were protective, especially for women, suggesting women in high tobacco use communities may benefit from early detection.

CONCLUSIONS

Results support using area-level behavioral data as tools for identifying high risk communities suitable for more resource-intensive research or interventions. Findings also suggest that area-level social resources are consistent drivers of lung cancer disparities, and merit continued research attention.

Spatiotemporal Analysis of Oklahoma Tobacco Helpline Registrations Using Geoimputation and Joinpoint Analysis

OBJECTIVE

Tobacco quitlines provide free smoking cessation telephone services to smokers interested in quitting tobacco. We aimed to explore spatial and temporal analyses of registrations to the Oklahoma Tobacco Helpline including those of any racial group and American Indians (AI) from January 1, 2006, to June 30, 2017. This will allow tribal and community organizations, such as the Oklahoma Tribal Epidemiology Center, to better implement and evaluate public health prevention efforts at a smaller geographic area using the larger geographic units that are publicly available.

DESIGN

Retrospective, descriptive study.

SETTING

Oklahoma.

PARTICIPANTS

Registrants to the Oklahoma Tobacco Helpline.

MAIN OUTCOME MEASURES

To evaluate the spatial distribution of Helpline participants using geoimputation methods and evaluate the presence of time trends measured through annual percent change (APC).

RESULTS

We observed increased density of participants in the major population centers, Oklahoma City and Tulsa. Density of AI registrations was higher in the rural areas of Oklahoma where there is a larger tribal presence compared with participants of any racial group. For all racial groups combined, we identified 3 significant trends increasing from July 2008 to March 2009 (APC: 10.9, 95% confidence interval [CI], 0.8-21.9), decreasing from March 2009 to May 2014 (APC: -0.8, 95% CI: -1.1 to -0.4), and increasing from May 2014 to June 2017 (APC: 0.8, 95% CI: 0.0-1.6). The number of AI registrations to the Helpline increased significantly from July 2008 to March 2009 (APC: 12.0, 95% CI: 2.0-22.9) and decreased from March 2009 to June 2014 (APC: -0.7, 95% CI: -1.0 to -0.3).

CONCLUSIONS

Results of this project will allow the Helpline to efficiently identify geographic areas to increase registrations and reduce commercial tobacco use among the AI population in Oklahoma through existing programs at the Oklahoma Tribal Epidemiology Center.

Geographic Imputation of Missing Activity Space Data from Ecological Momentary Assessment (EMA) GPS Positions

This research presents a pilot study to develop and compare methods of geographic imputation for estimating the location of missing activity space data collected using geographic ecological momentary assessment (GEMA). As a demonstration, we use data from a previously published analysis of the effect of neighborhood disadvantage, captured at the U.S. Census Bureau tract level, on momentary psychological stress among a sample of 137 urban adolescents. We investigate the impact of listwise deletion on model results and test two geographic imputation techniques adapted for activity space data from hot deck and centroid imputation approaches. Our results indicate that listwise deletion can bias estimates of place effects on health, and that these impacts are mitigated by the use of geographic imputation, particularly regarding inflation of the standard errors. These geographic imputation techniques may be extended in future research by incorporating approaches from the non-spatial imputation literature as well as from conventional geographic imputation and spatial interpolation research that focus on non-activity space data.

Evaluation of geoimputation strategies in a large case study

BACKGROUND

Health data usually has missing or incomplete location information, which impacts the quality of research. Geoimputation methods are used by health professionals to increase the spatial resolution of address information for more accurate analyses. The objective of this study was to evaluate geo-imputation methods with respect to the demographic and spatial characteristics of the data.

METHODS

We evaluated four geoimputation methods for increasing spatial resolution of records with known locational information at a coarse level. In order to test and rigorously evaluate two stochastic and two deterministic strategies, we used the Texas Sex Offender registry database with over 50,000 records with known demographic and coordinate information. We reduced the spatial resolution of each record to a census block group and attempted to recover coordinate information using the four strategies. We rigorously evaluated the results in terms of the error distance between the original coordinates and recovered coordinates by studying the results by demographic sub groups and the characteristics of the underlying geography.

RESULTS

We observed that in estimating the actual location of a case, the weighted mean method is the most superior for each demographic group followed by the maximum imputation centroid, the random point in matching sub-geographies and the random point in all sub-geographies methods. Higher accuracies were observed for minority populations because minorities tend to cluster in certain neighborhoods, which makes it easier to impute their location. Results are greatly affected by the population density of the underlying geographies. We observed high accuracies in high population density areas, which often exist within smaller census blocks, which makes the search space smaller. Similarly, mapping geoimputation accuracies in a spatially explicit manner reveals that metropolitan areas yield higher accuracy results.

CONCLUSIONS

Based on gains in standard error, reduction in mean error and validation results, we conclude that characteristics of the estimated records such as the demographic profile and population density information provide a measure of certainty of geographic imputation.

A geographic information system-based method for estimating cancer rates in non-census defined geographical areas

PURPOSE

To address locally relevant cancer-related health issues, health departments frequently need data beyond that contained in standard census area-based statistics. We describe a geographic information system-based method for calculating age-standardized cancer incidence rates in non-census defined geographical areas using publically available data.

METHODS

Aggregated records of cancer cases diagnosed from 2009 through 2013 in each of Chicago's 77 census-defined community areas were obtained from the Illinois State Cancer Registry. Areal interpolation through dasymetric mapping of census blocks was used to redistribute populations and case counts from community areas to Chicago's 50 politically defined aldermanic wards, and ward-level age-standardized 5-year cumulative incidence rates were calculated.

RESULTS

Potential errors in redistributing populations between geographies were limited to <1.5% of the total population, and agreement between our ward population estimates and those from a frequently cited reference set of estimates was high (Pearson correlation r = 0.99, mean difference = -4 persons). A map overlay of safety-net primary care clinic locations and ward-level incidence rates for advanced-staged cancers revealed potential pathways for prevention.

CONCLUSIONS

Areal interpolation through dasymetric mapping can estimate cancer rates in non-census defined geographies. This can address gaps in local cancer-related health data, inform health resource advocacy, and guide community-centered cancer prevention and control.

Neighborhood Factors and Fall-Related Injuries among Older Adults Seen by Emergency Medical Service Providers

Falls are serious health problems among older adults, and are the leading cause of fatal and nonfatal injuries treated by emergency medical services (EMS). Although considerable research has examined the risk factors of falls at the individual level, relatively few studies have addressed the risk factors at the neighborhood level. This study examines the characteristics of neighborhood environments associated with fall injuries reported to EMS providers. A total of 13,163 EMS records from 2011 to 2014 involving adults aged 65 and older in the city of San Antonio (TX, USA) were analyzed at the census tract level (n = 264). Negative binomial regression was used to identify significant census tract-based neighborhood environmental variables associated with the count of fall injuries in each census tract. Adjusting for exposure variable and the size of the census tract, neighborhoods with higher residential stability, captured as the percent of those who lived in the same house as the previous year were associated with decreased count of fall injuries. Neighborhoods with higher residential density and having a higher vacancy rate were associated with increased count of fall injuries. The study highlights the importance of stable and safe neighborhoods in reducing fall risks among older adults, which should be considered a prerequisite for promoting age-friendly environments.

Using Small-Area Estimation to Calculate the Prevalence of Smoking by Subcounty Geographic Areas in King County, Washington, Behavioral Risk Factor Surveillance System, 2009-2013

Introduction

King County, Washington, fares well overall in many health indicators. However, county-level data mask disparities among subcounty areas. For disparity-focused assessment, a demand exists for examining health data at subcounty levels such as census tracts and King County health reporting areas (HRAs).

Methods

We added a “nearest intersection” question to the Behavioral Risk Factor Surveillance System (BRFSS) and geocoded the data for subcounty geographic areas, including census tracts. To overcome small sample size at the census tract level, we used hierarchical Bayesian models to obtain smoothed estimates in cigarette smoking rates at the census tract and HRA levels. We also used multiple imputation to adjust for missing values in census tracts.

Results

Direct estimation of adult smoking rates at the census tract level ranged from 0% to 56% with a median of 10%. The 90% confidence interval (CI) half-width for census tract with nonzero rates ranged from 1 percentage point to 37 percentage points with a median of 13 percentage points. The smoothed-multiple–imputation rates ranged from 5% to 28% with a median of 12%. The 90% CI half-width ranged from 4 percentage points to 13 percentage points with a median of 8 percentage points.

Conclusion

The nearest intersection question in the BRFSS provided geocoded data at subcounty levels. The Bayesian model provided estimation with improved precision at the census tract and HRA levels. Multiple imputation can be used to account for missing geographic data. Small-area estimation, which has been used for King County public health programs, has increasingly become a useful tool to meet the demand of presenting data at more granular levels.

The association of area-level social class and tobacco use with adverse breast cancer characteristics among white and black women: evidence from Maryland, 1992-2003

Background

In breast cancer, worse disease characteristics are associated with fewer social resources and black race. However, it is unknown whether social gradients have similar impact across race, and whether behaviors, including tobacco use, may explain a portion of the social gradient.

Methods

We modeled relationships between area-level social class, tobacco spending and tumor characteristics, using 50,062 white and black cases diagnosed from 1992–2003 in Maryland, a racially and economically diverse state on the east coast of the United States. Multi-level models estimated the effect of area-level social class and tobacco consumption on tumor grade, size, and stage at diagnosis.

Results

Adjusting for race, age and year of diagnosis, higher social class was associated with lower risk for tumors with histological grade 3 or 4 (O.R. 0.96, 95% C.I. 0.94,0.99), those diagnosed at SEER stage 2 or later (O.R. 0.89, 95% C.I. 0.86, 0.91), and tumor size >2 cm (O.R. 0.87, 95% C.I. 0.84, 0.90). Higher tobacco spending was associated with higher risk for higher grade (O.R. 1.01, 1.00, 1.03) and larger tumors (O.R. 1.03, 95% C.I. 1.01, 1.06), but was not statistically significantly related to later stage (O.R. 1.00, 95% C.I. 0.98, 1.02). Social class was less protective for black women, but tobacco effects were not race-specific.

Conclusions

Results suggest that in one U.S. geographic area, there is a differential protection from social class for black and white women, supporting use of intersectionality theory in breast cancer disparities investigations. Area-level tobacco consumption may capture cases’ direct use and second hand smoke exposure, but also may identify neighborhoods with excess cancer-related behavioral or environmental exposures, beyond those measured by social class. Given the growing global burden of both tobacco addiction and aggressive breast cancer, similar investigations across diverse geographic areas are warranted.

Inaccuracy, uncertainty and the space-time permutation scan statistic

The space-time permutation scan statistic (STPSS) is designed to identify hot (and cool) spots of space-time interaction within patterns of spatio-temporal events. While the method has been adopted widely in practice, there has been little consideration of the effect inaccurate and/or incomplete input data may have on its results. Given the pervasiveness of inaccuracy, uncertainty and incompleteness within spatio-temporal datasets and the popularity of the method, this issue warrants further investigation. Here, a series of simulation experiments using both synthetic and real-world data are carried out to better understand how deficiencies in the spatial and temporal accuracy as well as the completeness of the input data may affect results of the STPSS. The findings, while specific to the parameters employed here, reveal a surprising robustness of the method's results in the face of these deficiencies. As expected, the experiments illustrate that greater degradation of input data quality leads to greater variability in the results. Additionally, they show that weaker signals of space-time interaction are those most affected by the introduced deficiencies. However, in stark contrast to previous investigations into the impact of these input data problems on global tests of space-time interaction, this local metric is revealed to be only minimally affected by the degree of inaccuracy and incompleteness introduced in these experiments.

Geostatistical analysis of health data with different levels of spatial aggregation

This paper presents a geostatistical approach to combine two geographical sets of area-based data into the mapping of disease risk, with an application to the rate of prostate cancer late-stage diagnosis in North Florida. This methodology is used to combine individual-level data assigned to census tracts for confidentiality reasons with individual-level data that were allocated to ZIP codes because of incomplete geocoding. This form of binomial kriging, which accounts for the population size and shape of each geographical unit, can generate choropleth or isopleth risk maps that are all coherent through spatial aggregation. Incorporation of both types of areal data reduces the loss of information associated with incomplete geocoding, leading to maps of risk estimates that are globally less smooth and with smaller prediction error variance.