Geographic data provide opportunities to conduct public health surveillance, analyze differences in risk factors and health outcomes, and evaluate and plan public health interventions.1–4 Improvements in scientific methodologies and the widespread availability of Geographic Information System technology have allowed public health researchers to examine spatial data to understand geographic variation in health behaviors and outcomes. Researchers use address-based geocoding to precisely examine potential geographic clustering and explore associations between environmental exposures and health outcomes.5,6 However, some addresses cannot be geocoded or are imprecisely coded (eg, due to postal code or rural route-based addresses). Furthermore, privacy concerns have limited the availability of public data, and data related to small population subgroups are often available only at larger geographical units, such as county-level.7 The potential of spatiotemporal analysis is limited when data are solely based on large geographical units resulting in spatial aggregation and elimination of small area differences and abnormalities. With the aim of increasing the spatial resolution of the data, this study implements a random point allocation geoimputation method in 2 steps to generate data points from county-based event data to the smaller census tract units. The results of this geoimputation are then applied to evaluate the geographic distribution of Oklahoma Tobacco Helpline (referred to from now on as Helpline) registrations in Oklahoma.
To explore this methodology, we examined the geographic variation of Helpline registrations, with a special emphasis on American Indian (AI) participants, using the geoimputed data. We also evaluated temporal trends of Helpline registration using the original data provided by the Helpline. The State of Oklahoma has consistently had a higher smoking prevalence than many other states in the nation and AIs have a significantly higher smoking prevalence than whites.8,9 Oklahoma has invested considerable resources toward tobacco control in an effort to reduce the substantial burden of smoking-related morbidity and mortality. The Oklahoma Tobacco Settlement Endowment Trust established the Helpline in 2003 to provide free smoking cessation telephone services to any Oklahoman interested in quitting tobacco.10 Tobacco quitlines are proven to be an effective strategy for tobacco cessation,11–19 and the Helpline has ranked in the top 20% of all state quitlines for reach, investment, and quit rates since the North American Quitline Consortium began its benchmarking activities.20 Although the Helpline reach and quit rates fluctuate slightly over time, the Helpline consistently reaches about 4% of tobacco users in the entire state.20 Approximately 35% of Helpline registrants report 30-day abstinence, based on a follow-up survey of a sample of registrants 7 months postintervention.10
Our initial study examining the effectiveness of the Helpline in serving AI tobacco users found that the Helpline was equally effective for AIs and that utilization, satisfaction, quit success, and reach were comparable with whites in Oklahoma. However, it is unclear whether the effectiveness is geographically consistent across the entire state and whether rural AIs utilize the Helpline or have equal success compared with urban AIs.10 Oklahoma has a high proportion of the population residing in rural areas, and rural residence is often associated with an increased risk of adverse health behaviors, morbidity, and mortality compared with urban residence.21,22 Individuals in rural areas often have a higher smoking prevalence, lower socioeconomic status (SES), reduced access to health care, and worse health outcomes than those in urban areas.21
Evaluations of public health programs and services are often being challenged when population data are available only at the county level. The Helpline routinely reports call volume, registrations, and services received for population subgroups of interest at the county level. However, county-level information is not conducive to implementing and interpreting sophisticated spatial analyses. Improving the spatial resolution of data with coarse resolution can be achieved using geoimputation methods.5,23–27 These methods utilize additional information to assign imprecise observations (such as at the county level) to more precise prediction (such as at the census tract level) using the underlying population. Such methods incorporate random processes, deterministic computations, or a hybrid of the 2 to predict higher spatial resolution information. For example, a deterministic geoimputation algorithm might assign the record to the census tract with the highest population or it might assign the record to the geographic center of the census tracts weighted by the population. A random geoimputation algorithm on the contrary might assign the records on the basis of a random function weighted by the populations of the census blocks. Geoimputation techniques often make use of secondary data, such as population or subsets of the population (including a combination of age, gender, and race), to improve the prediction.
Our goal for this study was 2-fold: first, we describe an important tool for understanding relatively large area ecological data; second, we examine differences at a finer spatial resolution in Helpline utilization (census tract) and temporal trends of registrations among all racial groups and AIs in Oklahoma. We aimed to explore this methodology (1) to better evaluate public health prevention efforts at a smaller geographic area using the larger geographic units that are publicly available and (2) to provide tribal and community organizations, such as the Oklahoma Tribal Epidemiology Center (OKTEC), with more precise information on the geographic distribution of Helpline use for AIs in Oklahoma.
Data sources and study population
Helpline data are collected by the Helpline provider, Optum Wellbeing, using an initial intake survey conducted at registration among tobacco users who call the 1-800-QUIT NOW number, register for services online, or are referred to the Helpline via fax or electronic referral by a community or health care provider. Registration captures demographic variables, which include self-reported race and ethnicity, in addition to baseline information related to tobacco use and motivation to quit. These data are provided to The University of Oklahoma Health Sciences Center monthly and used to conduct the evaluation of the Helpline, which includes utilization patterns and outcomes. For this study, we used the total number of adult tobacco users registering for Helpline services by county and month for all participants and those self-identifying as AIs. All tobacco users aged 18 years or older who registered for Helpline services from January 1, 2006, to June 30, 2017, were included in the totals. We included those of any racial group to evaluate overall trends of the Helpline and separately analyzed those identifying as AI to better understand trends for this unique population. The study was reviewed by the Institutional Review Board of The University of Oklahoma Health Sciences Center.
The geoimputation method in this study uses stochastic processes. First, we assigned a registration made in a county to a census tract on the basis of the ratio of the population of each tract within the county. Tracts with a larger population within a county had a proportionately higher probability of having the participant assigned. Second, we generated a random point within the census tract and assigned the participant to that point. We conducted the geoimputation in 3 steps for each Helpline registration using Visual Basic for Applications and Python scripts. These are (1) data initialization, (2) geoimputation to the census tract level, and (3) geoimputation to a specific point.
Data initialization involved reorganization of the records so that the data set can be read and processed by the actual geoimputation algorithm. Dimensions of the original tables are 77 records for each county by 138 columns of month and year, representing 11.5 years. The data in these tables denoted the number of registrations made in each county in each month. Data for each month in a separate column are not ideal for the geoimputation algorithm. A Visual Basic for Applications script was written to append the data into new records (ie, for every month), so that instead of the multitude of columns, we obtained 3 columns: county, number of Helpline registrations and the date (month and year), and 10 626 records (which is the multiplication of 77 counties and 138 months).
After the data were initialized, the first geoimputation operations were executed by a Python script. This step involved geoimputation of the county-level information to the census tract level and included extraction of the fraction of the population from each census tract within the entire county (as shown in Table 1a). Next, the Python script calculated cumulative probabilities, ranging from 0 to 1 (as shown in Table 1a) and assigned a random number, ranging from 0 to 1 (Table 1b) to each record. Finally, the script selected the census tract with a range that encompassed the random number (as shown in Table 1b).
Following the assignment of records to census tracts, the third step of the geoimputation involved increasing the resolution to a random point in the selected census tract. We assigned the X and Y coordinates within each selected census tract by a random point generation using the CreateRandomPoints_management function in the arcpy library within the same Python script. Selecting a random point as opposed to a centroid avoids generating artificial clusters.5
The same procedures were repeated for the AI population using the additional Census information to identify the proportion of AI in each census tract from the 2010 Tiger Maps.28 As a result, the script produced 35 942 data points for AI registrants across the State of Oklahoma and 327 987 data points for registrations from all racial groups. These points were analyzed in the subsequent spatial analyses.
Point densities were calculated on a 1 by-1 mile raster grid using the results of the geoimputation from January 2006 through June 2017 for registrations from all racial groups combined and AI registrants only using ArcGIS Pro (version 2.0; ESRI, Redlands, California). All geoimputed locations that fall within each grid are totaled and divided by the area of the neighborhood. This method allows examining overall geographical patterns of point distributions through aggregations.
Time trends analysis
In a separate analysis using the original Helpline data regardless of geographic location (ie, not including the geoimputed data), we used the Joinpoint Regression Program (version 184.108.40.206; National Cancer Institute, Bethesda, Maryland) to analyze monthly time trends among registrations to the Helpline from January 2006 through June 2017. Trends in Helpline registrations over time were characterized using annual percent change (APC). If the slope of the trend was significantly different from zero, an APC estimate was reported as increased or decreased; however, if there was no statistically significant difference, the trend was reported as stable. To account for transition in trends, we also estimated the average annual percent change (AAPC) for the prespecified fixed interval (January 2006 through June 2017).
We analyzed data on 327 987 registrations to the Helpline and 11% of these were AI (N = 35 942). Registrations originated from all 77 Oklahoma counties (all racial groups: minimum = 82 registrations, maximum = 64 839 per county; AI: minimum = 1, maximum = 5428 registrations per county). Registrations to the Helpline fluctuate over time, with peaks frequently observed with new products, services, media campaigns, and Federal/State tax increases (Figure 1). While registrations from AI are on a smaller scale, the percentage of AI registrations is approximate to the percentage of AIs in Oklahoma (13% alone or in combination with other racial groups).29 Although the fluctuations in Helpline usage are less pronounced among AI compared with all racial groups, similar trends are observed compared with the total number of registrants.
Our analyses generated 2 point-density maps for (a) registrations from all racial groups and (b) the AI registrations from January 2006 through June 2017. In our analyses of point density for all racial groups (Figure 2), we observed higher density in the major urban areas of Oklahoma City and Tulsa and slightly less density in suburban and other medium-sized cities. For AIs, we observed a denser and larger distribution around the Tulsa area, followed by Oklahoma City, which is the opposite of what we observed for those of any racial group, though the differences are subtle (Figure 3). In addition, we observed a higher number of registrations in areas where there is a large tribal presence in the eastern, south-central, and east-central Oklahoma for AIs.
Time trends analysis
The results from the Joinpoint analysis are presented in Table 2 (see Supplemental Digital Content Figures S1 and S2, available at http://links.lww.com/JPHMP/A571 and http://links.lww.com/JPHMP/A572). Among all races, the AAPC for number of registrations for all racial groups over the entire study period was 0.5 (95% confidence interval [CI]: −1.8-2.8) and was also 0.5 for AI registrations (95% CI: −1.7 to 2.8), though this was not a significant change. For all racial groups combined, we identified 6 joinpoint trends, with the first 3 being stable from January 2006 to January 2008 (APC: −1.4; 95% CI: −2.8 to 0.1), from January 2008 to April 2008 (APC: 59.2; 95% CI: −21.8 to 224.0), and from April 2008 to July 2008 (APC: −29.2; 95% CI: −65.2 to 44.0). However, in later years, we identified 3 significant trends increasing from July 2008 to March 2009 (APC: 10.9, 95% 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: 0.03-1.6).
For AI registrants to the Helpline, we also identified 6 joinpoint trends, which were stable from January 2006 to January 2008 (APC: −1.0; 95% CI: −2.4 to 0.5), January 2008 to April 2008 (APC: 48.8, 95% CI: −25.9 to 199.0), April 2008 to July 2008 (APC: −29.2, 95% CI: −64.7 to 42.3), and June 2014 to June 2017 (APC: 0.7; 95% CI: −0.04 to 1.5). However, the number of AI registrations 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).
Discussion and Conclusion
The number of Helpline registrations fluctuated on the basis of many factors including tax increases, promotions, and media campaigns. For those of any racial group and AIs, we observed clustering in the primary population centers of Oklahoma City and Tulsa. For AIs, registrations also originated from tribal areas of Oklahoma, which are located outside of the metropolitan areas. In our temporal analysis, we observed a significant increase in Helpline use for AIs and all racial groups from July 2008 to March 2009, which is likely due to the spike in registrations in anticipation of the Federal tax increase that was implemented on April 1, 2009. We also observed an increase for all racial groups from May 2014 to June 2017, potentially related to expanded services during this period.30
We expect that total registrations would be higher than that of AIs due to the lower percentage of AI population in Oklahoma than that of the total population. American Indian race alone or in combination with other racial groups is approximately 13% of the population of Oklahoma and the percentage of AI participants was 11%.29 However, because of the higher rates of smoking among AIs, we would expect to observe a higher proportion of registrants who are AI than the overall population. Although the majority of the population resides in urban areas, there is a higher percentage of AIs in rural areas (12.3%) than in urban areas (6.7%)22; thus, efforts to reduce commercial tobacco use among AIs should include both urban and rural programs. Many tribes in Oklahoma have long-standing efforts to promote the Helpline, in addition to the use of the fax or electronic referral system through many tribal health systems, which may have increased the number of registrations outside of the metropolitan areas.
The prevalence of smoking among AIs still remains the highest of any racial group at 24% in Oklahoma and 31% nationally, though this may be under or overestimated because of the underrepresentation of AIs in surveys and both geographic and cultural differences.8,31 With 38 federally recognized tribal nations, Oklahoma has the second largest population of AIs in the nation.32 One of the goals of the Oklahoma tobacco control program is to reduce the health disparities related to tobacco use among the AI population. Tobacco use among AIs remains a complex issue due to the traditional and ceremonial uses of tobacco in many tribal nations. Steps have been taken to specifically promote the Helpline to AIs and to ensure that quit coaches are aware of the differences between traditional and commercial tobacco use. To evaluate the effectiveness of the Helpline for AIs, we previously compared the utilization patterns, satisfaction, quit outcomes, and reach for AI and white tobacco users across the state from July 1, 2010, to June 20, 2013.10 In that study, we examined more than 11 000 AIs who used Helpline services, which represented nearly 11% of registrants. Although the quit rate at 7-month follow-up for AIs (31.7%) was slightly lower than that for whites (36.5%), the difference was not statistically significant and the quit rates for both racial groups exceeded the North American Quitline Consortium benchmark of 30%.10 The study also found that the Helpline reached 3.5% of AI tobacco users in the state, compared with 3.3% of white tobacco users.10
A recent study evaluated socioeconomic disparities in long-term treatment outcomes in Arkansas, though this did not include results for AI.33 The authors observed that among quitline users from 2005 to 2008, which included a highly motivated group of smokers of low to middle SES (measured through a composite of education and income), the odds of abstinence for the highest SES participants were significantly greater (odds ratio: 1.75; 95% CI: 1.44-2.13) than those for the lowest SES participants. The authors noted that while these callers were highly motivated, they were highly dependent on nicotine and had a long smoking history. In addition, callers of the lowest SES more frequently worked in settings that allowed smoking and were more frequently referred to quitline services by a health care provider instead of media compared with those of higher SES. This emphasizes the need to work closely with populations who may be more susceptible to smoking relapse. Furthermore, quitlines should continue targeting services to reach those who may access only quitline services through a health care provider, which may be infrequent for those of the lowest SES.
In summary, we conducted geoimputation based on county and race (AI and all racial groups combined) since we did not have access to higher-resolution geographic data to estimate Helpline registrants at the census tract level. This method allowed us to explore the spatial distribution and provided a general understanding of the location of registrants for programmatic evaluation and planning purposes. We consider the results extracted from county-level information to be plausible in the context of the distribution of the actual population. Additional information (such as the age, gender, education, and income), elimination of areas outside residential areas, and identification of tribal communities could be employed for more accurate geoimputation and will be incorporated in future analyses.
While we assumed that the population of registrants was distributed randomly in population centers (based on census tract population), there could be misclassification, which may have resulted in overestimated point density clusters in the metropolitan areas but underestimated density in rural areas. However, this was an initial analysis and will allow future studies to better target geographic locations for further analysis or intervention. Because we had limited information from Helpline and US Census data, we were able to incorporate only the distribution of the population by race in AIs and all racial groups combined.
It would be instructive to validate the geoimputation results using a sample of actual addresses, which is planned for future work. However, Hibbert et al23 used the same method, finer geography (ZIP areas), and validated their results using actual addresses, concluding that the geoimputation yielded accurate results depending on the underlying geography and population. Curriero et al24 used a range of strategies and levels of address-related information to geographically impute addresses at ZIP code level with reasonably acceptable outcomes and observed that the more detail in terms of the additional information an address has, the more accurate the results are. Dilekli et al26 utilized similar methods to conduct a sensitivity analysis using actual addresses at census block group, census tract, and county level. The authors observed that the geoimputation methods are useful across the geographic units, with increasing error terms with the larger geographical units. The observed error ranged (depending on the underlying population density) between approximately 10 000 m and approximately 20 000 m using county-level units when incorporating the distribution of the population density, as in the current study. Therefore, we can expect reasonably strong geoimputation performance, with larger error terms, even in the absence of the validation. In future studies, we first plan to add additional demographic information to get more accurate results and then to conduct sensitivity analyses using different spatial units such as ZIP code areas and multiple imputation to evaluate variability of each imputation result using a sample of full addresses, as done in the study by Dilekli et al.26
The OKTEC is committed to decreasing the burden of tobacco-related diseases experienced by the 43 federally recognized tribes in Oklahoma, Kansas, and Texas. The Helpline is a vital public health program for smoking cessation that does not depend on access to transportation or insurance coverage and is completely free. The information gathered from this study will serve as a spotlight for the OKTEC to focus its tobacco control efforts. This analysis will also serve as evidence to tribal decision makers of the need for the increased use of effective health communication interventions. The data highlight the importance of further examination of the differences in the commercial tobacco policies and the amount of exposure to Helpline messaging among the tribal nations in these rural areas.
In conclusion, we observed both spatial and temporal patterns in the Helpline data. Future studies should incorporate more precise geographic information to efficiently evaluate the accuracy of the geoimputation and provide more detailed spatial information to further target services and address gaps in the Helpline usage. The results will inform the Helpline, OKTEC, and Oklahoma tribes about the spatial and temporal distribution of Helpline registrations and have the potential to improve services and ultimately reduce the burden of tobacco among all Oklahomans.
Implications for Policy & Practice
- The results of this project will help the Oklahoma Tobacco Helpline (Helpline) identify areas of enhanced programing to increase registrations and reduce commercial tobacco use in Oklahoma, including the AI population where smoking prevalence is higher.
- Based on the observed spatial and temporal trends and the evaluation of past polices and events, the reach of the Helpline can be improved by addressing gaps.
- These results will allow the Oklahoma Tobacco Settlement Endowment Trust and the Oklahoma Area Tribal Epidemiology Center (OKTEC) to better work with tribal partners to enhance current programs related to commercial tobacco cessation and target geographical and contextual areas of need, including existing programs through the OKTEC.
- This information will serve as a spotlight for the OKTEC to focus its tobacco cessation efforts. This analysis will also serve as evidence to tribal decision makers of the need for the increased use of effective health communication interventions.
1. Candace IJN, Laura MF. Geographic Information Systems (GIS) for health promotion and public health: a review. Health Promot Pract. 2009;12(1):63–73.
2. Goovaerts P. Geostatistical analysis of health data with different levels of spatial aggregation. Spat Spatiotemporal Epidemiol. 2012;3(1):83–92.
3. Schootman M, Gomez SL, Henry KA, et al Geospatial approaches to cancer control and population sciences. Cancer Epidemiol Biomarkers Prev. 2017;26(4):472–475.
4. Krieger N. Follow the North Star: why space, place, and power matter for geospatial approaches to cancer control and health equity. Cancer Epidemiol Biomarkers Prev. 2017;26(4):476–479.
5. Henry KA, Boscoe FP. Estimating the accuracy of geographical imputation. Int J Health Geogr. 2008;7(1):3.
6. Clendenning M, Baze ME, Sun S, et al Origins and prevalence of the American Founder Mutation of MSH2. Cancer Res. 2008;68(7):2145–2153.
7. John MA, Ian MS, Lars V. Disclosure Limitation and Confidentiality Protection in Linked Data: Working Papers 18-07. Washington, DC: Center for Economic Studies, US Census Bureau; 2018.
8. Jamal A, Phillips E, Gentzke AS, et al Current cigarette smoking among adults—United States, 2016. MMWR Morb Mortal Wkly Rep. 2018;67(2):53–59.
9. Odani S, Armour BS, Graffunder CM, Willis G, Hartman AM, Agaku IT. State-specific prevalence of tobacco product use among adults—United States, 2014-2015. MMWR Morb Mortal Wkly Rep. 2018;67(3):97–102.
10. Martinez SA, Beebe LA, Campbell JE. Oklahoma
tobacco helpline utilization and cessation among American Indians
. Am J Prev Med. 2015;48(suppl 1):S47–S53.
11. Hollis JF, McAfee TA, Fellows JL, Zbikowski SM, Stark M, Riedlinger K. The effectiveness and cost effectiveness of telephone counselling and the nicotine patch in a state tobacco quitline. Tobacco Control. 2007;16(suppl 1):i53–i59.
12. Zhu SH, Anderson CM, Tedeschi GJ, et al Evidence of real-world effectiveness of a telephone quitline for smokers. N Engl J Med. 2002;347(14):1087–1093.
13. An LC, Schillo BA, Kavanaugh AM, et al Increased reach and effectiveness of a statewide tobacco quitline after the addition of access to free nicotine replacement therapy. Tob Control. 2006;15(4):286–293.
14. Tinkelman D, Wilson SM, Willett J, Sweeney CT. Offering free NRT through a tobacco quitline: impact on utilisation and quit rates. Tob Control. 2007;16(suppl 1):i42–i46.
15. Ossip-Klein DJ, McIntosh S. Quitlines in North America
: evidence base and applications. Am J Med Sci. 2003;326(4):201–205.
16. Lichtenstein E, Zhu SH, Tedeschi GJ. Smoking cessation quitlines: an underrecognized intervention success story. Am Psychol. 2010;65(4):252–261.
17. Stead LF, Perera R, Lancaster T. A systematic review of interventions for smokers who contact quitlines. Tob Control. 2007;16(suppl 1):i3–i8.
18. Lichtenstein E, Glasgow RE, Lando HA, Ossip-Klein DJ, Boles SM. Telephone counseling for smoking cessation: rationales and meta-analytic review of evidence. Health Educ Res. 1996;11(2):243–257.
19. Cummins SE, Bailey L, Campbell S, Koon-Kirby C, Zhu SH. Tobacco cessation quitlines in North America
: a descriptive study. Tob Control. 2007;16(suppl 1):i9–i15.
21. Smith KB, Humphreys JS, Wilson MG. Addressing the health disadvantage of rural populations: how does epidemiological evidence inform rural health policies and research? Aust J Rural Health. 2008;16(2):56–66.
23. Hibbert JD, Liese AD, Lawson A, et al Evaluating geographic imputation approaches for zip code level data: an application to a study of pediatric diabetes. Int J Health Geogr. 2009;8:54.
24. Curriero FC, Kulldorff M, Boscoe FP, Klassen AC. Using imputation to provide location information for nongeocoded addresses. PLoS One. 2010;5(2):e8998.
25. Walter SR, Rose N. Random property allocation: a novel geographic imputation procedure based on a complete geocoded address file. Spat Spatiotemporal Epidemiol. 2013;6:7–16.
26. Dilekli N, Janitz AE, Campbell JE, de Beurs KM. Evaluation of geoimputation strategies in a large case study. Int J Health Geogr. 2018;17(1):30.
27. Cooper MT, Campbell J, Dilekli N, Darden PM. Access to care for children under five in Oklahoma
: a geographic imputation application. J Oklahoma
State Med Assoc. 2018;111(8):784–789.
30. Centers for Disease Control and Prevention. Federal and state cigarette excise taxes—United States, 1995-2009. MMWR Morb Mortal Wkly Rep. 2009;58(19):524–527.
State Department of Health. Behavioral risk factor surveillance system 2016 on Oklahoma
statistics available for everyone (OK2SHARE). http://www.health.state.ok.us/scripts/broker.exe
. Published 2016. Accessed December 20, 2018.
32. United States Census Bureau. US Census Bureau, 2016 American community survey. http://factfinder.census.gov/
. Published 2018. Accessed March 12, 2018.
33. Varghese M, Sheffer C, Stitzer M, Landes R, Brackman SL, Munn T. Socioeconomic disparities in telephone-based treatment of tobacco dependence. Am J Public Health. 2014;104(8):e76–e84.
American Indians; North America; Oklahoma; spatial analysis; temporal analysis; tobacco use cessation
Supplemental Digital Content
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.