THE INTERACTION OF TOBACCO RETAILER DENSITY AND RURALITY WITH TOBACCO USE PREVALENCE IN KENTUCKY

Abstract

Purpose:

Tobacco retailer density is a key correlate of tobacco use prevalence in the US, but this relationship has typically been examined in urban areas. We examined the relationship between retailer density and tobacco use across Kentucky counties, a state with high smoking rates and a substantial rural population.

Methods:

We obtained a list of tobacco retailers from the 2021 Kentucky Synar program, used for minor’s access inspections. We merged these data with county-level prevalence estimates for cigarette and smokeless tobacco use for each county in Kentucky calculated using pooled 3-year estimates (2017–19) of Kentucky Behavioral Risk Factor Surveillance System data. We modeled county-level frequency of each outcome among adults as a function of tobacco retailer density and the Index of Relative Rurality, and interactions using ANCOVA, controlling for county percent of adult population and percent with an advanced degree.

Findings:

We observed significant interactions between retailer density and rurality for both cigarette and smokeless tobacco use. Post-hoc protected pairwise comparisons demonstrated that within the 40 most urban counties, cigarette and smokeless tobacco use were significantly higher among counties with the highest retailer density, compared to low density counties. Retailer density was not associated with prevalence of cigarette and smokeless tobacco use within the two more rural tertiles of counties.

Conclusions:

Consistent with prior research, policies limiting tobacco retailer density could be most beneficial in urban and urban-adjacent areas, while other policies may need to be considered for rural areas, even within a relatively rural state such as Kentucky.

Introduction

Tobacco use is the leading cause of preventable death in the US.¹ It is readily available through retail environments where people purchase everyday goods such as food, gasoline, and alcohol. Several expert reviews causally link marketing exposure with tobacco use.^{2, 3} Tobacco marketing is also prevalent in the tobacco retail environment.⁴ In 2022, the tobacco industry spent $8.58 billion on cigarette and smokeless tobacco marketing, with most spending occurring in retail environments.^{5, 6} Living in areas with higher tobacco retail density has also been associated with tobacco use across multiple studies.⁷ Additionally, in a recent meta-analysis of 27 studies, reducing exposure to tobacco retailer density was associated with a decreased risk of tobacco use (RRR 2.55 95% CI 1.91 to 3.19).⁸ Retailer density has been hypothesized to influence tobacco use through decreased travel costs, increased cues to use and purchase tobacco, and increased exposure to tobacco marketing where density is higher.⁹ However, fewer studies have examined the impact of retailer density on tobacco use outcomes in rural areas where population density is low.

Retail tobacco marketing strategies in rural areas may also differ from those in urban areas. Rural stores contain more smokeless tobacco marketing and charge lower prices for cigarettes and cigars per unit than urban stores.¹⁰ The rural tobacco marketing environment differentially impacts certain groups, with rural male adolescents reporting more exposure to tobacco marketing than their urban counterparts.¹¹ Tobacco retailer density is commonly calculated by dividing the number of retailers in a given geographic area by the number of residents or, less commonly, by land area. Prior research found that tobacco retailer density per 1000 residents is higher in rural vs. urban communities in Ohio and nationally in non-metro (mean density 1.36) vs. metro areas (mean density 1.11).¹² However, the relationship between tobacco retailer density and tobacco use may differ in rural vs. urban environments. In a national study, higher tobacco retailer density was associated with higher prevalence of cigarette use, but only in metro areas.¹²

To further explore the relationship between tobacco retailer density and tobacco use by rurality we examine this relationship across counties in Kentucky, a state in the top 10 US states by rural population and with one of the highest smoking rates in the US.¹³ The purpose of this study is to evaluate whether tobacco retailer density, relative rurality (a construct examining urbanicity and rurality across a continuum), or their interaction each measured at the county-level is associated with prevalence of cigarette or smokeless tobacco use, controlling for county-level percent of adults in the population and county socioeconomic status.

Materials and Methods

Data sources

Tobacco Retailer Density.

Information on retail locations where tobacco was available for purchase in each Kentucky county during the year 2021 was obtained from the state Synar program for minor’s access inspections. The Synar list includes all tobacco retailers who are included in statewide retail inspections. Because the state of Kentucky does not require tobacco retail licenses, the Synar list is the most comprehensive list readily available in the state. Tobacco retail density was calculated by dividing the number of tobacco retailers by the census-estimated population of each county in 2019, based on US Census estimates available from the Kentucky State Data Center and then multiplying by 1000.¹⁴ This calculation therefore represents the number of tobacco retailers per 1000 residents of each county. In addition to calculating tobacco retailer density by population, we also considered county-level retailer density per square mile to be included in separate models as a sensitivity analysis.

Tobacco Use.

Kentucky Behavioral Risk Factor Surveillance System (BRFSS) telephone survey data from 2017 through 2019 were combined and reweighted to estimated three-year county-level prevalence (%) estimates for adult cigarette smoking and smokeless tobacco use.¹⁵ We used established weighting procedures as recommended by CDC for the BRFSS for reweighting the data by dividing by weights from each year by the number of years included.¹⁶ Estimates for current cigarette smoking were derived from the question, “If you have smoked at least 100 cigarettes in your entire life, are you now smoking every day, some days, or not at all?” Current smoking was defined as those who answered every day or some days and had used at least 100 cigarettes in their lifetime. Current smokeless tobacco use was derived from the question “Do you currently use chewing tobacco, snuff, or snus every day, some days, or not at all?” Current use was defined as using smokeless tobacco some days or every day.¹⁵ These estimates were obtained for each of Kentucky’s 120 counties.

Rurality.

The Index of Relative Rurality, a continuous measure of rurality^{17, 18}, was used to characterize the rurality of each county in Kentucky in 2020, the most recent year available. This measure operationalizes ‘rurality’ as a multifaceted construct encompassing population size, population density, distance from a metropolitan area, and proportion of the area that has been developed or “built-up”. The index ranges from 0 (most urban) to 1 (most rural). These data were obtained from the original authors of this measure and merged with the other county-level data described previously.¹⁹

Data analysis

We summarized the data using means and standard deviations or frequency distributions, depending on the variable. We evaluated unadjusted associations among tobacco retailer density (both the per 1000 population and per square mile), relative rurality, cigarette smoking prevalence, and smokeless tobacco prevalence using correlation; due to non-normal distributions in several variables, we used Spearman’s rank correlation (ρ) throughout. We divided the continuous measures of retailer density per population (per 1000 residents), retailer density per square mile, and relative rurality (Index of Relative Rurality) into tertiles prior to analysis. This was chosen as a strategy since the distribution of retailer density per population is approximately normal, but the distribution of retailer density per square mile is left skewed and the distribution of relative rurality is right skewed, with a greater number of counties in the state being more rural.

Using tertile measures of retailer density and relative rurality, we conducted analyses of covariance (ANCOVA) models for both cigarette and smokeless tobacco use. We included percent of the population aged 18 and above and percent with an advanced degree as indicators of population distribution and county-level socioeconomic status, respectively. The latter variable was correlated with both household income and poverty status, so we only included this measure to avoid multicollinearity. Post-hoc pairwise comparisons were accomplished using Fisher’s least significant difference procedure. As a protection against overall Type I error, we only considered the preplanned pairwise comparisons relative to the interaction terms; since these were significant in both product models with retailers per population as the density measure, the main effects are not interpretable and were not considered further. The preplanned comparisons were the average use within each tertile of relative rurality, compared across the levels of retailer densities. Analysis was done using SAS for Windows, 9.4; and an alpha level of .05 was used throughout to determine statistical significance.

Results

Across the 120 counties in the state, the average retailer density per population was 1.20, with a range from 0.39 to 2.01 (see Table 1). The mean retailer density per square mile was 0.12, with a range from 0.010 and 1.48. Relative rurality varied from 0.19 to 0.59 (most urban to most rural in Kentucky), with an average of 0.51. Average county-level prevalence of cigarette use was 27%, with a range of 8–45%. Finally, smokeless tobacco use prevalence at the county level varied across the state between 0% and 25%, with an average of 9%. Retailer density per population and relative rurality were positively correlated (ρ = 0.31, p < .001), with a greater ratio of stores per population in more rural areas. Conversely, retailer density per square mile and relative rurality were negatively correlated (ρ = −0.89, p < .001), consistent with lower retailer density per square mile in more rural locations. Smoking prevalence was positively correlated with both retailer density per population (ρ = 0.23, p = .011) and greater relative rurality (ρ = 0.28, p = .0019), and negatively correlated with retailer density per square mile (ρ = −0.25, p = .0053). However, retailer density measures and relative rurality were not associated with smokeless tobacco use (ρ = 0.15 with p = .11 and ρ = −0.12 with p = .20 for the retailer densities by population and area, respectively, and ρ = 0.11 with p = .23 for relative rurality). Finally, as expected, there was a positive association between use prevalence estimates for smoking cigarettes and smokeless tobacco (ρ = 0.24, p = .0095).

Table 1.

Descriptive statistics for retailer density, relative rurality, and the use prevalence measures, with bivariate associations evaluated with Spearman’s rank correlation (N = 120 counties).

Variable	Mean (SD)	Median (range)	Correlations: Spearman’s rho (pvalue)
				Retailer density per square mile (2)	Index of relative rurality (IRR) (3)	Smoking prevalence (4)	Smokeless prevalence (5)
Retailer density per population	1.20 (0.33)	1.21 (0.39–2.01)	(1)	<0.10 (>.90)	0.31 (<.001)	0.23 (.011)	0.15 (.11)
Retailer density per square mile	0.12 (0.17)	0.072 (0.010–1.48)	(2)	--	−0.89 (<.001)	−0.25 (.0053)	−0.12 (.20)
Index of relative rurality (IRR)	0.51 (0.056)	0.52 (0.19–0.59)	(3)		--	0.28 (.0019)	0.11 (.23)
Smoking prevalence	0.27 (0.071)	0.27 (0.082–0.45)	(4)			--	0.24 (.0095)
Smokeless prevalence	0.087 (0.048)	0.081 (0.00–0.25)	(5)				--

Each of the tertile variables (for retailer density per population, retailer density per square mile, and relative rurality) included 40 counties within each category of the three-level variable. The chi-square test of association between the density per population and relative rurality tertile indicators was significant overall (Χ² =13.2, p = .010), with the counties having the lowest retailer density per population more frequently categorized in the most urban tertile by relative rurality. At the other extreme, counties with the highest retailer density per population were more likely to be in the more rural areas. However, there were at least 5 counties in each combination of levels for the two tertile indicators. As expected, based on the correlation analysis, the relationship between the tertiles of retailer density per square mile and relative rurality was significant but in the opposite direction (Χ² =96.0, p < .001). Among the 40 counties with the lowest degree of rurality, 32 (80%) were in the highest tertile of retailer density per square mile, while among the 40 with the highest degree of rurality, 32 (80%) were in the lowest tertile of retailer density. Unlike the 3×3 table obtained for the tertiles of density per population versus relative rurality, when retailer density was linked to a measurement of area, two of the cells had no values, namely the most urban and lowest retailer density per square mile and the most rural and the highest retailer density per square mile. This observation may suggest a degree of redundancy between rurality and retail density per square mile that does not exist in Kentucky between rurality and density per population.

Analysis of covariance models.

The ANCOVA model with retailer density per population, relative rurality, and their interaction as factors, and the covariates of percent adults in the population and percent with an advanced degree, was significant overall (F = 4.1, p < .001). In this model, the two-way interaction between tertile of retailers per population and tertile of relative rurality was significant (F = 3.7, p = .008). Post hoc comparisons of average use prevalence percentages demonstrated that within the first tertile of the relative rurality scores (i.e., the most urban counties in the state), the pattern of use varies by tertile of retailer density (see Figure 1). Within the 40 most urban counties in the state, the average smoking rate was 21% in the lowest retailer density areas, 25% in the middle retailer density areas, and 32% in the highest density areas. The overall Kentucky smoking rate for 2019 was 23.6%²⁰ so only those in the most urban counties that also had the lowest retailer density had a lower average smoking prevalence than the state as a whole. While the smoking prevalence between the lowest and middle retailer density counties was not different (p = .082), both groups were significantly lower in cigarette use prevalence from the group of counties with the highest retailer density by population (p < .04 for both comparisons). Within the other two relative rurality tertiles (i.e., moderately rural and most rural), there was no difference in cigarette use prevalence according to tertile of retailer density: all pairwise comparisons in these two higher rurality tertiles were not significant.

Figure 1.

Interaction of Tobacco Retailer Density per population and Relative Rurality on Adult Smoking Prevalence, controlling for percent of population 18+ and percent of population with an advanced degree.

The findings for smokeless tobacco were nearly identical, aside from the lower use prevalence of this product overall. The overall ANCOVA model was significant (F = 2.3, p = .016), as was the two-way interaction (F = 2.7, p = .035). Post hoc pairwise comparisons are summarized in Figure 2. Among the most urban counties in the state, smokeless tobacco use varied systematically according to density of retailers by population. In this most urban tertile, the use prevalence rates were 12%, 7% and 5%, respectively for the densest to least dense retailer areas. The smokeless tobacco use rates for the highest and lowest density counties within this IRR tertile were significantly different from each other (p = .01), but neither differed from the middle density group of counties in the most urban tertile. Within each of the more rural tertiles of counties (i.e., those in the middle and upper tertiles in terms of relative rurality), there were no significant differences in the pairwise comparisons of smokeless tobacco use prevalence, regardless of which retailer density counties were being compared. Percent adults in the population was also significant in this model and had a positive association with smokeless tobacco use; counties with greater percentages of adults in the population tended to have higher smokeless tobacco use rates.

Figure 2.

Interaction of Tobacco Retailer Density Per Population and Relative Rurality on Adult Smokeless Tobacco Use Prevalence, controlling for percent of population 18+ and percent of population with an advanced degree.

Discussion

This study provides novel insight into the relationship between tobacco retailer density and prevalence of cigarette and smokeless tobacco use by relative rurality in Kentucky, a state ranking among the top 10 US states for both tobacco use prevalence and relative percentage of state residents living in rural areas. Similar to prior work, tobacco retail density per 1000 residents was higher in rural areas.¹² Notably, in the bivariate analyses, we found that cigarette smoking prevalence was positively associated with greater retailer density per population and rurality, while smokeless tobacco use prevalence was unrelated to these outcomes; this is a novel finding not examined in prior studies. That greater retail availability of combustible cigarettes is associated with more use is concerning given the elevated health risks associated with cigarette vs. smokeless tobacco use. However, it appears that smokeless tobacco use prevalence overall may be driven by factors other than availability (e.g., social norms).

We also found that the relationship between retail density and tobacco product use prevalence was moderated by rurality. For both cigarette and smokeless tobacco use, urban counties with the highest tobacco retailer density per population had higher population-average use prevalence than urban counties with lower tobacco retailer density, controlling for percent of adults in the population and educational attainment,– highlighting the need for tobacco control policies to target the most retailer-dense areas among urban counties to prevent and reduce tobacco use. However, these differences were not observed for those counties in the second or third tertiles of rurality—that is, the 80 least urban counties—where cigarette and smokeless tobacco use were uniformly high regardless of tobacco retailer density. These findings thus suggest that tobacco retailer density in Kentucky might not be as correlated with rates of cigarette smoking and smokeless tobacco use in rural counties, corroborating previous national literature regarding smoking, but adding new information related to smokeless tobacco use.¹²

Tobacco control research and policy must therefore consider other factors driving high prevalence of use in rural counties. Retailer density might not be as relevant, perhaps due to other factors not assessed here, such as relatively far distances to retail locations regardless of density, contexts such as Kentucky’s tobacco growing history, positive or neutral social norms around tobacco use, or relative lack of comprehensive smokefree policies in these areas relative to urban ones.²¹ In the absence of tobacco retailer licensing, there could also be unmeasured residual confounding in that the comprehensiveness of tobacco retailers on the Synar list could differ between rural and urban environments in unknown ways.

Strengths and Limitations.

This is among the first studies to consider the relationship of tobacco retailer density and relative rurality and cigarettes within a state with a large rural population and high rates of tobacco use. This study is also one of the first to examine tobacco retailer density in relation to smokeless tobacco use. By including educational attainment and percent adults in the population in the models, we demonstrated these differences in tobacco use by retailer density and rural/urban context are robust and not attributable to factors known to predict increased tobacco use (lower education, older population). One limitation is that BRFSS estimates can be less robust due to small sample sizes in rural counties. To mitigate this concern, we used three-year aggregated estimates to pool observations, which narrowed confidence intervals considerably. Additionally, to enhance the comparability of these data over time we selected BRFSS data from pre-COVID years. KY BRFSS data were unavailable at county levels in 2020 and 2021 due to lower response rates and administration changes due to the pandemic. As such, we utilized the most recent data available from the state BRFSS (2017–2019) and Synar inspection lists 2021 at the time of the study. Unfortunately, state Synar lists are updated continuously over time so were not readily available for earlier years. Future analyses can examine such data prospectively to ensure better temporal alignment of data. Finally, as a cross-sectional study we cannot determine causality of these relationships and cannot identify if retailers tend to locate where there are more individuals using tobacco or whether tobacco retailer density is driving increased tobacco use. However, both demand (more individuals using tobacco) and supply (increased tobacco retailer density) factors have shown positive relationships demonstrating that the interplay between retail availability of tobacco and tobacco use are likely complex, reinforcing, and interrelated.²²

Conclusion

Overall tobacco retailer density is an important contributing factor associated with tobacco use risk and this relationship is most salient in urban environments, even in a relatively rural state like Kentucky. Rural communities in Kentucky had high tobacco use regardless of tobacco retail density suggesting the need for interventions and policies beyond retail reduction strategies to reduce geographic disparities in tobacco use.

Table 2:

Analysis of Covariance of the interaction of retailer density by population and rurality on the prevalence of Smoking and smokeless tobacco use among Kentucky counties, 2021

Variable	Smoking Prevalence		Smokeless Tobacco Use Prevalence
	F value	p-value	F value	p-value
Percent Age 18+	1.03	0.31	4.22	0.042
Percent with Advanced Degree	0.64	0.43	0.34	0.56
Retailer Density Per 1000	1.00	0.37	0.61	0.54
Index Relative Rurality 2010	2.04	0.13	2.17	0.12
Retailer Density per pop *Index Relative Rurality	3.67	0.0076 *	2.70	0.035 *

^*Post-hoc analysis based on Fisher’s Least Significant Difference procedure.