Community pharmacies and addictive products: sociodemographic predictors of accessibility from a mixed GWR perspective

Abstract

Community pharmacies selling potentially harmful products may contradict their role in health promotion. From a spatial analysis perspective, this study investigated the sale of alcohol, tobacco, and lottery tickets by community pharmacies in Passaic County, New Jersey, and assessed the relationship between sociodemographic factors of community residents and their potential accessibility to those community pharmacies. A mixed geographically weighted regression analysis revealed that census block groups with higher median household income tend to have less accessibility to pharmacies that sell addictive products. Relationships between Latino population and those pharmacies are mixed. No significant relationship was found for African American population.

Keywords:

• potentially addictive products
• community pharmacies
• geographic information systems
• accessibility measure
• geographically weighted regression

Introduction

Community pharmacies in the United States play a role in health promotion by acting as conduits of health information and retailing products aimed at improving the health of its consumers. Health-related institutions, such as hospitals, outpatient clinics, and pharmacies, are all professionally obligated to promote their patients’ health. However, whereas most hospitals and clinics have implemented policies prohibiting the sale of products such as tobacco and alcohol, most community pharmacies have continued to sell these potentially addictive products despite recommendations and resolutions of their professional organizations (Hudmon et al. 2006). The role of health promotion may be negated when pharmacies retail products that are harmful to consumers’ physical, mental, or financial well-being. The sale of alcohol, tobacco products, and lottery tickets in communities could represent a fundamental contradiction to the health promotion function of pharmacies. Overwhelmingly, pharmacists have stated their opposition to the sale of addictive products on their shelves (Hudmon et al. 2006; Kotecki et al. 2000). Many pharmacies, however, continue to carry a combination of these potentially addictive products despite the health and economic risks associated with their use. This type of economic activity has been described in the health promotion literature as creating an atmosphere of disease promotion rather than health promotion (Freudenberg 2005). Targeting community pharmacies selling potentially addictive products as opportunity zones for health promotion at a community level may be supported by disease prevention activities rooted in behavioral interventions that seek to reduce the accessibility to harmful retail products (Gordon et al. 2008; Kloos 2005; Lichtenstein et al. 1991; Mayers et al. 2012; Schneider et al. 2005). This article focuses on identifying the community context and defining the economic and demographic profiles associated with greater accessibility to addictive products in pharmacies.

In most outlet retail analysis literature, ease of access to, or community availability of, additive products has been measured in a variety of ways including population-based, roadway-based, and network distance measures (Gruenewald and Johnson 2010; Freisthler 2005; Hyland et al. 2003; Mayers et al. 2012; Schonlau et al. 2008; Weitzman et al. 2003; Wiggins et al. 2010). Such measures have the implicit assumption that access to addictive products is confined within a specific community whose boundaries are impenetrable. In addition, many existing studies have attempted to examine the locational strategies of retail outlets using traditional methods such as ordinary least squares (OLS) regression analysis. Only a few, recent studies (Mayers et al. 2012; Wheeler et al. 2006; Waller et al. 2007; Yu et al. 2009) have tested innovative methods that extend and address the limitations of traditional methods. One specific drawback of traditional methods involves the assumption of spatial homogeneity, which presumes that the tested relationships will be invariant from observation to observation. The assumption of spatial homogeneity, however, is often violated when geographic information is involved (Fotheringham et al. 2002). Geographically based relationships are often heterogeneous in that the tested relationships often vary across space (Lo 2008; Koutsias et al. 2010).

In addition, the few prior studies investigating the role that community pharmacies play in distributing harmful products have focused solely on the sale of tobacco (Hickey et al. 2006; Morton et al. 2010). Additionally, most existing scholarship on the community context of addictive retail has investigated only one specific product (Asumda and Jordan 2009; Mayers et al. 2012; Paschall et al. 2007; Schneider et al. 2005; Nielsen et al. 2010; Peterson et al. 2011; Wiggins et al. 2010). Nielsen et al. (2010) and Paschall et al. (2007), for example, focused only on the issue of alcohol availability, while Asumda and Jordan (2009), Mayers et al. (2012), Peterson et al. (2011), and Schneider et al. (2005) studied only tobacco availability, and Wiggins et al. (2010) investigated only the issue of gaming or lottery availability. These studies fail to recognize that outlets like community pharmacies engage in activities across domains. The consumption of any one of tobacco, alcohol, and lottery tickets has well-documented effects in terms of mortality, morbidity, and financial burden. Tobacco products are associated with 438,000 deaths per year in the United States and US$167 billion in annual health-related costs and loss of labor productivity (American Cancer Society 2007; Centers for Disease Control and Prevention 2005, 2007). Alcohol abuse is linked with US$68 billion in loss of productivity and US$26 billion in health costs annually (National Institute on Alcohol Abuse and Alcoholism 2000). Additionally, lottery ticket sales represent another threat to individual health in terms of addiction and financial loss. Among pathological gamblers, the use of lottery tickets is one of the most prevalent forms of gambling and has been linked to substance use and psychological disorders (Grüsser et al. 2007; Petry 2003). Based on these arguments, the current study extends prior work by examining outlet sales of different types of addictive products together rather than separately.

To address the above issues, we extend previous research by proposing first a different measure of accessibility to addictive products and then testing procedures that recognize the inherent spatial non-stationarity. We investigate the usefulness of these procedures by analyzing community pharmacies selling all potentially addictive products instead of focusing on just one of them, and their locational strategies. This article is organized in the following way. First, we review previous literature on retail outlets’ sales of addictive products, their locational strategies, and possible community effects. We then present our method with the measures of addictive product availability and accessibility, and propose an accessibility index that is based on geographic distance and derived via GIS. Data used in the current study and rationale of applying geographically weighted regression (GWR), a method that specifically addresses spatial non-stationarity, are discussed in the “Methods” section as well. Results from data analysis are then presented, and we conclude our study with a detailed discussion of the analytical results.

Locational strategies of retail outlets

The literature investigating the relationship between the retail of addictive products and community access has had different aims according to which product is under analysis. In the tobacco outlet literature, the focus has been on locational strategies of tobacco retailers or identifying the types of communities where tobacco sales are concentrated. Tobacco product retailers have generally been shown to be more densely located in areas with lower socioeconomic status and higher concentrations of racial and ethnic minorities (Asumda and Jordan 2009; Fakunle et al. 2010; Hyland et al. 2003; Peterson et al. 2011). In one of the first investigations of the community context of tobacco sales, Hyland et al. (2003) found that census tracts with lower median income and higher percentages of African Americans had a higher density of tobacco outlets, measured as the number of outlets per 10 km of roadway. Research that followed has had similar findings: census tracts with low income and high African American and Latino populations have greater availability of tobacco products (Schneider et al. 2005). More recent research incorporating spatial analyses has provided greater specification of the relationship between demographics and tobacco availability. Yu et al. (2010) found that a 1% increase in the Latino population in a census tract was associated with 6.3 unit increase in the number of tobacco retailers per 10 km of roadway in the state of New Jersey at census tract level. Identifying the types of communities where tobacco retail is highly concentrated is important as it informs strategies to contain the harm of tobacco use. One goal of policy interventions for tobacco control has been to increase the price of tobacco both in terms of the money spent on each pack of cigarettes and the search costs incurred as a result of decreased accessibility to tobacco retailers (Chapman and Freeman 2009; Cohen and Anglin 2009; Schneider et al. 2005). Moving beyond the analysis of tobacco outlet density, Reitzel and colleagues (Reitzel et al. 2011) demonstrated the importance of tobacco outlet proximity. In an evaluation of a smoking cessation program, they showed that participants were more likely to abstain from smoking if the nearest tobacco outlet was more than 500 meters from their home.

While the tobacco literature has focused mainly on locational strategies, the alcohol literature has been more concerned with the alcohol-related outcomes in communities where outlets are heavily concentrated (Gorman et al. 2001; Gruenewald and Johnson 2010; Komro et al. 2010; Livingston 2008; Pasch et al. 2009; Resko et al. 2010; Schonlau et al. 2008; Weitzman et al. 2003). In a review of the alcohol outlet density literature, Popova et al. (2009) found the density of alcohol retailers in an area to influence increased frequency of drinking, violent crime, motor vehicle accidents and rates of co-morbid disease. Researchers have consistently shown alcohol-related problems including motor vehicle accidents, problem behavior in adolescents, and violent crime to be higher in communities inundated with alcohol retailers (Chilenski and Greenberg 2009; Chilenski 2011; Gorman et al. 2001; Gruenewald and Remer 2006; Hahn et al. 2010; Resko et al. 2010). Among those examining locational strategies, researchers investigating sociodemographic predictors of increased alcohol availability have found measures of social disorganization (i.e., lower socioeconomic status, residential instability) to be related with alcohol outlet density (Nielsen et al. 2010). Policy interventions to reduce alcohol outlet density have been promoted in the alcohol literature as well for their potential to mitigate the harms associated with alcohol use (Campbell et al. 2009; Hahn et al. 2010).

The locational strategies and related outcomes for lottery outlets remain largely unexamined. Wiggins et al. (2010) found lottery outlets to be more densely located in majority-Hispanic neighborhoods in one New Jersey county, although income and race were unrelated to the outlet concentration. Proximity to any type of gambling facility (i.e., casino, sports betting) was related to increased gambling behavior or problem gambling (Pearce et al. 2008; Wheeler et al. 2006). The research may be scant in terms of connecting lottery outlet density to specific sociodemographic profiles, but there exists considerable evidence that those who play and spend the most money on the lottery tend to be of lower socioeconomic status (Layton and Worthington 1999; McCrary and Condrey 2003).

Methods

Measuring access to addictive outlets

Research investigating the sociodemographic correlates of community access to addictive products has conceptualized accessibility or availability in several ways. These strategies have included density measures based on roadway length, land area, and network density, where the number of addictive product retailers is normalized by the population, roadway length, or square miles in an area (Freisthler 2005; Gruenewald and Johnson 2010; Schonlau et al. 2008; Weitzman et al. 2003). While the precise measurement may be different (e.g., outlets per 10 km of roadway vs. outlets per 1000 residents vs. outlets per square mile), accessibility measures share the feature of being bound by the geographical unit of analysis under study. Whether the community proxy is census block group, census tract, zip code, etc., the administrative boundaries are by design impermeable.

Though density measures may be suited to describe the concentration of addictive product selling outlets within geographic units, they may not be good measures for accessibility of the general population to addictive products considering people are mobile, not bound by administrative or neighborhood boundaries. By using density measures, the implicit assumption is that retail outlets that sell potentially addictive products can target only the specific population of the geography in which they are located (e.g., census tract, block group, or other geographic unit).

In addition, in one study (Yu et al. 2009) it was found that as the geographic scale gets finer, for instance, when the unit of analysis changes from census tract to census block groups, more and more units tend to have zero density measures due to the fact that there are no addictive product retail outlets located within the administrative boundaries. The increased count of zero entries would not only mask potential relationships between sociodemographic factors and the locational strategies of addictive product selling pharmacies, but also render statistical analysis unreliable due to the increased skewness of the density measure. Considering both the spillover effects and the potentially high number of census units with zero community pharmacies that retail addictive products, this study creates an alternative measure for the accessibility of community pharmacies that retail addictive products. This alternative measure is the distance between a geographic unit’s centroid (which represents an average accessibility of the residents in that particular geographic unit) to the nearest addictive product selling pharmacy regardless of which product it is selling. We term such measure the inversed accessibility index. Here, the longer the distance between the geographic unit’s centroid and the nearest pharmacy (the larger the inversed accessibility index), the lower the general accessibility for the particular unit’s population to addictive product selling pharmacies. The measure can be derived via a distance spatial join in a geographic information system such as ArcGIS® between the geographic unit layer (in our study, this is the census block group layer) and the addictive product selling pharmacy layer. As pointed out by Dai et al. (2013), such practices in general tend to produce more reasonable results.

Data

In this study, we attempt to investigate the locational strategies of community pharmacies selling potentially addictive products in Passaic County, New Jersey. Data for this analysis come from five sources. We gathered retail license information for Passaic County, New Jersey, from the following sources: 2008 listing of licensed community pharmacies from the New Jersey Division of Consumer Affairs Board of Pharmacy; 2008 listing of licensed tobacco outlets and lottery outlets from the New Jersey Department of the Treasury; 2008 listing of licensed alcohol outlets from the New Jersey Department of Alcoholic Beverage Control.

The addictive products datasets were merged with the licensed pharmacy listing to create a final dataset that indicated which pharmacies retailed addictive products. The address for each pharmacy was geocoded using ArcGIS®. Finally, the 2000 US Census was used to create the sociodemographic indicators at the census block group level including the median income and the percentage of residents who were African American or Latino in each block group. Although other sociodemographic factors might also be relevant to accessibility to addictive products, the selection of these three factors to characterize the sociodemographic profile of a community follows the rationale that the decision of making addictive products available in pharmacies is usually assumed to be based on these factors, as seen in many previous studies (for instance, Hyland et al. (2003), Morton et al. (2010), Peterson et al. (2005, 2011), Schneider et al. (2005), and Yu et al. (2009), to name but a few).

Addressing potentially variant relationships: GWR

The primary goal of this research is to determine the relationship between sociodemographic factors (proxies for locational strategies) and accessibility to community pharmacies selling potentially addictive products. Regression analysis is well suited for this type of task in which the sociodemographic factors are used to explain the variance of the accessibility measure. It was found, however, that the accessibility measure displayed a strong spatial autocorrelation pattern (the Moran’s I is 0.664, and significant at 95% confident level using randomization test). This is to be expected considering the nature of how it is derived. The inversed accessibility index explicitly recognizes and measures the sphere of influence for a retail outlet as crossing administrative or neighborhood boundaries, which will lead to dependence among the units of measurement, violating the assumption of independent observations (Yu et al. 2010). Thus, a classical regression analysis such as OLS estimation will be biased and inefficient. Any inference from OLS analyses would be potentially unreliable. In addition, many recent studies have found that when geographic data are used for relationship analysis, the relationship doesn’t seem to stay constant from place to place, hence the so-called spatial non-stationarity of regressed relationships (Fotheringham et al. 2002; Yu and Wu 2004; Yu 2006, 2007; Lo 2008; Yu et al. 2009; Koutsias et al. 2010; Cardozo et al. 2012; Mayers et al. 2012; Su et al. 2012). The GWR approach, among many other newly emerging spatial data analytical methods (Gelfand et al. 2003), takes care of such spatial non-stationarity well. It is also a method that models both spatial non-stationarity and autocorrelation directly in the model specification and structure. The methodology is fairly mature and applied in many studies. To avoid repetition, we do not present the technical details of GWR here, but refer interested readers to Fotheringham et al. (2002) or Yu (2006, 2007).

After deriving the inversed accessibility index, we found that the index is highly negatively skewed. As discussed in Yu et al. (2009), such skewness would make results from any ensuing analysis unreliable. A 10-based logarithm transformation is then applied and the transformed variable follows a normal distribution fairly well. Three sociodemographic factors, the percentage of Latino, the percentage of African American, and median household income, were selected from the census data for the regression analysis. A summary of the data is reported in Table 1.

Table 1. Data description.

	Distance (m)	Ten-based logarithm transformation of distance (dependent variable)	Median household income	Percentage of African American	Percentage of Latino
Min.	166.3	2.221	0	0.000	0.000
First Qu.	1001	3.000	36,234	0.000	5.803
Median	1734	3.239	47,908	3.421	21.013
Mean	2781	3.259	50,991	12.879	29.184
Third Qu.	3134	3.496	64,768	13.735	48.907
Max.	28,150	4.450	138,782	100.000	100.000
Std. Dev.	3427.362	0.383	22,430.550	20.608	26.782

Though we argue that the OLS estimator might not be appropriate for our data, for comparison purposes, we still fit the data with an OLS, and report the results and a cross-model comparable statistics, AIC value in Table 2. Then a preliminary GWR analysis was conducted on the data. The GWR estimation was done in R (R Development Core Team 2012); 46 nearest neighbors were used to create the adaptive bandwidths that produces the minimal corrected AIC score (Fotheringham et al. 2002; Páez et al. 2002). The non-stationarity test (Fotheringham et al. 2002), however, indicates that only the coefficients of the percentage of Latino is significantly varying across the space (Table 3). The test result hence warrants a mixed GWR analysis (Fotheringham et al. 2002; Yu 2006) in which the coefficients of median household income and percentage of African American are kept stationary (constant across the space), but the coefficients of percentage of Latino are allowed to vary across the space. The mixed GWR is often called for when not all the explanatory variables vary across space as revealed by the non-stationary test. Calibrating a mixed GWR model is done while first calibrating the stationary part of the model, and then the residual of the stationary part will be used as the dependent variable to calibrate the non-stationary part of the model (Yu 2006). The mixed GWR was calibrated with codes written in R by the authors. The stationary part of the mixed GWR and the cross-model comparable AIC value are reported in Table 4, while the non-stationary part is expressed in Figure 1.

Figure 1. The non-stationary part of the mixed GWR model.

Table 2. Ordinary least squares estimation results.

	Estimate	Std. Error	t-Value	Pr(>|t|)
Intercept	3.155 (3.024, 3.286)	0.067	47.233	0.000
Median household income	5.51E–06 (3.72E–06, 7.30E–06)	9.13E–07	6.040	0.000
Percentage of African American	−1.93E–03 (–3.52E–03, –3.42E–04)	8.10E–04	−2.380	0.018
Percentage of Latino	−5.23E–03 (–6.60E–03, –3.86E–03)	6.97E–04	−7.507	0.000
AIC for OLS: 139.087; adjusted R^2 = 0.433

Note: Numbers within the parentheses are 95% confidence interval.

Table 3. Stationarity test for preliminary GWR analysis.

	Intercept	Median household income	Percentage of African American	Percentage of Latino
F-statistic	0.913	1.183	1.563	2.179
Numerator degree of freedom	66.146	144.696	13.485	41.291
Denominator degree of freedom	366.247	366.247	366.247	366.247
p-Values	0.668	0.107	0.090	0.000

Results

A total of 115 community pharmacies in Passaic County, NJ, were included in this analysis (Figure 2). Of the 115 pharmacies, the majority (n = 75; 65.2%) sold at least one of the addictive products under analysis. Specifically, a large percentage of pharmacies (n = 41; 35.7%) sold tobacco alone, while 14.8% (n = 17) sold both tobacco and lottery tickets. Smaller percentages of pharmacies (n = 13; 11.3%) sold lottery tickets alone, while only 1.7% (n = 2) of pharmacies sold both tobacco and alcohol, and another 1.7% (n = 2) of pharmacies sold all three products, tobacco, alcohol, and lottery tickets. Figure 2 also suggests that the majority of pharmacies located in the southern end of the County, where the percentage of Latinos concentrates (Figure 3).

Figure 2. Pharmacies in Passaic County.

The cross-model comparable AIC values of the OLS and mixed GWR models (Tables 2 and 4) indicate that the mixed GWR model fits the data far better than the non-spatial OLS model, even considering the added complexity of the former (since a drop of 3 in the AIC value is deemed a significant improvement; see Fotheringham et al. (2002) for a detailed discussion). Moreover, unlike the non-spatial OLS analysis, our mixed GWR analysis produces some distinctive and rather anti-intuitive results.

First, the stationary part of the model (Table 4) is fairly straightforward, in which we see that the accessibility to community pharmacies selling potentially addictive products decreases (distance variable increases) as a particular census block group’s median household income increases. On the other hand, the accessibility index is not significantly related with the percentage of African American population in census block groups, which is in contrast to our preliminary OLS results where the accessibility index increases as the percentage of African American population increases (Table 2), and other studies with similar settings (Morton et al. 2010; Peterson et al. 2005; Yu et al. 2010). Controlling for spatial effects creates a rather different picture of the relationships between accessibility to community pharmacies selling potentially addictive products and demographic factors than analysis in which spatial effects are ignored.

Table 4. The stationary part of the mixed GWR model.

	Estimate	Std. Error	t-Value	Pr(>|t|)
Intercept	3.203 (3.081, 3.325)	0.062	51.964	0.000
Median household income	4.29E–06 (2.67E–06, 5.97E–06)	8.29E–07	5.170	0.000
Percentage of African American	−9.71E–04 (–2.78E–03, 8.40E–04)	9.24E–04	−1.052	0.294
AIC for the mixed GWR model: 67.6; adjusted R^2 for the stationary part = 0.350

Note: Numbers within the parentheses are 95% confidence interval.

Second, the relationship between the accessibility index and percentage of Latino population in the census block groups is more complex, however, as can be seen in Figure 1. There are two very clear clusters of block groups in Passaic County (the north and middle to south end; Figure 1) in which the two variables are significantly related based on the pseudo-t tests from the mixed GWR analysis. Their relationship, however, is not uniform. The relationship not only varies in magnitude (the coefficients vary from –0.024 to 0.157, which is to be expected in a GWR modeling scheme), but more interestingly, it varies in directions as well. In the north end of the County, where there is a relatively low percentage of Latino population (generally less than 10% of the total population in those block groups; Figure 3) living in rather affluent census block groups (Figure 4), the accessibility index decreases (distance variable increases) as percentage of Latino population increases. This directly contradicts the results obtained from the OLS estimator in which the accessibility index increases as the percentage of Latino population increases. This relationship, however, still holds in the south end of Passaic County where the majority of Latino population (and African American population as well) resides, and where the median income is lower (Figures 3 and 4).

Figure 3. Distribution of Latino population in Passaic County.

Figure 4. Wealth distribution in Passaic County.

Discussion

The current study extends existing research by advancing an accessibility measure not limited by administrative boundaries and by investigating the sale of addictive products together rather than in isolation with spatially explicit approaches (mixed GWR). Based on the literature and our previous investigations, we expected to observe greater access to harmful retail in communities with lower median income and higher percentages of racial and ethnic minorities. The results of our geospatial analyses show that the relationships between accessibility to addictive products and sociodemographic factors were not necessarily constant from place to place, especially at a finer scale of analysis. Widely adopted non-spatial and global analytical procedures, such as the OLS regression, tend to average such variation and hence mask subtle local patterns that might have significant policy implications. To effectively control general access to addictive products, researchers and policymakers need to employ methods that specifically take into consideration spatial effects and are able to detect detailed, local relationships and patterns such as GWR in addition to conventional methods.

The assumption that relationships among variables vary across space is not taken as is in GWR analysis. Instead, such assumption can be statistically tested. As we see in our particular analysis, the GWR stationarity test suggests that the relationships between the accessibility index and median household income, percentage of Africa Americans, do stay constant across space in Passaic County, NJ. The relationship between the accessibility index and percentage of Latino, on the other hand, varies significantly across space. This result indicates that as a relatively mature local approach, GWR is quite versatile and flexible in dealing with both stationary and non-stationary relationships at the same time and in the same model structure. Such flexibility provides a robust policy-making implication in practice. In our particular study, the method suggests wealth status has a relatively uniform impact on community pharmacies selling potentially addictive products’ locational strategies. Racial/ethnic factors, on the other hand, have different impacts on such strategies and need to be examined locally.

It should be noted that although OLS is not an appropriate estimator while using this dataset, it does provide a starting point to understand the relationship that we intend to build between accessibility to community pharmacies selling potentially addictive products and sociodemographic factors. Comparing the OLS to the GWR results confirms that the more affluent a census block group is, the less accessible its residents will be to any community pharmacies selling potentially addictive products. This relationship not only holds in both OLS and GWR analyses, but also seems to carry over and influence the relationship between the accessibility index and the percentage of Latino population. Specifically, in the relatively affluent block groups where there is often less Latino presence, even if they contained relatively higher percentages of Latino residents, their accessibility indexes were lower. This suggests a possible moderation effect for ethnicity and income that future studies could attempt to untangle. The relationship revealed by the global OLS estimator where higher percentages of Latino residents was related to greater access of addictive products stays true in the GWR analysis only in the less affluent census block groups (and usually with a higher percentage of Latino presence) in the south end of Passaic County. Such an interaction is often modeled with an interaction term in traditional moderation analysis. Yet the traditional analysis by design can only produce spatially homogeneous products. Detailed patterns of interactions could potentially change from observation to observation. More importantly, via GWR, we are able to present a more detailed and mappable pattern of how wealth and ethnicity interact with each other across space in their relationship with accessibility to addictive product selling outlets. Our analysis reveals a rather subtle locational strategy of community pharmacies selling potentially addictive products. In Passaic County, where income inequality is relatively large as in the rest of New Jersey, it is the economically disadvantaged neighborhoods that are targeted by pharmacies to sell addictive products, and the Latinos seem to be especially targeted in those neighborhoods. Previous studies establishing the targeting of neighborhoods with high percentages of minority groups for the sale of addictive products may reflect the potent intersection of race and class where economically depressed areas with large ethnic and racial minority populations may be at particular risk for an inundation of harmful retail. Reducing the number of community pharmacies selling potentially addictive products could be an environmental intervention embedded into the process of promoting local development as a strategy to improve the health of a community. Further analyses with other locations would provide greater understanding of the need for community-level prevention strategies.