Introduction The availability of tobacco is thought to influence smoking behaviour, but there are few longitudinal studies examining if the location and number of tobacco outlets has a prospective impact on smoking cessation.
Methods The Ontario Tobacco Survey, a population-representative sample of Ontario adult smokers who were followed every 6 months for up to 3 years, was linked with tobacco outlet location data from the Ontario Ministry of Health. Proximity (distance), threshold (at least one outlet within 500 m) and density (number of outlets within 500 m) with respect to a smokers’ home were calculated among urban and suburban current smokers (n=2414). Quit attempts and risk of relapse were assessed using logistic regression and survival analysis, adjusted for neighbourhood effects and individual characteristics.
Results Increased density of tobacco outlets was associated with decreased odds of making a quit attempt (OR: 0.54; 95% CI 0.35 to 0.85) in high-income neighbourhoods, but not in lower income ones. There was an increased risk of relapse among those who had at least one store within 500 m (HR: 1.41 (95% CI 1.06 to 1.88). Otherwise, there was no association of proximity with quit attempts or relapse.
Conclusions The existence of a tobacco retail outlet within walking distance from home was associated with difficulty in succeeding in a quit attempt, while the increased density of stores was associated with decreased attempts in higher income neighbourhoods. The availability of tobacco may influence tobacco use through multiple mechanisms.
- Public policy
- Tobacco industry
Statistics from Altmetric.com
Tobacco use is the number one cause of preventable mortality. Five million deaths each year are attributable to smoking, with an estimated rise to as much as 10 million deaths per year by the 2030s.1 Yet, despite the widespread awareness of the harms of smoking, tobacco is still widely available for sale with few restrictions or requirements. Limited restrictions on availability, notably age restrictions on purchasing tobacco, have been implemented as part of many comprehensive tobacco control programmes but tobacco is still widely available. Accordingly, there have been calls for reducing the availability of tobacco.2–6
Based on fundamental laws of supply and demand and the experience with policies that limit or liberalise the sale of other consumer goods,7 primarily alcohol,8 there is substantial theoretical reason to suspect that the availability of tobacco contributes to keeping the prevalence of tobacco high. Competition among many retailers results in more individuals and corporations with a vested interest in increasing sales, leading to increased marketing.9 10 Increased number of vendors also requires a larger sales force dependent on tobacco consumption.11 For former smokers, receiving cues to purchase tobacco in places where they regularly shop may contribute to high levels of recidivism.12 13
Indeed, there has been shown a consistent association between retail availability and prevalence of smoking. Novak et al 14 found that retail density was significantly associated with smoking prevalence and that the magnitude of the estimate increased after controlling for socio-economic status and other neighbourhood factors.14 Studies have found a relationship between high density of tobacco retailers (ie, the number of retailers per unit of a geographic area) around the individual’s home and tobacco use,15–21 as well as the proximity (the straight line or walking line distance between the closest tobacco retail outlet) to their home.16 18 19 22 A third joint characteristic should also be considered—that of a threshold effect. A threshold effect might suggest that as long as there is at least one store within a certain proximity or area it would be sufficient to produce the association but that additional distance or number of stores have minimal additional effect.
Reitzel et al 23 found that smokers in a smoking cessation trial who lived within 250 or 500 m of a convenience store were less likely to remain abstinent after a quit attempt than other smokers who lived further away; there was no effect of density on cessation.23 However, the effect was apparent only among men, and it is not clear if these results would be generalisable to smokers not participating in smoking cessation studies. A similar study in England yielded null results.24
Cantrell et al 16 found, in a representative sample, an association between high density and living fewer than 500 m from an outlet and reduced 30 days abstinence.16 A study of Finnish public servants found effects for both proximity and density within 500 m from home on the likelihood of having quit smoking at follow-up on average 5.5 years after baseline among male moderate to heavy smokers25; however, a related study by Pulakka et al 19 employed a case cross-over design that allowed for within subject comparisons of stopping smoking as well as relapse.
This study attempts to address some of the previous limitations of the still limited literature of tobacco retail availability and smoking behaviour by comparing proximity, density and threshold effect; and by distinguishing between making a quit attempt and relapse to smoking. This study also derives its sample from a population-representative cohort that assessed smoking behaviour at 6-month intervals for up to 3 years and used administrative licensing data to provide an accurate identification of tobacco outlets.
Data were from the Ontario Tobacco Survey (OTS) panel study of smokers.26 27 Briefly, the OTS is a representative sample of adults (18+) of the Ontario (eligible population aged 18+, 10.4 million), interviewed by telephone, who had smoked in the past 6 months at the time of recruitment. Participants were stratified by region and recruited in waves over 6-month intervals between 2005 and 2008. Up to six follow-up interviews were conducted at 6-month intervals until 2011. Technical reports with more details are available.26 The response rate at baseline was 61% for the full sample of 4501 smokers. Technical reports presenting details of the study design, questionnaires, demographic characteristics and smoking behaviour of the overall OTS samples are available online (http://otru.org/research-evaluation/ontario-tobacco-survey/ots-technical-documentation/).26 Participants in the full sample had characteristics consistent with census data from Ontario (detailed data available in the technical documentation).26 The OTS was approved by the institutional review boards of the University of Toronto and the University of Waterloo.
The analytic sample for this study included a subset of the full sample: past-month daily smokers who had smoked more than 100 cigarettes in a lifetime and lived in an urban or suburban area (n=2414). Among eligible participants at baseline, only 148 (6.1%) were lost to follow-up. Attrition was unrelated to key variables (see table 1).
The Ontario Ministry of Health and Long-Term Care Tobacco Information System database of tobacco-selling vendors contains licensing records pertaining to compliance with the provincial law prohibiting the sale of tobacco to minors and includes the vendor address, store type and last date visited by public health inspectors. There were 11 361 unique tobacco retailer locations, of which 11 113 could be mapped. The list was from June 2011.
Ground truthing of the list was conducted using a subsample of all stores and restaurants within four randomly selected Forward Sortation Areas, which comprises an area represented by the first three digits of the postal code. At each vendor visited, Global Positioning System coordinates were documented using a mobile phone device. Sensitivity of the list was extremely high at 98% (57 of the 58 stores sold tobacco). Specificity was 88% (57 of the 65 stores found to be selling tobacco were on the list). There were no systematic differences noted for the stores not on the list except for a tendency to be located in recently built developments.
The full address of participants and tobacco retail outlets were geocoded using ARCGIS V.9.0. Participants who lived in rural areas were excluded as rural addresses do not identify the physical location of the home as accurately as addresses in urban and suburban areas in Canada.28–30 Walking distance to the nearest tobacco outlet for each participant was calculated using the New Closest Facility tool (proximity). The number of outlets within a circular buffer with a straight-line radius of 500 m from home was used to calculate density. The square root divided by 100 of the measure of proximity was calculated to rescale to adjust for the positively skewed distribution of the proximity data (skewness was calculated as 3.5 prior to rescaling and 0.96 after). A measure of the threshold effect was coded as 1 for having a store within 500 m walking distance and 0 otherwise. The cut point of 500 m was selected for consistency with previous research.16 19 23
Quit attempts were assessed by self- report at each follow-up by asking if the participant had made a serious quit attempt to quit smoking for good. Only those who reported quit attempts were included in the relapse analysis (n=921). Time to relapse was defined as the reported longest duration of quit attempt in days (not smoking even a single cigarette). Time to relapse was a cumulative measure where at each follow-up, smokers were asked if they have smoked even a single cigarette and the total amount of time since the last cigarette calculated.
Neighbourhood and census data
Individual neighbourhood variables were calculated using the full six-digit postal code for each individual linked to census data for the full postal code (household income quintile, tertile of percentage of immigrants and municipal size). Household income quintile and immigrant tertile are the variables released in the public use file available for linking postal code to 2006 census data by Statistics Canada.28
Most data in this study were compiled at the individual level, minimising error related to the modifiable areal unit problem;23 however, further unmeasured clustering by neighbourhood was controlled by identifying a larger area neighbourhood as a cluster variable for control in the analysis. Neighbourhood was defined using the first three digits of the postal code. In Canada, the use of the first three digits of the postal code is a common proxy for a neighbourhood jurisdiction covering 8000 households on average compared with 19 households for the full six digit postal code.28 In this study, there were a range of 7.1 participants per three-digit postal code-defined neighbourhood with a range of 1–56 participants per area.
Potential confounders included age (categorised as 18–29, 30–39, 40–49, 50–59, 60–69, 70+ for descriptive analysis but included as a continuous variable otherwise), gender, marital status (married vs other), having children under 18 in the home (yes/no), level of education completed (less than high school, completed high school, some college or university, completed college or university), region of Ontario (North, Southwest, East, Greater Toronto Area), perceived addiction (Question: How addicted to smoking do you feel yourself to be: very, somewhat, not at all addicted), use of pharmaceutical quit aids (ie, patch, gum, spray, buproprion, varenicline (yes/no)) and use of behavioural quit aids (ie, self-help, counselling, group therapy (yes/no)). For the present study, high and moderate nicotine dependence was based on the Heaviness of Smoking Index (HSI) and defined as a score of greater than .31 Demographic characteristics were measured at baseline; smoking characteristics were assessed at the survey prior to the quit attempt.
Descriptive characteristics of the sample by having made a quit attempt were assessed at baseline. Point estimates in descriptive analyses were weighted to reflect the population sampling probabilities and target population of Ontario smokers. Variance estimates and bivariate tests of significance reflected the complex survey design using the Taylor series linearised method using Stata V.14.
To assess quit attempts, three models were run adding proximity, then density, then threshold, adjusted for covariates. Logistic regression was used to predict making at least one quit attempt during the period of the study. In all models, the association was adjusted for individual demographic characteristics as assessed at baseline (age, gender, region of Ontario, level of education, children living in the home and marital status), neighbourhood characteristics (neighbourhood income, neighbourhood level of education, municipal size) and smoking history (cigarettes per day, perceived addiction and heaviness of smoking, ever use of pharmaceutical cessation aids, ever use of behavioural cessation aids, number of previous quit attempts).
To assess the association between retail availability and risk of relapse, three hierarchical models including proximity, density and threshold were run adjusted for covariates. An interval censored survival model (Stata procedure—intcens-) was used that assumed a Weibull distribution consistent with rate of relapse decreasing over time was used. A hierarchical model was necessary to identify the independent components of density and threshold variables (ie, density include a proximity element (within 500 m), and threshold includes both a density element (at least one store) and proximity (within 500 m). Entry into this analysis occurred at the follow-up where the eligible participant reported a quit attempt. The association was adjusted for the same set of covariates as the quit attempt analysis except that smoking-related covariates were obtained from the survey prior to the reporting of the quit attempt (ie, lagged).
Previous studies have found interactions with gender23 25 and income.16 32 Interactions with gender and neighbourhood income were assessed by evaluating the statistical significance of interaction term of either gender (male vs female) or neighbourhood income (highest vs other quintiles) with quit attempts and risk of relapse.
All models were adjusted for potential clustering at the neighbourhood level (three digits of the six-digit postal code).
The sample was 55% male, 36% were under 40 years old and 67.5% lived within 500 of a tobacco outlet (table 2). Of the 2414 eligible participants (daily urban or suburban smokers with follow-up data), 921 reported a quit attempt over the period of the study. Smokers who made a quit attempt were more likely to have prior use of pharmaceutical or behavioural quit aid and lower level of nicotine dependence.
Interactions between the proximity, density and threshold variables were tested. There were no significant interactions with gender. There was an interaction between attempting to quit, living in a high-income area and proximity to a retail outlet where proximity only affected those living in high-income areas. Results of the quit attempt analysis were also shown stratified by neighbourhood income.
There were no significant effects of any variables in the unstratified analysis between retail availability and quit attempts (table 3). There was no effect of proximity on the odds of making a quit attempt among either those in the highest income quintile neighbourhoods or those in the lower income neighbourhoods (see table 4). There was an effect of density where increased number of stores within 500 m reduced the odds of making a quit attempt (OR: 0.54; 95% CI 0.35 to 0.85). The effect of density continued after adding the threshold variable. There was no independent effect of threshold on making a quit attempt.
There was no effect of proximity (HR 1.00; 95% CI 0.99 to 1.01) on risk of relapse; however, there was an effect of density (HR 1.11; 95% CI 1.00 to 1.23) (see table 5). When the threshold indicator was added to the model, living within 500 m of a retail outlet was associated with an increased HR (risk) of 1.41 (95% CI 1.06 to 1.88) compared with those living further from an outlet. In this model, there was no residual effect of density.
This study examined the independent effects of proximity, density and a threshold effect and found that the key criterion determining quit failure with respect to tobacco availability was having at least one outlet within walking distance of home. Tobacco retail availability around a smoker’s home increases the risk of relapse among smokers attempting to quit while increased density was associated with decreased risk of making a quit attempt among smokers living in lower income neighbourhoods.
This study suggests a different mechanism of action of retail availability with respect to making a quit attempt and success in that quit attempt where quit attempts. The increased risk of relapse among those who lived within 500 m of a tobacco outlet was consistent across people living in a range of neighbourhoods. This may imply that relapse is associated with the ability to easily access tobacco, but the distance to the store within that range, or having other stores within that range, has little to marginal impact. The outcomes of relapse and making a quit attempt have been assessed in various ways in the literature. Some (eg, refs 16 and 25) used a short-term or long-term measure of abstinence which includes both relapse and having made a quit attempt, while others (eg, refs 23 and 24) used smoking in a cessation trial and could assess only relapse. Pulakka assessed relapse and quitting on a long-term scale that examined use before and after changes in retail availability and found that quitting but not relapse was associated with proximity and partially with density.19
On the other hand, the association of density with making a quit attempt, but not proximity or threshold effects, may suggest that density is associated with a normalising effect of high availability of tobacco. A reduction in the number of retail outlets could influence the environmental cues that smoking is a normative behaviour and thus make it more likely that a person chooses to try to quit. Given that this effect was only apparent in the more affluent neighbourhoods, increased density may indicate a higher prevalence of smoking than a more affluent smoker may be exposed to otherwise. The interaction contrasts however with Cantrell et al 16 who found that increased density was associated with decreased protobacco attitudes and quitting only in high poverty areas.16 The impact of retail density only being apparent in high-income neighbourhoods is consistent with Chuang et al 32 who found in a study of four counties in Northern California that distance to the closest retailer and density of retailer was associated with increased prevalence of smoking, with the effect limited to those living in high socio-economic status neighbourhoods.32 Cantrell et al 16 suggest that the differences found between their study and Chuang’s study may be due to the differences on the measure retail types, as this study uses a list of retail outlets taken from administrative licensing data rather than assuming sales at particular outlet types which could also explain the differences between this study and that of Cantrell et al 16 Additionally, there may also be different effects of the interaction of income on consumption,32 making a quit attempt (this study) and 30-day abstinence, a combination of making a quit attempt and short- term abstinence.
Proximity was not found to have a significant independent effect in any model. The effect of proximity may be more impacted by differences in mobility patterns which were not assessed as this study relied on proximity to a smoker’s home only. Kirchner et al 13 examined total contact with tobacco retailers as they moved through the city, and found associations with increased cravings and relapse.13
This study has implications for understanding the potential impact of changes to tobacco availability. It suggests that there would likely be marginal impact of small changes in tobacco retail availability on quitting successfully, given the massive oversupply of tobacco in most areas. Two thirds of urban or suburban smokers in Ontario lived within 500 m of a tobacco outlet. Second, it suggests that large changes in availability could have dramatic impacts in use. Pearson et al 33 examined the effect of reducing tobacco availability in New Zealand using a model that predicted behaviour change based on a continuous function of travel times translated into increased costs and found limited effects of a 95% reduction in the number of outlets. However, a threshold model would suggest that the impact of a major reduction would vary depending on the details of how that reduction was achieved. If changes that increase the average distance to an outlet to just beyond the threshold would have much larger effects than would be expected under a continuous model. Variability of findings in previous studies could potentially be partial explained by natural variation in the number of smokers within 500 m relative to the number of stores within that distance.
This paper focused on the effects of retail availability on risk of relapse and attempts to quit, but retail availability may have different mechanisms of action on different aspects of smoking behaviour. For instance, smoking initiation may be driven by normalisation of tobacco sales and increased ability to access tobacco underage.34 35 It is possible that both of these aspects may be driven by the number of stores (ie, the density) rather the threshold effect found here.
The threshold value of 500 m was chosen based on previous literature and the actual threshold number may be more or less than this value. The measure of density used was a simple one for the purposes of interpretability where it reflects the count per unit where a unit is a buffer of a circle with radius of 500 m. There was some identified misclassification of retail outlets in the list provided by the Ministry of Health and there may have been changes over time in those outlets. Spatial autocorrelation was control for only by administrative cluster. This study also excluded occasional smokers and there may be different effects among this population; similarly, there may be different effects and different threshold in more rural areas where expectations of availability of any product may differ; results should not be generalised to effects in rural areas or among occasional smokers. Response bias due to the response rate of 61% may have also led to reductions in generalisability. While loss to follow-up among the analytical sample was low, length of time under observation did not appear to be associated with relapse but was associated with likelihood of making a quit attempt (see online supplementary tables).
Valuable data were also missing with respect to the actual exposure of the participants to retail availability. Both daily exposure13 and exposure to tobacco retail close to work and other important locations would have provided additional and valuable information.
Public health implications
Tobacco retail availability around a smoker’s home influences changes in smoking behaviour among a representative population sample of daily smokers. Both effects of density and having an outlet with close distance were important predictors of behaviour change, but they appeared to have different ways in which they made their influence. Tobacco is vastly oversupplied in Ontario as it is in many jurisdictions. Changes to zoning and licensing of tobacco has the possibility of reducing the number of outlets substantially.3–6 For instance, limiting the sale of tobacco within 500 m of schools would have the effect of reducing the number of outlets by over 60% in Ontario.15 This study adds to the growing literature demonstrating an effect of tobacco retail availability and the potential of restrictions on availability as an effective tobacco control intervention.
What this paper adds
Limited longitudinal observational data exist suggesting a relationship between tobacco availability and smoking behaviours, and the mechanism of this relationship is not clear.
This paper uses a longitudinal population representative cohort linked to state-level administrative data to examine changes in quit attempts and relapse up to 3 years of observation.
Quit attempts appear to be influenced by the density of tobacco outlet within high-income neighbourhoods only; risk of relapse is primarily affected by having at least one outlet within a close proximity.
Contributors MC and JC conceived the study. GM performed the geographical analyses and retail analyses. MC performed the statistical analysis. JC designed and implemented the OTS survey. All authors contributed to writing and interpretation.
Funding The Ontario Tobacco Research Unit and the Ontario Tobacco Survey is supported by the Ontario Ministry of Health and Long Term Care. This work is supported by Canadian Cancer Society grant #702160 (MC) and the Canadian Institute for Health Research. The funders had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript or the decision to submit the paper for publication.
Competing interests None declared.
Ethics approval University of Toronto; University of Waterloo.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Research teams may apply for access to OTS data through one of the following university-based data libraries: Propel Centre for Population Health Impact – Population Health Data Repository at the University of Waterloo or the University of Toronto Data Library.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.