This systematic review described the methods used in the existing literature to capture and measure TRD and smoking behaviour. The methods for measuring TRD varied across the literature, however the most common approach assessed density directly through applying circular buffers at varying distances from specific geocoded locations such as homes or schools using GIS software (n = 14). Smoking behaviour was also described in many ways, however once reclassified using the available information, past-month smoking behaviour was most frequently captured (n = 17).
Smoking behaviour
It was evident in the literature that measuring smoking behaviour was complex and often inconsistent,(28) making it difficult to interpret. The studies included in this systematic review used a wide range of smoking descriptors, such as experimentation, occasional, non-daily, regular, and daily smoking. For example, several studies described ‘Current smoking’ as ‘Any smoking in the past 30 days’, while other studies described it as ‘Currently smoke daily or occasionally’, with the former considered to reflect past-month smoking and the latter current smoking once re-classified by the researchers in the current review.
The most common way that studies measured smoking behaviour was past-month smoking. Self-reported smoking behaviour was the most common data collection method used in the included studies, but this may under-report true smoking behaviour due to potential social desirability bias(36) or recall decay, particularly if respondents are asked to remember their smoking behaviour over a long period of time (e.g. on how many days in the past 30 days respondents smoked and the average number of cigarettes smoked on those days).
A consistent approach to capturing and classifying different smoking behaviours is important so that comparisons across different studies can be undertaken. One reason for methodological inconsistencies in capturing smoking behaviour is the absence of clear guidance on best-practice approaches when exploring outcome measures. Although previous research on TRD and smoking behaviour has acknowledged these inconsistencies, possible solutions are limited. In 2011, the Global Adult Tobacco Survey Collaborative Group developed a guide for surveys that include questions focusing on tobacco use,(37) however many of the survey instruments used to collect data in the included studies appear not to have followed its recommendations. More widespread usage of this resource would strengthen future research into TRD and smoking behaviour.
This analysis identified subtle differences in wording across various types of smoking behaviour within the included studies. For example the study by Chan et al.(38) defined daily smoking as ‘smoking every day or almost every day in the past 30 days’ and defined occasional smoking as ‘smoking some days or only 1 or 2 days in the past 30 days’. Both smoking behaviours were reclassified as past-month smoking according to the classification tool developed for the present analysis since it focuses on the recency of the smoking behaviour, not its intensity. A strength of this tool, therefore, is that it provides an objective basis for summarising and sorting the most commonly used questions about smoking behaviour. Its application would facilitate comparison between TRD and different smoking behaviours in future systematic reviews and meta-analyses.
Tobacco retailer density
Most included studies focused on assessing TRD and smoking behaviour from participants’ homes or schools primarily using circular buffers at varying distances. Reasons for choosing a particular buffer size included practical constraints,(39) the buffer distance was approximately a 10 minute walk from school,(40) or the distance was assumed to be the outer limit that most students would walk or cycle regularly to school in urban settings and the minimum distance necessary to encompass at least one tobacco retailer amongst students who attend a rural school.(32) Although the use of circular buffers is a simple and reliable method for measuring TRD, studies using this approach did not take into account the built environment, which could increase the true travel time or distance to retailers. For example, a participants’ home or school could be geographically close to several tobacco retailers using circular buffers; however buildings, fences, highways, waterways or other physical features might prevent direct travel to those retailers, thus increasing the true distance and/or travel time. It is possible that earlier studies may have resorted to using circular buffers to measure TRD as GIS software did not provide access to KDE measures. GIS software has evolved over time, giving researchers more nuanced measures of exposure to retailer presence in the community than was available historically.
Therefore studies that measured density using network service areas (e.g. the street network or footpaths) represented as polygons around geocoded locations appear to be a more valid and reliable measure of TRD, as they take the physical environment into account. Valiente et al.(14) recommend that TRD is measured through both length-distance and travel time using the street network and footpaths, and weighted by the size of an area, population, or road length, or measured using KDE.(14)
The Uncertain Geographic Context Problem also represents an inherent limitation in the definition of exposure measures used to analyse the real influence of the environment on population health, and few included studies captured data on the time and duration that participants spent within defined areas (e.g. homes or schools) and/or the length of time spent exposed to tobacco retailers whilst in these areas.(41, 42) Studies focusing beyond participants’ home and school environments would likely add value to this field of research and provide a better understanding of how regular interactions with tobacco retailers in the broader environment (i.e. activity spaces) may influence smoking behaviours.(43)
Few studies included in this systematic review assessed possible associations between TRD within daily activity spaces and smoking behaviour. Studies focusing only on home or school environments may not take into account adolescents’ increased autonomy, mobility and social networks that extend beyond these two settings.(44, 45) Only two studies in the current systematic review analysed TRD and smoking around activity spaces.(29, 34) The inability to collect data on respondents’ movements across activity spaces may be due to the complex technical nature of tracking participants over an extended period of time and over geographically large areas.(41) However smartphone technologies allow real-time data to be easily captured from participants, providing much more detailed information on individuals’ interactions with tobacco retailers in the broader environment during day-to-day activities, instead of relying on self-report data.(41, 46) Research using Ecological Momentary Assessment in defined populations (e.g. adolescent smokers) to identify where and when they make retail tobacco purchases may be useful. The literature assumes that TRD and TRP are major predictors of where adolescents purchase tobacco products, however a more important criterion may be a retail source not likely to be observed by parents or neighbours for example. As better GIS software and GPS-enable devices become available, more nuanced investigations may be able to explore observed associations between a community's tobacco retailers and tobacco use within that community.
The included literature also highlighted differences in measuring TRD across varying geographical areas, such as urban, regional and rural districts.(31) It has been suggested(47) that TRD measures in non-metropolitan areas might need to be adjusted to account for lower population densities and greater travel distances to tobacco retailers. For example, McCarthy et al.(32) explicitly justified their buffer in part to facilitate capturing rural students' exposures to rural tobacco retailers. Although this systematic review did not compare studies by metropolitan or non-metropolitan locations, most studies focused on metropolitan areas, with only four studies(31, 32, 48, 49) analysing associations between TRD and smoking behaviour in non-metropolitan settings. Existing research suggests that TRD may be greater in non-metropolitan areas,(50, 51) therefore it is important for future research to explore associations between TRD and smoking behaviour in these areas.(49)
Previous research has also identified mixed sensitivity when auditing existing tobacco retailer databases, such as those generated through licensing or registration systems or through commercial sources.(52–54) For large geographical areas such as major cities, it is recommended that researchers conduct random sample ground-truthing to establish the proportion of tobacco retailers likely to have been overcounted (e.g. no longer selling tobacco) or undercounted (e.g. new businesses that have not yet registered) to accurately describe how well a database represents the population of interest, based on just a small random sample of the areas being studied. For research in smaller geographical areas, future studies could attempt to verify all tobacco retailers in existing databases and to identify other potential tobacco retailers operating through field visits within defined geographical areas prior to collecting data on smoking behaviour.(54) This may increase the accuracy of TRD and provide a more precise representation of the true exposure to tobacco retailers amongst participants.
The current systematic review did not include longitudinal studies. Longitudinal studies are important to determine whether causal relationships exist between TRD, smoking behaviour and cessation amongst both youth and adults, and to identify protective factors that might reduce or prevent associations between tobacco availability and smoking behaviour. Currently it is unclear whether TRD has a particular effect on certain types of smoking behaviour, however longitudinal or cohort studies could examine these relationships over time. A recent systematic review(13) included one longitudinal study and found a positive association between TRD and smoking. It should be noted that although longitudinal or cohort studies are superior to cross-sectional studies in identifying causal direction, they are subject to similar model misspecification due to the uncontrolled, unmeasured confounders that may explain observed associations. Nuyts et al.(12) highlights the importance of quasi-experimental studies, which take advantage of natural experiments when, for example, jurisdictions adopt tobacco retailer density and proximity restrictions near schools.
We agree with Marsh et al. that inconsistencies in how the exposure and outcome measures are defined are likely to play a role in the reported associations between TRD and smoking behaviour. The results from existing systematic reviews and meta-analyses should therefore be interpreted with caution, and future systematic reviews would be strengthened by a more consistent approach to measuring TRD and smoking behaviour in the literature. We note, however, that as more editors and funders demand that authors place the data on which their work is based in publicly-accessible repositories, it will become increasingly possible to overcome at least some of these inconsistencies by undertaking secondary analysis of the publicly available data.
Studies were not grouped by country or location and were not categorised according to existing tobacco retail policy approaches, such as minimum pricing legislation, PoS display bans, tobacco advertising bans, tobacco retailer licensing systems or minimum-distance laws. Jurisdictions have taken different approaches to retail tobacco sales legislation and these factors may play important roles in the promotion and normalisation of tobacco products in the community (through advertising, for example).(39) It is also important to recognise that much of the existing literature tends to focus on determining whether density (and/or TRP) contributes to smoking behaviour, however it would be appropriate for future research to determine whether certain policies that address density (and/or proximity) may prevent smoking behaviour and/or improve smoking cessation outcomes for existing smokers who are attempting to quit.