Objective To investigate the potential impacts of several tobacco control interventions on adult daily smoking prevalence in the Australian state of Queensland, using a system dynamics model codeveloped with local and national stakeholders.
Methods Eight intervention scenarios were simulated and compared with a reference scenario (business as usual), in which all tobacco control measures currently in place are maintained unchanged until the end of the simulation period (31 December 2037).
Findings Under the business as usual scenario, adult daily smoking prevalence is projected to decline from 11.8% in 2017 to 5.58% in 2037. A sustained 50% increase in antismoking advertising exposure from 2018 reduces projected prevalence in 2037 by 0.80 percentage points. Similar reductions are projected with the introduction of tobacco wholesaler and retailer licensing schemes that either permit or prohibit tobacco sales by alcohol-licensed venues (0.65 and 1.73 percentage points, respectively). Increasing the minimum age of legal supply of tobacco products substantially reduces adolescent initiation, but has minimal impact on smoking prevalence in the adult population over the simulation period. Sustained reductions in antismoking advertising exposure of 50% and 100% from 2018 increase projected adult daily smoking prevalence in 2037 by 0.88 and 1.98 percentage points, respectively.
Conclusions These results suggest that any prudent approach to endgame planning should seek to build on rather than replace existing tobacco control measures that have proved effective to date. Additional interventions that can promote cessation are expected to be more successful in reducing smoking prevalence than interventions focussing exclusively on preventing initiation.
- end game
- public policy
Statistics from Altmetric.com
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.
As smoking prevalence in an increasing number of countries has declined to less than 20%, many public health researchers have argued that the focus of tobacco control efforts should shift from minimising tobacco use to eliminating it entirely. These researchers have proposed a range of ‘endgame’ strategies for phasing out tobacco products, including regulating nicotine levels in cigarettes, prohibiting the sale of tobacco to people born after a specified date, requiring smokers to obtain a prescription before purchasing tobacco, and introducing quotas on the manufacture and importation of tobacco products.1 Many of these proposed approaches aim to promote a relatively rapid decline in smoking prevalence to a level where legislation prohibiting the sale of tobacco products could (if necessary) be enacted with overwhelming public support (we assume here that this will be the usual goal of an endgame strategy). Most approaches have never been implemented, however, and some researchers have expressed concern that novel interventions are being overemphasised in endgame planning, at the expense of approaches that have already proved effective in reducing smoking prevalence (eg, antismoking mass media campaigns, taxation, smoke-free legislation).2 3
Dynamic simulation models, codeveloped with key stakeholders, provide a means of exploring the potential effectiveness of alternative policy options for endgame planning in tobacco control. Such models enable multiple sources of evidence, including published research, expert and local knowledge, practice experience and quantitative data (eg, from administrative health records, population surveys) to be integrated in a logically-consistent, interactive ‘what-if’ tool that can be used to improve understanding of the possible effects of proposed policy interventions before they are implemented in the real world.4–6 This paper presents a dynamic simulation modelling study of the potential impacts of several tobacco control policy approaches on adult daily smoking prevalence in the Australian state of Queensland. Primary aims of the study were: (1) to assess the capacity of existing tobacco control measures to sustain a rapid decline in smoking prevalence; (2) to explore the potential for further reducing smoking prevalence through additional policy interventions (specifically, tobacco wholesaler and retailer licensing and increasing the minimum age of supply); and (3) to examine the potential consequences of redirecting resources from existing tobacco control measures to novel endgame approaches.
Queensland is the third most populous state in Australia, with an estimated resident population of 5.01 million (as at 30 June 2018), comprising 20.1% of the total Australian population.7 Approximately 424 000 Queensland adults (18 years and above), including 12.2% of men and 10.0% of women, were daily smokers in 2018.8 The Queensland government has invested in a range of smoking reduction initiatives designed to prevent initiation, encourage and support cessation and minimise exposure to secondhand smoke as part of an overarching tobacco control strategy aimed at creating a culture in which smoking is not accepted as a social norm. Although significant progress has been made in reducing smoking prevalence, with the proportion of Queensland adults smoking daily declining from 17.9% in 2002 to 11.1% in 2018, there is a concern that without well-directed, ongoing investment in tobacco control initiatives, it will become increasingly difficult to achieve further gains. Several policy options for promoting a continued decline in smoking prevalence are available, including (but not limited to) scaling up existing programmes; however, which options are likely to be most effective in reducing prevalence—ultimately, to a point where no Queenslander smokes—is currently unclear.
A system dynamics model9 was developed using a participatory modelling approach that involved a range of local and national stakeholders, including representatives from state governments, health and social policy agencies, research institutions and community and consumer groups. Stakeholder input was elicited in a series of workshops, held between July and December in 2017, during which participants identified and mapped the primary causal pathways underlying the dynamics of smoking in Queensland, as well as the mechanisms by which a set of prioritised interventions influence those dynamics. The detailed causal map generated in these workshops was used to construct an initial computational model that was expanded and refined through an iterative process of stakeholder review and model revision and analysis. Model parameter estimates and other numerical inputs were derived from published research articles, publicly available data and unpublished data provided by stakeholders (see online supplementary appendix 1). The participatory modelling approach employed has been described in detail elsewhere.4 Model construction and analysis were performed using Stella Architect V.1.6.2 (www.iseesystems.com).
Model structure, outputs and calibration
The full system dynamics model comprises a set of interconnected submodels, or sectors, that includes: (1) a population sector, capturing changes in population size and structure resulting from births, migration, ageing and mortality; (2) a smoking status sector that captures changes in the numbers of adult daily smokers and ex-smokers, disaggregated by age group and sex; (3) an initiation sector that models smoking initiation rates among adolescents (aged 12–17 years) and young adults (18–29 years); and (4) a cessation and relapse sector that models cessation attempt and failure rates, mean cigarette consumption (which directly influences the cessation failure rate) and the rate of relapse among ex-smokers. Additional sectors capture the dynamics of e-cigarette use among adolescents and young adults, the prevalence of psychological distress (which directly affects multiple smoking-related behaviours), and hospitalisations for smoking-related diseases (lung cancer, chronic obstructive pulmonary disease, acute myocardial infarction, and stroke). The principal model components and their inter-relationships are depicted in figure 1. Detailed descriptions of all model sectors are provided in online supplementary appendix 1.
Primary model outputs (ie, estimated health outcomes) include daily smoking prevalence and numbers of daily smokers in the total Queensland population aged 18 years and above, daily smoking prevalence and numbers of daily smokers by age group and sex (including adolescent current smoking prevalence and numbers of adolescent current smokers), past smoking prevalence and numbers of ex-smokers in the total population and by age group and sex, initiation rates for adolescents and young adults, cessation attempt and success rates, and numbers of hospitalisations and hospitalisation rates for smoking-related diseases. All outputs are calculated every quarter over a period of 41 years, starting from 1 January 1997, so that the impacts of interventions are modelled from the time of implementation to the end of 2037 (simulations were run from 1997 to permit comparisons of model outputs with historical data). The analyses presented here focus on adult daily smoking prevalence in the total population. Additional analyses, using subpopulation-specific versions of the whole-population model, examined the potential impacts of interventions on smoking prevalence in Aboriginal and Torres Strait Islander, socio-economically disadvantaged, and regional and remote communities; the results of these analyses will be presented elsewhere.
Parameter values that could not be derived directly from published research or available data were estimated via constrained optimisation, using historical time series data on adult daily smoking prevalence and past smoking prevalence (disaggregated by age group and sex), current smoking prevalence among female and male adolescents, and sex-specific cessation attempt and success rates; this approach essentially involves adjusting the unknown parameter values to minimise discrepancies between the model outputs and historical data. A weighted sum of the mean absolute error for each time series (ie, the mean of the absolute differences between the observed data values and the corresponding model outputs) served as the objective function for the optimisation, where the weights were equal to the inverse of the time series mean (this weighting was performed to accommodate substantial differences in scale between, for example, adolescent smoking prevalence and the prevalence of past smoking). The set of parameter values minimising this function was obtained using Powell’s method with 1000 additional starts.10
Simulation experiments and sensitivity analyses
The analyses presented here focus on three policy interventions for reducing smoking-related harms that have been the subject of recent debate in Queensland, although the impacts of a range of other interventions can also be explored using the model. The three focal interventions involve: (1) modifying the current level of antismoking mass media campaign exposure; (2) introducing a tobacco licensing scheme; and (3) increasing the minimum age of legal supply of tobacco products from 18 to 21 (table 1). Eight alternative intervention scenarios (table 2) were compared with the baseline (business as usual) scenario, in which all tobacco control measures currently in place remain unchanged until the end of the simulation (a description of existing tobacco control initiatives and legislation is provided in ref).8
The impact on model outputs of uncertainty in estimates of the direct effects of each intervention on rates of smoking behaviour change was assessed via sensitivity analysis. Latin hypercube sampling was employed to generate 100 sets of values for all parameters determining the direct effects of the three interventions of interest on initiation, cessation attempt, cessation failure and relapse rates (as applicable) from a relatively broad distribution of values (online supplementary appendix 2). The effects of antismoking advertising exposure on initiation and cessation attempt and failure rates, which are described using graphical functions, were varied independently by multiplying the input level of advertising exposure for each function by a normal random variable with mean 1 and SD 0.2 (this operation effectively compresses or stretches the function along the horizontal axis; online supplementary appendix 2). Exposure to antismoking advertising influences the dynamics of smoking from the start of the simulation (1 January 1997), so altering the effects of a given level of exposure on rates of smoking behaviour change affects the fit of the model to historical data. Thus, for each set of parameter values sampled, we recalibrated the model before running the baseline and eight intervention scenarios described above. Differences in projected adult daily smoking prevalence at the end of 2037 between the business as usual and intervention scenarios were calculated for each set of parameter values and summarised using simple descriptive statistics.
Figures 2 and 3 present the projected prevalence of adult daily smoking in the total Queensland population for the baseline and eight intervention scenarios. Under the baseline scenario (business as usual), adult daily smoking prevalence declines from 11.77% (95% equal-tail interval, 11.74%–11.81%) in 2017 to 5.58% (5.45%–5.71%) in 2037 (note that all intervals reported in this paper are derived from the distributions of model outputs calculated in the sensitivity analyses; they indicate the impact of uncertainty in selected parameter estimates, but should not be interpreted as confidence intervals).
When compared with the baseline scenario, a sustained 50% increase in antismoking advertising exposure from 2018 reduces the prevalence of adult daily smoking in 2037 by 0.80 (0.59–1.00) percentage points (figure 4). Similar reductions are projected for both tobacco licensing scenarios; adult daily smoking prevalence declines to 4.93% (4.45%–5.40%) in 2037 with the introduction of a licensing scheme that permits the supply of tobacco products by alcohol-licensed venues (a reduction of 0.65 percentage points), and to 3.85% (3.20%–4.54%) in 2037 with the introduction of a licensing scheme prohibiting tobacco sales by alcohol-licensed venues (a 1.73 percentage-point reduction). The greatest reduction in adult daily smoking prevalence is achieved when a sustained 50% increase in antismoking advertising exposure is combined with a tobacco licensing scheme that prohibits tobacco sales by alcohol-licensed venues; under this scenario, adult daily smoking prevalence is projected to decline to 3.18% (2.49%–3.78%) in 2037, 2.40 (1.75–3.01) percentage points lower than the baseline prevalence (figure 4).
An increase in the minimum age of supply of tobacco products to 21 commencing in 2018 substantially reduces the initiation rate among adolescents, but has very little impact on daily smoking prevalence in the adult population (figure 3). Under the business as usual scenario, the adolescent initiation rate declines from 0.00366 (0.00339–0.00391) smokers per person per year in 2017 to 0.00186 (0.00174–0.00197) smokers per person per year in 2037. Increasing the age of supply to 21 reduces the projected adolescent initiation rate in 2037 by nearly 30% to 0.00133 (0.000827–0.00172) smokers per person per year; however, the projected prevalence of adult daily smoking in 2037 is reduced by only 0.08 (0.02–0.16) percentage points (1.49% of the baseline value; figure 4).
Significant reductions in antismoking mass media campaign exposure have a substantial impact on adult daily smoking prevalence. A sustained reduction in antismoking advertising exposure of 50% from 2018 increases the projected prevalence of adult daily smoking in 2037 by 0.88 (0.66–1.07) percentage points (ie, compared with the business as usual scenario), and complete disinvestment in mass media campaigns, resulting in a 100% reduction in exposure, increases projected prevalence in 2037 by 1.98 (1.46–2.50) percentage points (figure 4).
The simulation results presented above support a number of general conclusions about the impacts of potential tobacco control policy interventions on the dynamics of smoking in Queensland over the next two decades. First, ongoing investment in the strong tobacco control measures currently in place—including tobacco advertising and sponsorship bans, smoke-free environment legislation, taxation on tobacco products, and significant investment in antismoking mass media campaigns and cessation support services—is sufficient to sustain a decline in adult daily smoking prevalence, which is projected to decrease to 6% in 2037 under the business as usual scenario. Second, there is considerable scope for reducing smoking prevalence further through additional policy interventions, scaling up existing tobacco control measures, or a combination of both of these options. Nevertheless, additional interventions focussing exclusively on preventing initiation may not be very effective in reducing adult smoking prevalence; although increasing the minimum age of legal supply of tobacco products has the intended effect of substantially reducing the adolescent initiation rate, this intervention has minimal near-term impact on daily smoking prevalence in the adult population. And third, disinvestment in current tobacco control measures, including mass media campaigns, may substantially reduce the rate of decline in smoking prevalence, so that any redirection of resources from existing measures to novel endgame interventions involves a potentially significant amount of risk.
Several researchers have proposed that in countries that have substantially reduced smoking prevalence through effective public health interventions, the population of remaining smokers will become increasingly less responsive to those interventions as prevalence continues to decline (since smokers who could be prompted to quit will have already done so).11–13 According to this ‘hardening hypothesis’, future progress in reducing smoking prevalence in these countries will depend on shifting the emphasis of tobacco control interventions towards intensive treatment of individual smokers.12 However, recent analyses of population survey data provide no evidence that significant declines in smoking prevalence in the USA, Europe and New Zealand have been associated with a ‘hardening’ of the smoking populations of these regions, suggesting that current tobacco control strategies will continue to be effective in reducing smoking prevalence.14 15 Our results are consistent with the results of these studies, indicating that existing tobacco control measures are sufficient to maintain a substantial rate of decline in adult smoking prevalence in Queensland. Nevertheless, it should be emphasised that, although our model captures changes in mean cigarette consumption (one measure of ‘hardening’) and its influence on cessation failure rates, we have not modelled the population of ‘hardened’ smokers explicitly.
Our conclusion that tobacco control interventions focussing exclusively on preventing adolescent initiation are unlikely to be effective in reducing smoking prevalence in the adult population (at least in the time frame considered here) is consistent with the results of previous modelling studies.16 17 The number of adolescent smokers currently entering the Queensland adult smoking population each year is considerably less than 1% of the total number of adult daily smokers (0.54% based on model outputs for 2017), so that even a substantial reduction in the adolescent initiation rate will have minimal immediate effect on population smoking prevalence. Although a significant reduction in adolescent initiation would be expected to eventually produce a substantial decrease in adult smoking prevalence (because fewer adult smokers removed from the smoking population by cessation or mortality are replaced by new smokers), the delay between implementing an effective prevention intervention and observing a significant reduction in population smoking prevalence may be several decades.16 Policy options such as increasing antismoking advertising campaign expenditure and tobacco licensing, which directly affect the behaviour of established smokers, are able to actively reduce the number of adult smokers in the population, and so provide a more effective means of promoting a rapid decline in smoking prevalence.
The substantial impact of disinvestment in antismoking advertising campaigns on projected smoking prevalence observed in our simulations substantiates concerns that de-emphasising conventional tobacco control interventions in endgame planning could jeopardise future progress in reducing smoking-related harms.2 3 Studies examining the effectiveness of state-level tobacco control programmes in the USA and Australia provide empirical evidence that disinvestment in existing tobacco control measures may adversely affect population health outcomes, consistent with our simulation results.18 19 Relatively rapid initial declines in per capita adult cigarette consumption and adult smoking prevalence associated with the California Tobacco Control Program, introduced in 1989, stalled abruptly following an approximately 40% reduction in funding in 1993–1994.19 An abrupt decrease in the rate of decline in smoking prevalence was also observed in South Australia after government investment in antismoking mass media campaigns was suspended in July 2013, prompting a partial reinstatement of funding in July 2014.18 Although it is generally accepted that novel tobacco control policies will be integral to any effective endgame strategy, these results suggest that redirecting resources from conventional tobacco control measures to untested interventions entails a considerable level of risk.
This study has a number of important limitations that should be pointed out. First, our model captures only the aggregate dynamics of smoking-related behaviours in Queensland, and so effectively disregards potentially significant heterogeneity in relevant exposures and their effects among demographically-similar individuals; the model assumes, for example, that the level of exposure to antismoking advertising campaigns does not vary across the population, and that adult daily smokers of a given age and sex respond uniformly to a specified level of advertising exposure (this focus on aggregate, rather than individual-level, dynamics is an inherent feature of system dynamics models). Second, the data used in deriving estimates for parameters in the model vary considerably in quality and were not always directly applicable to our study population (online supplementary appendix 1). Where possible, estimates were derived from systematic reviews or population-based, preferably Australian cohort studies; however, in many cases, the only available data were from smaller cohort studies or cross-sectional analyses. Third, a number of potentially significant influences on the future dynamics of smoking in Queensland, including trade in illicit tobacco and the use of e-cigarettes as cessation aids, were not considered, due mainly to time and data availability constraints. Fourth, there is currently only limited evidence indicating the probable sizes of the direct effects of tobacco licensing on smoking-related behaviours, so the effects specified for this intervention should be considered provisional (see table 1). Despite these limitations, however, comparisons of model outputs and historical data indicate that our model is able to reproduce important aspects of system behaviour with reasonable accuracy (online supplementary appendix 1). Moreover, sensitivity analyses demonstrate (reassuringly) that our primary conclusions about the potential effectiveness of alternative interventions do not depend on the particular values specified for key model parameters (see figure 4).
Although our findings are generally consistent with those of previous empirical studies and simulation analyses, it should be emphasised that the tobacco control policies in place in Queensland are far more stringent than in many parts of the world, which may restrict the broader applicability of our conclusions. The limited effectiveness of interventions focussing on preventing initiation in reducing population smoking prevalence indicated here, for example, is attributable primarily to the very low rate of initiation in the adolescent population, and therefore may not be expected in other contexts where the adolescent initiation rate is significantly higher. Additionally, the vast majority of novel approaches to phasing out tobacco products that have been proposed were not modelled as part of this project, and some of these approaches could have considerably greater projected impacts than the restricted set of interventions analysed here. Nevertheless, it should also be noted that these caveats do not detract from the general conclusion that maintaining (or increasing) investment in effective tobacco control measures such as those currently in place in Queensland should be a principal concern in endgame planning.
The results of our simulation analyses indicate that the comprehensive set of tobacco control policies currently in place in Queensland is capable of sustaining a substantial rate of decline in daily smoking prevalence (currently less than 12% in the adult population). However, there is also considerable potential for accelerating this decline through additional policy interventions that can encourage cessation and reduce the chance of relapse among existing and past smokers. Although additional interventions focussing exclusively on prevention may substantially reduce smoking uptake among adolescents, the potential near-term impact of such interventions on population smoking prevalence is limited, given the already low rate of adolescent initiation. Perhaps most importantly, our simulation results indicate that disinvestment in existing tobacco control measures (eg, antismoking advertising campaigns) has the potential to impact substantially on future trends in smoking prevalence, suggesting that any prudent approach to endgame planning should seek to build on rather than replace conventional policy interventions that have proved effective in reducing smoking prevalence to date.
What this paper adds
A range of ‘endgame’ strategies have been proposed to effect relatively rapid declines in smoking prevalence to levels where legislation prohibiting the supply of tobacco products could be enacted with overwhelming public support.
Many of these proposed approaches have never been implemented, however, and there is some concern that novel interventions are being overemphasised in endgame planning at the expense of more conventional measures that have proved effective in reducing smoking prevalence to date.
Our simulation results indicate that existing tobacco control policies are capable of sustaining a substantial rate of decline in population smoking prevalence to near 5% (at least).
Although there is considerable potential for accelerating this decline through additional policy interventions that can encourage cessation and reduce the chance of relapse among current and past smokers, the near-term impact of novel interventions focussing exclusively on preventing initiation may be limited.
Disinvestment in conventional tobacco control measures has the potential to impact substantially on future trends in smoking prevalence, suggesting that any prudent approach to endgame planning should seek to build on rather than replace existing policy measures.
The system dynamics model presented in this paper was developed by Decision Analytics (Sax Institute) in partnership with Queensland Health’s Preventive Health Branch and The Australian Prevention Partnership Centre. The authors are immensely grateful to all members of the Queensland Tobacco Modelling Group for their contribution to the model development process: Rachael Bagnall (Cancer Council Queensland), Billie Bonevski (University of Newcastle), Simone Braithwaite (Queensland Health), Susan Clemens (Queensland Health), David Cockatoo-Collins (Department of Aboriginal and Torres Strait Islander Partnerships, Queensland), Anne Curtis (Health Consumers Queensland), Alison Durham (National Heart Foundation of Australia), Coral Gartner (University of Queensland), Marita Hefler (Menzies School of Health Research), Madonna Kennedy (Queensland Health), Louise Mahoney (Department of the Premier and Cabinet, Queensland), Jane Martin (Queensland Health), Irene McCarthy (Queensland Health), Roger Meany (Queensland Health), Richard Mills (Institute for Urban Indigenous Health), Regina Mullins (Department of Housing and Public Works, Queensland), Shelley Peardon (Queensland Health), Kaye Pulsford (Queensland Health), Barbra Smith (Northern Queensland Primary Health Network), Tanea Smith (Department of Aboriginal and Torres Strait Islander Partnerships, Queensland) and Dishan Weerasooriya (Department of Health, Western Australia).
Contributors RW, MW, J-AA, AS and PW conceived the study. PW, RW, MW and J-AA organised and ran the workshops. AS built the system dynamics model, performed the analyses and drafted the paper. TR and MB provided historical data used for model calibration. All authors contributed to model development and preparation of the final manuscript.
Funding This research was funded by the Queensland Department of Health.
Disclaimer The views expressed in this paper are those of the authors and do not represent the views of the Queensland Department of Health or Queensland Government policy.
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information.