In this study we evaluated the impact of several smoking prevalence reduction and cessation scenarios on average life expectancy in Australia, with interventions targeting increasingly older age groups. It was appropriate to relate the gains by a reduced smoking prevalence to 10 %, to the estimated gains in the case of total smoking cessation because the inclusion of complete cessation as a scenario allowed us to quantify the maximal gain in average life expectancy which can be used as a benchmark with which to compare gains from more realistic prevalence reduction targets.
We found that reducing the prevalence of smoking in the Australian population to 10 % could increase average life expectancy of all men by approximately 0.1 to 0.4 years, and all women by approximately 0.1 to 0.3 years. These are at best 54 % and 49 % for men and women respectively of the maximal life expectancy gains that could be achieved by complete smoking cessation. Amongst smokers the potential gains are greater, with an increase in average life expectancy amongst men smokers of approximately 0.4 to 2 years, and approximately 0.74 to 2 years amongst women smokers. However the gains in life expectancy for men and women smokers associated with 10 % smoking prevalence were at best 46 % and 38 % respectively of the maximal gains achievable through complete elimination of smoking. The maximal gains in life expectancy for men and women associated with 10 % smoking prevalence at the population level occurred when both aged less than 60 years were targeted. This indicates that including older people in the targeted population dilutes the effect of smoking reduction interventions, and therefore concentrating smoking cessation programs on populations aged less than 60 years represents a more optimal strategy.
As has been noted previously it is difficult to interpret gains in life expectancy from preventive interventions as any gains are usually averaged across an entire target population and may range from just weeks to months [13]. The target population will include individuals who gain little additional life expectancy as well as individuals who gain years of extra life. After reviewing 83 published reports of gains in life expectancy from a range of medical interventions, including both preventive measures and disease treatments, Wright and Weinstein [13] concluded that “a gain in life expectancy of a month from a preventive intervention targeted at populations at average risk and a gain of a year from a preventive intervention targeted at populations at elevated risk can both be considered large”. Since our estimated maximal gains in life expectancy to achieve a smoking prevalence of 10 % is at best 0.4 years for men and 0.3 years for women, for a population at an elevated risk of mortality such as Australia, the results we have obtained are consistent with potentially moderate or medium gains in life expectancy if smoking prevalence could be reduced to 10 %, in line with the Government set targets. The moderate gains in life expectancy at the population level are plausible given the already relatively “low” smoking prevalence in the population as the non-smokers dilute them. However, the gains are large when considering smokers only. In other settings or countries with higher baseline prevalence the gains would be larger at both the population level and for smokers.
Previous studies often defined smoking reduction as at least 50 % reduced from baseline at a population level without targeting of specific age or sex groups [14, 15]. In contrast, our study targets specific age groupings for each sex for reductions in smoking prevalence. Also, our targets are defined according to policies adopted by successive Australian governments to reduce smoking prevalence to 10 % by 2018. This is the first study in Australia which quantified improvements in survival due to reductions in smoking prevalence amongst different age and sex groups. It demonstrated a clear and positive gradient in life expectancy with increasing reduction in smoking prevalence in the Australian population.
We restricted our analysis to all-cause mortality as it is an objective endpoint and is available for all individuals in the AUSDIAB study. However, it is unclear whether the relative impact of the interventions we studied on mortality translates to other health outcomes such as major smoking related diseases and their associated disability burden. Further work is needed in this area.
Whilst the analysis was performed in a cross-sectional setting, the changes in the smoking behavior have primarily long-term effects. The life table method enabled us to estimate the long-term effect of smoking reduction on life expectancy, on the assumption that the mortality rate for each risk percentile which was estimated on the basis of all baseline risk factors (for the baseline scenario) and the simulated distribution of smoking having a prevalence of 10 % and the actual baseline distribution of other risk factors (for each intervention scenario), would continue for the rest of the life of each member of a risk percentile. We did not directly model the lag or delay in the health benefits of smoking cessation. However, if there is any lag or delay in health benefits of quitting, we do not anticipate that to be very significant at the population level.
Although it is possible to estimate the number of deaths prevented or reductions in mortality rate due to reductions in smoking prevalence using the simpler population attributable risk method, our research question was to estimate improvements in life expectancy for which the population attributable risk method alone is not sufficient as an additional method, for example, the life table method will also be required to estimate life expectancy using the mortality rate estimated by the population attributable risk method. Since the risk percentiles method already incorporates the life table, we can estimate life expectancy as a direct output.
We gradually targeted ages <30 years, <40 years and so on below higher ages to determine the age after which there is no further gain in life expectancy or there is a slowing down in the gain. Performing the analyses for ages 30+, 40+ and so on would not have allowed us to achieve this objective (and so were not performed in this study). Also, targeting the population by below a certain age would be helpful to target fewer people compared to targeting the entire AUSDIAB sample based on the full age range of 25-80 + .
We targeted the current smokers to quit smoking in order to reduce smoking prevalence to 10 %. The issue of ex-smokers having higher risk of mortality than never smokers could not be modelled because the EURO SCORE [8] did not make any distinction between ex-smokers and never smokers while estimating risk scores as the risk equation which was used to estimate these had only two categories for smoking-current smokers and nonsmokers (nonsmokers included both ex-smokers and never smokers). In effect, the mortality rates among never and former smokers were averaged. The inability to distinguish between never and former smokers while estimating absolute risk of individuals is a limitation of this study. Despite this limitation of the EURO SCORE, it was found to recalibrate well to the AUSDIAB sample [9].
Since the smoking prevalence estimated using the 1999-2000 (baseline) AUSDIAB survey is approximately 16 %, the projected target of 10 % by 2018 seems superficially achievable. However, using dynamic forecasting modelling Gartner et al. [16] found that as the initiation rate has been declining in Australia, to achieve 10 % prevalence by 2020, the current cessation rate should be doubled. However, latest estimates in Australia indicate that the smoking rate among adults aged 14 years and over was 12.8 % and for 18 years and over was 13.3 % in 2013 [17]. There was a statistically significant decline between 2010 and 2013-the period during which the government target to achieve 10 % prevalence by 2018 was well in force. With 5 years remaining to reach 2018 since 2013, the target of achieving 10 % prevalence by 2018 seems to be achievable.
Since we have shown that 10 % smoking prevalence gives at best half of maximum possible gains under the scenario of complete cessation, further modelling will help to answer questions such as whether reduction of smoking prevalence to 5 % will remediate most of, or little of, the differential between reduction to 10 % prevalence and complete cessation. To achieve a target of 5 % we modelled for men and women gains in life expectancy for a few selected scenarios (results not shown), like the random reduction to a smoking prevalence of 5 % for the whole population and random reduction to 5 % prevalence for those below 60 years, and found that there was almost a proportionate decrease in gains in life expectancy compared to the corresponding scenario with a target of 10 % prevalence. The potential gains for men and women were at best 0.63 and 0.5 years which are 73 % and 68 % respectively of that obtained by complete smoking cessation. Thus, setting a target of 5 % prevalence would remediate most of the differential between reduction to 10 % prevalence and complete cessation. The analysis and issues presented in this paper are highly complex and as is the case in all scientific endeavours, there is always a possibility that methodological problems may affect the results and interpretation of the findings. One major concern is obviously the cross-sectional approach.