This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

This study aimed to describe past time trends of the prevalence of former smokers in Italy and to estimate prevalence projections using a Bayesian approach. An age-period-cohort (APC) analysis has been carried out in order to investigate the effect of the age, period and birth cohort on the prevalence of former smokers during 1980–2009. A Bayesian APC model with an autoregressive structure for the age, period and cohort parameters has been used to estimate future trends. Results showed that awareness of harm from smoking occurred at younger ages with each advancing cohort, and that women were more likely to attempt to stop smoking during pregnancies and breastfeeding, whereas men attempted to quit only when smoking-related diseases became evident. Projections of future trend recorded a further increase in the number of former smokers in future decades, showing an estimate of the “end of smoking” around years 2060 and 2055 in men and women, respectively. The application of the APC analysis to study the prevalence of former smokers turned out to be a useful method for the evaluation of past smoking trends, reflecting the effects of tobacco control policies on time and generations, and to make projections of future trend.

Smoking is a major risk factor for many tumours and other chronic diseases, and reduces length and quality of life [

In Italy, several policies have been already implemented since the 1970s. A smoking ban in hospitals, schools, cinemas, and public transportations was introduced in 1975, followed by a smoking ban in front-offices of public administrations in 1995, and finally by a comprehensive smoking ban in all workplaces and in the hospitality sector in 2005 [

In response to all these efforts, smoking rates in Italy have progressively declined. Male smoking prevalence declined from 41.6% in 1986 to 29.5% in 2009, an average annual drop of 1.2%. Meanwhile, female smoking prevalence declined from 19.2% in 1986 to 17.0% in 1993, and stalled at that level [

The effect of smoking cessation policies may be highlighted by studying the prevalence of former smokers. Variations in prevalence of former smokers over time may be due to age, to a period effect, which is a reflection of changes in pro-tobacco efforts and smoking cessation policies in time, as well as to a cohort effect that may reflect different cumulative exposures to pro- and anti-tobacco efforts [

This study investigated the effects of age, time period, and birth cohort on trends of Italian former smokers as a reflection of smoking cessation policies, by using the APC models on the trends of prevalence of former smokers. In addition, we predicted future trends of the prevalence of former smokers in Italy using a Bayesian APC approach, which incorporates prior information about smoothness on each time scale to reduce random variation and improve the precision of the projections [

The purpose is to use methods typically used in other contexts in order to advance our understanding of the trends of tobacco smoking and to provide data supporting evidence-based planning for more effective tobacco control. In addition, an estimate of the end of tobacco smoking may be given using this approach.

Data on smoking prevalence by gender and age were available from the Italian National Institute of Statistics (ISTAT) Multipurpose Surveys “Health conditions and access to health services” and “Aspects of daily living”. The two surveys are a surveillance system carried out almost every year between 1980 and 2009, and were based on similar scientific design providing comparable data on smoking habits over the years [

Time trends of the prevalence of former smokers were evaluated through an APC analysis. Data were arranged in six 5-year periods (1980–1984, 1985–1989, 1990–1994, 1995–1999, 2000–2004, and 2005–2009) and twelve 5-year age groups (25–29 to 80–84 years). These age groups and calendar periods involved 17 overlapping 5-year cohorts identified by central year of birth, and defined using the relationship cohort = period − age [

We assessed the age, period, and cohort effects by means of generalized linear models assuming that the number of former smokers follows a Binomial distribution [

In the APC analysis we obtained similar results for both genders (

In

Age-specific prevalence of former smokers (

APC model assessment for prevalence of former smokers in Italy in periods 1980–2009, and 1995–2009 for males and females.

1980–2009 | 1995–2009 | |||||
---|---|---|---|---|---|---|

Males | AIC | DEV | Δ-Dev | AIC | DEV | Δ-Dev |

APC | 627 | 43,391 | - | 168 | 5276 | - |

AP | 1642 | 115,482 | 72,091 | 687 | 23,494 | 49,635 |

AC | 5,689 | 402,846 | 359,455 | 181 | 5,734 | 392 |

Females | ||||||

APC | 559 | 38,613 | - | 196 | 6,293 | - |

AP | 2,829 | 199,872 | 161,259 | 1338 | 46,301 | 40,007 |

AC | 5,398 | 382,230 | 343,618 | 239 | 7,791 | 1,497 |

Age, period, and cohort effects from the APC model over the period 1980–2009 for (

Bayesian projections showed a global convergence with the Geweke statistic over 1.96 for 3 and 8 parameters out of 132 for males and females, respectively. Sensitivity analyses confirmed the stability of posterior prevalence values over different values of the prior hyper parameters, producing posterior prevalence values that fell inside the credible interval. In particular, the maximum difference among posterior projections was of 0.44 and 2.32 points percentage in males and in females respectively, in both cases observed for the last year of projections.

Age-specific former-smoker prevalence fitted and projected by cohort using the APC Bayesian model with the 1980–2009 period as basis for projections are reported in

In

Age-specific prevalence of former smokers by period, fitted in 1980–2009 (solid line) and projected for 2010–2030 (dotted line) for males and females.

Fitted and projected prevalence of former smokers cumulated by age (25–85 years) with 95% credible intervals by period for males and females. Dots are the observed cumulated prevalence.

We described time trends of the prevalence of former smokers and projected trends for the next decades in Italy. In 1980–2009 younger cohorts showed higher prevalence values for all ages in both genders. Rise in prevalence occurred earlier in life, as for more recent cohorts. This is likely to reflect awareness of harm from smoking that occurred at younger ages in each advancing cohort [

The APC analyses showed that the period effect reached its maximum in 1992–2002, as an effect of the laws implemented in that period, such as smoking ban on TV advertisement (1991), smoking ban on public administrations (1995), and rise in cigarettes taxes (from 1990). At the same time, the cohort effect showed a progressive continuous increase since the first generation, suggesting that awareness of harm from smoking progressively became stronger. The analysis restricted to the time period 1995–2010 revealed an increasing importance of the cohort.

Projections of the cumulated prevalence showed that former smokers will reach 42.7% of men and 32.1% of women in 2030 (

Projections on the smokers prevalence (%) for males and females.

Year | Prevalence (%) | |
---|---|---|

Males | Females | |

2012 | 23.44 | 14.55 |

2017 | 22.10 | 14.24 |

2022 | 19.37 | 12.06 |

2027 | 16.96 | 9.58 |

2032 | 14.72 | 7.05 |

2037 | 11.73 | 6.37 |

2042 | 9.09 | 4.58 |

2047 | 6.45 | 2.80 |

2052 | 3.80 | 1.01 |

2057 | 1.16 | 0.00 |

2062 | 0.00 | 0.00 |

The novelty of this work is the use of APC methods, which are usually used for the study of trends in deaths rates, to study trends in the prevalence of former smoker, and to make projections in a Bayesian perspective.

This study has some limitations. First, prevalence projections resulted highly uncertain. The credible intervals encompass both uncertainty associated with the choice of the model and uncertainty associated with projecting beyond the range of the data. This is necessarily reflected by the rapidly increasing width of the intervals as the length of the projection increases [

In conclusion, this study showed a constant increase in prevalence of former smokers in Italy since 1980s, and regarding all generations and age-groups. Projections of future trend recorded a further increase in the number of former smokers in future decades, showing an estimate of the “end of smoking” around years 2060 and 2055 in men and women, respectively.

This work was published with the contribution of Istituto Toscano Tumori (ITT), grant proposal 2008.

The authors declare no conflict of interest.

Following Breslow and Clayton [

According to this model each point (except the first two on each scale) is predicted by linear extrapolation from its two immediate predecessors, plus a random error from a normal distribution with mean 0. The first two parameters for age, period and cohort effects are given non-informative priors. The distributions for the α_{1},…, α_{A} age, β_{1},…, β_{P} period, and γ_{1},…, γ_{C} cohort effects, can be defined as follows:

The precision of the normal distributions can be represented as a hyperparameter for which the prior distribution reflects prior belief concerning the smoothness of the parameters. In our analyses we use non-informative hyperprior distributions so that the hyperparameters are estimated solely from the data:

The count of former smokers _{ap}_{ap}_{ap}_{ap}_{ap}

We carried out a sensitivity analysis on the prior distributions, to evaluate the robustness of the results. In particular, we changed the mean of the normal distributions for the first two parameters for age, period, and cohort from 0 to 0.01. Then we changed the shape and inverse scale parameters of the Gamma distributions for the age, period and cohort one at time from 0.001 to 0.005.