Methods

We analysed individual data on the basis of the two most recent Italian national health interview surveys, which were carried out by the Italian National Statistical Institute in 1999–2000 and 2004–2005. These surveys were representative of the entire non-institutionalised national population. Both surveys included a random sample of non-institutionalised subjects, and used interviewer as well as self-administered questionnaires. Interviews were conducted by civil servants (appointed by each municipality included in the study) within the homes of the families identified. Each survey collected data in different time periods throughout one year in order to reduce the effect of seasonal variability. Sample sizes were about 140 000 and 120 000, while response rates were 87.0% and 85.7%, for the oldest and for the most recent survey, respectively. The design of these surveys was previously described.23 24

From the electronic data files, we extracted information on age, gender, education, smoking status, age at starting smoking and age at quitting smoking. Educational attainment was used as an indicator of socioeconomic position, since it was found to be a relatively strong predictor of smoking prevalence in adult populations.25 Furthermore, among adults, this indicator is much more stable over an individual lifetime than income or occupation. We classified all individuals who completed upper secondary education as higher educated, while those with a lower degree or with no formal educational were classified as lower educated.

We included in the analysis individuals who reported to have ever smoked at least one year. Questions on smoking behaviour were the following: “do you currently smoke?”, “how old were you when you started smoking regularly?”, “how old were you when you quit smoking?” and “how frequently do you currently smoke (for former smokers: how frequently did you use to smoke)?”. Occasional smokers (that is, those who did not smoke on a daily basis), as well as cigar or pipe smokers and those with a duration of smoking less than one year were excluded from the analysis. Smoking cessation was defined as abstinence from smoking for 12 months or longer, similar to a previous study.26

Smokers aged 20–62 at the time of the oldest survey (n = 34 789), and smokers aged 20–67 at the time of the most recent survey (n = 33 135) were included in the analyses. On the basis of respondents’ age, age at starting and age at quitting smoking, we computed age-specific smoking cessation rates during each 3-year period between 1982 and 1996 for the oldest survey, and rates between 1982 and 2002 for the most recent survey. Smoking cessation rates were computed for smokers aged 20–44 years at the time of possible smoking cessation, with distinction by 5-year age group. Rates were calculated by dividing the number of individuals who quit smoking during a defined 5-year age group and 3-year period by the total number of person-years lived during the same age and period by respondents who were smokers.

The analysis was limited to young adult smokers (that is, 20–44 years old) between 1982 and 2002 in order to reduce the potential bias due to selective mortality. Since we used data of two recent cross-sectional surveys, data analysis was performed retrospectively on a “survivor” cohort—that is, on subjects who were still alive at the time of the survey.

From the age-specific cessation rates, we derived two summary measures of smoking cessation—age-standardised rate and the cumulative probability of smoking cessation. The former was computed by means of the direct method, using a “flat” reference population; in other words, equal weight was given to each age category. The latter summary measure was calculated with a period life table using age-specific probabilities of smoking cessation as input. These were derived from the age-specific rates of smoking cessation according to the formula

p(x, x+5) = 1−exp[−5r(x, x+5)]

where p(x, x+5) is the probability of quitting between age x and age x+5 and r(x, x+5) is the cessation rate in the same period.27 The cumulative probability of smoking cessation at age 45, describes, for smokers aged 20, the expected probability of smoking cessation up to age 45, if the age-specific cessation rates observed cross-sectionally apply over their own lifetime.

Absolute and relative measures of inequalities were computed calculating the age-standardised rate difference (RD) and age-standardised rate ratio (RR) between higher and lower educated respondents. Poisson regression models were computed in order to estimate the effect of educational level, adjusting for age and period. Period was entered in the models as a nominal variable, since smoking cessation rates did not show a linear relation with time. Interaction terms were added to the model in order to evaluate whether the effect of education varied with time. All analyses were stratified by sex. Stata 10 was used for all statistical analyses.

Results

Table 1 describes the characteristics of ever smokers included in the present analyses. Although there were between-survey differences in the proportion of former smokers and in the median age of quitting smoking, in both surveys the proportion of former smokers was generally higher among males, and among the higher educated subjects. Furthermore, median age at quitting smoking was lower among the higher educated subjects and among females in both surveys.

Tables 2 and 3 show age standardised rates of smoking cessation among higher educated and lower educated subjects aged 20–44 years by 3-year period for males and females, respectively. Higher educated subjects constantly showed higher quit rates than lower educated subjects. Among males, smoking cessation rates were quite stable at a figure of about 2 per 100 person-years until 2000, when they rapidly increased up to 3–4 per 100 person-years (table 2). In 2000–2002, the RR of smoking cessation for a higher educated compared to a lower educated subject was 1.47 (95% CI 1.28 to 1.69), while the RD per 100 person-years was +1.29 (95% CI +0.81 to +1.78). In the early 1980s, similar figures were found for RRs, whereas RDs were smaller, although confidence intervals overlapped. Quit rates estimated from the oldest survey (for the period 1982–1996) were different for some periods compared to those estimated from the most recent survey, generally showing slightly higher values. However, a similar pattern was observed for the two data sources.

Among women, smoking cessation rates were stable until the last period. Among higher educated females, however, a temporary increase in smoking cessation occurred in the late 1980s (table 3). For both higher educated and lower educated women, a sudden increase in smoking cessation rate is evident in the period 2000–2002. Between 1997–1999 and 2000–2002, the rate ratio (RR) of smoking cessation for higher educated compared to lower educated women remained stable at a figure of about 1.3, while the rate difference (RD) per 100 person-years increased from +0.55 (95% CI +0.11 to +0.98) to +1.00 (95% CI +0.39 to +1.61). Again, data from the oldest survey show a similar pattern, although quit rates estimated from the oldest survey were in some cases higher than those estimated from the most recent survey.

Over the whole time period, the age-standardised RR of smoking cessation for a higher educated compared to a lower educated smoker was 1.33 (95% CI 1.25 to 1.41) among males, and 1.41 (95% CI 1.30 to 1.53) among females. Poisson regression modelling (which was performed separately for men and women) showed that a significant increase in the rates of smoking cessation was present only in the early 2000s among males, while among females rates increased during the mid and late 1980s, decreased afterwards and largely rose after 1997. Significant interactions were found between educational level and period among both males and females: among males, higher educated smokers showed a larger increase in smoking cessation than their lower educated counterparts during the early 2000s, whereas among females a significant increase in educational inequalities was found during late 1980s.

The life-table analysis allowed us to estimate the cumulative probability of smoking cessation up to the age of 45. This is shown in figure 1, separately for males and females of lower and higher educational level, for the initial and final 3-year period. In all graphs, the curve for the lower educated subjects lies below that for the higher educated, reflecting a lower cumulative probability of smoking cessation before the 45th birthday. In 1982–1984, the cumulative probability of smoking cessation before age 45 was 35.5% and 44.0% for lower educated and higher educated males, respectively, while two decades later, it increased to 49.8% (lower educated males) and 63.7% (higher educated males). Among females, corresponding figures for lower educated versus higher educated subjects were 27.3% and 33.5% in 1982–1984, and 54.1% and 64.2% two decades later. Thus, the absolute difference in the cumulative probability of smoking cessation increased from the first to the last period for both men (9–14 percentage points) and women (6–10 percentage points).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Cumulative probability of smoking cessation in 1982–1984 and 2000–2002.

Discussion

Summary of the main findings

Between 1982 and 1999, smoking quit rates were approximately constant among both higher and lower educated males and females, whereas they rapidly increased afterwards. Higher educated smokers constantly showed higher cessation rates than lower educated subjects. The relative size of educational difference in smoking cessation did not significantly vary over time. However, there was a tendency to an increase in absolute inequalities in cessation rates in 2000–2002.

Limitations of the study

To retrospectively reconstruct individuals’ smoking histories has several limitations. First, the smoking experience of individuals who died before they could take part in the survey cannot be taken into account. Smokers may have had higher mortality rates than both never smokers and former smokers, and this may have inflated our estimate of quit rates. This is particularly true for the elderly, who have higher mortality rates compared to younger subjects. In our study, we therefore limited the analysis to subjects aged 20–44 in each 3-year period. For the period further back in time (1982–1984), quit rates were estimated on the basis of data reported by subjects aged 37–62 at the time of the first survey (1999–2000), and by subjects aged 42–67 at the time of the second survey (2004–2005). The study groups for which quit rates were estimated were progressively younger for the following 3-year periods. By limiting the analysis to relatively young age groups, we aimed to restrict the potential bias due to selective mortality.

Second, smoking relapses are not taken into account in the survey, because respondents report only the “final” date of smoking cessation. Subjects who have just recently stopped smoking may have defined themselves as “former smokers” at the time of the survey and may have relapsed shortly after the survey. If this was the case, smoking cessation rates shortly before the survey would have been artificially inflated. To overcome this limitation, we adopted a stringent criterion, defining smoking cessation as abstinence from smoking of at least 1 year. Given that most relapses occur within 6–12 months,26 our estimates of quit rates are not affected by this problem to a large extent.

Third, under-reporting of smoking was found in several studies.6 28 These studies compared information on legal sales of tobacco products in Italy with data on self-reported smoking collected in health interview surveys. They showed that subjects are currently more likely to under-report smoking than they were in previous years. This may produce both an artificial downward trend in smoking prevalence over time, and an underestimate of the average daily number of cigarettes reported by smokers. However, it is unlikely that under-reporting influenced estimates of smoking cessation as well, because these estimates were measured through retrospective survey questions.

A fourth problem relates to biased recall of age at smoking cessation. Previous studies found a good level of agreement between contemporaneous and retrospective reports of smoking status.29 30 If respondents postponed their actual date of smoking cessation (a systematic error known as forward telescoping bias31), this would produce an artificial increase in quit rates over time. However, a similar bias would affect both surveys: contrary to the most recent data (2004–2005), quit rates derived from the oldest survey (1999–2000) did not show a sudden increase over time (see tables 2 and 3). It is therefore unlikely that this bias explained the increase in quit rates observed in 2000–2002.

Comparison with other studies

Whereas a large number of trend studies focused on inequalities in smoking prevalence,8 32 very few reported time-trends in smoking cessation rates by socioeconomic indicators. Among subjects aged 20–34 in the US, quit rates increased in all educational groups, but the tendency was more marked among the higher educated.17 In Spain, Schiaffino et al found diverging trends in quit rates between higher educated and lower educated subjects, with widening inequalities in smoking cessation in the early 1990s and afterwards.20

A higher socioeconomic position was strongly associated with a higher likelihood of smoking cessation in many studies. Psychosocial stress, lack of social support, greater nicotine dependence and a lower ability to benefit from smoking cessation services are among the factors that may contribute to the socioeconomic differential in smoking cessation.33 34 35 Smoking may have a functional value especially among people from lower socioeconomic groups, who may perceive smoking as a way of coping with everyday life and work difficulties. Finally, knowledge of hazards of smoking may have diffused later to lower educated people than to higher educated people.36

In Italy, (relative) inequalities in smoking cessation were generally smaller compared to other European countries.37 Our estimate of relative effect (30–40% higher quit rate among higher educated smokers) is consistent with that study. This finding may be because the determinants of health inequalities are multiple, and they act at different levels.38 “Downstream” determinants of successful smoking cessation are proximate factors, which include, among others, easy-to-reach smoking cessation services as well as workplace interventions, while “upstream” determinants include public policies that reduce material disadvantage and social exclusion.

A few studies used the quit ratio (that is, the proportion of smokers who have quit), as measure of smoking cessation.39 40 41 However, this measure cannot convey information about the timing of smoking cessation in relation to age or period. The different approach adopted in the present analysis allowed us to measure quit rates by age and period, and brought to light a temporary increase in inequalities in smoking cessation among women in the late 1980s, and larger cessation rates among both high and low educated smokers in 2000–2002 than in previous years.

A recent study showed that in Italy inequalities in the likelihood of quitting smoking were constant among males and increased among females across three successive birth cohorts.21 A “cohort” effect may explain the increase in inequalities in smoking cessation that we observed in Italy during the late 1980s among females. Since for the following periods the educational gap in smoking cessation was reduced, it is likely that educational differences in smoking quit rates were smaller among younger cohorts.

The observed pattern for 2000–2002 might possibly reflect the effect of anti-tobacco policies. When anti-smoking programmes are directed at the general population, inequalities in smoking may become even larger, because higher socioeconomic groups may be more responsive to these initiatives.42 However, a recent comparative study in 18 European countries showed a good positive correlation between the enforcement of anti-tobacco policies and quit ratios, among both higher educated and lower educated individuals37: the wider the spectrum of tobacco control policies, the higher the quit ratios, irrespective of education. Recent studies have shown that smoking cessation services that directly target the most disadvantaged subjects may determine a reduction in inequalities in smoking cessation.43

In Italy, a comprehensive anti-tobacco policy was not present until recently.44 Several policy measures were instead introduced during the two decades under study, such as a ban on smoking advertising (1983), introduction of health warnings on cigarette packages (1990) and a ban of smoking in public places (1995).45 46 An important step forward was the release of the National Health Plan 1998–2000, which set among the main objectives that of containing the diffusion of tobacco, and fixed as a specific objective the reduction of health disparities. It is possible that this latter initiative contributed to the increase in smoking cessation rates among both higher and lower educated subjects, although earlier policies may have played a part as well. Of note, smoking cessation services were poorly structured until the early 1990s, while in the following years, the involvement of governmental agencies and local health units progressively grew.47

Conclusion

Smoking quit rates among young adults have been constant during the past two decades and increased only recently, while educational inequalities in smoking cessation persisted on both a relative and an absolute scale. The increase in smoking cessation rates observed in 2000–2002 suggests that anti-tobacco initiatives may have reached more disadvantaged smokers, although smokers of higher socioeconomic groups seem to have benefited the most. Anti-tobacco initiatives should explicitly focus on the most disadvantaged because, in addition to equity concerns, this will determine the largest benefit for the whole population.

Future studies need to investigate whether the comprehensive policy against smoking adopted recently in Italy, and similarly in other countries,48 49 did bring the greatest benefits to smokers from disadvantaged groups. In order the evaluate these as well as other policies, we advocate a more frequent use of “dynamic” processes of smoking behaviour—that is, measures of smoking initiation and cessation, in addition to measures of smoking prevalence.50