Volume 107, Issue 9 p. 1677-1686
RESEARCH REPORT
Full Access

Impact of the 2005 smoke-free policy in Italy on prevalence, cessation and intensity of smoking in the overall population and by educational group

Bruno Federico,

Corresponding Author

Bruno Federico

Department of Health and Sport Sciences, University of Cassino, Cassino, Italy,

Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands

Bruno Federico, Department of Health and Sport Sciences, University of Cassino, Via S. Angelo snc (località Folcara), 03043 Cassino (FR), Italy. E-mail: b.federico@unicas.itSearch for more papers by this author
Johan P. Mackenbach,

Johan P. Mackenbach

Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands

Search for more papers by this author
Terje A. Eikemo,

Terje A. Eikemo

Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands

Search for more papers by this author
Anton E. Kunst,

Anton E. Kunst

Department of Public Health, Academic Medical Center, University of Amsterdam, Amsterdam, the Netherlands

Search for more papers by this author
First published: 23 February 2012
Citations: 29

ABSTRACT

Aims  To estimate the immediate as well as the longer-term impact of the 2005 smoke-free law on smoking prevalence, cessation and intensity both in the overall population and separately by educational level.

Design  Interrupted time–series analyses of 11 cross-sectional nationally representative surveys.

Setting  Italy, 1999–2010.

Participants  Adults aged 20–64 years.

Measurements  For each year we computed the prevalence of current smoking, the quit ratio and the mean number of cigarettes smoked per day. All measures were standardized by age. Segmented linear regression analyses were performed for each smoking variable separately by sex.

Findings  Among males, smoking prevalence decreased by 2.6% (P = 0.002) and smoking cessation increased by 3.3% (P = 0.006) shortly after the ban, but both measures tended to return to pre-ban values in the following years. This occurred among both highly and low-educated males. Among low-educated females, the ban was followed by a 1.6% decrease (P = 0.120) in smoking prevalence and a 4.5% increase in quit ratios (P < 0.001). However, these favourable trends reversed over the following years. Among highly educated females, trends in smoking prevalence and cessation were not altered by the ban. Among both males and females, long-term trends in the daily number of cigarettes, which were already declining well before the implementation of the policy, changed to a minor extent.

Conclusion  The impact of the Italian smoke-free policy on smoking and inequalities in smoking was short-term. Smoke-free policies may not achieve the secondary effect of reducing smoking prevalence in the long term, and they may have limited effects on inequalities in smoking.

INTRODUCTION

In Europe, smoking restrictions in public and work-places were first introduced in 2004 in Ireland 1, and progressively in many other countries 2. Smoke-free policies have been shown to reduce exposure to environmental tobacco smoke among hospitality workers and the general population, as well as to reduce the hospitalization rate for acute myocardial infarctions and asthma exacerbations 3. In addition, smoke-free policies may have important ‘side effects’ on smokers' behaviour, inducing them either to quit altogether or to reduce the number of cigarettes smoked per day 4.

A recent systematic review of 21 studies reported a median decrease in smoking prevalence of 3.4% after the introduction of smoke-free legislation 5. Conversely, a more refined time–series analysis of data from the United States, Canada and some European countries showed that in most countries pre-existing trends in smoking were unaffected 6. Increased rates of smoking cessation may follow the ban: in England attempts to quit smoking were greater during the 2 months after the introduction of the ban compared to an analogous 2-month period in the following year 7.

In Italy, a policy which prohibited smoking in all public and work-places came into effect on 10 January 2005. The policy received wide support both before and after its implementation 8-10. Environmental tobacco smoke in hospitality premises reduced up to 97%, suggesting that the ban was widely respected 11, although a recent study showed that the degree of observance of the legislation in these venues decreased during the following years 12.

In the overall population, smoking prevalence decreased from 26.2% in 2004 to 24.3% in 2006, and there was a parallel decrease in the mean number of cigarettes smoked per day 13. However, it is not known if the reduction in smoking prevalence observed immediately after the implementation of the ban continued over the following years. Recent studies showed contrasting results, but they lacked statistical power 9, 14.

It is also unknown if the effect of the Italian smoke-free policy differed among high and low socio-economic groups. Two recent reviews did not find any evidence of a differential effect by socio-economic characteristics for smoking restrictions 15, 16. It has been argued that unless the more disadvantaged subjects are targeted, population-level interventions may increase smoking inequalities 4, 17. However, only a handful of studies have assessed if smoke-free legislation had a different impact by socio-economic status 18, 19, and very few studies have assessed the differential impact of nation-wide smoke-free policies. In England, there was no difference by socio-economic status in the increase in quit attempts observed shortly after the smoke-free legislation 7.

In this study, we aimed to estimate the immediate as well as the longer-term impact of the 2005 smoke-free law in Italy on the smoking behaviour of adult subjects, and to assess if the impact differed by educational group. More specifically, we aimed to answer the following research questions:

  • 1

    Was the 2005 ban associated with an immediate change in the prevalence, cessation and intensity of smoking?

  • 2

    Was the 2005 ban associated with a change in long-term trends in the prevalence, cessation and intensity of smoking?

  • 3

    Did the changes in the prevalence, cessation and intensity of smoking associated with the ban (if any) differ according to level of education?

METHODS

We used data from 11 multi-purpose yearly surveys ‘Aspects of everyday life’, carried out by the National Institute of Statistics [Indagine sui Consumi delle Famiglie (Survey of Consumption by Families: ISTAT)] between 1999 and 2010. These surveys were conducted on large representative samples of the Italian non-institutionalized population, and independent samples were drawn each year. Two sampling schemes were adopted, depending on the size of municipality. For bigger municipalities cluster sampling was used, with households being the primary sampling units. A two-stage sampling was instead used for smaller municipalities. Details of the sampling design are described elsewhere 20. Each observation was given an individual weight that corresponded to the inverse of the probability of being sampled.

For each household sampled, data of all members of the household were collected with the use of an interviewer-administered form as well as a self-compiled questionnaire. The following questions about smoking behaviour were included in the self-compiled questionnaire: ‘Do you currently smoke?’, ‘What type of tobacco product do you use?’ and ‘How many cigarettes do you smoke daily?’

These surveys were carried out each year between 1999 and 2010, with the exception of year 2004. For this year, we obtained data derived from the Health Interview Survey 2004–05, relative to the December 2004 wave. This latter survey was also carried out by ISTAT, and analogous sampling and data collection methods were used. Data for 2004 are shown for descriptive purposes only, and were not used in the statistical models.

For each year between 1999 and 2010 we computed the prevalence of current smoking, the prevalence of former smoking among ever smokers (also known as the quit ratio) and the mean number of cigarettes smoked per day. All measures were age-standardized by means of the direct method of standardization, using as reference the Italian population in 2005. Weights provided by ISTAT were applied to adjust both prevalence rates and means. Education was categorized into two classes: high and low. Highly educated subjects were those who held at least a high school degree (level 3 of the International Standard Classification of Education), while the remaining subjects were classified as low-educated.

All analyses were stratified by sex and they were restricted to subjects aged 20–64 years at the time of each survey. We performed an additional analysis for those aged 20–24 years. We hypothesized that the smoking ban would have a stronger impact among young subjects, as they are more likely to attend hospitality venues 21, 22.

After data were collapsed in a data set with years as units of observation containing the age-standardized summary measures of smoking separately for sex and educational level, segmented linear regression analysis was performed 23. This statistical method allows breaking one independent variable (time in our case) into several subdomains in which the relationship between the dependent variable (smoking) and the independent variables (time and educational level) may change. In our case, the subdomains are the period before and the period after the implementation of the policy. The regression model allows estimating separate slopes as well as separate intercepts for the two periods. A recent paper by Bajoga et al. applied this method to evaluate smoke-free legislation in several countries, and provides a clear pictorial representation of the method 6.

The results of five nested models are shown. Model 1, which is the reference model, contains three parameters: one for time, one for education and one for the intercept. Time, in years, was entered into all models as a continuous variable, and was centred on January 2005, when the smoke-free policy was introduced, whereas education was entered into all models as a dichotomous variable, coded 0 for the low-educated and 1 for the highly educated. As a result, the coefficient for the intercept in model 1 describes the estimated value of smoking (either prevalence, quit ratio or mean number of cigarettes) among low-educated subjects just before the introduction of the policy. The coefficient for time describes instead the change in smoking over a 1-year period, which is assumed to be constant over the whole time-period in this model. The coefficient for education describes the difference in smoking between highly and low-educated subjects, which is also assumed to be constant over time in this model. In addition to the parameters included in model 1, model 2 contains a parameter for the change in intercept in 2005 which allows assessment of whether or not the policy was associated with an immediate impact on the smoking patterns in the overall population.

Model 3 adds a parameter for the change in slope to model 2, which allows assessment of whether or not the policy was associated with a change in trends in the smoking patterns of the overall population. In this model, as well as in models 4 and 5, the coefficient for time describes the estimated change in smoking among low-educated subjects over a 1-year period before the ban. The slope of the regression line after the ban can be computed by summing up algebraically the coefficients for time and for change in slope. In both models 2 and 3, the changes associated with the ban are assumed to be the same for highly and low-educated subjects.

Model 4 adds the interaction between education and time to model 3, which allows evaluation of whether or not trends in smoking over the whole time-period differ between highly and low-educated subjects. Finally, model 5 allows assessment of whether or not the changes associated with the policy differ by education, by adding two parameters to model 4. The first one is the interaction between education and the change in intercept, which allows evaluation of whether or not the immediate change associated with the policy differed between highly and low-educated subjects, and the second is the interaction between education and the change in slope, which allows evaluation of whether or not the change in long-term trends associated with the policy differed between highly and low-educated subjects.

Models 1–5, which are nested models, were compared using two measures: the adjusted R2 and the partial F-test. The adjusted R2 describes the proportion of the variability explained by each model, taking into account the number of covariates in the model, whereas the partial F-test evaluates if a simpler model (with k parameters) is improved significantly by a more complex model (with p + k parameters). The test statistic is computed as follows: inline image, where SSER is the error sum of squares of the reduced model and SSEC is the error sum of squares of the more complex model. The degrees of freedom of the numerator are p, whereas those of the denominator are n − (k + p + 1).

All analyses were carried out using the software package STATA version 11.

RESULTS

The characteristics of the surveys used in this study are shown in Table 1, together with descriptive statistics on smoking behaviour among subjects aged 20–64 years. The prevalence of current smoking in the overall population decreased over time, while the quit ratio increased. Changes in both prevalence and cessation of smoking were particularly marked immediately before or just after the introduction of the 2005 policy, whereas in the following years values tended to be similar to those of the period before the policy was introduced. A clear decline over the whole time-period is observed for the number of cigarettes smoked daily, from 15.0 in 1999 to 13.1 in 2010.

Table 1. Descriptive statistics of the surveys.
Survey Period of data collection n Highly educated (%) Current smokers (%) Mean no. cig/day Quit ratio (%)
M F M F M F M F
1999 November 1999 34 953 46.8 46.2 37.8 21.5 16.7 12.3 40.2 40.4
2000 November 2000 36 639 47.8 47.7 37.0 22.2 16.8 12.0 40.1 40.4
2001 December 2001–March 2002 32 949 49.0 49.1 36.4 21.0 16.8 12.1 40.8 41.8
2002 November 2002 34 330 49.2 49.2 35.7 21.1 16.3 12.1 42.0 42.4
2003 October 2003 33 389 50.1 50.5 36.0 22.2 16.0 11.7 41.9 42.1
2004a December 2004 19 488 48.9 50.1 33.0 20.8 16.8 12.8 43.6 44.1
2005 March 2005 30 321 51.0 52.6 33.2 20.2 15.1 11.1 45.7 45.3
2006 February 2006 29 696 51.6 53.5 34.2 21.3 15.4 11.3 44.5 45.6
2007 February–March 2007 29 131 52.6 54.4 33.7 21.2 15.2 11.3 44.8 44.7
2008 February–March 2008 29 360 52.8 55.0 34.2 20.7 15.4 11.3 44.3 45.6
2009 March 2009 28 979 54.8 56.8 34.7 21.6 14.9 11.3 43.5 45.0
2010 March 2010 29 342 56.6 58.9 34.4 21.2 14.5 11.0 44.1 46.9
  • a Health Interview Survey 2004–05.

Table 2 shows that there was an immediate decline associated with the ban in the prevalence of current smoking among males (β = −2.6% for the change in intercept in model 2, p = 0.002). However, trends in smoking prevalence reversed after 2005 (β = 0.7% per year for the change in slope in model 3, p < 0.001). No difference in this pattern is apparent by educational level (P-value for the multiple partial F-test that compares model 5 with model 4: 0.561). The absolute difference in smoking prevalence between highly and low-educated males widened slightly over the whole time-period (Fig. 1).

Table 2. Coefficients of the regression analyses of smoking prevalence (%).
Males Model 1 Model 2 Model 3 Model 4 Model 5
Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 41.8 <0.001 43.2 <0.001 41.8 <0.001 41.8 <0.001 41.4 <0.001
Time (years)a −0.2 0.003 0.1 0.284 −0.3 0.010 −0.3 0.022 −0.4 0.025
Educationb −11.1 <0.001 −11.1 <0.001 −11.1 <0.001 −11.1 <0.001 −10.4 <0.001
Change in intercept −2.6 0.002 −1.8 0.002 −1.8 <0.001 −1.6 0.020
Change in slope 0.7 <0.001 0.7 <0.001 0.8 <0.001
Education × time −0.2 0.010 0.0 0.860
Education × change intercept −0.5 0.565
Education × change slope −0.3 0.304
Adjusted R2 0.970 0.982 0.992 0.994 0.994
P-value* 0.002 <0.001 0.010 0.561
Females Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 23.0 <0.001 23.0 <0.001 23.0 <0.001 23.0 <0.001 23.3 <0.001
Time (years)a −0.1 0.145 0.1 0.660 −0.1 0.529 0.0 0.880 0.1 0.697
Educationb −1.0 0.027 −1.0 0.024 −1.0 0.024 −1.0 0.007 −1.6 0.178
Change in intercept −1.2 0.189 −0.8 0.360 −0.8 0.246 −1.6 0.120
Change in slope 0.3 0.273 0.3 0.167 0.4 0.224
Education × time −0.3 0.003 −0.5 0.189
Education × change intercept 1.6 0.273
Education × change slope −0.1 0.727
Adjusted R2 0.225 0.259 0.270 0.554 0.550
P-value* 0.189 0.273 0.003 0.415
  • aTime is the slope coefficient and it is centred on the date of the implementation of the ban (January 10, 2005). bEducation is coded 1 for highly educated and 0 for low-educated. *P-value of the partial (and multiple partial) F-test comparing model 2 with model 1, model 3 with model 2, model 4 with model 3 and model 5 with model 4.
Details are in the caption following the image

Age-standardized rates of current smoking by educational level and sex. Triangles represent low-educated subjects, circles represent highly educated subjects. Grey hollow markers are used for the Health Interview Survey held in December 2004

Among females, both the immediate change and the change in trends in smoking prevalence associated with the ban did not reach statistical significance, as shown by the large P-values of the partial F-tests for models 2 and 3. A non-significant decrease in smoking prevalence was associated with the ban among the low educated females (−1.6%, P = 0.120). However, long-term trends favoured the highly educated (β = −0.3% per year for the interaction term between education and time in model 4). Figure 1 shows that in the study period the relationship between education and smoking prevalence changed from a positive to a negative association.

Time trends in the quit ratio (Table 3) mirror those in smoking prevalence. Among males, the policy was associated with a sudden increase in quit ratio (β = 3.3% for the change in intercept in model 2, p = 0.006). Quit ratios decreased after 2005 (β = −1.1% per year for the change in slope in model 3, p < 0.001). No difference in this pattern is apparent by educational level (P-value for the multiple partial F-test that compares model 5 with model 4: 0.858). Between 1999 and 2010, the absolute difference in quit ratio between highly and low-educated males increased (Fig. 2).

Table 3. Coefficients of the regression analyses of quit ratios (%).
Males Model 1 Model 2 Model 3 Model 4 Model 5
Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 37.3 <0.001 35.5 <0.001 37.6 <0.001 37.6 <0.001 37.6 <0.001
Time (years)a 0.3 0.004 −0.1 0.426 0.6 0.009 0.4 0.014 0.4 0.075
Educationb 6.5 <0.001 6.5 <0.001 6.5 <0.001 6.5 <0.001 6.5 <0.001
Change in intercept 3.3 0.006 2.2 0.013 2.2 0.002 2.4 0.016
Change in slope −1.1 <0.001 −1.1 <0.001 −1.1 0.001
Education × time 0.3 0.001 0.3 0.321
Education × change intercept −0.4 0.745
Education × change slope 0.1 0.756
Adjusted R2 0.849 0.896 0.950 0.973 0.969
P-value* 0.006 <0.001 0.001 0.858
Females Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 39.4 <0.001 38.0 <0.001 39.2 <0.001 39.2 <0.001 38.2 <0.001
Time (years)a 0.4 <0.001 0.1 <0.663 0.4 0.115 0.3 0.269 0.1 0.758
Educationb 6.4 <0.001 6.4 <0.030 6.4 <0.001 6.3 <0.001 8.3 <0.001
Change in intercept 2.6 0.050 2.0 0.090 2.0 0.058 4.5 <0.001
Change in slope −0.6 0.104 −0.6 0.069 −0.8 0.016
Education × time 0.3 0.028 0.7 0.043
Education × change intercept −4.9 0.003
Education × change slope 0.5 0.283
Adjusted R2 0.865 0.891 0.902 0.924 0.962
P-value* 0.050 0.104 0.028 0.003
  • aTime is the slope coefficient and it is centred on the date of the implementation of the ban (January 10, 2005). bEducation is coded 1 for highlyeducated and 0 for low-educated. *P-value of the partial (and multiple partial) F-test comparing model 2 with model 1, model 3 with model 2, model 4 with model 3 and model 5 with model 4.
Details are in the caption following the image

Age-standardized quit ratios by educational level and sex. Triangles represent low-educated subjects, circles represent highly educated subjects. Grey hollow markers are used for the Health Interview Survey held in December 2004

A different pattern emerged for the female quit ratio: the policy was associated with an immediate increase in quit ratio (β = 2.6% for the change in intercept in model 2, p = 0.050), but the change in time trends (β = −0.6% per year for the change in slope in model 3) was not significant at the 0.05 level. However, the immediate effect of the policy was more favourable among low-educated females than among the higher educated, with a 4.5% increase in quit ratios among low-educated females, p < 0.001. Long-term trends clearly favoured the higher educated (β = 0.7% for the interaction term between education and time in model 5). As a result, educational differences in quit ratios widened over time (Fig. 2).

Table 4 shows that the 2005 policy was associated with a sudden but modest downward shift in the mean daily number of cigarettes among males (β = −0.6 for the change in intercept in model 2). No significant change in trends was observed. The immediate impact of the ban did not differ by education (P-value for the multiple partial F-test that compares model 5 with model 4: 0.274). Educational differences in the mean daily number of cigarettes among males widened over time (Fig. 3).

Table 4. Coefficients of the regression analyses of number of cigarettes.
Males Model 1 Model 2 Model 3 Model 4 Model 5
Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 16.7 <0.001 17.1 <0.001 16.9 <0.001 16.9 <0.001 17.0 <0.001
Time (years)a −0.2 <0.001 −0.2 0.002 −0.2 0.012 −0.2 0.028 −0.1 0.145
Educationb −1.9 <0.001 −1.9 <0.001 −1.9 <0.001 −1.9 <0.001 −2.0 <0.001
Change in intercept −0.6 0.042 −0.5 0.094 −0.5 0.058 −0.4 0.339
Change in slope 0.1 0.381 0.1 0.317 −0.0 0.899
Education × time −0.1 0.022 −0.1 0.270
Education × change intercept −0.4 0.501
Education × change slope 0.2 0.234
Adjusted R2 0.962 0.939 0.938 0.953 0.955
P-value* 0.042 0.381 0.022 0.274
Females Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 12.2 <0.001 12.4 <0.001 12.3 <0.001 12.3 <0.001 12.4 <0.001
Time (years)a −0.1 <0.001 −0.1 0.111 −0.1 0.157 −0.1 0.383 −0.0 0.630
Educationb −1.0 <0.001 −1.0 <0.001 −1.0 <0.001 −1.0 0.023 −1.1 0.015
Change in intercept −0.4 0.123 −0.4 0.209 −0.4 0.129 −0.6 0.122
Change in slope 0.1 0.539 0.1 0.453 0.1 0.452
Education × time −0.1 0.007 −0.1 0.305
Education × change intercept 0.4 0.449
Education × change slope −0.1 0.732
Adjusted R2 0.811 0.825 0.819 0.880 0.872
P-value* 0.123 0.539 0.007 0.618
  • aTime is the slope coefficient and it is centred on the date of the implementation of the ban (January 10, 2005). bEducation is coded 1 for highlyeducated and 0 for low-educated. *P-value of the partial (and multiple partial) F-test comparing model 2 with model 1, model 3 with model 2, model 4 with model 3 and model 5 with model 4.
Details are in the caption following the image

Age-standardized mean number of cigarettes smoked daily by educational level and sex. Triangles represent low-educated subjects, circles represent highly educated subjects. Grey hollow markers are used for the Health Interview Survey held in December 2004

Among females, neither the immediate change nor the change in trends associated with the ban reached statistical significance at the 0.05 level (Table 4). This was apparent among both highly and low-educated females (P-value for the multiple partial F-test that compares model 5 with model 4: 0.618). Educational differences widened over time, as shown in Fig. 3.

Among young subjects (Table 5 and Fig. 4), the ban was associated with a downwards trend in smoking prevalence only for low-educated males (P = 0.088), while there was hardly any change for highly educated males and for females.

Table 5. Coefficients of the regression analyses of smoking prevalence (%) among subjects aged 20–24 years.
Males Model 1 Model 2 Model 3 Model 4 Model 5
Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 50.0 <0.001 49.3 <0.001 50.1 <0.001 50.1 <0.001 50.0 <0.001
Time (years)a −0.2 0.030 −0.4 0.052 −0.1 0.657 −0.1 0.817 0.0 0.977
Educationb −18.8 <0.001 −18.8 <0.001 −18.8 <0.001 −18.8 <0.001 −18.8 <0.001
Change in intercept 1.4 0.293 1.0 0.485 1.0 0.492 2.7 0.121
Change in slope −0.4 0.345 −0.4 0.353 −1.0 0.055
Education × time −0.1 0.495 −0.3 0.581
Education × change intercept −3.4 0.155
Education × change slope 1.3 0.088
Adjusted R2 97.9 97.9 97.9 97.8 98.4
P-value* 0.293 0.345 0.495 0.005
Females Beta P-value Beta P-value Beta P-value Beta P-value Beta P-value
Intercept 27.9 <0.001 27.7 <0.001 28.2 <0.001 28.2 <0.001 28.6 <0.001
Time (years)a 0.1 0.214 0.1 0.729 0.2 0.518 0.5 0.159 0.6 0.229
Educationb −8.1 <0.001 −8.1 <0.001 −8.1 <0.001 −8.1 <0.001 −9.0 <0.001
Change in intercept 0.4 0.777 0.1 0.932 0.1 0.920 −0.8 0.691
Change in slope −0.3 0.579 −0.3 0.518 −0.2 0.705
Education × time −0.5 0.017 −0.7 0.314
Education × change intercept 1.8 0.514
Education × change slope −0.1 0.934
Adjusted R2 87.5 86.9 86.3 90.0 90.0
P-value* 0.777 0.579 0.017 0.587
  • aTime is the slope coefficient and it is centred on the date of the implementation of the ban (January 10, 2005). bEducation is coded 1 for highly educated and 0 for low-educated. *P-value of the partial (and multiple partial) F-test comparing model 2 with model 1, model 3 with model 2, model 4 with model 3 and model 5 with model 4.
Details are in the caption following the image

Age-standardized rates of current smoking by educational level and sex among subjects aged 20–24 years. Triangles represent low-educated subjects, circles represent highly educated subjects. Grey hollow markers are used for the Health Interview Survey held in December 2004

DISCUSSION

The 2005 smoke-free policy altered smoking patterns only in the short term, but did not affect pre-existing trends. Among males, there was no impact of the policy on inequalities in both smoking prevalence and smoking cessation, whereas among females inequalities in smoking cessation decreased shortly after the ban. A small and immediate effect of the policy was found on daily number of cigarettes. Long-term trends in the prevalence, cessation and intensity of smoking, which favoured the highly educated, were not affected by the policy.

Before discussing the implications of our findings, we first acknowledge the strengths as well as the limitations of this study. A major strength of this study is its high statistical power, because each of the 11 cross-sectional surveys that we used was based on a large and representative sample of the Italian population. Throughout the years, these surveys were carried out in different months, ranging from November to March. The seasonal variation in smoking behaviours 24 may have somewhat altered our findings. However, this influence is likely to be small, because of the limited variation in the timing of the surveys.

Data for 2004, which were used only to obtain descriptive statistics, were derived from a different survey. The large differences observed in the mean number of cigarettes smoked per day are due probably to incomparability among the surveys. Conversely, trends in smoking prevalence and cessation suggest that there may have been an anticipation of the effect of the smoking ban in the period just before its implementation.

Information on smoking behaviour was self-reported. However, self-reported information on smoking is generally considered to be acceptable 25, 26, although it may determine an underestimation of smoking prevalence 27. Recently, Gallus et al. highlighted a tendency to increased under-reporting of smoking in Italy 28. This may partly explain the decline in the prevalence of smoking observed over time, but cannot explain the sudden changes observed in 2005. In addition, there is no evidence that misclassification of smoking is correlated with socio-economic status 29. Thus, despite the reliance on self-reported data, educational differences in smoking trends are unlikely to be affected substantially.

Other factors may have influenced smoking trends in Italy over the past decade in addition to the smoke-free policy. The price of cigarettes is perhaps the most important one 30. In Italy the price of cigarettes rose by about 65% between 1999 and 2010, and the largest relative increase occurred between 2003 and 2005 31. In the regression analyses, price was strongly collinear with time, and was thus not used in the final models. Other findings also suggest that price did not confound the observed associations between the implementation of the smoke-free policy and smoking behaviour. The daily number of cigarettes smoked, which is perhaps the most sensitive indicator to rising price of tobacco, did not show any marked changes in the 2003–05 period. In addition, because subjects of lower socio-economic status may be more sensitive to price influences than subjects of higher status 32, if the influence of price had been large a decrease in inequalities in smoking would have occurred in the period 2003–05. Again, this was not the case.

There is no consistent evidence on the effectiveness of smoking bans on active smoking 33. In some countries there was either an immediate drop or a change in trends in the prevalence of smoking after the implementation of a nation-wide smoke-free ban, whereas in other countries long-term trends in smoking prevalence were unaltered by the ban 6, 34. In England, there was no change in smoking prevalence whereas levels of cigarette consumption decreased 35. In Spain, smoking prevalence had already decreased among subjects aged 25–44 before the implementation of the smoke-free policy, but this favourable trend reversed shortly afterwards 36. Our findings are consistent with these results, and they do not support the hypothesis that smoking restrictions affect smoking prevalence in the long term, while they may have a transient effect.

Logically, changes in smoking prevalence associated with the Italian smoke-free policy should reflect changes in smoking cessation. The implementation of a smoke-free policy may provide a strong incentive to quit smoking, because of increased perceived normative pressures not to smoke 37. Contrary to Hahn et al. 38, we showed that smoking quit ratios clearly increased shortly after the ban, whereas they decreased in the following years. This decrease may be attributed to relapse into smoking. In order to sustain smokers' attempts to quit and avoid relapse, smoking cessation services are essential 39. Unfortunately, in Italy their organization is still fragmented, with a general lack of adequate resources in terms of funding, infrastructures and trained personnel 40. Fewer than 2% of smokers who made a quit attempt reported to have received support from cessation services 41.

Another explanation for the decrease in quit ratios is the relaxation of the observance of the ban in working and public places. The population perception that the ban was respected decreased from 90.5% in 2005 to 83.9% in 2008 12. In addition, the number of hospitality venues which created partially covered outdoor dining and drinking areas, where smoking was allowed, increased after the ban. In these areas, second-hand smoke exposure was found to be similar to that recorded indoors before the implementation of the ban 42. The reduced compliance or adaptation to the ban may have contributed to reduced social pressure to quit smoking among current smokers as well as to smoking relapse among former smokers.

No intensive anti-tobacco publicity campaign was implemented around the time of the smoking ban. An action plan devoted to fight smoking as well as other behavioural risk factors was implemented in 2007, while a national mass-media anti-smoking campaign was carried out only in 2009 43.

Only few published studies have assessed whether or not the impact of smoke-free policies varied by socio-economic status, and they provide little evidence that smoke-free policies may reduce inequalities in smoking 15, 44. In Scotland, the impact of the smoke-free policy on smoking cessation did not differ by socio-economic status, although better-off subjects had more positive attitudes about the ban 45. A more recent study showed that socio-economic differences in quitting smoking reduced after the implementation of the smoke-free policy in Scotland, but not in England 46. The high compliance with the ban observed shortly after its implementation in Italy may have particularly benefited low socio-economic groups, as low-educated men and especially women more often work in hospitality premises 47.

In conclusion, the implementation of a smoke-free policy in Italy altered smoking patterns only in the short term, but did not affect pre-existing trends. Inequalities in smoking reduced among females, but only in the short term. While the Italian smoke-free policy was successful in improving the air quality of working and public places, its impact on smoking and inequalities in smoking was limited.

Declarations of interest

None.

Acknowledgements

We thank Dr Caspar Looman of the Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, the Netherlands, for his advice on regression modelling.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.