Scolaris Content Display Scolaris Content Display

Mass media interventions for smoking cessation in adults

This is not the most recent version

Collapse all Expand all

Abstract

available in

Background

Mass media tobacco control campaigns can reach large numbers of people. Much of the literature is focused on the effects of tobacco control advertising on young people, but there are also a number of evaluations of campaigns targeting adult smokers, which show mixed results. Campaigns may be local, regional or national, and may be combined with other components of a comprehensive tobacco control policy.

Objectives

To assess the effectiveness of mass media interventions in reducing smoking among adults.

Search methods

The Cochrane Tobacco Addiction Group search strategy was combined with additional searches for any studies that referred to tobacco/smoking cessation, mass media and adults. We also searched the Cochrane Register of Controlled Trials (CENTRAL) and a number of electronic databases. The last search was carried out in February 2013.

Selection criteria

Controlled trials allocating communities, regions or states to intervention or control conditions; interrupted time series.
Adults, 25 years or older, who regularly smoke cigarettes. Studies which cover all adults as defined in studies were included.
Mass media are defined here as channels of communication such as television, radio, newspapers, billboards, posters, leaflets or booklets intended to reach large numbers of people, and which are not dependent on person‐to‐person contact. The purpose of the mass media campaign must be primarily to encourage smokers to quit. They could be carried out alone or in conjunction with tobacco control programmes.
The primary outcome was change in smoking behaviour. This could be reported as changes in prevalence, changes in cigarette consumption, quit rates, odds of being a smoker.

Data collection and analysis

Two authors independently assessed all studies for inclusion criteria and for study quality (MB, LS, RTM). One author (MB) extracted data, and a second author (LS) checked them.
Results were not pooled due to heterogeneity of the included studies and are presented narratively and in table form.

Main results

Eleven campaigns met the inclusion criteria for this review. Studies differed in design, settings, duration, content and intensity of intervention, length of follow‐up, methods of evaluation and also in definitions and measures of smoking behaviour used. Among nine campaigns reporting smoking prevalence, significant decreases were observed in the California and Massachusetts statewide tobacco control campaigns compared with the rest of the USA. Some positive effects on prevalence in the whole population or in the subgroups were observed in three of the remaining seven studies. Three large‐scale campaigns of the seven presenting results for tobacco consumption found statistically significant decreases. Among the seven studies presenting abstinence or quit rates, four showed some positive effect, although in one of them the effect was measured for quitting and cutting down combined. Among the three that did not show significant decreases, one demonstrated a significant intervention effect on smokers and ex‐smokers combined.

Authors' conclusions

There is evidence that comprehensive tobacco control programmes which include mass media campaigns can be effective in changing smoking behaviour in adults, but the evidence comes from a heterogeneous group of studies of variable methodological quality. One state‐wide tobacco control programme (Massachusetts) showed positive results up to eight years after the campaign. Another (California) showed positive results during the period of adequate funding and implementation and in final evaluation since the beginning of the programme. Six of nine studies carried out in communities or regions showed some positive effects on smoking behaviour and at least one significant change in smoking prevalence (Sydney). The intensity and duration of mass media campaigns may influence effectiveness, but length of follow‐up and concurrent secular trends and events can make this difficult to quantify. No consistent relationship was observed between campaign effectiveness and age, education, ethnicity or gender.

Plain language summary

Can tobacco control programmes that include a mass media campaign help to reduce levels of smoking among adults

Mass media interventions involve communication through television, radio, newspapers, billboards, posters, leaflets or booklets, with the intention of encouraging smokers to stop, and of maintaining abstinence in non‐smokers. It is likely that they contribute to a reduction in smoking when used as part of a complex set of interventions, but it is difficult to establish their independent role and value in this process. Eleven studies are included in this review, but they are of variable scale and quality. Five large studies out of the nine which reported smoking prevalence found some positive changes in smoking behaviour. Three large studies out of seven that measured the quantity of tobacco smoked found reductions. Four of the seven studies which measured quit rates reported significant increases in abstinence, but this finding was difficult to interpret because studies used different definitions of smoking, smokers and quit attempts. The intensity and duration of mass media campaigns may influence effectiveness, but length of follow‐up and concurrent events in the community can make this difficult to verify. We found no consistent patterns between the effects of the campaigns and age, education, ethnicity or gender of those taking part.

Authors' conclusions

Implications for practice

  • Tobacco control programmes that include mass media campaigns may change smoking behaviour in adults, but the evidence comes from studies of variable quality and scale. The specific contribution of the mass media component is unclear.

  • The duration and intensity of an intervention may affect its impact on smoking behaviour, but evaluations need to last long enough to detect lasting changes, and to allow for confounders and for secular trends.

  • No consistent relationship was observed between campaign effectiveness and age, education, ethnicity or gender.

Implications for research

  • Evaluations of mass media campaigns should include adequate planning.

  • Evaluations should include control groups matched to intervention groups, or at least with between‐group baseline differences noted and adjusted for in the analysis.

  • Formative research is also essential to test the value of targeting intended populations.

Background

It is estimated that nearly 20% of the world's population smokes cigarettes, accounting for about 800 million men and 200 million women (Eriksen 2012). World wide, tobacco is the leading cause of preventable death, killing nearly 6 million people each year, with 80% of deaths occurring in low‐ and middle‐income countries (WHO 2011). It is estimated that if no additional measures are introduced the smoking prevalence in 191 countries included in the Global InfoBase Database of the World Health Organization (WHO) will be 22.7% in 2020 and 22% in 2030, accounting for 872 million smokers (Mendez 2012). In 2003 the World Health Assembly adopted the Framework Convention on Tobacco Control (FCTC), which now includes 174 countries, covering 85% of the worlds population (WHO FCTC 2003). To assist those countries in implementation of effective interventions to reduce the demand for tobacco in 2008, the WHO disseminated recommendations called MPOWER, which included the following strategies: monitor tobacco use, protect people from tobacco smoke, offer help to quit tobacco use, warn about the dangers of tobacco, enforce bans on tobacco advertising promotion and sponsorship, and raise taxes on tobacco (WHO 2008). Article 12 of the FCTC requires signatories to promote and strengthen public awareness of tobacco control issues using available communication tools, including mass media campaigns (WHO 2011).

Mass media interventions consist of the dissemination through television, radio, print media and billboards, of cessation‐related messages, informing smokers and motivating them to quit. Mass media campaigns can be effective in keeping tobacco control on the social and political agenda, in legitimising community action and in triggering other interventions. Campaigns are designed either directly to change individuals' smoking behaviour (the risk factor model), or to catalyse other forces of social change (the social diffusion model), which may then lead to a change in social norms about smoking (Wellings 2000). Social diffusion campaigns, such as those run in Australia, Canada, Thailand, the United Kingdom and some US states, are designed to de‐normalise smoking, so counteracting the tobacco industry's message that smoking is desirable and harmless (WHO 2001).

Research into the effectiveness of mass media campaigns is generally conducted through community trials. This term includes both randomised and non‐randomised studies which involve whole communities as the unit of assignment, with data collection from individuals within the communities. Murray 1998 has identified four main features which differentiate the designs of such studies, and which are generally determined by the nature of the research question:
(i) Main effects (to assess the impact of a single intervention) versus factorial (to assess the impact and sometimes the interaction of two or more variables).
(ii) Data collection schedules: these can range from a single post‐intervention measurement of the groups, through two or more series of assessments (including pre‐intervention baseline measures), to continuous surveillance.
(iii) Cross‐sectional versus cohort: Cross‐sectional designs are appropriate when the investigators are concerned with the impact of the intervention on the population as a whole, while cohort designs are more suited to measuring behaviour change in individuals over time.
(iv) A priori matching and stratification may help to limit bias and improve precision.
Analysis 1.1, 'Response and Retention Rates', summarises the design characteristics of the studies included in this review.

Mass media tobacco control campaigns in the USA began in 1967, following the publication of the 1964 Surgeon General's report on smoking and health (Surgeon General 1964). The Federal Communications Commission enforced the Fairness Doctrine, obliging radio and television stations to broadcast one tobacco control message for every three cigarette commercials (equivalent to a media value today of USD 300 million (WHO 2001)). This policy lasted until 1970, when a ban on broadcast cigarette advertising came into effect. Cigarette consumption had declined by 37% during the campaign, but began to rise again after the advertising ban ended free access to broadcast time for tobacco control messages (USDHHS 1991; Warner 1977).

Mass media campaigns in the 1970s tended to be based on the premise that information and heightened public awareness would of themselves effect changes in cultural norms and smoking behaviour. This assumption was challenged by the social learning theory approach, which held that public attitudes to smoking were more successfully translated into behaviour change if the mass media campaign was combined with well‐targeted interpersonal interventions conducted by healthcare workers or other credible agencies. This approach also acknowledged the importance of role models and peer‐group pressure and support in changing behaviour (Bandura 1977; Flay 1987b; NCI 1991). Later campaigns, the 'second generation' model, concentrated more on developing personal skills to cope with social and media pressure, and to recognise and resist tobacco industry advertising; they were also more likely to help smokers and non‐smokers to improve their decision‐making and problem‐solving abilities (Logan 1999). Such an approach, however, did not address the continuing dissonance between 'expert' opinion (scientists, healthcare providers, policy‐makers, many non‐smokers) and those smokers who resented the paternalism of campaigners, and who may have had complicated cultural and emotional attachments to their smoking, independently of their physical addiction (Yankelovich 1991; Logan 1999; Hastings 2002).

A number of studies suggest that media‐supported cessation campaigns can be an effective part of comprehensive and synergistic tobacco control programmes, reaching individuals directly with cessation messages and influencing their knowledge, attitudes and behaviour. Campaigns have been run and evaluated in a number of countries, including Australia, Canada, France, Iceland, New Zealand, the Philippines, Poland, Singapore and the UK, and in Arizona, California, Kansas, Massachusetts, Florida, Minnesota and Oregon within the USA (Flay 1987a; USDHHS 1991; WHO 2001). According to a 2011 WHO report, 23 countries, covering 28% of the world's population, run national anti‐tobacco media campaigns fulfilling all the WHO requirements (WHO 2011), while a further 30 countries conduct campaigns fulfilling most but not all of those requirements.

Previous reviews of the literature lend some support to tobacco control media campaigns as a component of comprehensive tobacco control programmes (World Bank 1999; Fiore 2000). Much of the literature is focused on the effects of tobacco control advertising on young people (Reid 1995; Tyas 1997; Pechmann 2000; Wakefield 2000; Wakefield 2003), but there are also a number of evaluations of campaigns targeting adult smokers, which show mixed results. Some national and state‐wide interventions have been shown to be effective in reducing smoking rates, while the outcomes are less consistent for community and local campaigns (Flay 1987a; Flay 1987b; Levy 2000; NCI 2000; Hopkins 2001; WHO 2001; Friend 2002; Siegel 2002a; Siegel 2002b).

Objectives

To assess the effectiveness of mass media interventions in reducing smoking among adults.

We addressed the following questions:

1. Do mass media campaigns reduce smoking (measured by prevalence, cigarette consumption, quit attempts and quit rates) compared with no intervention in comparison communities?

2. Do mass media campaigns run in conjunction with tobacco control programmes reduce smoking, compared with no intervention or with tobacco control programmes alone?

3. Which characteristics of these studies are related to their efficacy?

4. Do mass media tobacco control campaigns cause any adverse effects?

Methods

Criteria for considering studies for this review

Types of studies

  • Randomised or quasi‐randomised controlled trials allocating communities, regions or states to intervention or control conditions.

  • Controlled trials without randomisation, allocating communities, regions or states to intervention or control conditions.

  • Interrupted time series.

Uncontrolled before‐and‐after studies, and uncontrolled studies with post‐intervention measurements but no baseline measurement were excluded.

Types of participants

Adults, 25 years or older who regularly smoke cigarettes. Studies which cover all adults as defined in studies were included. Studies addressing only adolescents and 18‐ to 25‐year‐olds are covered in a separate Cochrane review (Brinn 2010).
Interventions for pregnant women were excluded, since this topic is covered by a separate Cochrane review (Lumley 2009).

Types of interventions

Mass media are defined here as channels of communication such as television, radio, newspapers, billboards, posters, leaflets or booklets intended to reach large numbers of people, and which are not dependent on person‐to‐person contact. The purpose of the mass media campaign must be primarily to encourage smokers to quit. They could be carried out alone or in conjunction with tobacco control programmes. Studies of comprehensive programmes were included, provided that the comparison was structured in such a way that the contribution and efficacy of the mass media component could be assessed.

Interventions comprising competitions and incentives or quit and win contests are covered by other Cochrane reviews (Cahill 2008; Cahill 2011). Internet‐based interventions and mobile‐phone interventions are covered by other Cochrane reviews (Civljak 2010; Whittaker 2012).

Types of outcome measures

Measures of smoking behaviour:

  • Primary: Tobacco cessation, covered by prevalence rates, quit rates

  • Secondary: Tobacco reduction, covered by changes in the number of cigarettes purchased or smoked, prevalence of daily smoking, quit attempts

We prefer outcomes measured at the longest follow‐up, and at least six months from the beginning of the intervention. It is generally not feasible for community trials to conduct biochemical validation of their smoking cessation results, and we do not require this of the included studies in this review.

Intermediate measures:

  • Attitudes to smoking

  • Knowledge about smoking, including smoking norms, and effects of tobacco on health

  • Adverse side effects

Process measures:

  • Descriptions of formative research, pilot studies and ongoing evaluation and modification of the intervention

  • Media weight (reach, frequency and duration), campaign awareness/exposure

  • Dose‐response relationships (e.g. volume of calls to telephone helplines)

  • Maintenance of programmes after the interventions were completed

  • Intervention costs

Mass media campaigns that have only been reported in terms of intermediate outcomes or process measures are excluded.

Search methods for identification of studies

The Cochrane Tobacco Addiction Group search strategy was combined with ad hoc searches for any studies that referred to tobacco/smoking cessation, mass media and adults. We also searched the Cochrane Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, CINAHL, PsycINFO, Dirline, Hstat, Healthstar, Science Direct, EIFL Direct, IBZ, IDEAL, Addiction Abstracts, ASSIA, ISI, ERIC, IBSS, Sociological Abstracts, Conference Paper Index, ProQuest, Springer Link, Swetsnet, and the ASH (Action on Smoking and Health) Database. In addition, we searched the reference lists of identified studies and checked the web sites of included mass media campaigns for additional data. The most recent search was carried out in February 2013.

Data collection and analysis

One reviewer prescreened all search results (abstracts), for possible inclusion or as useful background. Three reviewers (MB, LS and RTM) independently assessed relevant studies for inclusion. Discrepancies were resolved by consensus. The editorial base would have resolved any persistent disagreements. Reasons for the non‐inclusion of studies were noted.
One reviewer (MB) extracted data, and a second reviewer (LS) checked them. This stage included an evaluation of quality. Two reviewers independently assessed each study according to the presence and quality of the randomisation process, whether the analysis was appropriate to the study design, and the description of withdrawals and drop‐outs. Interrupted time series studies were assessed according to criteria defined by the Cochrane Effective Practice and Organisation of Care Group (EPOC 2006). These require a minimum of three assessments before the intervention and three afterwards for the study to be included in the review.

We extracted data on:

  • Country and community status (e.g. state, city)

  • Participants (baseline demographic, clinical and smoking characteristics)

  • Intervention (duration, intensity, message development)

  • Outcomes, and how they were measured

  • Length of follow‐up

  • Completeness of follow‐up

  • Definition of smoking cessation

  • Biochemical confirmation of abstinence, if present.

In cases of missing data, we contacted the authors of the study where possible.
Results were not pooled due to expected heterogeneity of included studies, and are presented narratively and in the Results Tables. Eligible studies differed significantly in design, setting, duration, content and intensity of intervention, length of follow‐up, methods of evaluation and also in definitions and measures of smoking behaviour used.

We include in this review the Tobacco Addiction Group glossary of tobacco‐related terms in Appendix 1.

Results

Description of studies

Results of the search

From our update searches, the latest of which was performed in February 2013, we identified 1443 studies. Based on titles and abstracts, 1307 references were excluded, resulting in 136 full papers to be retrieved. In addition, we found 10 references through handsearching and from the websites of included mass media campaigns. We therefore retrieved 146 full‐text papers. Based on these, we excluded 142 papers.

For this update we identified three reports updating the California TCP campaign (California TCP 2003), and ten new excluded studies (Figure 1).


Prisma flow diagram of search results

Prisma flow diagram of search results

Included studies

We identified 11 campaigns meeting our inclusion criteria. Detailed information about each is shown in the table Characteristics of included studies. Sixty‐five excluded studies and reasons for exclusion are described in the Characteristics of excluded studies table.

Eight studies had strong designs comparing the effect of mass media campaigns in exposed areas to control areas, using indices of smoking behaviour (mainly prevalence) in the whole population (Stanford 3 City 1977; North Coast QFL 1983; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997; McVey 2000; McAlister 2004). One study (Mogielnicki 1986) had a similar design, but assessed the effect of the intervention by examining the abstinence rates achieved in one clinic each in the intervention and control areas. Two studies (California TCP 2003; Massachusetts 2003) had strong controlled designs, but the intervention areas were exposed to mass media campaigns plus a range of other tobacco control interventions as compared to no special interventions in control areas. Study sites ranged from the USA (Stanford 3 City 1977; Mogielnicki 1986; McPhee 1995; Jenkins 1997; California TCP 2003; Massachusetts 2003; McAlister 2004) to the UK (McVey 2000), South Africa (CORIS 1997) and Australia (North Coast QFL 1983; Sydney QFL 1986).

Characteristics of study populations:
Study populations were diverse. All studies involved adults, although the age of adulthood varied between studies. Eight studies targeted both men and women, and three targeted men only (Mogielnicki 1986; McPhee 1995; Jenkins 1997). Two studies targeted male Vietnamese immigrants in the USA (McPhee 1995; Jenkins 1997). Two state‐wide campaigns targeted adults, adolescents and the general population (California TCP 2003; Massachusetts 2003).

Characteristics of interventions:
In six studies the theoretical basis for the development of the intervention was described. In Stanford 3 City 1977 social marketing theory, social learning theory and communication theory underpinned the intervention, while McAlister 2004 specified social learning theory, the transtheoretical model and elements of modelling, social reinforcement for behaviour change and emotional arousal. California TCP 2003 and Massachusetts 2003 both used social diffusion theory with social marketing and social policy change. North Coast QFL 1983 cited social marketing and communication theory, and Mogielnicki 1986 specified current marketing methodology in developing its mass media campaign. In the remaining five studies no theoretical basis was specified.

Two campaigns conducted as part of tobacco control programmes involved TV, radio, print media and billboard advertising (California TCP 2003; Massachusetts 2003). One study used only TV advertising (McVey 2000); one used TV and radio advertising (Mogielnicki 1986); one study used only billboards, posters, mailings and local newspapers (CORIS 1997). The remaining studies used TV, radio and print media, billboards and/or posters (Stanford 3 City 1977; North Coast QFL 1983; Sydney QFL 1986; McPhee 1995; Jenkins 1997; McAlister 2004). Two studies aimed at the reduction of cardiovascular risk factors, such as smoking, diet, blood pressure, cholesterol, sedentary lifestyle, and stress (Stanford 3 City 1977; CORIS 1997), while the remaining nine studies aimed specifically at changing smoking behaviour (reducing smoking prevalence, reducing the number of cigarettes smoked or increasing quit rates).

The two state‐wide campaigns (California TCP 2003; Massachusetts 2003) were each part of a wider tobacco control programme which implemented a number of public policy measures to reduce smoking, but we have not included those outcomes in our review.

Assessments:
Cross‐sectional independent surveys were used in five studies (North Coast QFL 1983; McPhee 1995; Jenkins 1997; California TCP 2003; Massachusetts 2003), cohort follow‐up in two studies (Mogielnicki 1986; McVey 2000) and both methods in four studies (Stanford 3 City 1977; Sydney QFL 1986; CORIS 1997; McAlister 2004). Three studies reported follow‐up beyond the duration of the intervention and the immediate post‐intervention assessment, at 12 years for CORIS 1997, at 18 months for McVey 2000, and at one year for Sydney QFL 1986. Stanford 3 City 1977 reported one additional year of follow‐up for the high risk group only. Interviews were conducted in person in six studies (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; Sydney QFL 1986; CORIS 1997; McVey 2000), by telephone in four studies (McPhee 1995; Jenkins 1997; Massachusetts 2003; McAlister 2004), and by both methods in California TCP 2003. In addition to the interviews, physical examination and blood tests were carried out in three studies (Stanford 3 City 1977; North Coast QFL 1983; CORIS 1997). In most studies smoking abstinence was self‐reported, and was biochemically confirmed in all participants in only two studies: Stanford 3 City 1977 checked plasma thiocyanate, and Mogielnicki 1986 checked plasma thiocyanate and exhaled carbon monoxide. Subsamples of participants were tested for salivary cotinine in Sydney QFL 1986, and for plasma thiocyanate in North Coast QFL 1983.

Outcomes measures:
Difference in smoking prevalence was the main outcome measure in seven campaigns (North Coast QFL 1983; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997; California TCP 2003; Massachusetts 2003). One study reported it for the high risk group only (Stanford 3 City 1977). Changes in the number of cigarettes (or grammes of tobacco) smoked were reported in five studies (Stanford 3 City 1977; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997) and was the main outcome for Stanford 3 City 1977. In seven studies quit rates or abstinence rates were reported (Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997; ; McVey 2000; California TCP 2003), and were the main outcome in two studies (Mogielnicki 1986; McVey 2000). In McAlister 2004, point prevalence of daily smoking (ceasing to smoke at all or every day) was reported as the main outcome. In the two state campaigns per capita cigarette consumption based on aggregate sales data was also presented (California TCP 2003; Massachusetts 2003).

The number of quit attempts was reported in six studies (Sydney QFL 1986; McPhee 1995; Jenkins 1997; California TCP 2003; Massachusetts 2003; McAlister 2004).

Intermediate measures:
Attitudes to smoking were assessed at baseline in nine studies (North Coast QFL 1983; Mogielnicki 1986; McPhee 1995; CORIS 1997; Jenkins 1997; McVey 2000; California TCP 2003; Massachusetts 2003; McAlister 2004) with follow‐up data provided in five of them (North Coast QFL 1983; McPhee 1995; Jenkins 1997; California TCP 2003; Massachusetts 2003) but in one study only in graphical form (North Coast QFL 1983). Social pressure/influences in the decision to quit were assessed in four studies at baseline and at follow‐up (North Coast QFL 1983; Sydney QFL 1986; McPhee 1995; Massachusetts 2003). Norms concerning smoking behaviour were assessed in three studies at baseline and at follow‐up (McPhee 1995; California TCP 2003; Massachusetts 2003): In McPhee 1995 this was assessed by the number of friends or household members smoking and giving advice or being advised to stop smoking. In the two state‐wide campaigns norms were assessed by measuring support for tobacco control legislative measures and exposure to environmental tobacco smoke at work and at home, but only within the intervention communities and not for the controls.
Information‐seeking behaviour in the population was assessed at baseline and at follow‐up in one study (Sydney QFL 1986).
Knowledge and/or beliefs about cardiovascular risk factors or the effects of smoking were assessed at baseline in eight studies (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; Sydney QFL 1986; CORIS 1997; California TCP 2003; Massachusetts 2003; McAlister 2004). Follow‐up data were provided in six studies (Stanford 3 City 1977; North Coast QFL 1983; Sydney QFL 1986; CORIS 1997; California TCP 2003; Massachusetts 2003), but in one study only in graphical form without elucidation in the text (North Coast QFL 1983).

No study reported adverse effects of the campaigns.

Process measures:
Formative research or pilot studies were used in nine campaigns (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; Sydney QFL 1986; Jenkins 1997; McVey 2000; California TCP 2003; Massachusetts 2003; McAlister 2004). No information was provided for two campaigns (McPhee 1995; CORIS 1997). Ongoing evaluation and modification of the intervention was reported in five studies (Stanford 3 City 1977; North Coast QFL 1983; McVey 2000; California TCP 2003; Massachusetts 2003).

Detailed information regarding media weight (numbers of TV and radio spots, newspaper articles etc.) was provided in five studies (Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; Jenkins 1997; California TCP 2003), and summary information was given in four studies (Stanford 3 City 1977; CORIS 1997; McVey 2000; Massachusetts 2003). Little or no information was given in McAlister 2004 or North Coast QFL 1983.
Awareness and reach of the intervention was measured in seven studies (Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; Jenkins 1997; California TCP 2003; Massachusetts 2003; McAlister 2004) with only McAlister 2004 not reporting numerical data for this outcome.

Dose response: Some evidence on possible dose‐response relationships was mentioned, either as numbers of calls to quitlines (Sydney QFL 1986; McVey 2000), as increasing knowledge with increasing intensity of the intervention (Stanford 3 City 1977), or as increasing cessation rates, number of quit attempts, or changes in social norms with increasing numbers of channels or intensity of intervention (California TCP 2003; McAlister 2004). In McVey 2000 abstinence rates were compared in two study areas which received single‐ or double‐weight interventions. No study presented a formal dose‐response analysis.

Maintenance: The campaigns were maintained beyond the intervention period in six studies (Stanford 3 City 1977; Sydney QFL 1986; CORIS 1997; McVey 2000; California TCP 2003; Massachusetts 2003). In one study (Stanford 3 City 1977) the campaign was continued beyond the final assessment point for one year, but with reduced intensity and no additional evaluation. The CORIS 1997 campaign was subsequently maintained by the community, but no details are given. The McVey 2000 campaign was continued nationally after a controlled evaluation. In Sydney QFL 1986 the mass media campaign was run in later years for a few weeks, with assessment of long‐term success. The two state‐wide campaigns (California TCP 2003; Massachusetts 2003) were established as constitutional amendments, so they remained in place.

Intervention costs: costs per capita were reported in four studies (CORIS 1997; California TCP 2003; Massachusetts 2003; McAlister 2004), and total campaign costs in four studies (Sydney QFL 1986; Mogielnicki 1986; California TCP 2003; Massachusetts 2003). Some cost effectiveness analysis was performed for the two state‐wide campaigns (California TCP 2003; Massachusetts 2003). No economic evaluation was reported for five campaigns (Stanford 3 City 1977; North Coast QFL 1983; McPhee 1995; Jenkins 1997; McVey 2000).

Risk of bias in included studies

Overall, we assessed the risk of bias of the included studies as high. For an overview of the methodological quality of the included trials see Figure 2.


Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Allocation

Mogielnicki 1986 randomised clinic attenders to different interventions within each clinic assigned to intervention or control conditions. It is normally not feasible or affordable to use true randomisation in community studies of this kind. Nine were of quasi‐experimental design, with intervention and control status non‐randomly allocated (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997; McVey 2000; McAlister 2004). The two state‐wide campaigns were assessed through an interrupted time series design comparing the effects of the campaigns as part of tobacco control programmes with other American states where there were no campaigns.

Selection of participants for outcome measures:
In most studies samples of participants were selected through random digit dialling, or through selecting from enumeration lists or telephone books. In one study the total population was surveyed (CORIS 1997).

Response rates and retention rates:
Response rates (applicable to nested cross‐sectional analyses) were reported in nine campaigns. Three of them reported the combined rates for the intervention and control communities (Sydney QFL 1986; California TCP 2003; McAlister 2004) and five reported them separately (Stanford 3 City 1977; North Coast QFL 1983; McPhee 1995; CORIS 1997; Jenkins 1997). Massachusetts 2003 reported combined and separate response rates. In those studies reporting separately for intervention and control communities, the response rates at baseline ranged from 70% to 84% in the intervention communities, and from 64% to 82% among the controls. The follow‐up response rates for those studies ranged from 68% to 94% for intervention communities, and from 62% to 88% for the controls. The response rates combined for intervention and control community varied between 42.7% and 99.4%.

Retention rates (applicable to cohort analyses) were reported in six studies (Stanford 3 City 1977; Mogielnicki 1986; Sydney QFL 1986; CORIS 1997; McVey 2000; McAlister 2004), and ranged from 53.5% to 76% in the intervention communities and from 52% to 73% in the controls. Two studies (McVey 2000; McAlister 2004) provided some information on participants lost to follow‐up, and Sydney QFL 1986 surveyed non‐responders in Sydney and Melbourne at baseline, for evidence of selection bias. Among the cohort studies, CORIS 1997 reported on demographic differences between cohort and non‐cohort participants (i.e. those who had only responded to one of the surveys), and Stanford 3 City 1977 identified higher smoking prevalence and heavier daily smoking among non‐cohort participants than in the cohorts. Study design, with detailed response and retention rates, are shown in Analysis 1.1.

Comparability of intervention and control community at baseline:
Nine studies described the demographic characteristics of participants at baseline (Stanford 3 City 1977; Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; CORIS 1997; Jenkins 1997; McVey 2000; Massachusetts 2003; McAlister 2004), with six of them conducting statistical tests for comparability (Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; Jenkins 1997; Massachusetts 2003; McAlister 2004). In North Coast QFL 1983 the statistical comparisons are reported without numerical detail. California TCP 2003 did not provide any information on comparability of the population. The intervention and control communities were shown to be demographically disparate in three studies, with analyses controlling for those differences (McPhee 1995; Jenkins 1997; Massachusetts 2003). Sydney QFL 1986 tested for possible confounding by sex, age, education, marital status, and socio‐economic status, but found none of them to be predictive of quitting.

Blinding

Only one study reported blinding participants and personnel (McVey 2000), in other studies it was unclear or they were not blinded. Three studies reported on blinding outcome assessors (McAlister 2004; McVey 2000; Mogielnicki 1986), in other studies it was unclear or they were not blinded.

Incomplete outcome data

None of the included studies reported attrition rates or losses to follow‐up.

Selective reporting

Only one trial did not report the outcomes as prespecified in the Methods section (McAlister 2004), with all the other studies reporting outcomes as prespecified.

Other potential sources of bias

Other sources of bias were not identified in most studies. The California TCP 2003 and the Massachusetts 2003 campaigns were interrupted time series studies. Both had clearly defined time points when the intervention was conducted, with at least three data points before and three after the intervention. It was unclear whether other quality assessment criteria for interrupted time series had been fulfilled.

Evaluation process:
In six campaigns the evaluation was done by study investigators (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; McPhee 1995; CORIS 1997; Jenkins 1997). In the remaining five included studies independent organisations conducted the evaluation or surveys or some parts of the work (Sydney QFL 1986; McVey 2000; California TCP 2003; Massachusetts 2003; McAlister 2004). For the Californian and Massachusetts state‐wide campaigns national surveys were conducted.

Statistical analysis:
Two of the 11 included studies (McPhee 1995; McVey 2000) reported sample size calculations. However, because of the variability of the effect sizes, we do not consider that the absence of a power calculation should be interpreted as a marker of lower quality. All the studies except CORIS 1997, which reported use of t‐tests with and without covariance adjustments, used regression analyses to produce their results. Two studies reported one‐sided P value tests (Stanford 3 City 1977; McAlister 2004) and CORIS 1997 used two‐sided P value tests.

The Californian and Massachusetts campaigns were described using an interrupted time series design. We assessed the reports according to the Cochrane Effective Practice and Organisation of Care Group criteria (EPOC 2006). All the studies identified a clearly defined point in time when the intervention occurred, and at least three assessment points before and after the intervention. However, the impact of the intervention independent of other changes was not clearly established. The TCP studies used regression models for data analysis. The intervention was deemed unlikely to affect data collection, and sources and methods of data collection were consistent before and after the intervention. Smoking prevalence was self‐reported and unvalidated. In two of the studies (one for each campaign) aggregated cigarette sales data were used as an objective measure. None of the studies reported on completeness of data sets, but did report response rates for outcome surveys.

Effects of interventions

Detailed results are presented in 'Intermediate measures' (Analysis 1.2) and 'Primary measures of smoking behaviour' (Analysis 1.3). Summary findings on theoretical orientation, costs and outcomes are reported in 'Study summary by type of outcome' (Analysis 1.4). Baseline differences and possible confounders are reported in Analysis 1.5.

Smoking prevalence:
Among the nine campaigns reporting smoking prevalence, two studies (North Coast QFL 1983; CORIS 1997) reported smoking prevalence separately for men and women, and by age group in North Coast QFL 1983. Two studies reported prevalence for the whole population and for men and women separately (Sydney QFL 1986; Massachusetts 2003). Two studies targeted and reported on men only (McPhee 1995; Jenkins 1997), and the Stanford 3 City 1977 study reported in detail only on the high risk cohort. In two studies smoking prevalence was not reported (Mogielnicki 1986; McVey 2000).

Decreases in smoking prevalence were observed in both the state‐wide programmes (California TCP 2003; Massachusetts 2003) compared with the rest of the USA, but in California the decrease was statistically significant only during the early period of the campaign, before cuts in funding. The final assessments of the California TCP showed greater decline of smoking prevalence since the beginning of the programme compared with the rest of the USA and lower smoking prevalence, but the statistical significance of the differences was not reported. In the Massachusetts campaign the decrease was statistically significant for the population as a whole and for men, but not for women. It should be noted that the mass media campaigns in both states were part of a comprehensive programme of tobacco control measures.

Among those studies which analysed men and women separately, both the Australian studies (North Coast QFL 1983; Sydney QFL 1986) detected significantly decreased prevalence among men and women, while two studies found significant reductions in men's but not women's smoking at long‐term follow‐up (Sydney QFL 1986; CORIS 1997 at 12 years). For the two studies which examined Vietnamese men's smoking, Jenkins 1997 detected a significant decrease in prevalence at two year follow‐up, while McPhee 1995 failed to detect a significant reduction. Stanford 3 City 1977 failed to detect a significant effect of the media‐only intervention on prevalence compared with controls at three years, although the declining trend favoured the control community.

Cigarette consumption:
Among the seven campaigns reporting cigarette consumption, one study (CORIS 1997) presented cigarette consumption separately for men and women, two studies (McPhee 1995; Jenkins 1997) for men only, two studies (Stanford 3 City 1977; Sydney QFL 1986) for the population as a whole. Stanford 3 City 1977 detected a significant reduction in cigarette or tobacco consumption for the high risk group, but not in the media‐only intervention community compared with controls. The remaining studies failed to detect significant differences.

In the two state‐wide campaigns cigarette consumption was measured on the basis of aggregated sales data. In California a significant decline was observed compared with the rest of the USA. In Massachusetts, declines in consumption were reported, but without statistical comparisons.

Quit attempts:
Of the five studies which assessed quit attempts, two found no significant differences between the intervention and control communities (McPhee 1995; Jenkins 1997) and McAlister 2004 only assessed quit attempts among continuing smokers. The two state‐wide campaigns assessed quit attempts only in the intervention community. In Massachusetts they increased in line with campaign duration but not to a statistically significant extent, while in California rate changes were reported, but without statistical comparisons.

Quit rates:
Seven studies reported quit rates or abstinence rates, with only one study (CORIS 1997) reporting separately for men and women. Three of the studies included men only (Mogielnicki 1986; McPhee 1995; Jenkins 1997). The California TCP 2003 presented the quit ratio, i.e. the percentage of ever‐smokers (current and former) who were ex‐smokers in a given year, and McVey 2000 gave quit rates for smokers and abstinence rates for ex‐smokers. McAlister 2004 reported point prevalence of daily smoking (ceasing to smoke at all or every day). This study also detected a benefit of 3% in quit rates (8% versus 5%) between study areas that received a media campaign and no cessation services, compared with areas without either intervention. This difference, however, was not statistically significant.

CORIS 1997 found significant differences in intervention quit rates for women but not for men, compared to controls. Two studies found significant intervention effects (Mogielnicki 1986; Jenkins 1997), while in Sydney QFL 1986 the quit rate was not reported separately from the smoking reduction rate. Significant differences were not detected in California TCP 2003 or McPhee 1995. McAlister 2004 reported no significant change in the point prevalence of daily smoking. McVey 2000 detected a significant intervention effect of the media campaign on abstinence rates for smokers and ex‐smokers combined at 18 months.

Intermediate measures:
Five of six studies presenting follow‐up results for knowledge or beliefs data found some increases in knowledge about smoking or cardiovascular risk factors (Stanford 3 City 1977; Sydney QFL 1986; CORIS 1997 [women only]; California TCP 2003;; Massachusetts 2003), with only North Coast QFL 1983 failing to detect a significant change.

Among five studies presenting follow‐up data on attitudes to smoking or cardiovascular risk factors, three found no significant differences between the intervention and control communities (North Coast QFL 1983; McPhee 1995; Jenkins 1997). The Californian and Massachusetts TCP campaigns found significant change compared with the period before the campaign, but there were no comparisons with other states.

Information‐seeking behaviour, measured as the number of calls to quitline, enrolments in 'quit centres' and the number of 'quit kits' sold, was reported as increased in Sydney QFL 1986.

Social influences or pressure to quit compared with baseline rates did not change in one study (North Coast QFL 1983), and increased in three studies (Sydney QFL 1986; McPhee 1995; Massachusetts 2003); in McPhee 1995 there was no difference between the intervention and control communities at the follow‐up, while in the Massachusetts and Sydney studies no comparison with the control community was reported. Norms concerning smoking changed compared to baseline in all three studies which measured them (McPhee 1995; California TCP 2003; Massachusetts 2003), but McPhee 1995 found no difference between the intervention and control communities, and in the state‐wide campaigns no comparison was made with non‐intervention states.

Process measures:
Details of mass media campaign awareness in the intervention community comparison with the control community was presented in two studies (McPhee 1995; Jenkins 1997) and in both was significantly higher in the intervention community. None of the studies presented formal dose‐response analysis of intervention effects.

Intervention costs:
Cost‐effectiveness data were presented for two state‐wide campaigns (California TCP 2003; Massachusetts 2003). They indicated that both campaigns brought benefit in terms of decreases in the number of cigarette packs sold per capita per year for each per capita dollar spent on the media campaign. The effect was more pronounced in California than in Massachusetts. Per capita media expenditure was reported in four studies (CORIS 1997; California TCP 2003; Massachusetts 2003; McAlister 2004) and ranged from USD 0.25 and USD 3.35 per person per year. Two studies reported raw campaign costs (Mogielnicki 1986: copy development and production USD 7480, broadcast time USD 15,150; Sydney QFL 1986: AUD 620,000 for media and a 'Quit Centre'), but without attempting any cost benefit analysis.

Discussion

Summary of main results

Eleven mass media campaigns met our inclusion criteria for this review. Two state‐wide tobacco control programmes with mass media campaigns (California TCP 2003; Massachusetts 2003) and six out of nine community studies (Stanford 3 City 1977; North Coast QFL 1983; Mogielnicki 1986; Sydney QFL 1986; CORIS 1997; Jenkins 1997) showed some positive effects on smoking behaviour, and at least one significant change in smoking behaviour (McVey 2000), although the increase in abstinence was for smokers and ex‐smokers combined.

Overall completeness and applicability of evidence

The inclusion criteria for this review (requiring a comparison community) were designed to help us to identify and assess the specific contribution of the mass media to changing smoking behaviour. However, mass media campaigns are rarely the only component of a community‐based smoking cessation intervention, and it is often difficult or impossible to disentangle the contribution that the separate elements make to the overall impact of a comprehensive tobacco control programme. The advantage of assigning a community to control (no campaign) status should be that broad secular trends in smoking behaviour may more easily be distinguished from the specific effect of the intervention being tested. Despite this methodological strength, several of the studies in our review reported confounding of their findings by extraneous or concurrent events. For example, the findings of the baseline survey for the CORIS 1997 study directly contributed to the establishment of the Heart Foundation of Southern Africa, which then set up a number of tobacco control initiatives that may have contaminated CORIS's subsequent findings. The North Coast QFL 1983 campaign came under attack from the tobacco industry and was briefly suspended, with the resulting national publicity assumed to have influenced smoking behaviour in the control community. During the Sydney QFL 1986 programme, which led to a 2.8% decrease in local prevalence, cigarette prices in Sydney rose less than in the rest of Australia, which may have masked some of the positive effects of the campaign. While such interactions may demonstrate the synergy between campaigns and societal changes, they compromise our ability to measure the impact of such research.

Mass media campaigns are inherently difficult to evaluate, since large samples are required to detect relatively small effects on individual members of the target community. However, even small changes may deliver significant benefits at the population level. McVey 2000 extrapolated that the odds ratio (1.53) for being a non‐smoker following the HEA tobacco control TV campaign would yield a decline in prevalence of 1.2% in a stable population with a smoking rate of 28% (the approximate prevalence rate in the UK at that time (ASH 2007)). There is also evidence that a memorable media campaign, particularly a TV‐based one, may increase calls to quitlines, the distribution of quit kits and enrolment in treatment programmes, but that these may be transient responses, and do not necessarily translate to an increase in successful quit attempts. Low success rates and high drop‐out rates may be a consequence of unrealistically high expectations raised by a successful campaign (Sydney QFL 1986).

None of our included studies tested simply a mass media intervention. Some compared groups receiving a mass media intervention alone with groups receiving mass media and community interventions. In these cases only the mass media groups were included in our review (Stanford 3 City 1977; North Coast QFL 1983; CORIS 1997; McVey 2000; McAlister 2004). In some studies the intervention was led by a mass media programme but also included components such as quit lines, physician involvement and clinics (Mogielnicki 1986; Sydney QFL 1986; McPhee 1995; Jenkins 1997). We decided to include all studies in which a mass media programme led the intervention, but this has involved complex processes of evaluation and comparison.

The definitions of smoker, ex‐smoker and quitter varied from study to study, making between‐study comparisons problematic. In addition, the surveys used in both the state‐wide tobacco control programme (TCP) campaigns modified their definition of a smoker during the course of the campaigns. Because of those differences, people who were defined as smokers, ex‐smokers or quitters might not have fallen into those categories in another study. Some studies included both smoking cessation and smoking reduction as primary objectives, and in one study (McAlister 2004) the criterion for success changed retrospectively from complete cessation to no longer smoking every day.

The state TCP campaigns were both introduced as constitutional amendments, and are funded from tobacco excise tax increases. They are to be continued, but with reduced funding and less aggressive advertising. Compared with smoking prevalence before the programme and in other states before and after the programme, the early campaign in California was associated with significant decreases in smoking prevalence. However, the early success was not sustained through the later stages of the campaign. The final evaluation up to 2008 showed a faster decline in smoking prevalence compared with the rest of the USA population. In addition smoking‐attributable cancer mortality rates declined more in California than in the rest of the USA (18.8% vs 2.4%) during 1979‐2002 (California TCP 2003)

Fichtenberg 2000 reported a correlation between changes in funding, per capita cigarette consumption and mortality from heart disease in California during the 1990s, and estimated that over the first eight years of the campaign 33,000 deaths from heart disease were avoided. Goldman et al calculated that a decline of 12.2% following the second wave of media activity could be directly attributed to the campaign rather than to fiscal pressures (California TCP 2003). In Massachusetts the programme budget was also cut. Friend 2002 notes that the gradual tapering off of the effectiveness of the California and Massachusetts campaigns may be an inevitable reduction in impact over time, irrespective of fluctuating trends in funding.

The impact of campaign duration and intensity is difficult to ascertain. The Flay 1987b and Friend 2002 reviews both detected an effect of longer duration and higher intensity campaigns, but our own assessment has less clearcut findings. McVey 2000, comparing the impact of single‐ and double‐weight TV campaigns on quit rates, found no significant differences at six months, with the single‐weight region (Granada) at 6.3% and double‐weight (Tyne Tees, Yorkshire) at 6.6%, yielding an adjusted OR of 1.02 (P = 0.94). The Vietnamese‐American men studies I (McPhee 1995) and II (Jenkins 1997) ran for a total of 24 and 39 months respectively, with the latter producing positive effects, but other factors in the study design (a 15‐month pilot phase, physician input, Saturday schools, student and family involvement) may have confounded the mass media effect. North Coast QFL 1983 lasted for three years, but after early success the campaign was scaled down and prevalence rose again during the second year; the intervention effects may have been masked by a concurrent substantial decline in smoking in the control community. In the Stanford 3 City 1977 study, it is difficult to disentangle the effects of the mass media campaign from face‐to‐face counselling in the intervention communities. However, significant declines in cigarette consumption were observed only within the community in which the whole population was exposed to a mass media campaign with high‐risk smokers receiving intensive counselling, and not in the community that received the media campaign alone. In the Sydney QFL 1986 study, a significant decline in smoking prevalence was observed in the intervention city at the end of the first year, but from that point on both cities (Sydney and Melbourne) over four years received complex interventions which included mass media campaigns. It was not possible to separate out the independent effect of the mass media from co‐interventions such as physician input, smoking clinics, school programmes and shopping mall displays. From the studies in our review, there was no consistent relationship discernible between campaign duration and effectiveness.

Differences by age, gender, ethnicity and education presented similar problems. During the assessment period (1989 to 2000) for the California TCP 2003, the decline was significant for women but not for men during the last phase only. The trend was significant for older smokers (45+) throughout the campaign. Changes in male smoking prevalence were similar between ethnic groups, with the highest smoking prevalence reported in African Americans. In women, significant declines in prevalence were observed for Hispanic and non‐Hispanic white women. The greatest decline in male smoking prevalence was observed for college graduates, while in women the largest declines were noted in those who did not graduate from high school. In the Massachusetts campaign, the decline in smoking prevalence between 1990 and 2000 was significantly different from the rest of the USA for men but not for women. The effect was also more pronounced for people aged between 18 and 34, for those who graduated from high school but not college, and for white non‐Hispanics.

After four years follow‐up CORIS 1997 detected significantly higher quit rates for women than for men, but after 12 years this difference was no longer apparent, with prevalence for men significantly lower in the intervention than in the control community, but not for women. The North Coast QFL 1983 study found no difference between men and women, but a significant trend by age, with greater declines in prevalence among younger smokers. The Sydney QFL 1986 study detected no significant associations between changes in prevalence and demographic characteristics, other than a long‐term decline in prevalence for men but not for women. The impact of age was contradictory, with three campaigns detecting positive effects among older smokers, and three among younger smokers (up to 34 years). Gender indications were also inconclusive, with three studies showing positive long‐term effects for men, and one for women. This mixed picture casts doubt upon the widely‐held assumption that targeted campaigns are likely to be the most effective. Chapman 2007 points out that the tobacco industry, although nuancing much of its promotion to appeal to different subgroups within a population, does not tailor its packaging or advertising of the major brands (e.g. Camel, Marlboro) to different cultural groups or countries.

Quality of the evidence

Overall the included studies were assessed as being at high risk of bias. All the included studies used some kind of control group, but did not randomise communities to intervention and control conditions. Baseline demographic characteristics were statistically compared in five studies (three with positive outcomes and two with a negative outcome). Studies with declared baseline differences between compared groups controlled for those differences in the analyses. Six studies did not conduct statistical tests for comparability between groups or did not describe details of demographic characteristics of the population. Since comparison groups were not randomised, there could be baseline differences between them which could have confounded the results. Most of the studies with positive findings had problems with drop‐outs and missing data. Response rates ranged from 42.7% to 99.4%, with retention rates between 52% and 76%. Most studies with positive findings did not provide information on participants lost to follow‐up. Any of these limitations could have confounded the results of the studies. Analysis 1.5 gives information on those studies which identified possible confounders, the analytical measures taken to control for these, and the changes in effect where reported.

Agreements and disagreements with other studies or reviews

There is a broad consensus that comprehensive tobacco control interventions which include mass media campaigns can be effective in reducing smoking consumption and prevalence (Flay 1987a; Flay 1987b; COMMIT 1995; Levy 2000; Friend 2002; Biener 2006; NICE 2007; NCI 2008; Niederdeppe 2008; Ontario 2009; Durkin 2012; Wilson 2012). The NICE 2007 review found little high quality evidence of the effectiveness of targeting mass media interventions at high‐risk groups such as pregnant women, men only or young smokers. Durkin 2012 assessed mass media interventions in adults on the basis of 26 population‐based studies (cohort, cross‐sectional and time series, with no control group). They concluded that the evidence supports the use of mass media campaigns as a part of comprehensive tobacco control programmes. In addition the authors evaluated different message types and different media channels using subgroup analysis. NCI 2008 reported on the role of the media in promoting and reducing tobacco use. They included studies of various designs, and concluded that mass media campaigns may encourage adult smoking cessation and lead to decreases in smoking prevalence. Niederdeppe 2008 conducted a systematic review of the effectiveness of media campaigns in smoking cessation in socio‐economically disadvantaged populations. They included 50 studies, and concluded that media campaigns are less or equally effective in populations with low socio‐economic status compared with populations with higher socio‐economic status. In addition they observed that media campaigns in disadvantaged populations are most effective when combined with tobacco control programmes. The Ontario 2009 Health Technology Assessment report examined the efficacy and cost effectiveness of selected population‐based interventions for smoking cessation, including mass media interventions. They agreed with the conclusions of our review. Wilson 2012 evaluated several tobacco control interventions outlined in the WHO MPOWER Package, including mass media interventions in adults and youth. They included 19 studies with and without control groups, some of which were also included in our review, but excluded studies assessing multicomponent interventions. They concluded that there is moderate evidence that mass media can be effective in reducing smoking prevalence in adults.

Prisma flow diagram of search results
Figures and Tables -
Figure 1

Prisma flow diagram of search results

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Figures and Tables -
Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Study

Design

Intervention

Control

California TCP 2003

Nested cross‐sectional

1990‐1 75.1%1 (CA only)
CTS: 71.3 ‐ 99.4% (CA only)
NHIS 79.5 ‐ 87.8% across all USA
BRFSS: 77% ‐ 84%
CPS: 87.9%

NHIS, BRFSS and CPS rates apply.

CORIS 1997

Nested cross‐sectional

Nested cohort

69.5% at baseline, 67.5% at resurvey
Cohort identified retrospectively at 2nd survey, i.e. 4087 (56.3%) of those originally surveyed.
In men 46.4% of cohort smoked vs 50% of drop‐outs; in women 15.4% of cohort smoked vs 21.4% of drop‐outs.
Non‐cohort (i.e 1 survey only) were younger, less educated and higher smoking, but differences applied equally across all groups.

64% at baseline, 63% at resurvey


Cohort control not reported separately

Jenkins 1997

Nested cross‐sectional

Pretest: 84%, post‐test: 94%

Pretest: 82%, post‐test: 88%

Massachusetts 2003

Nested cross‐sectional

Median (1995): 60.4%, median (1999): 42.7%6

Median (1995): 68.4%, median (1999): 55.2%

McAlister 2004

Nested cross‐sectional


Nested cohort

Baseline survey 9407, 7m follow‐up 8974; response rate approximately 60%.
Cohort identified retrospectively at 2nd survey, i.e. 622 (58%) of the 1069 baseline daily smokers. Conditions not reported separately. 835 valid phone numbers at follow‐up, giving retention rate of 74.5%.

Not reported separately

Not reported separately

McPhee 1995

Nested cross‐sectional

Pretest: 81%, post‐test: 82%

Pretest: 85%, post‐test: 88%

McVey 2000

Nested cohort

Smokers: 6m: 73%, 18m: 70%
Ex‐smokers: 6m: 76%, 18m: 75%

Smokers: 6m: 74%, 18m: 66%
Ex‐smokers: 6m: 80%, 18m: 76%

Mogielnicki 1986

Randomized cohort

Mail: Yr2 follow‐up: 17%
Clinic: Yr2 follow‐up: 54%

Mail: Yr2 follow‐up: 15%
Clinic: Yr2 follow‐up: 52%

North Coast QFL 1983

Nested cross‐sectional


Nested cohort

Baseline: 71%
Yr 2: 73%
Yr 3: 73%

Not reported here

Baseline: 72%
Yr 2: 74%
Yr 3: 74%

Not reported here

Stanford 3 City 1977

Nested cross‐sectional
(cohort + non‐cohort)

Nested cohort

Gilroy 116 at baseline.
Non‐cohort had higher baseline % rates of smoking (74.2/62.4) and more cpd (20/13.8) than cohort.

73.2% at 2yr follow‐up

Tracy 115 at baseline.
Non‐cohort had higher baseline % rates of smoking (78/52.8) and more cpd (17.4/14) than cohort.

72.2% at 2yr follow up

Sydney QFL 1986

Nested cross‐sectional

Nested cohort

Sydney: 66%
Melbourne: 67.9%
Sydney: 76%
Melbourne: 73%

Australia‐wide: 60%

Figures and Tables -
Analysis 1.1

Comparison 1 Mass media versus no mass media, Outcome 1 Response and retention rates.

Study

Intermediate measure

Process measures

California TCP 2003

Between 1996 and 1999 slight increases in agreements that smoking causes cancer (82.2% and 83.3%) and that passive smoking harms children's health (93.2% and 94%).

Between 1992 and 2002 increases in agreement that passive smoking causes cancer in nonsmokers (1992: 62.4%, 1996: 66.8%, 1999: 68.9%, 2002: 72.1%) and that smoking harms children's health (1992: 85.5%, 1996: 87.7%, 1999: 90.1%, 2002: 90.9%).

In 2008, 67.2% of daily smokers, 80.2% of non‐daily smokers, and 85.8% of nonsmokers perceived secondhand smoke (SHS) as a cause of cancer and 88.6% of daily smokers, 94.1% of non‐daily smokers and 94.7% of non‐smokers believed that SHS can harm the health of children and babies.

Between 1990 and 1991 increase in % of smokers thinking about quitting (from 38.6 to 42%; NS); health‐enhancing attitude score decreased (from 68.4% to 66.3%; SS).

Further increase in % of smokers thinking about quitting between 1994 and 2005, in the next 30 days (1994: 30.9, 2000:37.8%, 2005: 43.9%) and in the next 6 months (1994: 65.6%, 2000: 74%, 2005: 75%).

Between 1992 and 2002 increases in: % of indoor workers reporting a smokefree work site (1990: 35%, 1992: 46.3%, 1993: 65%, 1996: 90.5%, 1999: 93.4%, 2002: 95.4%, 2005: 94.5%, 2008: 96.4%).

In 2008, 95.2% of smokers and 96.6% of nonsmokers report having a completely smoke‐free workplace.

Between 1992 and 2008 increase in % of homes with smoking ban (1992: 48.1, 1993: 50.9%, 1996: 64.5%, 1999: 72.8%, 2002: 76.9%, 2005: 78.4%, 2008: 80.8%).

Between 1992 and 2008 increase in % of smokers with a total home ban on smoking (1992: 19.4 ± 1.6%, 2008: 59.3 ± 2.6%; increase of 204.9%).

Between 1990 and 2008 decrease in % of nonsmokers exposed to ETS at work (1990: 29%, 1993: 22.4%, 1996: 11.8%, 1999: 15.6%, 2002: 12.0%, 2005: 13.9%, 2008: 13.5%).

Increases in: support for further increase in tax on tobacco (1992: 74%, 1993: 78%; 2008: 78%), support for ban on tobacco advertising (1990: 52%, 1996: 65% [read from the graph]), support for ban on tobacco sponsorships (1998: 56%, 2000: 60%), % of adults preferring non‐smoking bars (1996: 75%, 2000: 81%), support for banning smoking outside the entrance of buildings (2002: 62.8%, 2008: 72.1%) and in restaurant outdoor patios (2002: 62.4%, 2008: 75%), outdoor public places (2002: 52.3%, 2008: 60.4%), Indian casinos (2002: 59.9%, 2008: 66.5%), inside cars when children are in them (2005: 92.3%, 2008: 95.2%).
Support for smoking restrictions in public places in at last 4 out of 6 venues significantly higher in CA than the rest of USA (1992/3: 58.5% vs 46.5%; 1995/6: 70.2% vs 51.5%; 1998/9: 75.8% vs 57.3%).
1998 survey showed that multicomponent exposure was significantly associated with reductions in smoking prevalence, increases in home smoking bans and reductions in perceived violations of workplace no‐smoking rules (P < 0.05).

Systematic monitoring of the campaign and independent evaluations of the programme.

Media weight: 1990/1: 50+ TV spots, 50 radio spots (69 radio stations), 20 outdoor ads (775 outdoor venues), 40 newspaper ads (130 newspapers); 1992/3: 40 spots (20 TV, 12 radio, 8 in outdoor locations), 44% focused on reducing ETS, 34% on countering pro‐tobacco influences, 20% smoking cessation and 2% on reducing youth access. 1990/4: local health departments conducted 10,000+ multi‐session programmes focusing on prevention (61%), cessation (37%) and ETS (2%). July 1995 ‐ Dec 1996: 19 ads (11 TV, 4 radio, 4 outdoor); 1997 ‐ 1998: 40 general audience ads for distribution (20 TV, 12 radio, 8 outdoor); July 1998 ‐ Aug 1999: 50 general audience anti‐tobacco ads (22 TV, 13 radio, 15 print ‐ mostly billboards).

Awareness/reach: 1990/1: unaided awareness in adult smokers 38.4% (similar for most ethnic groups), unaided and aided awareness almost 78.7%. In 1992 > 50% of the adult respondents and > 80% of the adolescent respondents recalled having seen or heard anti‐tobacco messages. In 1992 recall of anti‐tobacco campaign highest among the youngest age group (18 ‐ 24: M 75%, W 70%, 25 ‐ 44: M 65%, W 60%, 45+: M 58%, W 50% [read from the graph]).

In 1996, 67% of adults reported seeing antismoking message on TV, 44% hearing on radio, 41% seeing antismoking billboard. In 1998 80% of adults were exposed to tobacco control programme through two or more components.
In 1999 vs 1996 more respondents reported being exposed to lots of tobacco messages on TV, radio, billboards. 91% of adults reported seeing at least one anti‐tobacco ad in 1996, 1998 and 2000. Between 1996 and 2002 the % of smokers reporting seeing a lot of anti‐smoking ads on TV in last month increased, but then decreased between 2002 and 2008 (18 ‐ 24 yrs: 1996: 16.1%, 1999: 29.9%, 2002: 37.9%, 2005: 21.2%, 2008: 20.8%; 25 ‐ 40 yrs: 1996: 13%, 1999: 20.1%, 2002: 23.2%, 2005: 15.7%, 2008: 10.8%; 41+ yrs: 1996: 10.3%, 1999: 14.9%, 2002: 13.6%, 2005: 8.9%, 2008: 5.6%).

Dose‐response: 1990 ‐ 1996 smokers reporting recall of media spots more likely to make a quit attempt in the last year than those who did not. The more channels recalled between 1990 and 1996 the higher increase in cessation was observed. 1996 ‐ 2000 in counties with the highest multicomponent exposure rates, there were greatest reductions in adult smoking prevalence, workplace no‐smoking policy violations and the largest increases in home smoking bans.

Maintenance: The campaign was established as Proposition 99, as a constitutional amendment and mandated the conduct of a mass media campaign. The campaign has been running since 1990.

Intervention costs: CA TCP mass media campaign funding/ total expenditures targeted at tobacco use in millions of US dollars: 1989/90: 14.3/85.8; 1990/1: 14.3/132.0; 1991/2: 16.0/55.9; 1992/3: 15.4/84.0; 1993/4: 12.9/61.1; 1994/5: 12.2/56.3; 1995/6: 6.6/41.8; 1999‐2000: 19.6/60.8; 2000/1: 45.3/88.2; 2001/2: 45.3/108.1; 2002/3 21.1/61.7; 2003/4: 16.8/61.8; 2004/5: 15.7/56.8; 2005/6: 15.7/58.5; 2006/7: 20/65; 2007/8: 15.7/56.6; 2008/9: 15.7/56.5. Total expenditures for mass media campaign and for the programme between 1989 and 1996 were in millions of US dollars: 91.7 and 516.9. Average annual expenditure was USD 3.35 per capita per year, but from mid‐1993 to mid‐1996 when funding was decreased annual expenditure was USD 2.08 per capita per year.
The CTCP media funding was USD 1.33 per capita in 2001/2 and it decreased from 2002/3 (USD 0.6 per capita) to 2007/8 and 2008/9 (USD 0.43 per capita).

The analysis on the basis of per capita consumption of cigarettes and average per capita media expenditures gave estimates of a fall of 3.9 packs per capita per year for each per capita USD spent on the media campaign.

CORIS 1997

At baseline, knowledge scores higher in Swellenden than in Riversdale (both cross‐sectional and cohort surveys; no statistical comparisons given); at 4 yrs more increase in women in Swellenden (both cross‐sectional and cohort surveys; in cohort survey net change statistically significant); at 12 yrs increase in both communities and no difference.
Attitudes were assessed at baseline, but the results were not reported.

Participation and reach of activities recorded in mass media and community intervention town (not included in this analysis).

Media weight: Limited data. 1st yr: 6 different billboards, 6 posters, 8 mailings, frequent news items, health messages on electricity accounts, 1 special supplement in local newspaper. 2nd and 3rd yr frequency of billboards, posters and mailings about half of the initial rate, but news items frequent and annual special supplement in a local newspaper was added.

Awareness/reach: no evidence found.

Dose‐response: no evidence found.

Maintenance: after 4 yrs of active intervention a maintenance programme was run by community.

Intervention costs: per capita costs given (USD 5 over 4 yrs in intervention community).

Jenkins 1997

At baseline no differences in motivation to quit and self efficacy (SF/Houston: 29%/23%; 29%/25%). At follow‐up significant increase in motivation in both communities, but no difference (SF/Houston: 45%/ 44%), no significant change in self efficacy (SF/Houston: 33%/26%).

A 15‐month uncontrolled pilot anti‐tobacco campaign.

Media weight: newspaper articles ‐ 465,000 print media exposures; 15,000+ copies of brochure, 4600 copies of self‐help quit kit distributed; billboard and newspaper ads ‐ 2.8 million exposures, paid TV ads ‐ 100 mins of air time. Short anti‐tobacco presentations at 25 community events, 68 Vietnamese physicians took part in smoking cessation course and 400 Vietnamese students participated in anti‐tobacco activities.

Awareness/reach: Participants were asked if they had ever read, seen or attended any of five elements of media intervention. Recall of each was significantly greater in intervention than in control community (P < 0.05), except for newspaper articles. In both communities smokers were more likely than non‐smokers to recall the elements of the campaign.

Dose‐response: no evidence found.

Maintenance: no evidence found.

Intervention costs: no evidence found.

Massachusetts 2003

1993 ‐ 2000: the majority of non‐smokers believed that second‐hand smoke (SHS) can harm children (96%) and can cause lung cancer (89%); increase in % of smokers believing that environmental tobacco smoke (ETS) can harm their children (1993: 78%, 1999: 91%, 2000: 93%) and can cause cancer (1993: 58%, 1999: 72%, 2000: 74%);
1993 ‐ 1995: decrease in: % believing that ETS is harmful (1993: 90%, 1995: 84%), support for ban on support of sports and cultural events by tobacco companies (1993: 59%, 1995: 53%, P < 0.05); increase in: support for further tobacco tax increase with funds devoted to tobacco control (1993: 78%, 1995: 81%), support for ban on vending machines (1993: 54%, 1995: 64%, P < 0.05).
1993 ‐ 2000: Increase of % of workers reporting a smokefree work site (1993: 53%, 1995: 65%, P < 0.05; 2000: 75%, 2001: 82.2%), % of homes with smoking ban (1993: 41%, 1995: 51%, P < 0.05; 2001: 71.2%). Decrease in: ETS exposure at work (1993: 44%, 2000: 29%), at home (1993: 28%, 2000: 18%) and in restaurants (1995: 64%, 2000: 39%, 2002: 37%), mean hours of ETS exposure during prior week at work (1993: 4.2, 1995: 2.3, P < 0.05, 1997: 2.2, 2001: 1.4), and at home (1993: 4.7, 2000: 3.3). Increase in: % of population covered by smoking restrictions in restaurants (1995: 26%, 2000: 69%, 2001: 78%); % of population living in a town with some form of smoking restrictions in public places (1993: 22%, 2000: 78%, 2001: 85%); % of smokers reporting that fewer than half their friends and relatives smoke (1993: 40.8%, 2000: 46.9%; P for trend = 0.03); most non‐smokers reported so both in 1993 and 2000 (1993: 90.1, 2000: 89.9).
1995‐2000 significant increase in: support for complete ban on smoking in restaurants (1995: 42%, 2000: 51%), public buildings (1995: 46%, 2000: 58%), at indoor sporting events (1995: 52%, 2000: 67%), in shopping malls (1995: 53%, 2000: 67%), but not at outdoor sporting event (1995: 15%, 2000: 19%).
1993 ‐ 1999: decline in the proportion of people who allow visitors to smoke in their homes (1993: 57%, 1999: 37%, P < 0.01), increase in the proportion of people who asked a colleague not to smoke increased (1993: 29%, 1999: 36%).

Messages were developed through formative research including focus groups. Systematic evaluation of the programme.

Media weight: Oct 1993 ‐ Dec 1996: 66 spots aired; 35 TV ads (14,901 total gross rating points [GRP]), 27 radio spots (13,644 total GRP) on smoking cessation in adults; 38 TV (17,800 total GRP) and 22 radio (13,950 total GRP) ads targeted youth; 6 TV (17,727 total GRP) and 2 radio spots (6,308 total GRP) on ETS.

Awareness/reach: 88% respondents saw any TV ads, among quitters ‐ 97%. On average 4.48 ads recognised. In 1995, 65% of the adult population were aware of the statewide anti‐tobacco campaign, and 76% recognised the tag line of the media campaign "It's time we made smoking history". Around 89% of adult smokers reported having heard, read or seen information about quitting smoking within the past year.

Dose‐response: no evidence found.

Maintenance: The campaign is ongoing.

Intervention costs: On average USD13 million per year spent on anti‐tobacco advertising. About 48% expenditure (up to Dec 1996) spent on TV time (much prime time viewing hours). MTCP activities expenditures: 1st yr ‐ USD 43 million, 3rd yr ‐ USD 35 million. Mean per capita cost for media campaign was USD 2.42.
The analysis on the basis of per capita consumption of cigarettes and average per capita media expenditures gave estimates of a fall of 0.5 packs per capita per yr for each per capita dollar spent on the media campaign.

McAlister 2004

Knowledge, attitudes, beliefs, intentions, brand preferences were assessed at baseline. Process of change variables were measured by rating by responders how much they agree or disagree (1 = strongly disagree, 5 = strongly agree) with the statements regarding attitudes to smoking and occurrence of behaviours associated with smoking cessation. At follow‐up, mean level of agreement was significantly higher among those who reported daily smoking cessation compared to those who maintained daily smoking with regard to having support for quitting and behaviours associated with smoking cessation and dealing with tension. Participants still smoking at follow‐up were significantly more likely to agree with the statements about attitudes and ability to quit . There are no follow‐up results in comparison with control group.

In message development community forums, focus groups and pre‐testing were used. No information regarding ongoing monitoring of media campaign was provided.

Media weight: no evidence found.

Awareness/reach: At follow‐up correlations between media exposures and process variables were calculated for all groups. No separate result for mass media were given.

Dose‐response: Not given for mass media campaigns without community programmes separately.

Maintenance: no evidence found.

Intervention costs: per capita media spending reported, high level and low level media represented about USD 0.25 versus USD 1.00 per capita for media spending during the year of the campaign.

McPhee 1995

Motivation to quit and self efficacy significantly more common in Santa Clara (SC) at baseline (SC/Houston: 36%/23%; 34%/ 25%). At follow‐up motivation to quit increased significantly in both communities, but did not differ (SC/Houston: 44%/ 44%), self efficacy did not change significantly and did not differ at follow‐up (SC/Houston: 32%/ 26%). 'Any household member smokes' responses similar at baseline (SC/Houston: 31%/29%) fell significantly in both communities (SC/Houston: 26%/24%). 'No friend smokes' responses not different at baseline (SC/Houston: 4%/3%) increased significantly in both communities (SC/Houston: 15%/13%). 'All or most friends smoke' responses similar at baseline (SC/Houston: 43%/46%) increased in Houston (SC/H: 44%/51%). Never‐smokers or former smokers advising family or friends to quit smoking increased from baseline in both communities (SC vs Houston: from 51% to 62% (P < 0.05) vs from 34% to 66% (P < 0.05)). Current or former smokers being advised to quit by family or friends increased from baseline in both communities (SC vs Houston: from 53% to 63% (P < 0.05) vs from 51% to 62% (P< 0.05)).

Media weight: newspaper articles ‐ 562,000 print media exposures, videotape copies distributed to 60+ Vietnamese doctors' offices and 20 community agencies in SC; nearly 42,000 brochures at 500+ location in SC, nearly 6000 self‐help quit kit copies at 250+ locations; 1140 signs with adhesive backings saying "Please do not smoke" in Vietnamese distributed; 50 copies of 3 billboards posted each month and printed in newspapers and magazines (8,000,000 print media exposures). Paid TV ads ‐13,000 secs of air time; short anti‐tobacco presentations at 30 community events. CME courses on smoking cessation ‐ 68 Vietnamese physicians.

Awareness/reach: Participants were asked if they had ever read article or seen advertisement in Vietnamese language newspaper or seen television programme or billboard in Vietnamese or heard a speech at a Vietnamese community meeting. Significantly more respondents from SC reported exposure to anti‐smoking activities in Vietnamese language (except for newspaper articles and public speaking). Smokers and non‐smokers in SC recalled significantly more intervention elements (SC vs Houston: out of a possible 5: 3.0 vs 1.6, P < 0.01 and 2.8 vs 1.4, P < 0.01). In SC significantly more physicians reported using antismoking brochures in Vietnamese, providing self‐help quit kits and referring patients to smoking cessation programmes (SC vs Houston: 86.4% vs 32.5%, P = 0.001; 66.7% vs 43.2%, P = 0.03; 35.9% vs 13.5%, P = 0.02). Quality of health education materials assessed: at follow‐up 96.4% of physicians reported they were helpful.

Dose‐response: no evidence found.

Maintenance: no evidence found.

Intervention costs: no evidence found.

McVey 2000

Smoking and health‐related attitudes assessed at baseline, no follow‐up data given.

Qualitative pilot research studies using focus groups and in‐depth interviews with smokers and ex‐smokers were conducted before and during the campaign.

Media weight: It was estimated that over the course of the campaign in the double‐weight regions each viewer could see at least 20 screenings, and in single‐weight regions about 15.

Awareness/reach: no evidence found.

Dose‐response: nearly 20,000 calls to the quit line during the campaign. No evidence of an effect of intensity of ads measured at 6 month follow‐up found.

Maintenance: After study completion the TV advertisements were shown nationally in all TV regions in England.

Intervention costs: evidence not found.

Mogielnicki 1986

Attitudes and beliefs regarding cigarette smoking were assessed on enrolment on a 5‐point rating scale (1 = disagree strongly, 5 = agree strongly). No separate results for 2nd yr follow‐up (mass media) given.

Surveys, interviews, copy tests among patients fulfilling study inclusion criteria but not included in the final study group.

Media weight: One 60‐sec main commercial and 2 x 30‐sec variants used. A series of 3‐week 'flights'; main spot on TV was broadcast 40 times, and variants 106 times, radio spot ‐ 90 times. Each flight lasted 2 ‐ 3 weeks.

Awareness/reach: Participants were asked about the recall of advertising campaign, clinic participants in media group (White River Jct): 61% did and 39% did not recall the advertising campaign at 6 month follow‐up.

Dose‐response: Among those who recalled, 43.5% were abstinent at 6 months, while among those who did not 26.7% abstinent.

Maintenance: no evidence found.

Intervention costs: total costs of commercials development and production: USD 7480, broadcast time cost: USD15,150.

North Coast QFL 1983

Attitudes to smoking and knowledge of the effects of smoking assessed in a questionnaire (6 questions each). No significant differences were found. Influences in decision to quit smoking reported ‐ the most common: health concerns (Coffs Harbour/Tamworth: 53%/ 59%), the least common: social pressure and media advertisement (Coffs Harbour/Tamworth: 1%/1%, 2%/1%), other were friends/family (Coffs Harbour/Tamworth: 15%/ 13%) and doctor's advice (Coffs Harbour/Tamworth: 7%/5%). Technique of quitting: most common ‐ 'cold turkey' (87% in both towns), quitting with help was rare (Coffs Harbour: self‐help kits 2%, group programme 1%, hypnosis 2%).

Focus groups and spot surveys were used to assess the effectiveness of the intervention techniques used.

Media weight: Information collected only on community programme exposure.

Awareness/reach: no evidence found.

Dose‐response: no evidence found.

Maintenance: no evidence found.

Intervention costs: no evidence found.

Stanford 3 City 1977

25‐item behavioural interview on participants' knowledge about risk factors (diet, weight, exercise and smoking ‐ 3 items). Significant increases in knowledge scores in intervention cities compared to control city after 2 yrs.

Mass media campaign was monitored and revised; instructional content, development, pretesting, application and reformulation co‐ordinated by study staff.

Media weight: limited data; TV programming ‐ 3 hours, 50 TV spots, 100 radio spots, several hours of radio programming, weekly newspaper columns, no details about newspaper and ads stories, billboards, printed materials sent via direct mail to participants, posters in buses, stores and work sites.

Awareness/reach: no evidence found.

Dose‐response: increases in risk factor knowledge related to increases in intervention intensity, even in Watsonville‐reconstituted group (mass media only, but a group in community received intensive face‐to‐face counselling) the gains were bigger than in Gilroy (mass media only group) and in both intervention communities the gains were bigger than in Tracy.

Maintenance: After intensive 2 yrs, 1 yr reduced level campaign.

Intervention costs: no evidence found.

Sydney QFL 1986

Additional random samples of Sydney and Melbourne smoker populations surveyed in 1983 and 1986 (Sydney/Melbourne: 1983: 271/217; 1986: 557/ 550). Health beliefs and social influences increased between 1983 and 1986 (Sydney/Melbourne: 40+ years old: males 10% to 29%/21% to 29%; females 12% to 27%/22% to 32%; < 40: males 34% to 53%/30% to 43%, females 30% to 44%/31% to 41%).
Information‐seeking behaviour was measured as the number of calls to quitline and the number of enrolments to quit centre ‐ see dose‐response.

Formative research on messages' effectiveness among target audience.

Media weight: 1983: 389 primetime ads spots (3 commercials used) ‐ the intensity alternated in 2‐wk phases: between heavy and nothing for 1st 3 months, follow‐up ad campaign of half the intensity after 5 months. After 1st year spots during prime or fringe time for approx 4 wks at the start of the campaign yr ‐ each yr nearly 40 spots/wk.

Awareness/reach: During 1st yr each month 750 people in Sydney and 200 in Melbourne interviewed on recall of campaign messages and response to the question on likelihood of giving up smoking in the next 12 months. The recall rates for the commercial most frequently shown were 87% in smokers, 82% in non‐smokers and 85% in ex‐smokers, the second in the frequency: 73%, 69% and 67% respectively. The third commercial recall rates were: smokers: 39%, non‐smokers: 31%, ex‐smokers: 33%. During campaign more smokers in Sydney indicated they were likely to quit smoking, especially after 4 months (difference: 18.6%); the difference disappeared by the end of the campaign. Long‐term assessment: TV ads recall measured each yr in separate random samples of the population of at least 1000 persons higher in Sydney (Sydney/Melbourne: 1984: males: 72%/70%, females: 80%/67%; 1986: males: 92%/68, females: 94%/67%).

Dose‐response: number of calls to quitline was measured as a direct relationship to the number of TV spots aired (presented on the graph); peak response ‐ 11,000 calls in wk 8; 50,000+ calls to the quitline in the 1st 3 months. 'Quit centre' stop smoking programmes enrolments: almost 3000 in 1983 (previous yr 500); in wk 8 with peak quitline calls number ‐ 352. 19,196 quit kits sold.

Maintenance: Since 1984 the campaigns continued in both Sydney and Melbourne till 1986, ads shown on primetime TV for 6 ‐ 8 weeks during winter months.

Intervention costs: The budget for the campaign given (AUD 620,000), most of which was used for purchase of newspaper space and radio and TV time (AUD 500,000).

Figures and Tables -
Analysis 1.2

Comparison 1 Mass media versus no mass media, Outcome 2 Intermediate measures.

Study

Study population

Smoking prevalence

Tobacco consumption

Other outcomes

California TCP 2003

Selection: population‐based surveys conducted nationally and in California (CA):

National Health Interview Surveys (NHIS) 1978 ‐ 1994: household survey of a stratified, multistage, probability sample of the US population administered to all adults 18+ yrs;

Behavioral Risk Factor Surveillance System (BRFSS) 1984 ‐ 1994: state surveillance of behavioural risk factors among adults, data collected through random digit dial telephone interviews (with multistage cluster sampling method); with California Adult Tobacco Survey from 1994, data up to 2008.

Current Population Surveys (CPS) 1985, 1989, 1992/3 ‐ 1996: national survey of the US household population 15+ yrs, interviews in person; data up to 2008

California Tobacco Survey 1990‐2008

N of participants:
NHIS 1978 ‐ 1994,

response rate 79.5% to 87.8%; CA, rest of US: from 1112 to 5747, from 9205 to 39,203;

BRFSS 1984 ‐ 1994, response rate from 77% to 84%: from 1081 to 3959; with CATS from 1994 over 10,000 in CA and 80,000 to 400,000 in the rest of US.

CPS 1985 ‐ 1996: response rate from 86% to 89%: CA, rest of US: from 4076 to 8272, from 70,164 to 97,856; CPS 1992/3 to 2006/7, CA, rest of US: from 16,000 to 24,000, from 210,000 to 290,000; CTS, CA only between 44,000 and 93,000 respondents per survey.
Comparability of demographic data at baseline: not analysed.

Final data (From 1990 programme start to 2008):

All surveys (CTS, CPS and BRFSS pooled analysis years 1990 ‐ 2008, decline of smoking prevalence:

CA 0.32% points per year, rest of US 0.24% points per year.

2008 smoking prevalence: CA 13.1%,

rest of US: 19%

Interim reported data:

According to the analysis of the results of NHIS,

adult smoking prevalence decreased more rapidly from 1985 to 1990 than from 1978 to 1985 in CA and in the rest of USA . Increase in rate of decline between 2 periods in CA was 0.62% points per year (95% CI ‐1.27 to ‐0.03), and in rest of USA 0.43 (‐0.70 to ‐0.16).

Adult smoking prevalence decreased less rapidly from 1990 to 1994 than from 1985 to 1990 in CA (0.39% points per year) and in the rest of USA (0.05%, non‐sig different from zero).

Between 1978 and 1994 adult smoking prevalence was 2 ‐ 5 percentage points lower in CA than in the rest of USA, but the estimated rate of decline in smoking prevalence was no‐sig different from that of the rest of USA during any of the 3 time periods.
Combined estimated rate of decline before CA TCP began was similar in CA to the rest of USA. Rates of decline were not statistically different but estimated prevalence in CA was lower than in the rest of USA. The rate of decline increased significantly in CA at the early period of the programme but not in the rest of USA. In the late phase of the programme the rate of decline in CA and in the rest of USA was non‐sig different from zero and in both cases it was less than in the previous period.
BRFSS trends similar, but consistently around 2% lower.
CPS, CTS and CATS trends similar to BFRSS (all telephone surveys, while NHIS is face‐to‐face).

NHIS only: rate of change in smoking prevalence 1978 ‐ 1985, CA vs rest of US:
Adults aged 18+ yrs:

‐0.6 (95% CI ‐0.79 to ‐0.40) vs ‐0.5 (‐0.67 to ‐0.33);
1985 ‐ 1990: ‐1.22* (‐1.51 to ‐0.93) vs ‐0.93* (‐1.13 to ‐0.73);
1990 ‐ 1994: ‐0.39* (‐0.76 to ‐0.03) vs ‐0.05* (‐0.34 to +0.24);
Adults aged 25+ yrs:
1978 ‐ 1985: ‐0.61(‐0.83 to ‐0.39) vs ‐0.43 (‐0.63 to ‐0.24); 1985 ‐ 1990: ‐1.11* (‐1.37 to ‐0.84) vs ‐0.88* (‐1.10 to ‐0.66);
1990 ‐ 94: ‐0.61 (‐0.99 to ‐0.23) vs ‐0.20* (‐0.52 to +0.12); *significant difference (P < 0.05) between the estimated rate of change for that period and that for the previous period.

All surveys combined analysis: CA, rest of US rate of decline % (SE) ‐ smoking prevalence %:
pre‐1989 (preprogramme): ‐0.74 (0.12) ‐ 23.3, ‐0.77 (0.09) ‐ 26.2;
1989 ‐ 1993 (early programme): ‐1.06 (0.17) ‐ 18.0, ‐0.57 (0.14) ‐ 23.3;
1994 ‐ 1996 (late programme): 0.01 (0.21) ‐ 18.0, ‐0.28 (0.26) ‐ 22.4; P < 0.05 CA vs rest of US, P < 0.001, change from previous period.

Final data (1970 ‐ 2008), per capita taxable sales:

1970 per capita sales similar in CA and rest of US.

In 1989, CA sales 26.1% lower than sales for the rest of the US (108.8 versus 147.2 packs per year).

Faster rate of decline in CA vs rest of the US.

2002 CA vs rest of the US lower sales (48 versus 101 packs per capita per year).

From 2002 slower rate of decline in CA vs rest of the US (from 44.6 to 40.4 packs per year, a decline of 9.4%), but still lower sales than in the rest of the US. 2008: 40 versus 77 packs per capita per year, or 3.37 versus 6.42 packs per capita per month).

1989/90 to 2006/7, per capita consumption

greater decline in CA vs rest of the US: 60.8% vs 41.0%

(40 vs 92 packs per person per year).

Per capita consumption based on aggregated sales data declined faster in CA in the early period of the campaign than before it started, and the decline was significantly greater than in the rest of USA. During late programme the decline slowed but remained greater than in the rest of USA.
Rate of decline in per capita consumption based on aggregated sales data CA/rest of US pack (SE), packs/month:
Baseline 1989: ‐0.42* (0.03), 9.7/ ‐0.36 (0.02), 12.5;
Early programme 1989 ‐ 1993: ‐0.64#^ (0.03), 6.5/ ‐0.42 (0.03), 10.4;
Late programme 1994 ‐ 1996: ‐0.17 *# (0.07), 6.0/ 0.04 # (0.06), 10.5.

In another analysis presenting the results up to 1999 the rate of decline of tobacco consumption (tobacco consumption pack/month), CA/rest of US: Baseline 1/1983 ‐ 12/1988: ‐0.46 (9.5)/ ‐0.35 (12.4);
Early programme 1/1989 ‐ 12/1993: ‐0.58 (6.6)/ ‐0.4 (10.3);
Mid‐programme 1/1994‐10/1998: ‐0.16 (5.8)/ ‐0.07 (10);
Recent programme 10/1998‐12/1999: ‐1.56 (4.1)/ ‐0.78 (9.1).

In analysis presenting the results for tobacco consumption up to 2002, CA,/rest of US packs/month
1988: 9.8/12.5;
2002: 3.9/7.5.

* CA vs rest of US P < 0.01.
^CA vs rest of US P < 0.001.
#change from previous period P < 0.001.

Quit ratio, defined as % of ever‐smokers (current and former) who were former smokers in a given year, accelerated non‐sig in CA and in the rest of US between 1985 and 1990. 1990 to 1994, rate of increase in quit ratio decreased, with no significant changes in CA and in the rest of US compared to the period of 1985 to 1990.

Quit ratio: NHIS only, 1978 ‐ 1994, CA, rest of US, adults 18+years: 1978 ‐ 1985: 0.73 (0.22 to 1.24), 0.73 (0.40 to 1.05);
1985 ‐ 1990: 1.36 (0.74 to 1.97), 1.04 (0.62 to 1.46);
1990 ‐ 94: 0.18 (‐0.8 to 1.15), 0.15 (‐0.47 to 0.77);

Quit attempts: mean number of quit attempts in last yr decreased between 1990 and 1992 from 48.9% to 38.1%, then increased in 1996 and 1999 to 56.0% and 61.5%.
% of smokers attempting quitting and abstinent for 3m at time of survey increased (1990:15.3%, 1992:18.6%, 1993: 20.2%).
% of smokers with quit attempt in last yr lasting 7 days+ increased between 1990 and 2002 (1990: 29.2%, 1992: 25.1%, 1996: 36.1, 1999: 41.4%, 2002: 40.5%).

% of smokers who are currently quit for 3+ months did not change significantly (1990: 8.5 ± 1.0%, 1999: 5.4 ± 0.5%, 2008: 8.0 ± 2.5%)

% of recent smokers currently in a quit attempt of 6+ months did not change significantly (1990: 5.6 ± 0.7%, 1999: 4.2 ± 0.5%, 2008: 6.4 ± 2.5%)

CORIS 1997

Selection: Total population included in baseline and 4‐yr follow‐up surveys (15 ‐ 64 yrs at baseline and 15 ‐ 68 at 4‐yr follow‐up).
At 12‐yr follow‐up a random sample of white participants aged 15 ‐ 64 was selected ‐ about 50 people per sex‐ and age‐specific decile were selected, excluding people living < 2 yrs in the community.
High‐risk cohort was identified within cohort sample ‐ people who had at least 1 of: high cholesterol (> 20th percentile for age), high blood pressure (systolic > 160 or diastolic > 95 mm Hg or antihypertensive drug use), regular smokers.

N of participants:
Total population Swellendam [Intervention]/Riversdale [Control] [M: men; F: women]. Baseline 1979 (% 1980 census): M: 1224 (65)/ 1082 (60); F: 1396 (74)/ 1208 (68); Follow‐up: 1983 (% of 1980 census): M: 1171 (65)/ 1109 (62); F: 1323 (70)/ 1150 (64); 12‐yr follow‐up 1991: random sample Swellendam/Riversdale: M: 273/269; F: 267/274; response rate not reported.
Cohort Swellendam/Riversdale (% of total population): M: 710 (58)/595 (55); F: 821 (59)/710 (59);
High‐risk cohort, Swellendam/ Riversdale (% of cohort): M: 388 (55)/320 (53); F: 234 (29)/224 (32). Comparability of demographic data at baseline: Authors state that age and sex distributions were very similar.

In total population % of smokers was lower in men and women in intervention compared to control city (non‐sig difference).
Baseline prevalence (1979) and (% change) at 4 yrs (1983): Swellendam/ Riversdale: M: 49.5 (‐9.1)/45.7 (‐7.6); F: 17.6 (‐3.6)/16.1 (‐0.5).
At 4 yr follow‐up in the intervention cohort compared to control cohort the % of smokers decreased (non‐sig) in women but not in men.
Cohort: baseline Swellendam/ Riversdale, and (net change): M: 46.5/44.4 (+0.9); F: 14.5 /14.4 (‐3.0);
At 4 yr follow‐up in the high‐risk cohort in the intervention group compared to control group the % of smokers decreased (non‐sig) in women but not in men.
High‐risk cohort baseline Swellendam/ Riversdale (net change): M: 85.3/82.5 (+1.3), F: 50.9/45.5 (‐7.4);
At 12 yr (1991) follow‐up smoking prevalence was lower in intervention compared to control city.
1991 random sample smoking prevalence Swellendam/ Riversdale: M: 25.3*/34.2; F: 12.4/12.8; *significantly lower than Riversdale.

In total population tobacco consumption was lower in men and women in intervention compared to control city (non‐sig difference).
Baseline per capita consumption of tobacco grammes/day (1979) and % change at 4 years (1983): Swellendam/Riversdale: M: 11.2/‐2.6, 8.8/‐1.8; F: 2.6/‐0.4, 2.3/0.1;
At 4 yr follow‐up in the intervention cohort compared to control cohort tobacco consumption decreased in men and women (non‐sig difference).
Cohort baseline Swellendam/ Riversdale and (net change): M: 11.3/9.2 (‐0.4); F: 2.1/2.3 (‐0.2).
At 4 yr follow‐up in the high‐risk cohort in the intervention group compared to control group tobacco consumption decreased in men and women (non‐sig difference). High‐risk cohort baseline Swellendam/Riversdale (net change): M: 20.5/17.3 (‐0.8), F: 7.5/7.3 (‐0.4).
At 12 yr (1991) follow‐up tobacco consumption was lower in intervention compared to control city. 1991 random sample per capita consumption Swllendam/Riversdale: M: 4.8/5.8, W: 1.9/1.8.

In women quit rates were significantly lower in intervention compared to control city.
Quit rates: Swellendam/Riversdale: after 4 yrs: % of all smokers: M: 16.9/20.1; F: 28.3*/15.5; Light smokers: M: 26.5/32.6**; F: 44.4**/28.6**; Heavy smokers: M: 15.8/17.7; F: 23.7/10.7.

*significant difference compared to control; **significant compared with heavy smokers.

Jenkins 1997

Selection: Phone surveys of randomly selected Vietnamese men, with numbers chosen randomly from 23 most common Vietnamese surnames listed in area phone books. After enumerating all men aged 18+ living in the household and speaking Vietnamese, subjects for interview were selected according to random age rankings in the household.

N of participants:
San Francisco (SF)/Houston: Baseline (% response rate):
1990: 1133 (84)/1581 (82); Follow‐up (response rate %): 1992: 1202 (94)/1209 (88). Comparability of demographic data at baseline: significant differences in educational level, English language proficiency, income, unemployment and mean immigration year.

Smoking prevalence did not differ significantly at baseline. Post‐intervention, smoking prevalence fell significantly (P = 0.004) in SF and increased in Houston. Current smokers % among Vietnamese men: SF/Houston:
Baseline (1990): 36.1/39.6;
Follow‐up (1992): 33.9/40.9. Significant intervention effect (P ⋝ 0.01). The odds of being a smoker at follow‐up were significantly lower in intervention than in control community; OR: 0.82, 95% CI: 0.68 to 0.99). The odds of being a smoker after the intervention were lower for following characteristics: age groups 18 ‐ 24, 45+; at least a college education; good or fluent English; being a student. The odds of being a smoker were higher for unemployed and more recent immigrants.

Significant difference in cpd at baseline (Houston higher), significant decrease in Houston, but not in SF. At follow‐up mean cpd was significantly higher in Houston.
Mean cpd among smokers: SF/Houston, Baseline 1990: 11.1/13.2;
Follow‐up 1992: 10.3/11.9.

Significant increase in quit rate in SF compared to Houston (P = 0.017).
% Quit rates in past 2 yrs: SF/Houston:
Baseline 1990: 7.2/5.8;
Follow‐up 1992: 10.2/7.4; significant increase in SF, but not in Houston.
The odds of being a quitter in 1992 were significantly higher in intervention compared to control community (OR: 1.65, 95% CI: 1.27 to 2.15). Predictors for quitting included: being a student (OR: 2.19, 95% CI 1.45 to 3.33); more recent year of immigration (OR: 1.03, 95% CI 1.0 to 1.05); each additional year of age (OR: 1.03, 95% CI 1.02 to 1.04); at least high school education (OR: 1.33, 95% CI 1.04 to 1.7).
Mean number of quit attempts was similar in both communities at baseline and increased significantly in Houston but not in SF. Mean number of quit attempts: SF/Houston:
Baseline 1990: 1.4/1.1
Follow‐up 1992: 2.1/2.5.
At baseline % reporting any quit attempt in the past was significantly higher in SF. At follow‐up it increased significantly in both communities, but did not differ. Percentage reporting any quit attempt in the past: SF/Houston:
Baseline 1990: 61%/49%
Follow‐up 1992: 73%/77%.

Massachusetts 2003

Selection: population‐based survey in Massachusetts (MA) and in the rest of USA (excluding CA) ‐ Behavioral Risk Factor Surveillance System (BRFSS): annual, state‐based, standardised, random digit dialled telephone survey of non‐institutionalised US adults aged 18+; 42 states including MA and CA participated consistently in BRFSS between 1989 and 1998; MA tobacco surveys conducted since 1993‐4 (baseline) in adults, and every month since March 1995, estimates of adult smoking prevalence are derived from household screening interviews with an adult informant who reported on smoking status for all adult members of the household.

N of participants:
1990/9: 22,309 responses from MA, 946,241 from the rest of USA;
BRFSS 1989/98 sample sizes in MA from 1221 to 4944; in rest of USA (excluding CA) from 63,255 to 113,214.
MA surveys 1994/98 respondents: from 5736 to 21,909. Response rates for BRFSS were 77% to 84% between 1984 and 1994; Between 1995 and 1999 response rates for USA fell from 68.4% to 55.2% and in MA from 60.4% to 42.7%.
Comparability of demographic data at baseline: MA respondents were more likely to be white non‐Hispanic and more likely to be college graduates than the respondents from the rest of USA.

According to the analysis based on BRFSS 1989 to 1998 and MA Tobacco Surveys 1993 to 1999 the slope of smoking prevalence after 1992 was significantly different from zero and from the slope for the rest of USA. Based only on BRFSS 1990 ‐ 1999 controlling for age, sex, race, and education there was a greater decline in current smoking between 1990 and 1999 among MA men than among MA women, and the decline was greater in MA than in the rest of the USA for men and for both sexes combined.

Based on BRFSS 1989 to 1998 and MA Tobacco Surveys 1993 to 1999: for MA the slope of smoking prevalence after 1992 was ‐0.43*% a year ( 95% CI ‐0.66% to ‐0.21; P = 0.001); for the rest of USA the slope after 1992 was +0.03% a year (95% CI ‐0.06% to 0.12%; P = 0.46), * significantly different from zero and from the slope for the rest of USA (P < 0.001).
BRFSS only 1990 ‐ 1999: Baseline % prevalence (95% CI) total pop 1990, MA/rest of USA: 23.5 (21.0 to 26.1)/24.2 (23.7 to 24.7);
Follow‐up 1999: 19.4 (18.0 to 20.8)/23.3 (22.9 to 23.7); significantly lower than in the rest of USA (P < 0.001); crude prevalence OR of current smoking in MA in 1999 compared to 1990: 0.78 (0.66 to 0.92, P trend = 0.01); adjusted for sex, age, race, education OR: 0.83, (0.70 to 0.99, P trend = 0.08); in rest of USA prevalence of current smoking in 1999 compared to 1990: OR 0.95 (0.92 to 0.99, P trend = 0.99); adjusted for sex, age, race, education OR: 1.01 (0.97 to 1.05, P trend < 0.001). Between 1990 and 1999 average change in the log odds was ‐1.3% in MA and in rest of USA +0.6% (sig difference between slopes, P = 0.01). Men/women baseline prevalence % (95% CI) 1990, MA/rest of USA: M: 25.9 (22.0 to 29.8)/26.0 (25.2 to 26.7); W: 21.5 (18.2 to 24.8)/22.5 (21.9 to 23.2);
Follow‐up 1999: M: 19.5* (17.3 to 21.6)/25.6 (24.9 to 26.2); *significantly different from the rest of USA, P < 0.001; W: 19.3* (17.5 to 21.1). 21.2 (20.7 to 21.7), * significantly different from the rest of USA P = 0.04;
Crude prevalence OR (95% CI) of current smoking in MA/rest of USA in 1999 compared to 1990:
M: OR: 0.69 ( 0.54 to 0.89; P trend = 0.03)/0.98 (0.93 to 1.03; P trend = 0.07); Adjusted for age, race/ethnicity. education OR: 0.73 (0.56 to 0.94; P trend = 0.09)/1.03 (0.97 to 1.08, P trend < 0.001);
W: crude OR: 0.87 (0.70 to 1.09; P trend = 0.09)/0.93 (0.88 to 0.97, P trend = 0.03);
Adjusted for age, race/ethnicity, education OR: 0.95 (0.75 to 1.2; P trend = 0.43)/0.99 (0.95 to 1.04, P trend = 0.03);
Average decline per year in the log odds of current smoking between 1990 and 1999 in MA was M: 1.8% (SE: 0.011) and W: 0.7% (SE: 0.010), and in rest of USA there was an average annual increase of : M: 0.8% (SE: 0.002), W: 0.4% (SE: 0.002); Men: significant difference between MA and USA, P = 0.016; W: non‐sig difference between MA and USA P = 0.243.

Annual per capita tobacco consumption based on aggregated sales data:
1988 ‐ 1992 ‐ before tax increase declines in consumption for MA adults and for rest of USA were similar: 15%, 14% ‐ this = annual decline 3 ‐ 4%.
In 1993 consumption declined by 12% in MA and by 4% in rest of USA.
After 1993 there was consistent annual decline of more than 4% in MA and less then 1% in rest of USA.

Quit ratio: not reported.
Quit attempts: % of smokers planning to quit within the next 30 days increased between 1993 and 1997 (from 22% to 42%). Percentage of past year smokers making quit attempt increased non‐sig between 1993 and 1996 (from 47.5% to 54.2%). Percentage of quitters making successful quit attempt significantly increased between 1993 and 2000 (from 18% to 25%), % of smoking pregnant women decreased (from 25% in 1990 to 11% in 1999, while in the rest of US these %s were 18% and 12% respectively).

McAlister 2004

Selection: telephone surveys of randomly selected (random digit dialing) adults from Texas (including individuals not resident in treatment areas); respondent in each household with the most recent birthday aged 18+ selected. Cohort of smokers identified in baseline cross‐sectional sample was followed.

N of participants:
1069 daily cig smokers identified in baseline survey; 622 available for follow‐up.
Baseline: High media 133, low media 274, control 232: Follow‐up (% of original sample): 87 (65), 158 (58), 137 (59). Comparability of demographic data at baseline: not analysed.

Not reported ‐ only levels of daily smoking given (15.7 at baseline, 17.5 at follow up)..

Not reported.

Complete cessation achieved by approx 2% of panel of daily smokers.
% quitting daily smoking, i.e. reduction, not complete cessation (data estimated from bar graph): non‐significantly higher in high and low media groups compared to group without any intervention, both in the whole sample and in followed‐up subsample.
% of whole sample quitting daily smoking: high media 4.5, low media 4.7, control 3.0 (taken from graph); Followed group 7.0%, 8.0%, 5%.
Quit attempts: Among the participants still smoking at follow‐up 27.3% had made quit attempt.

McPhee 1995

Selection: Phone surveys of randomly selected Vietnamese men, chosen randomly from 23 most common Vietnamese surnames in area telephone books. After enumerating all men aged 18+ living in the household and speaking Vietnamese, subjects for interview were selected according to modified Kish procedure.

N of participants:
Santa Clara(SC)/Houston: Baseline (% response rate):
1990: 1322 (81)/1581 (82); Follow‐up (% response rate)
1992 (response rate): 1264 (85)/1209 (88).
Comparability of demographic data at baseline: sig differences in educational level, English language proficiency, mean year of immigration and unemployment.

At baseline smoking prevalence in SC differed significantly from Houston (control community). It slightly increased in control area and remained unchanged in intervention area ‐ there was no intervention effect. Current smokers among Vietnamese men, SC/ Houston: Baseline 1990: 36%/40%;
Followup 1992: 36%/41%.

At baseline SC cig consumption differed significantly from Houston. It fell significantly in the control area , but there was no significant intervention effect. Mean cpd among smokers: SC/Houston: Baseline 1990: 9.9/13.2;
Follow‐up1992: 9.6/12.0.

Quit rates in past 2 yrs differed significantly at baseline, but there was no significant intervention effect.
SC/Houston: Baseline 1990: 8.0%/6.0%,
Follow‐up 1992: 10%/7%.

Mean number of quit attempts was significantly higher in SC at baseline. At follow‐up it increased significantly in both communities, but did not differ. Mean number of quit attempts: SC/Houston:
Baseline 1990: 1.3/1.1.
Follow‐up 1992: 2.3/2.5.
At baseline percentage reporting any quit attempt in the past was significantly higher in SC. At follow‐up it increased significantly in both communities, but did not differ. Percentage reporting any quit attempt in the past: SC/Houston:
Baseline 1990: 64%/49%.
Follow‐up 1992: 79%/77%.
Adjusted OR for intervention in regression model was 1.1 (95% CI 0.9 to 1.4), i.e. no sig effect of programme.

McVey 2000

Selection: In each TV region a random sample of enumeration districts were selected (1 in 40) and within each district a random sample of households (1 in 20) using computer‐generated lists. Selected households were visited by interviewers and a resident member was selected for the sample with the use of pseudo‐random Kish‐grid method. Only adults at least 16 yrs were sampled. Only smokers and ex‐smokers were included in the study. Attempts were made to re‐interview the same respondents 6m and 18m later with similar structured questionnaire.

N of participants:
TV‐media (smokers/ex‐smokers), Baseline sample: 1744/1256;
Followed at 6m: 1064/854; Followed at 18m: 673/598; Control (smokers/ex‐smokers),
Baseline sample: 719/775;
Followed at 6m: 475/571;
Followed at 18m: 288/402.
Comparability of demographic data at baseline: no statistical analyses.

Not reported other than varying in different target regions.

Not reported.

By 18m, more successfully followed smokers in TV media group gave up smoking than in control group (non‐sig). Quit rates in smokers (TV media/control): 9.7%/8.7%; OR: 1.27 (95% CI 0.77 to 2.08), adjusted for predictors of giving up cigarettes: sex, age, manual/non‐manual job, daily cig consumption, "want to reduce smoking" and sex by manual/non‐manual work interaction. More ex‐smokers not relapsing were observed in TV media than in control group (significant): 96.3% vs 94.5%, adjusted OR: 2.21 (95% CI 1.11 to 4.40, P = 0.025); adjusted for predictors of remaining off cigs: number of cpd before stopping, length of time since stopping, whether persuaded to stop smoking by somebody else. TV media compared to no intervention was estimated to have increased the odds of not smoking.
Pooled common effect on not smoking: adjusted OR: 1.53 (1.02 to 2.29, P = 0.04).

Mogielnicki 1986

Selection: reception clerks asked all patients registering for clinic visits about smoking and interest in stopping.

N of participants:
In 2nd yr clinic response rates not reported (media group/control group) Baseline: 71/33; Follow‐up (% retention rate): 38 (54)/17 (52);
Mailing group: Baseline: no information given;
Follow‐up: 12/5, response rates not reported.
Comparability of demographic data at baseline for clinic participants: non‐sig differences in age, number of cigs smoked, thiocyanate level, % smoking high‐tar cigarettes. No details for mailing group participants given.

Not reported.

Not reported.

In clinic participants, number abstinent was higher in group receiving mass media compared to group not receiving mass media (calculated for all participants included in the study at baseline). In the mailing group in both media and no‐media groups no participant was abstinent and no number of participants included in the study at baseline was given.
Abstinence rates % (media group vs control group), clinic participants, total sample: 19.7 vs 3;
Followed sample: 36.8 vs 5.9.
Mailing group participants, followed sample: 0% vs 0%.

North Coast QFL 1983

Selection: systematic random sample of men and women 18+ from each town, up to 2 adults per household.

N of participants:
Coffs Harbour (CH)/Tamworth (control):
Baseline 1978 (% response rate): 612 (71)/589 (72);
1980: 1272 (73)/1239 (74); 1981: 1195 (73)/1195 (74). Comparability of demographic data at baseline: age and sex differences across towns and yrs.

Significant effect of the intervention was found.
Prevalence reported for 1978 (baseline), 1980, 1981 by town, age, sex and yr. In both towns younger people tended to have bigger declines and 65+ smallest. % points change in prevalence between 1978 and 1981: in CH Men (M) from 7.1 (65+ yrs) to 11.2 (18 ‐ 25 yrs) decline, Women (F): from 6.0 (65+ yrs) to 11.1 (18 ‐ 25 yrs) decline; in Tamworth M 4.1 (65+ yrs) to 5.1 (36 ‐ 45 and 46 ‐ 55 yrs), F 2.1 ( 65+ yrs) to 5.1 (18 ‐ 25 yrs). Effect of intervention compared to control was significant (P < 0.05).

Not reported.

Quit rates not reported.

Stanford 3 City 1977

Selection: in each city random multistage probability sample of men and women 35 ‐ 59 yrs old.

N of participants:
Watsonville/Gilroy/Tracy; Baseline (% of original sample): 605 (73)/542 (82)/532 (81);
Completing yr 2 follow‐up (% of baseline sample): 423 (70)/397 (73)/384 (72).
High‐risk subjects ‐ Watsonville media only/Gilroy/Tracy, Baseline: 56, 139, 136; Completing baseline and 2 yr follow‐up: 40, 94, 95. Comparability of demographic data at baseline: no statistical comparisons made.

Reported only for high‐risk group (baseline/1st /2nd /3rd year follow‐ups):
Watsonville‐randomised control (media only): 56.8/no results given as difference in the direction contrary to prediction;
Gilroy: 62.4/‐15.1/‐15.1/‐11.3;
Tracy: 52.8/‐6.4/‐10.6/‐14.9,
all non‐sig.

Per capita cig consumption for adults 35 ‐ 59 yrs (baseline/% change 1973/1974):
Watsonville reconstituted: 6.8/‐6.9,‐13.7*; Gilroy: 6.8/‐2.3/‐7.3;
Tracy: 6.9/‐1.1/‐2.5;
* statistically sig difference (P < 0.05) for % change values compared to Gilroy and Tracy (control).
High‐risk subjects (baseline 1972/% change 1973/1974): Watsonville‐randomised control: 14.2/‐5.8/‐15.1;
Gilroy: 14.6/‐9.8/‐13.8;
Tracy: 13.7/‐8.5/‐17.2.
In the publication presenting results for high‐risk group only the results are given for 3 yrs follow‐ up and they differ from the main Stanford Three City publication (Maccoby 1977) as they present the results for subjects who completed baseline and all 3 annual follow‐up surveys (baseline 1972/% change; 1973/1974/1975), Watsonville‐randomised control: 15.4/‐5.8/‐15.1/‐16.0;
Gilroy: 13.8/‐7.0/‐12.3/‐11.8;
Tracy: 14.0/‐7.4/‐15.9/‐21.0; all non‐sig.

Quit rates not reported.

Sydney QFL 1986

Selection: random weekly selection from list of all Australian electoral subdivisions. Selected 10 separate households starting from randomly selected address from each electoral subdivision (clockwise direction around the residential block). Within household youngest person 14+ or 16+ interviewed.

N of participants:
Cross‐sectional surveys (Sydney/rest of Australia), Baseline 1983: 3978/ 5154; Follow‐up 1984: 4051/4318. Melbourne cross sectional survey ‐ 1518.
Cohort (Sydney/Melbourne), Baseline 1983 : 900/600;
Follow‐up 1984 (% retention rate after excluding people moved): 570 (76)/364 (73). Australia‐wide response rate of approximately 60%.
Long‐term follow‐up weekly surveys, Sydney and Melbourne 1981 ‐ 1987: 68,136 males, 70,634 females.
Comparability of demographic data at baseline: no statistical comparisons made.

At 1st cross‐sectional surveys prevalence decreased in Sydney compared to the rest of Australia (non‐sig).
Baseline prevalence 1983/% change at follow‐up 1984; cross‐sectional Sydney: Men (M): 35.9/‐4.2; Women (F): 30.4/‐1.6; Total: 33.1/‐2.8; Rest of Australia: M: 39.2/‐3.3; F: 29.3/‐0.2; Total: 34.1/‐1.6.
Difference between Sydney and rest of Australia: 1.2% (SE 1.49; non‐sig).
In cohort study significant decrease in smoking prevalence in Sydney compared to Melbourne.
Cohort % Sydney/Melbourne: 30.9 (‐3.4)/36.8 (+0.)%. Standardised (to baseline prevalence) difference Sydney vs Melbourne: 5.4% (SE: 1.89, P < 0.01).
Overall campaign effect assessed as a weighted mean of cross‐sectional and cohort assessments showed significant effect of campaign on smoking prevalence. Overall campaign effect weighted mean: 2.8% (95% CI: 0.5 to 5.1).
In model assessing long‐term effects of the campaign (1981 ‐ 1987) there was an immediate decrease in smoking prevalence after the beginning of the intervention (6m) in Sydney and Melbourne in males and females. Continuation of the campaign in subsequent years was associated with further drop in percentage points in Sydney and Melbourne men, but not women. Long‐term effectiveness model: Sydney (baseline estimated prevalence 1981 ‐ June 1983: immediate campaign effect 1983 ‐ 6m % points change/continuation of the campaign 1983 ‐ 1987 % points change per year): M: 38.7/‐2.52/‐1.12; F: 31.6/‐2.61/no sig decline. Melbourne (baseline estimated prevalence 1981 ‐ June 1984: immediate campaign effect 1984 ‐ 6m % points change/continuation of the campaign 1984 ‐ 1987 % points change per year): M: 40.1/‐2.87/‐1.9; F: 30.9/‐2.5/no decline.

Cross‐sectional studies after 1st issue of the campaign found decrease in cpd in Sydney compared to the rest of Australia (non‐sig).
Changes in mean cpd in Sydney and rest of Australia,
Baseline 1983/% change in 1984:
M: 20.1 (‐0.6)/20.3 (+0.2); F: 18.4 (‐1.3)/18.2 (‐0.8).

Significantly larger proportion of Sydney smokers compared to Melbourne smokers quit smoking or cut down on cigs. Relapse rate and initiation rate were similar in both cities.
Cohort Sydney/Melbourne 1983 ‐ 1984: initial smokers % quit: 23/9; initial smokers cut down: 12/9; total quit + cut down: 35*/18; initial ex‐smokers relapse: 10/11; initial non‐smokers started: 4/4.

Quit attempts: Failed attempts to quit or cut down were assessed in cohort of smokers in Sydney and Melbourne ‐ proportions were not significantly different: S vs M: quit: 11% vs 19%, cut down: 20% vs 23%.

* P < 0.05.

Figures and Tables -
Analysis 1.3

Comparison 1 Mass media versus no mass media, Outcome 3 Primary measures of smoking behaviour.

Study

Type of outcome

Orientation

Cost

Effect

California TCP 2003

Prevalence







Packs per day

Social diffusion theory, with social marketing and social policy change

April 1990 ‐ June 1993 USD 26m for media campaign.
1990 ‐ 5 total programme spend USD 694m

Pre‐programme prevalence 23.3 in CA vs 26.2 in rest of USA. By 1993 down to 18.0 in CA, 23.3 rest of USA; By 1996, prevalence still 18.0 in CA vs 22.4 in rest of USA.

Final data (2008):

CA 13.1%,

rest of US: 19%

Rate of decline % (SE): CA: Pre‐TCP: ‐0.74 (0.12) > early period : ‐1.06 (0.17) [P < 0.05 CA vs USA, P < 0.001 change from pre‐TCP] > late period 0.01 (0.21) P < 0.05 CA vs USA.
USA: ‐0.77 (0.09) > 0.57 (0.14) > 0.28 (0.26).

Final data (up to 2008):

CA 0.32% points per year, rest of US 0.24% points per year.

Media campaign alone (without other Prop 99 components) led to decline in cig consumption of 12.2% between April 1990 and March 1991.
Decline of 7.7 packs per capita (pc) attributable to media campaign alone, with 10% increased expenditure on media yielding 0.5% reduction in cig sales.
1989 ‐ 96, CA spent USD 0.50 pc per year on media, leading to a fall of 3.9 packs pc per yr for each USD spent on media.

Final data (1970 ‐ 2008), per capita taxable sales:

CA sales 26.1% lower than sales for the rest of the US (108.8 versus 147.2 packs per year).

Faster decline in CA vs rest of the US.

2002 48 versus 101 pc per yr

2008: 40 versus 77 pc per yr,

CORIS 1997

Prevalence




Cigarettes per day

Not reported

pc cost of USD 5 over 4 yrs in media‐only intervention town.

After 4 yrs, Intervention (Control) prevalence in men dropped from 49.5 (45.7) to 40.4 (38.1) , and in women from 17.6 (16.1) to 14.0 (15.6) . Data contains paired and unpaired observations, in total population.

After 4 yrs, Intervention (Control) tobacco per day (pd) in men moved from 11.2 (8.8) to 8.6 (7.0) grams, and in women from 2.6 (2.3) to 2.2 (2.4).

Jenkins 1997

Prevalence



Cigarettes per day


Odds of quitting

Not reported

Not reported

At 2 yrs, San Francisco (int) prevalence down from 36.1 to 33.9 (P ≤ 0.01), and Houston (control) up from 39.6 to 40.9. Net change was ‐3.5 % points (P = 0.004).

San Francisco (Houston) cpd pretest: 11.1 (13.2); post‐test: 10.3 (11.9)

OR 1.65 (1.27 to 2.15) in favour of San Francisco

Massachusetts 2003

Prevalence




Packs per day

Social diffusion theory, with social marketing and social policy change

USD 39m pa for whole programme.




MA pc USD 6.50 (in 2000); cf. CA USD 3.31, UK $0.89, Australia $0.48 (1997), South Africa $0.04, France $0.32, spent on tobacco control programmes.

1990 ‐ 9: MA prevalence declined from 23.5 (CI 21.0 to 26.1) to 19.4 (CI 18.0 to 20.8).
41 US states: 24.2 (CI 23.7 to 24.7) to 23.3 (CI 22.9 to 23.7)

1993 ‐ 6: reduction of 0.5 per annum (pa) for each pc USD spent on media.
From 1993 pc MA consumption declined > 4% pa, compared with < 1% pa in comparison states.

McAlister 2004

Prevalence

Daily smoking cessation

Cessation

Social learning theory, transtheoretical model. using modelling, social reinforcement for behaviour change, and emotional arousal.

USD 9m pa

Prevalence of daily smoking increased from 15.7 to 17.5.

% of baseline smokers ceasing daily smoking: Low‐level media: 4.7, High‐level media 4.5, No media 3.0

Media‐only areas achieved 8% cessation vs control 5%.

McPhee 1995

Prevalence



Cigarettes per day

Not reported

Not reported

No change in prevalence in Intervention community from baseline to 2 yrs: 36.0, while controls increased (non‐sig) from 40 to 41.

Intervention cpd down from 9.9 to 9.6, controls down from 13.2 to 12.0.

McVey 2000

Quit and relapse rates.

No formal attributing, but 'morbid humour' used.

Not reported

At 18m, 9.8% of intervention smokers vs 8.7% of control had quit, and 4.3% of intervention ex‐smokers vs 3.7% controls had relapsed. Pooled OR of not smoking was 1.53 (CI 1.02 to 2.29, P = 0.04).

Mogielnicki 1986

Cessation rates

Mass marketing techniques (situation analysis, objective definition, copy development, media plan, campaign launch) combined with traditional clinical trial approach.

Copy development and production: USD 7480, Air time: USD 15,150 (1980 USD)

Per protocol clinic cessation at 2nd year:
Media only 36.8%, No media 5.9%.
Mailing recipients: no quitters in either group.

North Coast QFL 1983

Prevalence

Social marketing and communication theory

Not reported

Mean declines in prevalence in Intervention community of 9.5% from baseline to Yr 3, and in controls of 4.4%.

Stanford 3 City 1977

Prevalence



Cigarettes per day

Social marketing, social learning and communication theory

Not reported

All participants: Baseline: Intervention 65.5, Control 55.6.
Cohort: Baseline Intervention (Control) 62.4 (52.8), declined by 11.3% (14.9%) after 3 yrs (P < 0.05).

Mean Intervention cpd down 7.3% over 2 yrs from 6.8 at baseline; controls down 2.5% from 6.9 at baseline.

N.B. Within treatment groups, cohort baseline cpd of Int 13.8, control 14.0.

Figures and Tables -
Analysis 1.4

Comparison 1 Mass media versus no mass media, Outcome 4 Study summary by type of outcome.

Study

Differences

Possible confounders

Adjusted effects

Jenkins 1997

San Francisco (Int) respondents significantly less proficient in English, of lower education, lower income, less likely to be employed and more recent year of immigration than Houston (control) respondents.

All these factors were included in multivariate analyses.

Unadjusted ORs not reported. Strongest predictor of quitting was being a student (adjusted OR 2.19, 95% CI 1.45 to 3.33); at least high school education (OR 1.33: CI 1.04 to 1.70); more recent year of immigration (OR 1.03: CI 1.00 to 1.05); each + yr of age (OR 1.03: CI 1.02 to 1.04).

Massachusetts 2003

MA respondents were more likely to be white non‐Hispanic and more likely to be college graduates than the respondents from the rest of USA.

Age, sex, race and education were treated as confounders.

Unadjusted OR for current smoking in MA in 1999 vs 1990 was 0.78 (CI 0.66 to 0.92, P trend 0.01).
Adjusting for sex, age, race and education, OR was 0.83 (CI 0.70 to 0.99).
Compare with USA: 1990 unadjusted OR 0.95 (CI 0.92 to 0.99). Adjusted 1.01 (CI 0.97 to 1.05).

Adjusted prevalence OR for current smoking in MA showed annual decline in log odds of 1.3% pa, whereas USA showed average increase of 0.6% (P = 0.01).

McAlister 2004

Responders and non‐responders to follow‐up survey differed significantly by age, race and gender.
Also significant differences in gender and education between baseline and follow‐up samples.

Daily smoking rate "adjusted for age, gender and educational level". No OR reported.

McPhee 1995

Small but significant differences in mean age, educational level, English language proficiency, mean year of immigration, between SC and Houston and/or between pre‐ and post‐test samples. Larger differences in employment (8% unemployed in SC vs 6% in Houston pre‐test) and income (32% below poverty level in SC vs 26% in Houston post‐test).

Analyses controlled for site, time, (pre‐, post‐test), intervention term (site x time), age, education, English language proficiency, year of immigration, employment status, income.

Differences in prevalence between SC and Houston did not persist after controlling for sociodemographic characteristics. Strongest predictor of current non‐smoking was age 65+. Age 18 ‐ 24, immigration before 1977, college education and English fluency all predicted non‐smoking, but employment and income did not. Adjusted OR for intervention term was 1.1 (95% CI 0.9 to 1.4). Adjusted OR for recent quitting 1.1 (95% CI 0.7 to 1.7).

North Coast QFL 1983

Age and sex differences across towns and years.

To counter known confounding, an AGE*SEX*TOWN term was constructed for the regression model, and a TOWN*YEAR factor.

Sydney QFL 1986

Location of interview (Sydney vs Melbourne) was significant predictor of quitting (P < 0.05).

Sex, age, education, marital status, SES did not predict quitting or act as confounders.

No ORs reported.

Figures and Tables -
Analysis 1.5

Comparison 1 Mass media versus no mass media, Outcome 5 Baseline differences and possible confounding.

Comparison 1. Mass media versus no mass media

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Response and retention rates Show forest plot

Other data

No numeric data

2 Intermediate measures Show forest plot

Other data

No numeric data

3 Primary measures of smoking behaviour Show forest plot

Other data

No numeric data

4 Study summary by type of outcome Show forest plot

Other data

No numeric data

5 Baseline differences and possible confounding Show forest plot

Other data

No numeric data

Figures and Tables -
Comparison 1. Mass media versus no mass media