School‐based programmes for preventing smoking

Roger E Thomas; Julie McLellan; Rafael Perera

doi:10.1002/14651858.CD001293.pub3

. 2013 Apr 30;2013(4):CD001293. doi: 10.1002/14651858.CD001293.pub3

School‐based programmes for preventing smoking

Roger E Thomas ^1,^✉, Julie McLellan ², Rafael Perera ²

Editor: Cochrane Tobacco Addiction Group

PMCID: PMC7028068 PMID: 23633306

Abstract

Background

Helping young people to avoid starting smoking is a widely endorsed public health goal, and schools provide a route to communicate with nearly all young people. School‐based interventions have been delivered for close to 40 years.

Objectives

The primary aim of this review was to determine whether school smoking interventions prevent youth from starting smoking. Our secondary objective was to determine which interventions were most effective. This included evaluating the effects of theoretical approaches; additional booster sessions; programme deliverers; gender effects; and multifocal interventions versus those focused solely on smoking.

Search methods

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), the Cochrane Tobacco Addiction Group's Specialised Register, MEDLINE, EMBASE, PsyclNFO, ERIC, CINAHL, Health Star, and Dissertation Abstracts for terms relating to school‐based smoking cessation programmes. In addition, we screened the bibliographies of articles and ran individual MEDLINE searches for 133 authors who had undertaken randomised controlled trials in this area. The most recent searches were conducted in October 2012.

Selection criteria

We selected randomised controlled trials (RCTs) where students, classes, schools, or school districts were randomised to intervention arm(s) versus a control group, and followed for at least six months. Participants had to be youth (aged 5 to 18). Interventions could be any curricula used in a school setting to deter tobacco use, and outcome measures could be never smoking, frequency of smoking, number of cigarettes smoked, or smoking indices.

Data collection and analysis

Two reviewers independently assessed studies for inclusion, extracted data and assessed risk of bias. Based on the type of outcome, we placed studies into three groups for analysis: Pure Prevention cohorts (Group 1), Change in Smoking Behaviour over time (Group 2) and Point Prevalence of Smoking (Group 3).

Main results

One hundred and thirty‐four studies involving 428,293 participants met the inclusion criteria. Some studies provided data for more than one group.

Pure Prevention cohorts (Group 1) included 49 studies (N = 142,447). Pooled results at follow‐up at one year or less found no overall effect of intervention curricula versus control (odds ratio (OR) 0.94, 95% confidence interval (CI) 0.85 to 1.05). In a subgroup analysis, the combined social competence and social influences curricula (six RCTs) showed a statistically significant effect in preventing the onset of smoking (OR 0.49, 95% CI 0.28 to 0.87; seven arms); whereas significant effects were not detected in programmes involving information only (OR 0.12, 95% CI 0.00 to 14.87; one study), social influences only (OR 1.00, 95% CI 0.88 to 1.13; 25 studies), or multimodal interventions (OR 0.89, 95% CI 0.73 to 1.08; five studies). In contrast, pooled results at longest follow‐up showed an overall significant effect favouring the intervention (OR 0.88, 95% CI 0.82 to 0.96). Subgroup analyses detected significant effects in programmes with social competence curricula (OR 0.52, 95% CI 0.30 to 0.88), and the combined social competence and social influences curricula (OR 0.50, 95% CI 0.28 to 0.87), but not in those programmes with information only, social influence only, and multimodal programmes.

Change in Smoking Behaviour over time (Group 2) included 15 studies (N = 45,555). At one year or less there was a small but statistically significant effect favouring controls (standardised mean difference (SMD) 0.04, 95% CI 0.02 to 0.06). For follow‐up longer than one year there was a statistically nonsignificant effect (SMD 0.02, 95% CI ‐0.00 to 0.02).

Twenty‐five studies reported data on the Point Prevalence of Smoking (Group 3), though heterogeneity in this group was too high for data to be pooled.

We were unable to analyse data for 49 studies (N = 152,544).

Subgroup analyses (Pure Prevention cohorts only) demonstrated that at longest follow‐up for all curricula combined, there was a significant effect favouring adult presenters (OR 0.88, 95% CI 0.81 to 0.96). There were no differences between tobacco‐only and multifocal interventions. For curricula with booster sessions there was a significant effect only for combined social competence and social influences interventions with follow‐up of one year or less (OR 0.50, 95% CI 0.26 to 0.96) and at longest follow‐up (OR 0.51, 95% CI 0.27 to 0.96). Limited data on gender differences suggested no overall effect, although one study found an effect of multimodal intervention at one year for male students. Sensitivity analyses for Pure Prevention cohorts and Change in Smoking Behaviour over time outcomes suggested that neither selection nor attrition bias affected the results.

Authors' conclusions

Pure Prevention cohorts showed a significant effect at longest follow‐up, with an average 12% reduction in starting smoking compared to the control groups. However, no overall effect was detected at one year or less. The combined social competence and social influences interventions showed a significant effect at one year and at longest follow‐up. Studies that deployed a social influences programme showed no overall effect at any time point; multimodal interventions and those with an information‐only approach were similarly ineffective.

Studies reporting Change in Smoking Behaviour over time did not show an overall effect, but at an intervention level there were positive findings for social competence and combined social competence and social influences interventions.

Plain language summary

Can programmes delivered in school prevent young people from starting to smoke?

Increasing numbers of young people are smoking in developing and poorer countries. Programmes to prevent them starting to smoke have been delivered in schools over the past 40 years. We wanted to find out if they are effective.

We identified 49 randomised controlled trials (over 140,000 school children) of interventions aiming to prevent children who had never smoked from becoming smokers. At longer than one year, there was a significant effect of the interventions in preventing young people from starting smoking. Programmes that used a social competence approach and those that combined a social competence with a social influence approach were found to be more effective than other programmes. However, at one year or less there was no overall effect, except for programmes which taught young people to be socially competent and to resist social influences.

A smaller group of trials reported on the smoking status of all people in the class, whether or not they smoked at the start of the study. In these trials with follow‐up of one year or less there was an overall small but significant effect favouring the controls. This continued after a year; for trials with follow‐up longer than one year, those in the intervention groups smoked more than those in the control groups.

When trials at low risk of bias from randomisation, or from losing participants, were examined separately, the conclusions remained the same. Programmes led by adults may be more effective than those led by young people. There is no evidence that delivering extra sessions makes the intervention more effective.

Background

Children and adolescents in all cultures smoke, with increasing rates in many developing countries. Starting smoking usually leads to the behaviour lasting decades, with great difficulty in quitting. Few studies verify smoking by biochemical tests, and self reported rates probably underestimate true rates. Smoking uptake is associated with existing smoking by family and friends, and with risk‐taking behaviours. Researchers have implemented programmes to counteract these influences. Programmes in schools have evolved over four decades and include those providing information about smoking rates and harms from smoking; teaching children how to be more socially competent to avoid starting smoking; teaching skills to refuse offered tobacco and multimodal programmes with parents, teachers, and the community.

The incidence and prevalence of smoking among children and adolescents

Tobacco use is the main preventable cause of death and disease worldwide, and the five million deaths annually attributable to tobacco use are predicted to increase to eight million annually by 2030 (Warren 2009). Of the US population who were 17 or younger in 1995, it was estimated that five million would die prematurely of tobacco‐related causes, and that 20% of deaths could be avoided if smokers had either never started or had quit (Epstein 2000b).

The World Health Organization (WHO) 'Health behaviour in school‐aged children 1997‐8' survey of 11, 13 and 15 year olds in 29 countries (Europe, Canada and the USA) found that for the 15 year olds in 14 countries more than 20% of females, and in 11 countries more than 20% of males smoked daily (WHO 2000). Surveys of the smoking behaviour of 13 to 15 year olds were then conducted between 1998 and 2008 in all six WHO world regions with 100 initial, 100 second and nine third surveys involving 530,849 students. In 191 of the 209 surveys, more than 90% of the schools participated, and in 190 of 209 surveys, student participation was greater than 80%. The prevalence of both cigarette smoking and other forms of smoking such as water pipes, were both defined as at least monthly (Warren 2009).

For the 100 sites with follow‐up surveys, there were increases in the prevalence of smoking cigarettes at least one day per month at 27 sites and decreases at 10, and for other tobacco products (such as water pipes) at least one day per month there were increases at 33 sites and decreases at 13 (Warren 2009). Therefore, if poorer countries follow the trajectory of the more affluent countries, it is to be expected that 20% to 30% of 13 to 15 year olds may smoke, depending on the culture of the country and the activities of the tobacco companies.

Adolescent smoking remains a risk factor in adulthood. The 1995 US National College Health Risk Behavior Survey found that 70% had ever tried smoking a cigarette, and of these 42% were current smokers and 13% current daily smokers. Females were more likely to smoke than males (Pletcher 2000). Adolescents who begin smoking at younger ages are more likely to become regular smokers and less likely to quit (Tyas 1998). Of concern is the finding that the first use of tobacco after age 18 in the USA increased from 25% to 40% between 2002 and 2009 (SAMHSA 2009).

Villanti 2010 identified five types of smoking behaviour as adolescents become young adults: nonsmokers, early stable smokers, late starters, quitters, and 'light or intermittent smokers'. In adulthood, the early stable and late starter groups had the highest rates of smoking, but the light or intermittent smokers could go either way, and after two years had either temporarily quit or had become heavy smokers.

School‐based interventions

Over the past three decades the school environment has been a particular focus of efforts to influence youth smoking behaviour. The main perceived advantages are that almost all children can be reached through schools, and a focus on education fits naturally with the daily activities of schools. Researchers have used five types of interventions in schools, each based on a different theoretical orientation:

1. Information only curricula

Interventions that provide information to oppose tobacco use (also called normative education) are described by Griffin 2010 as "content and activities to correct inaccurate perceptions regarding the high prevalence of substance use." Griffin describes how many adolescents overestimate smoking prevalence and view smoking as normative behaviour. Normative curricula seek to inform students on actual rates of use and undermine inaccurate beliefs on the social acceptability of smoking. Normative materials are often used by programme deliverers in social resistance programmes. The assumption is that information alone will lead to changes in behaviour (Bangert‐Drowns 1988).

2. Social competence curricula

A group of interventions that aim to help adolescents refuse offers to smoke by improving their general social competence. Griffin 2010 recognises that poor personal and social skills can lead to development of drug use. Therefore, programmes benefit from including social learning processes or life skills such as problem‐solving and decision‐making, cognitive skills for resisting interpersonal or media influences, increased self control and self esteem, coping strategies for stress, and general social and assertive skills. These skills will also have broader applications for the students. The interventions are based on Bandura's social learning theory (Bandura 1977), which hypothesises that children learn drug use by modelling, imitation, and reinforcement, influenced by the child's pro‐drug cognitions, attitudes and skills. Susceptibility is increased by poor personal and social skills and a poor personal self concept (Botvin 2000).

3. Social influence curricula

Interventions that aim to overcome social influences promoting tobacco use by providing skills to adolescents (also called social skills interventions). Griffin 2010 describes these interventions as aiming to increase the "adolescents’ awareness of the various social influences that support substance use." Programmes adopt resistance skills training in which students are taught how to deal with peer pressure, high risk situations, how to effectively refuse attempts to persuade substance use from both direct and indirect sources. The interventions are based on McGuire's persuasive communications theory and Evans's theory of psychological inoculation (McGuire 1968; Evans 1976).

4. Combined social competence and social influences curricula

Methods that draw on both social competence and social influence approaches.

5. Multimodal programmes

These programmes combine curricular approaches with wider initiatives within and beyond the school, including programmes for parents, schools, or communities and initiatives to change school policies about tobacco, or state policies about the taxation, sale, availability and use of tobacco.

Why it is important to do this review

Tobacco education curricula are widely used in US schools, though few of those in use have been rigorously evaluated. The US 2000 National Youth Tobacco Survey national sample of 35,828 6th‐ to12th‐graders in 324 schools found that 70% of the middle schoolers and 50% of the high schoolers said they had received a programme that taught them the short‐term consequences of tobacco use. The percentages for receiving a normative programme were 40% and 18%; for programmes teaching why people smoke 64% and 38%; for programmes teaching refusal skills 51% and 17%; and for multi‐strategy programmes 38% and 17% (Wenter 2002). Wiehe 2005 identified eight programmes that followed participants to age 18 or the 12th grade and found little or no evidence of effectiveness. There is nevertheless continued uncertainty about both the relative and absolute effectiveness of school‐based programmes, and considerable variation in the extent to which they are implemented in other countries.

This review is important because there is no other systematic review of world literature on school‐based smoking prevention programmes without language or date restrictions. This review was first published in 2002. This update has refined how the included studies are categorised to provide analysis based on Pure Prevention cohort studies, Change in Smoking Behaviour over time studies and Point Prevalence of Smoking studies.

Objectives

The primary objective of this review is to assess the effectiveness of school‐based programmes in preventing children and adolescents from starting smoking. A secondary objective is to assess which programme elements, if any, are associated with effectiveness.

We considered one central question:

Are school programmes, categorised by intervention type, more effective than minimal or no intervention in preventing smoking? We considered the hypothesis that they are more effective separately according to the theoretical orientation of the prevention programme:

Information giving
Social competence
Social influence
Combined social influence and social competence
Multimodal programmes

If the review showed the effectiveness of one or more of these types of intervention, we proceeded to the secondary objective, i.e. to examine the direct evidence comparing different types of intervention, categorised by theoretical orientation, including:

Social influences versus information giving
Social influences versus social competence
Combinations of social influences, social competence and information versus single component interventions
Multimodal programmes versus single component interventions

We also aimed to consider the effect by gender and the method of programme delivery, including:

Peer‐led programmes versus those taught by researchers or teachers
Booster sessions after programme completion versus no booster
Tobacco‐focused interventions versus interventions focused on tobacco together with other substances such as alcohol and drugs

Methods

Criteria for considering studies for this review

Types of studies

We included studies in which individual students, classes, schools, or school districts were randomised to receive different programmes or to be the control, and in which baseline tobacco use was measured. We excluded studies if they did not state that allocation of individuals or groups to intervention and control groups was randomised. Random allocation of intervention was either to the individual or to individuals in clusters (in classes, in schools, in classes nested within schools, or in school districts). We assessed whether the studies were analysed using methods appropriate to the level of allocation and the level of measurement of the outcomes. No studies were excluded on the basis of publication status or language of publication.

Types of participants

Children (aged 5 to 12) and adolescents (aged 13 to 18) in school settings. We also included studies in which the participants were 5 to 18 during the intervention phase of the study, but were followed up in a few instances beyond 18.

Types of interventions

We included all school‐based programmes that had as one of their goals preventing tobacco use, irrespective of theoretical intervention. Some programmes aimed simply to provide information about tobacco. Others had more complex goals: teaching generic social skills to reinforce societal norms about individual behaviour; reinforcing the adolescent's self concept; and teaching social skills and specific tobacco refusal skills. Some focused on multiple addictions, and we included any programmes with any drug or alcohol focus provided outcomes for tobacco use were reported. Some focused on 'healthy schools.' We included these provided outcomes for tobacco use were reported. We classified programmes according to their dominant theoretical orientation and then allocated them to one of the five categories described in the Background section or to a sixth category, 'other'. Programmes that solely provided information were placed in the information only category, while recognising that all curricula provided information to participants.

For each study we determined whether the intervention programmes were compared with a control group, and whether the control group received no intervention, or the standard health education curriculum taught in the school, or the tobacco education curriculum in normal use in the school.

There were no restrictions on who delivered the intervention. These could include researchers, classroom teachers, health science teachers, healthcare professionals, undergraduate or graduate students, adolescent peers, or other personnel.

Types of outcome measures

The primary outcome was the effect of the intervention on the smoking status of individuals or cohorts who reported no use of tobacco at baseline. We recorded whether effects of the interventions were found at the conclusion of the programme, and whether such effects were sustained at follow‐up after completion of the programme. We required a minimum follow‐up of six months after the intervention.

We did not require biochemical validation (by saliva thiocyanate or cotinine or expired air carbon monoxide levels) of self reported tobacco use for inclusion, but recorded its use. If saliva samples were collected but not analysed (sometimes described as the 'bogus pipeline' procedure), this was recorded.

One problem in this field is that the studies often use different measures of tobacco use, either recording frequency (monthly, weekly, daily), or the number of cigarettes smoked, or an index constructed from multiple measures. Sometimes the variety of measures is intended to record the fact that young children begin smoking on a monthly basis, but as they get older may proceed to weekly and daily smoking. We excluded studies which did not report any measure of smoking behaviour, studies that did not assess baseline smoking status in the pre‐test survey, and studies that reported only changes in knowledge or attitudes about smoking.

Search methods for identification of studies

We searched the following databases using search strategies similar to those used in MEDLINE for each. Detailed search strategies are displayed in Appendix 1 (MEDLINE) and Appendix 2 (CINAHL):

MEDLINE 1966 ‐ 10/2012
EMBASE 1974 ‐ 10/2012
CINAHL ‐ 10/2012
PsycINFO 1967 ‐ 10/2005
ERIC 1982 ‐ 10/2005
Health Star
Tobacco Control 1992 ‐ 2005
Journal of Smoking Related Disorders 1990 ‐ 2005
Dissertation Abstracts 1960 ‐ [Search strategy = (Tobacco or smoking) and prevent? and (child or adolescent)]
US Department of Health Reviews
Proceedings of the World Conferences on Tobacco and Health
Cochrane Tobacco Addiction Review Group Specialised Register 10/2012
Reference lists of the articles selected in the above sources
Index of Scientific and Technical Proceedings
Conference Papers Index

In addition, we searched MEDLINE from 1966 to October 2012 for 133 individual authors who had published in the field. We also screened the reference lists of the included studies.   None of the previous meta‐analyses of the literature (listed in the additional references below) undertook a Cochrane search strategy.   The most recent searches were conducted in October 2012.

Data collection and analysis

Selection of studies

Two authors (RET and JM) independently assessed the search results for studies that met the inclusion criteria. Reference lists were checked for further relevant studies. The full text of each study was independently assessed, and the authors contacted for clarification in cases of uncertainty.

Data extraction and management

Two authors (RET and JM) independently extracted data, with disagreements resolved by recourse to co‐author RP. We categorised studies into six groups corresponding to the type of intervention (information; social competence; social influence; combined social competence and social influence; multi‐modal, and other). Information extracted included country of study, intervention focus, description of participants (numbers of participants, classes and schools, age, gender, ethnicity, existing smoking status), description of intervention (duration, nature, deliverer, outcome, follow‐up), quality of delivery, and statistical methods.

Assessment of risk of bias in included studies

Two authors independently assessed five aspects of risk of bias, with adjudication in case of disagreement by a third author. Each potential risk of bias was assessed to be either at low or at unclear risk (if no data were provided which could be judged to assess bias), or at high risk (study design or execution could cause over‐ or underestimation of the intervention effect). We contacted authors to verify any risk of bias information not presented in their publications.

Sequence generation (selection bias)
Allocation concealment (selection bias)
Blinding of outcome assessment (detection bias), which was assessed as unclear unless a specific reference was made to blinding of outcome assessors.
Incomplete outcome data (attrition bias) due to absence of some data for individuals or loss of all data for an individual after a certain time. We examined studies for systematic differences in the rate of loss to follow‐up among different groups. Where there was differential attrition between groups, we considered bias was more likely if there was no sensitivity analysis of the effect of this attrition on outcomes.
Selective reporting (reporting bias) due to authors either (1) not reporting all outcomes as determined by the objectives stated initially in their study protocol, or previous publications about the study or within the current publication, or (2) reporting only a subset of outcomes with significant results.

Data synthesis

We identified three groups of studies:

Pure Prevention cohorts (Group 1): Cohorts in which never‐smokers at baseline were followed and the number remaining never‐smokers at the various follow‐up intervals was ascertained. Where authors did not report these data we either computed them from the published articles or we contacted authors and requested that they compute these data. We obtained absolute numbers or odds ratios from individual randomised trials with the control group as comparator. Where the authors used a denominator which did not include all the participants originally randomised (e.g. a sample which the author described as the 'analysis sample,' which excluded drop‐outs and thus had smaller numbers at follow‐up), we recomputed the data using the numbers originally randomised. We calculated adjusted odds ratios based on the number of never‐smokers at specific time points. Adjustment was made for clustering by school/group based on either reported or estimated intraclass correlation coefficients (ICCs) and cluster sizes to determine design effects for each of the intervention groups. We then used this design effect to determine the effective sample size for each intervention group. We obtained a pooled estimate of the effect using the generalised inverse variance method and a fixed‐effect model. We conducted subanalyses for Group 1 based on gender, peer‐led (or substantially peer‐led) versus adult‐led studies, tobacco as the sole focus of the intervention versus multifocal interventions, and interventions that had subsequent booster sessions versus those with none.

Change in Smoking Behaviour over time (Group 2): Studies where the smoking behaviour was measured as change over time. These studies included those with growth curve analysis. We extracted summary measures for the change in smoking status/use from each study in this group. These were reported either for each study group (mean change or ß‐coefficient of change over time plus their associated standard error by study arm) or as an overall change measure attributed to the intervention (odds ratio (OR) with 95% confidence interval (CI), ß‐coefficients of linear change and associated standard error; one per study comparison). When overall effects were reported as ORs and 95% CI we transformed these into standard mean difference (SMD) by multiplying by √3/∏ = 0.5513 as recommended in the Cochrane Handbook (9.4.6 Combining dichotomous and continuous outcomes).

Point Prevalence of Smoking (Group 3): Studies reporting smoking prevalence at baseline and follow‐ups. Individuals were not followed individually to the follow‐up points, and thus the prevalence rates at baseline and follow‐up are cross‐sectional data. Measures included mean usage (indices and ever‐use), percentage in the past week, past month, lifetime usage, percentage smoker and percentage never‐ or nonsmoker. We calculated a summary measure by comparing the difference in smoking prevalence from baseline to follow‐up between the two arms. We obtained the standard error by estimating the correlation of smoking status from data available from a small selection of Group 1 studies, and using the total number of clusters as a proxy for sample size in each group.

For both Change in Smoking Behaviour over time (Group 2) and Point Prevalence of Smoking (Group 3) studies:

We extracted the most conservative smoking outcome, i.e. the lowest usage of smoking (ever‐smoked, and if not available then monthly smoking).
We contacted all authors (after 1995), asking them to identify a cohort of baseline never‐smokers, which would allow the study to be included in the Pure Prevention cohorts (Group 1)
We obtained a pooled estimate of the effects using the generalised inverse variance method and a standardised mean difference.

If a study provided data that were applicable to more than one group of studies, then the data were accepted for both groups. Data from all three groups were not pooled, but were analysed separately throughout the review.

The three groups (Pure Prevention cohort, Change in Smoking Behaviour over time, Point Prevalence of Smoking) were each analysed as an entire group, and then by the intervention used (information; social competence; social influences; combined social competence and social influences; multimodal; other interventions). Studies in the 'other interventions' group were sufficiently different from each other that, although they were presented within the meta‐analysis for the entire group, it would not be appropriate to combine them as a subgroup by intervention within the Results and Discussion sections.

For all groups, study results were analysed by outcomes of one year or less, and then by longest available follow‐up point. The raw data are tabulated in Appendix 3; Appendix 4; Appendix 5.

Where a study compared more than one intervention arm the control group was split equally between them for both outcome events and sample size. The additional intervention arms within the study were added to the review with a text link to the first.

All RCTs were cluster‐randomised trials (C‐RCTs), except for one trial (Werch 2005), and calculations to allow for the effects of clustering using intraclass correlation coefficients (ICCs) were either made by the study authors or were applied by the review authors.

All studies included in the review were assessed and placed into one of the three analysis groups above (Pure Prevention cohorts, Change in Smoking Behaviour over time, and Point Prevalence of Smoking). Studies were included in the review but excluded from the analysis if, once allocated to one of the three analysis groups, it was established that data were missing from studies, such as no baseline and follow‐up numbers, no control arm data, or the review authors were unable to reconcile the data. In these instances we contacted the study author. If there was no response or data were no longer available for these studies then it was not possible to include the studies in the analysis. In some instances if data were available, but only the total number of schools or classes was known and not the numbers allocated to each arm, then the number of schools or classes was estimated based on the proportion of individuals within the group.

Results are presented as: descriptive text, tables and forest plots (pooled data).

Subgroup analysis and investigation of heterogeneity

We used the I² statistics to assess inconsistency across studies and provide a measure of heterogeneity (Higgins 2003). Thresholds for interpretation of heterogeneity were adopted as outlined in the Cochrane Handbook : 0% to 40% ‐ low, 30% to 60% ‐ moderate, 50% to 90% ‐ possible substantial, 75% to 100% ‐ considerable heterogeneity. Where the heterogeneity was deemed to be considerable we did not pool the results and provided a narrative assessment instead.

We conducted subgroup analyses by theoretical approach in all three groups (Pure Prevention cohorts, Change in Smoking Behaviour over time, and Point Prevalence of Smoking). We completed further subgroup analyses on Pure Prevention cohorts only (Group 1). This group was selected for additional subgroup analyses because these studies followed individual baseline never‐smokers through to follow‐up, and were expected to provide both the clearest indication of intervention effects and to have the lowest heterogeneity between studies. These analyses examined differences by gender, peer‐led versus adult‐led interventions, interventions focusing solely on tobacco versus interventions covering multiple areas, and the effects of adding booster sessions.

Sensitivity analysis

We conducted sensitivity analyses for all groups, to compare the overall study results against those studies with low or unclear risk of bias from attrition. We also viewed only those studies at low risk of bias from sequence generation, to assess whether the quality of randomisation had any impact on the overall results. We did not conduct sensitivity analyses for selective reporting, since all studies were assessed to be at low risk of bias, except for five studies which were rated as being at unclear risk and were not included in any analysis because of lack of data.

Results

Description of studies

Full details of all the trials are given in the Characteristics of included studies, Characteristics of excluded studies, and Characteristics of ongoing studies tables. Each study is identified by the name of the first author and year of publication of the main results paper. Additional references are listed together with this main publication under the study ID.

Included studies

The Characteristics of included studies table provides detail on each of the included studies. Overall, 133 cluster‐randomised controlled trials (C‐RCTs) and one RCT, giving a total of 200 arms and involving 428,293 participants from 25 different countries were included and placed in three groups (Note: ‘arms’ refers to different intervention groups within the RCTs, see Figure 1 and Appendix 6):

Flow chart of retrieval and identification of Group 1, 2 and 3 studies.

Pure Prevention cohorts (Group 1): This group included 56 trials with 184,467 participants. Of these, 49 trials (73 arms) with 142,447 participants from 19 different countries provided analysable data. Twenty‐six were from the USA, four each from the Netherlands and the UK, three each from Canada, Germany and Italy, two each from China and Spain, and one each from Austria, Australia, Belgium, Czech Republic, Denmark, Finland, Greece, Portugal, South Africa, Sweden and Thailand. (N.B. Faggiano 2008 provided the comprehensive write‐up of results for a study set in Austria, Belgium, Germany, Greece, Italy, Spain and Sweden). See Appendix 7 for a list of Group 1 studies by country.

Change in Smoking Behaviour over time (Group 2): Studies which provided change data. This included 16 trials with 57,577 participants, of which 15 trials (27 arms) with 45,555 participants provided analysable data. These studies came from three countries: 12 from the USA, two from India, one from Canada. See Appendix 8 for a list of studies by country.

Point Prevalence of Smoking (Group 3): Studies which provided point prevalence data. This included one RCT and 65 C‐RCTs with 208,518 participants, of which one RCT and 24 C‐RCTs (39 arms) with 110,016 participants from 11 different countries provided usable data. Twelve were from the USA, two each from Australia, the Netherlands and the UK, and one each from France, Germany, India, Mexico, Norway, Romania and Sweden. See Appendix 9 for a list of studies by country. The only three studies (four arms) with intention‐to‐treat analysis are also in this group (McCambridge 2011; Sloboda 2009; Spoth 2002 (LST); Spoth 2002 (LST + SFP)).

Four studies (six arms) provided data to more than one group: Spoth 2001 (ISFP); Spoth 2001 (PDFY) to Pure Prevention cohorts (Group 1) and Change in Smoking Behaviour over time (Group 2), Ringwalt 2009a and Spoth 2002 (LST); Spoth 2002 (LST + SFP) to Pure Prevention cohorts (Group 1) and Point Prevalence of Smoking (Group 3), and Perry 2009 to Change in Smoking Behaviour over time (Group 2) and Point Prevalence of Smoking (Group 3). This is reflected in the total participant numbers and total trial numbers stated being reduced to take account of their multiple contributions.

Forty‐nine studies with 152,544 participants were not analysable for a variety of reasons:   (1) the publications did not provide data or only incomplete data on smoking status in the intervention and control groups at either baseline or follow‐ups;   (2) Numbers for intervention and control groups were not provided;   (3) the data were in an unusable format;   (4) the data were judged to be unreliable on closer scrutiny;   (5) the authors were not contactable to provide additional data;   (6) the authors were not able to provide these missing data.

Some studies focused on tobacco alone, and others on tobacco, alcohol, drugs, violence, cardiac health or policy change. The range of interventions was also heterogeneous. They included information about:

Short‐ and long‐term consequences of smoking;
Prevalence of smoking;
Generic social skills;
Tobacco‐, alcohol‐ and drug‐refusal skills;
Interventions about tobacco included with interventions about risk‐taking, violence and carrying weapons;
School interventions associated also with family and community interventions;
Interventions to change school and state policies about tobacco availability;
Classroom management and reading strategies for teachers;
Culturally sensitive programmes, for example programmes for native North Americans.

The educational techniques were varied, and included lectures, quizzes, skits, collages, puppet plays, debates, role‐plays, making videos, discussions of videotaped role‐plays, films, interactive internet programmes, and meetings with athletes. Some studies compared interventions without a control group, and some included a control group in their comparisons. Some compared different types of presenters (teachers versus peers), and some compared videotaped to lecture presentations.

The presenters were usually the classroom teachers, but also included researchers, health educators, science teachers, undergraduate and graduate students, community members, uniformed police, and same‐age and older peers. The trials identified in this review are also heterogeneous in terms of duration of intervention (one hour to 36 classes spread over three years), and time from completion of intervention to final follow‐up (six months to 12 years).

The outcome measures most frequently chosen by authors were never‐smoking, and lifetime, monthly, weekly or daily smoking. Some studies used Pechacek's (Pechacek 1984) or Botvin's (Botvin 1980; Botvin 1984) composite indices, or constructed their own. Some studies classified students as current nonsmokers (which included never‐smokers, quitters and sometimes experimenters), and this heterogeneous category was the most difficult to assess. The authors were therefore contacted for clarification and/or new data sets. Few studies biochemically confirmed self reports at all stages of the research.

Excluded studies

Two hundred and two studies are excluded from the review. The majority (114) are not randomised controlled trials. Other reasons are that the intervention(s) was not in schools (N = 14), follow‐up was less than six months (N = 27), there were no smoking outcomes (N = 34), there were no baseline data (N = 2), the study was outside the age limits (N = 6), the study goal was smoking cessation only and did not include prevention, or there was no intervention (N = 5). These studies are listed in the Characteristics of excluded studies table, because the title and/or abstract had appeared to be relevance to this review.

Ongoing studies

Six studies are classified as ongoing. In four, some details and data are known from the studies, but are insufficient at this time to confirm inclusion in the review. The remaining two are expected to be included in a future update of the review, but the full results are currently awaiting publication. All six are listed in the Characteristics of ongoing studies table.

Risk of bias in included studies

(See Figure 2)

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies, whether or not they had analysable data. The non‐coloured section of each bar represents different arms of multiple‐arm studies, for which risk of bias is assessed as a single measure for each study.

Selection bias: For the randomised control trials with analysable data, selection bias was assessed at low risk of bias in approximately half of the studies, and at unclear risk in almost all the remaining studies. Within the group of studies without analysable data, 12% were at low risk and 84% at unclear risk. The key area of uncertainty came from authors who mentioned only the word 'randomly', which resulted in a judgement of 'unclear.'

For the trials with analysable data, allocation concealment was assessed as being at unclear risk of bias in almost 95% of the studies. Those with no analysable data were either at unclear or at high risk of bias. This was predominantly because there was no comment in the study about allocation concealment. The Cochrane Handbook notes that:

"Cluster‐randomized trials often randomise all clusters at once, so lack of concealment of an allocation sequence should not usually be an issue. However, because small numbers of clusters are randomised, there is a possibility of chance baseline imbalance between the randomised groups, in terms of either the clusters or the individuals. Although not a form of bias as such, the risk of baseline differences can be reduced by using stratified or pair‐matched randomisation of clusters. Reporting of the baseline comparability of clusters, or statistical adjustment for baseline characteristics, can help reduce concern about the effects of baseline imbalance."

For each C‐RCT we verified, where possible, (1) if all clusters were randomised at the same time, (2) if samples were stratified on variables likely to influence tobacco‐use outcomes, (3) if clusters were pair‐matched, and (4) if there was baseline comparability between the intervention and control groups. Of the C‐RCTs with analysable data, 63% used pair matching and/or stratification.

Blinding: This was assessed as at unclear risk or unstated in almost all studies. Wood 2008, for 146 meta‐analyses involving 1346 trials, found that in trials with subjective outcomes, estimates of effect were exaggerated when there was unclear or inadequate concealment (ratio of odds ratios (ORs) 0.69, 95% confidence interval (CI) 0.59 to 0.82) and lack of blinding (ratio of ORs 0.75, 95% CI 0.61 to 0.93) but not in trials with objective outcomes. The outcomes in the studies in this review are objective smoking outcomes presented subjectively by adolescents. As Adams 2008 has shown, when adolescents' reports are objectively verified biochemically or they are asked to write their name on the questionnaire, their reports of weekly or monthly smoking rates significantly increase.

In this review, in most studies students were promised anonymity as they completed their questionnaires, but would most likely have known which study arm they were in, so that blinding was not feasible. In most studies the interventions were presented by classroom teachers, so that blinding of presenters was not possible. We cannot predict whether these factors would have increased or decreased the reporting of smoking rates.

Attrition bias: Across all study groups and also for those studies without analysable data, the percentage of studies assessed as being at low risk of attrition bias ranged from 40% to 50%, those at unclear risk from 40% to 58%, and those at high risk from 13% to 21%. There is no really satisfactory solution for missing data (Altman 2007). Patients excluded after randomisation are unlikely to be representative of those remaining (Nűesch 2009). The Cochrane Handbook advises mapping any methods for handling missing data closely to the known characteristics of the datasets, and to other datasets in the literature that are likely to have comparable outcomes. Adolescents who smoke may quit and re‐try, but are most likely to increase their frequency over time. There is thus some parallel with studies which tend to have worsening outcomes over time, such as lung cancer. Intention‐to‐treat solutions, such as baseline observation carried forward (BOCF), last observation carried forward (LOCF), and complete case analysis (excluding participants with incomplete outcome data) are therefore inappropriate because they require that the mechanisms governing drop‐out are independent of future unobserved measurements (Molenberghs 2004; Kenward 2009). Such independence is unlikely in this review because those who drop out are known to be more likely to be smokers and to have personal, family, friendship, social and cultural factors that promote smoking. Therefore, we did not replace missing data with our own estimates.

Selective reporting: For the trials with analysable data, the risk of bias from selective reporting was low for all the RCTs in Groups 1, 2 and 3, and for 90% of the studies which provided no analysable data.

Effects of interventions

Studies were classified into three groups according to how authors presented their data: Group 1 (Pure Prevention cohorts), Group 2 (Change in Smoking Behaviour over time), and Group 3 (Point Prevalence of Smoking). We contacted authors in Groups 2 and 3 and invited them to recompute their data to provide datasets of baseline never‐smokers; if they were unable to comply or did not reply we computed such datasets where we could. These results were then further analysed by duration of follow‐up and intervention category.

GROUP 1: PURE PREVENTION COHORT (49 C‐RCTs, 73 arms)

Comparison of all intervention curricula versus control, with duration of follow‐up of one year or less (See Analysis 1.1):   When the outcomes for all the trials testing any of the five different intervention curricula were pooled there was no overall effect (odds ratio (OR) 0.94, 95% confidence interval (CI) 0.85 to 1.05; I² = 0%; Analysis 1.1). The I² statistic for subgroup differences across all interventions was 44.1%, but within each intervention category heterogeneity was minimal.   One small trial (Howard 1996) which tested an information curriculum found no effect.   The combined social competence and social influences curricula (six RCTs/seven arms) showed a statistically significant effect in preventing the onset of smoking (OR 0.49, 95% CI 0.28 to 0.87; P = 0.01; I² = 0%; Analysis 1.1.3). However, for the social influences curricula (16 RCTs/25 arms) (OR 1.00, 95% CI 0.88 to 1.13; I² = 0%; Analysis 1.1.2) and the multimodal curricula (three RCTs/five arms) (OR 0.89, 95% CI 0.73 to 1.08; I² = 50%; Analysis 1.1.4), the results were not significant, with the 95% confidence interval including the line of no effect (= 1).   There was no RCT testing a social competence curriculum versus control with a follow‐up duration of less than one year.   One study with two arms, Figa‐Talamanca 1989 (F); Figa‐Talamanca 1989 (N.F), was included in the overall effect, but the intervention used did not fit into one of the five main intervention categories.

1.1 — Comparison 1 All curricula versus control, Outcome 1 Group 1: Pure Prevention cohort (adjusted) ‐ 1 year or less.

Sensitivity analyses:   Sensitivity analyses restricted to studies at low risk of bias in Group 1 found no differences from the all‐trials versions, apart from the trials of social competence and social influences curricula, which no longer demonstrated a significant effect, i.e. the all‐trials OR was 0.49 (95% CI 0.28 to 0.87), compared with the low risk of bias trial OR of 0.55 (95% CI 0.28 to 1.09; Analysis 2.1.3).

Comparison of all curricula versus control, with longest follow‐up period: [See Analysis 1.2]

1.2 — Comparison 1 All curricula versus control, Outcome 2 Group 1: Pure Prevention cohort (adjusted) ‐ longest follow‐up.

When the outcomes for all the trials testing any of the five different intervention curricula were pooled there was a significant effect favouring the intervention (OR 0.88, 95% CI 0.82 to 0.96; P = 0.002; I² = 0%), with a mean risk reduction of 12%. (See Figure 3): Heterogeneity was 0%, except for the multimodal curricula trials (I² = 50%).   One C‐RCT testing information curricula detected a nonsignificant effect (OR 0.12, 95% CI 0.00 to 14.87; P not applicable).

Forest plot of comparison: 1 All curricula versus control, outcome: 1.2 Group 1: Pure Prevention cohorts (adjusted) at longest follow‐up.

Social competence curricula (five C‐RCTs/seven arms) versus control showed a statistically significant result in favour of the intervention (OR 0.52, 95 % CI 0.30 to 0.88; P = 0.02; I² = 0%; Analysis 1.2.2), as also did the combined social competence and social influences versus control (eight C‐RCTs/10 arms), (OR 0.50, 95% CI 0.28 to 0.87; P = 0.01; I² = 0%; Analysis 1.2.4).

There were no statistically significant differences for social influences programmes or multimodal curricula.

Four trials (six arms) contributed to the overall results, but not to the individual curricula (Brown 2002; Figa‐Talamanca 1989 (F); Figa‐Talamanca 1989 (N.F); Johnson 2009; Kellam 1998 (GBG); Kellam 1998 (ML)).

Sensitivity analyses:

Sensitivity analyses restricted to trials at low risk of selection bias demonstrated no differences from the all‐trials findings. Ranking by risk of attrition bias made little difference to the findings, apart from a widening of the confidence interval to include the line of no effect, i.e. all‐trials OR 0.88 (95% CI 0.82 to 0.96) versus low risk of bias OR 0.89 (95% CI 0.78 to 1.02).

GROUP 2: CHANGE IN SMOKING BEHAVIOUR OVER TIME (15 C‐RCTs, 27 arms)

Comparison of all curricula versus control, with duration of follow‐up of one year or less: (See Analysis 1.3)

1.3 — Comparison 1 All curricula versus control, Outcome 3 Group 2: Change in Smoking Behaviour over time ‐ 1 year or less.

The eight studies (15 arms) demonstrated a small statistically significant effect favouring the control group (standardised mean difference (SMD) 0.04, 95 % CI 0.02 to 0.06; P = 0.00001; I² =27%). This is similar to Pure Prevention cohorts (Group 1) combined social competence and social influences curriculum (only one C‐RCT (one arm)) found a significant effect favouring the intervention (SMD ‐0.38, 95%CI ‐0.59 to ‐0.17; P = 0.0004), but unlike Group 1 social influences curricula found a small statistically significant effect favouring the controls (six C‐RCTs/10 arms) (SMD 0.04, 95% CI 0.03 to 0.06; P = 0.00001; I² = 0%). There were no significant effects for information and social competence curricula.

Sensitivity analyses:   A sensitivity analysis restricted to trials at low risk of attrition bias demonstrated a nonsignificant effect.

Comparison of all curricula versus control, with longest follow‐up period: (See Analysis 1.4)   Fifteen C‐RCTs (27 arms) demonstrated a nonsignificant effect (SMD 0.01, 95% CI ‐0.00 to 0.02; P = 0.18; I² = 57%). Two C‐RCTs (five arms) that tested social competence curricula favoured the intervention (SMD ‐ 0.04, 95% CI ‐0.06 to ‐0.01; P = 0.01; I² = 0%). Ten C‐RCTs (16 arms) testing social influences curricula (SMD 0.05, 95% CI 0.03 to 0.06; P = 0.00001; I² = 0%) favoured the controls. There was no effect for information, combined social competence and social influences or multimodal curricula.

1.4 — Comparison 1 All curricula versus control, Outcome 4 Group 2: Change in Smoking Behaviour over time ‐ longest follow‐up.

Sensitivity analyses:   Sensitivity analyses restricted to trials at low risk of attrition or selection bias demonstrated no important differences from the all‐trials findings.

GROUP 3: POINT PREVALENCE OF SMOKING (25 C‐RCTs, 39 arms): (See Analysis 1.5, Analysis 1.6)

1.5 — Comparison 1 All curricula versus control, Outcome 5 Group 3: Point Prevalence of Smoking ‐ 1 year or less.

1.6 — Comparison 1 All curricula versus control, Outcome 6 Group 3: Point Prevalence of Smoking ‐ longest follow‐up.

The heterogeneity in this group of studies (for all interventions and for both follow‐up durations) was extremely high (minimum I² = 99%) and beyond what would be expected by chance alone. We have, therefore, not pooled these trials, but display them for reference

In the 16 studies (21 arms) that provided data at one year or less, eight out of 21 comparisons significantly favoured the controls (Analysis 1.5). This trend continued through longest follow‐up, with 20 of 25 studies (39 arms) significantly favouring the controls (Analysis 1.6).

Sensitivity analyses restricted to trials at low risk of selection bias or at low and unclear risk of attrition bias had no impact on the results.

Subgroup analyses (Pure Prevention cohort, Group 1 only)

Differences by gender (Analysis 3.1, Analysis 3.3):   At one year for the limited number of studies which presented data by gender, there was both a significant effect (OR 0.69, 95% CI 0.49 to 0.96; P = 0.04; I² = 30%) for females (seven arms), and for males (six arms) (OR 0.66, 95% CI 0.44 to 0.98; P = 0.04; I² = 30%). The largest effect was found in one study (De Vries 2003 (Finland)) which tested a multimodal curriculum (OR 0.32, 95% CI 0.16 to 0.65; P = 0.002) in males.   At longest follow‐up there were no statistically significant differences for females (nine arms) or males (eight arms).

3.1 — Comparison 3 Group 1: Gender analysis, Outcome 1 Female ‐ 1 year or less.

3.3 — Comparison 3 Group 1: Gender analysis, Outcome 3 Male ‐ 1 year or less.

Peer‐ versus adult‐led interventions:   Adult‐led interventions (29 arms) were not shown to be more effective up to one year than controls in any of the programmes, except for combined social competence and social influences curricula (OR 0.46, 95% CI 0.26 to 0.84; P = 0.01; I² = 0%). There was no overall effect for the peer‐led interventions (8 arms) compared to controls, although this only included social influences curricula tested by a single study (Botvin 1982) which offered a combined social competence and social influences curriculum (Analysis 6.1, Analysis 6.3).

6.1 — Comparison 6 Group 1: Peer‐led analysis, Outcome 1 Peer‐led ‐ 1 year or less.

6.3 — Comparison 6 Group 1: Peer‐led analysis, Outcome 3 Adult‐led ‐ 1 year or less.

In contrast, at longest follow‐up there were significant overall effects for adult‐led interventions (56 arms) compared to the control groups (OR 0.88, 95% CI 0.81 to 0.96; P = 0.002; I² = 17%), and significant effects for two of the four curricula tested: social competence (7 arms) (OR 0.52, 95% CI 0.30 to 0.88; P = 0.02, I² = 0%) and combined social competence and social influences (7 arms) (OR 0.47, 95% CI 0.26 to 0.84; P = 0.01, I² = 0%), but not for social influences or multimodal curricula. For peer‐led programmes (11 arms) compared to controls (Analysis 6.2) there were no statistically significant differences overall, nor for the three curricula tested (social influences, combined social competence and social influences and multimodal).

6.2 — Comparison 6 Group 1: Peer‐led analysis, Outcome 2 Peer‐led ‐ longest follow‐up.

Four studies (six arms) which compared peer‐led and adult‐led interventions to controls were not included, either because it was not clear who delivered the programme (Conner 2010 (I); Conner 2010 (SE); Seal 2006) or because it was delivered online (Buller 2008 (Australia); Buller 2008 (USA); Prokhorov 2008).

Interventions focused on tobacco versus interventions covering multiple areas:   When the effectiveness of multifocal curricula (i.e. a combined focus on tobacco, drugs and alcohol prevention) was compared to control there was no overall effect at one year or at longest follow‐up. Only one curriculum, social competence (seven arms), showed a significant effect at longest follow‐up (OR 0.52, 95% CI 0.30 to 0.88; P = 0.02; I² = 0%; Analysis 5.2.2).

Curricula focused only on tobacco use prevention (26 arms) compared to controls showed no effect (OR 0.93, 95% CI 0.83 to 1.04) at one year, although there was an effect at longest follow‐up (42 arms) (OR 0.88, 95% CI 0.80 to 0.97; P = 0.01; I² = 20%; Analysis 5.4). None of the the three curricula tested at one year or at longest follow‐up (social influences, combined social competence and social influences, and multi‐modal) found significant differences.

5.4 — Comparison 5 Group 1: Tobacco focus, Outcome 4 Tobacco focused ‐ longest follow‐up.

Effect of adding booster sessions:   At one year or less there were no significant differences for curricula (36 arms) which did not include booster sessions, compared to controls (OR 0.94, 95% CI 0.85 to 1.05; Analysis 4.1), or at longest follow‐up (66 arms) (OR 0.90, 95% CI 0.83 to 0.97; P = 0.10; I² = 0%; Analysis 4.2).

4.1 — Comparison 4 Group 1: Booster sessions analysis, Outcome 1 No Booster sessions ‐ 1 year or less.

4.2 — Comparison 4 Group 1: Booster sessions analysis, Outcome 2 No Booster sessions ‐ longest follow‐up.

For curricula which included booster sessions, there were no significant differences from controls at one year or less (four arms) (OR 0.70, 95% CI 0.40 to 1.07), but at longest follow‐up (seven arms) there was a significant difference (OR 0.73, 95% CI 0.55 to 0.98; Analysis 4.4).

4.4 — Comparison 4 Group 1: Booster sessions analysis, Outcome 4 Booster sessions ‐ longest follow‐up.

The combined social competence and social influences curricula (OR 0.50, 95% CI 0.26 to 0.96; P = 0.04; I² = 0%) had a positive effect at one year or less (two arms) and also at longest follow‐up (three arms) (OR 0.51, 95% CI 0.27 to 0.96; P = 0.04; I² = 0%).

Discussion

Summary of Main Results   Outcomes are presented for three distinct groups: Pure Prevention cohorts of baseline never‐smokers, studies where authors presented results as Change in Smoking Behaviour over time, and studies where authors presented data as Point Prevalence of Smoking. Only four studies contributed to more than one group.

In the Pure Prevention cohort (Group 1), one might expect the clearest indication of whether smoking interventions prevent smoking, as studies followed the same cohort of never‐smoking individuals from baseline to follow‐ups. This group of cluster‐randomised controlled trials (C‐RCTs) with follow‐up of a year or less demonstrated no overall significant effect, with only the combined social competence and social influences curricula delivering positive results. Pooling the results from all the trials at longest follow‐up favoured the intervention groups (OR 0.88, 95% CI 0.82 to 0.96). This represents a risk reduction of 12% and suggests that interventions were more effective over a longer time period. The only intervention categories within this group that showed a statistically significant result were social competence and combined social competence and social influence curricula. This indicates that the success of the combined social competence and social influence curricula at one year was maintained over a longer period. There were no social competence intervention studies with one year or less of follow‐up for comparison.

Though pooled data suggest a significant effect in favour of the controls on Change in Smoking Behaviour over time (Group 2), the results are not incompatible with those of the Pure Prevention cohort studies (Group 1). Whilst the overall effect marginally favours the controls, there are similarities at intervention programme level to the results from the Pure Prevention cohort studies. This would be expected, since these studies, while measuring a change rate, follow the same groups of participants over time. Higher heterogeneity in this group could be explained by the differences between the participants (never‐smokers, experimenters and quitters) and between outcome measures.

Sensitivity analyses for Pure Prevention cohorts (Group 1) and Change in Smoking Behaviour over time (Group 2) for selection and attrition bias revealed no differences between studies at low risk and those at unclear or high risk.

In the Group 3 studies which present point prevalence smoking data, it was not possible to pool data due to the high level of heterogeneity, though the trends may have favoured the controls. The most likely explanation for the heterogeneity is that the same individuals are not consistently being measured over time, and thus point prevalence data are inadequate to measure the effectiveness of this type of intervention.

Subgroup analyses were only completed for the Pure Prevention cohorts (Group 1) data, and showed that:

Gender: For the few studies that reported results by gender, there were positive significant results for both females and males with one year or less of follow‐up. However, within both groups only one intervention category (multimodal) in one study for males found a positive significant result.

Peer‐led versus adult‐led interventions: There were no significant differences for studies at one year or less for peer‐led compared to adult‐led curricula, except for adult‐led combined social competence and social influences curricula. At longest follow‐up there were significant differences favouring adult‐led curricula, and for adult‐led social competence curricula and adult‐led combined social competence and social influences curricula.

Multifocal versus tobacco‐only interventions: At one year or less there were no differences between multifocal and tobacco‐only programmes. However, at longest follow‐up tobacco‐only curricula had a significant effect, and within multifocal interventions the social competence returned positive findings.

Booster sessions versus no boosters sessions: Major effort has been expended in many studies to provide booster sessions, expecting that they would reinforce the effects of the original programmes. At one year or less the presence or absence of boosters made no difference. Combined social competence and social influences curricula appeared to benefit from booster sessions in the medium and long term. This suggests that curricular orientation may be more important than providing booster sessions.

Overall completeness and applicability of evidence   The number of studies which provided no analysable data is large (49 C‐RCTs with 152,544 students), with seven C‐RCTs (42,020 students) from the Pure Prevention cohorts (Group 1), one C‐RCT (12,022 students) from the Change in Smoking Behaviour over time group (Group 2), and 41 C‐RCTs (98,502 students) from the Point Prevalence of Smoking group (Group 3). Twelve per cent of these trials are at low risk and 84% at unclear risk of selection bias, compared with approximately half at low risk and almost all the remaining studies at unclear risk in the trials with usable data. However, the percentages at low and unclear risk were similar for allocation, blinding, attrition and reporting biases. Our inability to include this large number of C‐RCTs and participants therefore excludes data of lower quality with respect to selection bias. A funnel plot (not shown) did not suggest publication bias in Pure Prevention cohorts or Change in Smoking Behaviour analyses.

Population:   Of the trials which provided analysable data, 56% were from North America (51% from the USA), 35% from Europe, 5% from Asia, 3% from Australia, and 1% from Africa. There is thus minimal representation from four of the six continents. In the US studies there is wide representation of urban and rural, socioeconomic, and ethnic groups. Few studies reported data separately by gender.

Interventions:   We placed no restrictions on the type of intervention that was included, provided it was school‐based. This resulted in a huge variety of interventions, which were analysed in six broad categories. A small number of interventions could not be classified, and although they are included in the overall analysis it was inappropriate to assess them as a separate category.

Social influence curricula were tested more than any other curricula in studies. In the Pure Prevention cohorts group, 63% of intervention arms at one year or less and 67% at longest follow‐up tested social influences interventions. The proportions in the Change in Smoking Behaviour over time group were 67% and 59% respectively. Only in the Change in Smoking Behaviour were social influence curricula found to be significant, and these favoured the controls.

Ideally, the review would have examined the positive effect of social competence or combined social competence and social influences further, by considering studies that made direct comparisons of these intervention types. However, although there were a few studies that explored comparisons between interventions, none of them considered these intervention types.

Outcomes:   The trials deployed a wide variety of outcome measures: never‐smoking; lifetime, monthly, weekly or daily smoking; numbers of cigarettes smoked during each of these time intervals; and indices such as Pechacek's (Pechacek 1984) or Botvin's (Botvin 1980; Botvin 1984). Some studies used the term 'current nonsmokers,' but this can include never‐smokers, experimenters and quitters, which can introduce a lack of clarity into any attempt to follow cohorts. The measures used most frequently are never‐smoking; smoking in the past 30 days and current nonsmoking.

Quality of the Evidence   The main strength of this review is the large number of included studies (134) and the number of participants (428,293). Although a large number of trials (85) with 275,749 participants provided analysable data, a limitation of this review is that 49 trials (152,544 participants; 37% of the total) were eligible, but did not provide sufficient data in their publications or did not provide the data after study authors were contacted. However, the data we could not include are deemed to be at greater risk of selection bias than the usable information.

For the Pure Prevention cohorts (Group 1) trials, it is worth noting that 49 studies (73 arms) with 142,447 participants were included in the analysis, representing 88% of all potential Pure Prevention cohorts trials.

Key methodological problems:   Key problems in some studies are a failure to describe robust methods of randomisation or allocation concealment, high rates of attrition, varying outcome measures for tobacco use, the use of 'current nonsmoker' as an outcome, failure to follow groups of never‐smokers, triers, and quitters separately over time, and failure to report basic data such as the numbers and smoking status in the intervention and control groups at baseline and follow‐ups. Our decision not to pool data from the Point Prevalence of Smoking trials arose from our assessment of point prevalence as an inadequate measure for reporting effects in these types of studies.

Consistency between the Pure Prevention cohorts (Group 1) and Change in Smoking Behaviour over time studies (Group 2) was good, but it was not possible to compare them with the Point Prevalence of Smoking studies (Group 3). Whilst many studies reported inadequately on their randomisation process and on attrition, sensitivity analyses suggest that these potential risks of bias did not have any real effect on the main findings for each group or intervention type.

Potential biases in the review process   One strength of this review is that the search was conducted across multiple electronic data bases, and included 'grey' literature, the searching of reference lists of articles, and consultation with experts. There were no limitations of date or language, and translations were obtained for any article as required. It is unlikely that this extensive search would have missed key trials.

Two authors independently reviewed all titles and abstracts and independently entered all data on Cochrane Tobacco Review Group data extraction forms. Extensive correspondence (over 600 emails) was undertaken with all study authors if data on risks of bias, the planning and conduct of the trial, numbers, stratification and pairing of clusters, baseline equivalence of intervention and control arms, and tobacco outcome status were not provided in the publications. Many study authors computed new databases of baseline never‐smokers for the review, or the reviewers computed this data.

Bias could have been introduced due to the high variability of outcome measures, although this has been reduced by dividing the studies into three groups and analysing the data for each group separately. The low heterogeneity in the Pure Prevention cohorts (Group 1) studies supports this approach. Bias may also have been introduced by certain assumptions made by the study authors in data extraction, and subsequent statistical analysis. This is particularly pertinent in the Point Prevalence of Smoking studies (Group 3), where we considered it inappropriate to pool the data.

Agreements and disagreements with other studies or reviews   There is no other comprehensive review of interventions in schools for comparison.

Authors' conclusions

Implications for practice.

There was a significant effect for the Pure Prevention cohorts studies which followed participants for more than one year, but not for shorter‐term outcomes: combined social competence and social influences interventions at all time points, and social competence interventions at longest follow‐up prevented smoking uptake compared with controls; social influence interventions did not appear to reduce uptake compared with controls.

Studies at low risk of selection and attrition bias did not deliver better results than the full mix of available trials.

Interventions delivered by adult presenters are more effective in the longer term than peer‐led programmes.

Adding booster sessions in subsequent years do not change outcomes.

Implications for research.

Further studies of social competence and combined social competence and social influences programmes could explore the potential of these interventions.

Further research is required to design and test programmes that will be optimally effective for both genders.

Further research is required to identify factors that can be tailored to the requirements of different ethnic groups.

Studies need to follow up participants for more than one year.

Studies should clearly identify and follow separately students in different stages of their smoking career (never‐smokers, experimenters, quitters, smokers of different frequencies and intensities), as composite change rates and point prevalence scores at baseline and follow‐up make the findings difficult to interpret.

Outcome measures should be standardised at trial design stage.

Studies are needed across all cultural areas of the world.

There is minimal information on the costs of designing and implementing these programmes. Economic evaluation is important, in view of the fact that many interventions have not proven their effectiveness.

What's new

Date	Event	Description
1 May 2013	Amended	Minor textual edit in the Discussion section, and the Author's conclusion section. Acknowledgement also added.

Date	Event	Description
22 March 2013	New search has been performed	Updated with 51 new studies. Latest search October 2012.
22 March 2013	New citation required and conclusions have changed	New analysis methods used. New author added. New categories and conclusions
22 June 2011	Amended	Additional table converted to appendix to correct pdf format
18 April 2008	Amended	Converted to new review format.
20 April 2006	New citation required and conclusions have changed	Substantive amendment

Study ID					Control arm			OR	Follow‐up
Study ID					No. lost to never‐smokers	Never‐smokers at baseline	Cluster No.	OR	Follow‐up

One year or less follow‐up

Howard 1996	I	0	51	3 classes	3	47	3 classes		1 yr

Armstrong 1990 (Teacher)	SI	74	358	15	106	339	15		1 yr
Armstrong 1990 (Peer)	SI	96	331	15	106	339	15		1 yr
Ausems 2004 (In school)	SI			9			9 baseline/7@1 yr	0.52 (adj)	1 yr
Ausems 2004 (Out School)	SI			8 baseline/6@1 yr			9 baseline/8@1 yr	0.44 (adj)	1 yr
Aveyard 1999	SI			27			26	1.14 (unadj)	1 yr
Buller 2008 (Australia)	SI	34	608	13	26	605	12		6 mths
Buller 2008 (USA)	SI	41	616	10	11	372	11		6 mths
Chou 2006	SI	142	862	7	175	975	7		1 yr
Coe 1982	SI	8	66	2	16	84	2		1 yr
De Vries 1994 (Voc)	SI	9	109	3	6	75	3		1 yr
De Vries 1994 (High)	SI	26	317	5	19	230	3		1 yr
De Vries 2003 (UK)	SI			22			21	1.06 (adj)	1 yr
Ellickson 1990 (Teen)	SI	527	2253	10	561	2175	10		1 yr
Ellickson 1990 (HealthEd)	SI	506	2099	10	561	2175	10		1 yr
Ennett 1994	SI			18			18	0.93 (adj)	1 yr
Figa‐Talamanca 1989 (F)	SI	10	99	8	1	108	8		1 yr
Figa‐Talamanca 1989 (N.F)	SI	0	88	8	1	108	8		1 yr
Gabrhelik 2012	SI	160	917	40	125	787	34		1 yr
Garcia 2005	SI	7	147	6	18	68	4		1 yr
Nutbeam 1993 (FSE)	SI	362	848	10	325	951	10		1 yr
Nutbeam 1993 (SAM)	SI	263	732	9	325	951	10		1 yr
Nutbeam 1993 (FSE+SAM)	SI	325	924	10	325	951	10		1 yr
Resnicow 2008 (LST)	SI	182	1161	12	226	1097	12		1 yr
Telch 1990 (Peers)	SI	4	117	4	27	199	7		6 mths
Telch 1990 (No peers)	SI	14	115	4	27	199	7		6 mths
Valente 2007 (TND)	SI	3	106	22	1	85	28		1 yr
Valente 2007 (TNDNetwork)	SI	4	113	25	1	85	28		1 yr

Botvin 1980	C	3	79	1	17	108	1		6mths
Botvin 1982	C	26	120	1	32	144	1		1 yr
Botvin 1983 (LST)	C	31	270	2	70	251	3		1 yr
Botvin 1983 (LST intensive)	C	13	170	2	70	251	3		1 yr
Botvin 1999	C	144	1263	29 total	173	912	29 total		1 yr
Resnicow 2008 (Harm Min)	C	126	1392	12	226	1097	12		1 yr
Seal 2006	C	0	52	1	1	59	1		6 mths

De Vries 2003 (Denmark)	MM			30			30	1.41	1 yr
De Vries 2003 (Finland)	MM	185	756	13	248	913	14		1 yr
De Vries 2003 (Portugal)	MM			14			11	0.73	1 yr
Simons‐Morton 2005	MM	333	1249	3	361	1080	4		1 yr
Wen 2010	MM	92	1162	2	89	840	2		1 yr

Longest Follow‐up (over 1 year)

Connell 2007	SC	95	196	3	100	222	3		11 yrs
Kellam 1998 (GBG)	SC	92	348	6	299	904	6		8 yrs
Kellam 1998 (ML)	SC	111	352	7	299	904	6		8 yrs
Spoth 2001 (ISFP)	SC	46	141	11	71	142	11		4 yrs
Spoth 2001 PDFY)	SC	50	128	11	71	142	11		4 yrs
Spoth 2002 (LST)	SC	64	462	12	68	408	12		1.5 yrs
Storr 2002 (CC)	SC	60	230	3	72	219	3		6 yrs
Storr 2002 (FSP)	SC	60	229	3	72	219	3		6 yrs
Walter 1986	SC	16	447	8	61	464	7		6 yrs

Armstrong 1990 (Teacher)	SI	116	358	15	70.5	169.5	7.5		2 yrs
Armstrong 1990 (Peer)	SI	132	331	15	70.5	169.5	7.5		2 yrs
Ausems 2004 (Out school)	SI			7 baseline/5@18 mths			8 baseline/7 @18 mths	0.42 (adj)	18 mths
Aveyard 1999	SI			27			26	1.06 (unadj)	2 yrs
Brown 2002	SI	176	1313	15	183	1201	15		2 yrs
Conner 2010 (I)	SI	65	297	15	104	373	19		2 yrs
Conner 2010 (SE)	SI	82	257	13	115	358	18		2 yrs
Crone 2011	SI	25	1311	62	33	1022	59		19 mths
De Vries 2003 (UK)	SI			22			21	0.94 (adj)	30 mths
Denson 1981	SI	8	256	6	49	272	6		2 yrs
Elder 1996	SI			56			40	1.01 (adj)	3 yrs
Ellickson 1990 (Teen)	SI	651	2253	10	338	1087.5	5		15 mths
Ellickson 1990 (HealthEd)	SI	642	2099	10	338	1087.5	5		15 mths
Ellickson 2003	SI	152	1765	34	191	1171	21		18 mths
Ennett 1994	SI			18			18	0.99 (adj)	2 yrs
Faggiano 2008	SI	245	2939	78	242	2791	65		18 mths
Gabrhelik 2012	SI	262	917	40	235	787	34		2 yrs
Hort 1995	SI	50	268	9	84	239	10		2 yrs
Johnson 2009	SI	381	891	10	459	1116	10		4 yrs
La Torre 2010 (A)	SI	22	135	8	23	119	7		2 yrs
La Torre 2010 (C)	SI	3	197	11	24	240	13		2 yrs
Peterson 2000	SI	1466	3684	20	1547	3756	20		12 yrs
Prokhorov 2008	SI	2	380	9	8	317	8		18 mths
Resnicow 2008 (LST)	SI	182	1161	12	162.5	548.5	6		2 yrs
Ringwalt 2009a	SI	368	2335	17	332	2475	17		3 yrs
Schulze 2006	SI	838	1205	89	596	872	83		18 mths
Unger 2004 (FLAVOR)	SI	194	933	8	115.5	538.5	4		18 mths
Unger 2004 (CHIPS)	SI	201	847	8	115.5	538.5	4		18 mths
Van Lier 2009	SI	52	349	16	51	279	15		4 yrs

Resnicow 2008 (Harm Min)	C	206	1392	12	162.5	548.5	6		2 yrs
Spoth 2002 (LST + SFP)	C	48	385	12	34	204	6		1.5 yrs

De Vries 2003 (Denmark)	MM			30			30	1.15 (adj)	30 mths
De Vries 2003 (Finland)	MM	404	756	13	419	913	14		30 mths
De Vries 2003 (Portugal)	MM			14			11	0.62 (adj)	30 mths
Piper 2000 (HFL)	MM	254	564	7	159.5	359.5	4		4 yrs
Piper 2000 (HFL Age)	MM	385	614	7	159.5	359.5	4		4 yrs
Weichold 2012 (Teacher)	SI & SC	9	45	3	3.5	7.5	0.5		2 yrs
Weichold 2011 (Peer)	SI & SC	5	9	1	3.5	7.5	0.5		2 yrs
Wen 2010	MM	77	571	2	59	449	2		2 yrs

Study ID	Intervention category	Intervention arm			Control arm			OR	Growth rate (SE)	Follow‐up period
Study ID	Intervention category	Baseline smoking measure	Follow‐up smoking measure	Cluster No.	Baseline measure	Follow‐up unit of measure	Cluster No.	OR	Growth rate (SE)	Follow‐up period

One year or less follow‐up

Sun 2008 (Cognitive)	I	19.92% cigarette use in the last 30 days 9 (using scale where 0 = none to 7 = 100+)		6	13.29% cigarette use in the last 30 days 9 (using scale where 0 = none to 7 = 100+)		3	1.35 (0.93,195) (adj)		1yr

Forman 1990 (SI)	SC	2.90 (1.49) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	3.02 (1.48) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	10	2.83 (1.65) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	2.93 (1.53) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	3.33			1 yr
Forman 1990 (SI ‐ NP)	SC	2.84 (1.71) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	2.81 (1.64) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	6	2.83 (1.65) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	2.93 (1.53) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	3.33			1 yr
Forman 1990 (SI ‐ P)	SC	2.81 (1.44) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	2.95 (1.47) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	4	2.83 (1.65) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	2.93 (1.53) mean (SD) Freq Cig. Use (1= never to 5 = everyday)	3.33			1 yr

Clark 2010	SI	1.97 (2.48) mean (SD) Average no. of days smoked in the last 30 days (scale from 0 = 0 to 10 = >38)	2.31 (2.67) mean (SD) Average no. of days smoked in the last 30 days (scale from 0 = 0 to 10 = >38)	7	2.16 (2.58) mean (SD) Average no. of days smoked in the last 30 days (scale from 0 = 0 to 10 = >38)	2.5 (2.70) mean (SD) Average no. of days smoked in the last 30 days (scale from 0 = 0 to 10 = >38)	7			1 yr
Kaufman 1994	SI		11.63 (3.98) mean (SD) for cigarette use (scale from 6 ‐ 32, higher = more use). Pretest smoking as covariate.	2		10.99 (2.51) mean (SD) for cigarette use (scale from 6 ‐ 32, higher = more use) Pretest smoking as covariate.	1			6 mths
Reddy 2002 (School + F)	SI	0.034 (0.0219, 0.0525) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	0.0366 (0.0264, 0.0504) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	10	0.0391 (0.251, 0.0605) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	0.0937 (0.0728, 0.1198) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	5			1 yr
Reddy 2002 (School only)	SI	0.0416 (000269, 0.0637) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	0.0571 (0.0422, 0.0768) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	10	0.0391 (0.251, 0.0605) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	0.0937 (0.0728, 0.1198) Mean (CI) ever tried (from SAS PROC Mixed and SAS Glimmix Macro). Based on yes/no to ever use	5			1 yr
Severson 1991(high, M)	SI	9.4 (48.6) mean cigarettes per month (SD)	24.9 (120.3) mean cigarettes per month (SD)	13 total for all four arms	3.2 (26.7) mean cigarettes per month (SD)	15.9 (83.3 mean cigarettes per month (SD)	13 total for all four arms			1 yr
Severson 1991 (High, F)	SI	5.7 (35.8) mean cigarettes per month (SD)	22.7 (97.1) mean cigarettes per month (SD)	13 total for all four arms	13.9 (72.6) mean cigarettes per month (SD)	17.9 (83.4) mean cigarettes per month (SD)	13 total for all four arms			1 yr
Severson 1991 (Middle, M)	SI	0.7 (5.4) mean cigarettes per month (SD)	9.1 (47.3) mean cigarettes per month (SD)	13 total for all four arms	1.3 (6.9) mean cigarettes per month (SD)	3.4 (23.1) mean cigarettes per month (SD)	13 total for all four arms			1 yr
Severson 1991 (Middle, F)	SI	1.9 (18.2) mean cigarettes per month (SD)	13.6 (59.0) mean cigarettes per month (SD)	13 total for all four arms	1.1 (5.7) mean cigarettes per month (SD)	12.4 (59.0) mean cigarettes per month (SD)	13 total for all four arms			1 yr
Shope 1996	SI	0.12 (0.61) mean (SD) cigarette use	0.27 (0.87) mean (SD) cigarette use	Estimate 25 classes total	0.12 (0.51) mean (SD) cigarette use	0.91 (1.73) mean (SD) cigarette use	Estimate 25 classes total			1 yr
Sun 2008 (Combined)	SI	12.24% cigarette use in the last 30 days 9 (using scale where 0 = none to 7 = 100+)		6	13.29% cigarette use in the last 30 days 9 (using scale where 0 = none to 7 = 100+)		3	0.91 (0.6 ‐ 1.37) (adj)		1 yr

Sussman 2007	C	30 day smoking prevalence		6			6	0.5 (0.34 ‐ 0.73) (adj)		1 yr

Longest Follow‐up (over 1 year)

Spoth 2001 (ISFP)	SC	slope growth curve							minus 0.0610 (0.02)	4 yr
Spoth 2001 (PDFY)	SC	slope growth curve							minus 0.01 (0.02)	4 yr

Flay 1985	SI	Measure = never‐smoker, tried once, quitted, experimenter, regular		11			11	1.22 (0.83,1.80)		6 yr
Perry 2003 (Dare boys)	SI		0.28 (0.05) growth rate (mean difference, SEM)	8		0.31 (0.05) growth rate (mean difference, SEM)	8			2.5 yrs
Perry 2003 (Dare girls)	SI		0.25 (0.07) growth rate (mean difference, SEM)	8		0.28 (0.07) growth rate (mean difference, SEM)	8			2.5 yrs
Perry 2009	SI		0.46 (‐0.19, 1.11) Linear rate of change (CI)	16		1.37 (0.72, 2.02) Linear rate of change (CI)	16			2 yrs
St Pierre 2005 (Adult)	SI			16			8		0.186 (0.255) logistic coefficent for the interaction of treatment with pre‐post contrast (SE)	3 yrs
St Pierre 2005 (Teen)	SI			16			8		0.069 (0.253) logistic coefficent for the interaction of treatment with pre‐post contrast (SE)	3 yrs

Brown 2005	C								minus 0.153 (0.105)	>1yr
Hecht 2003	C				slope growth curves. Use model				minus 0.016 (0.011)	14 mths

Perry 2003 (Dare+ boys)	MM		0.18 (0.05) growth rate (mean difference, SEM)	8		0.31 (0.05) growth rate (mean difference, SEM)	8			2.5 yrs
Perry 2003 (Dare+ girls)	MM		0.22 (0.07) growth rate (mean difference, SEM)	8		0.28 (0.07) growth rate (mean difference, SEM)	8			2.5 yrs

Group 1 Studies Included in analysis		Armstrong 1990 (Peer); Armstrong 1990 (Teacher); Ausems 2004 (In school); Ausems 2004 (Out School); Aveyard 1999; Botvin 1980; Botvin 1982; Botvin 1983 (Intensive); Botvin 1983 (LST); Botvin 1999; Brown 2002; Buller 2008 (Australia); Buller 2008 (USA); Chou 2006; Coe 1982; Connell 2007; Conner 2010 (I); Conner 2010 (SE); Crone 2011; Denson 1981; De Vries 1994 (High); De Vries 1994 (Voc); De Vries 2003 (Denmark); De Vries 2003 (Finland); De Vries 2003 (Portugal); De Vries 2003 (UK); Elder 1996; Ellickson 1990 (HealthEd); Ellickson 1990 (Teen); Ellickson 2003; Ennett 1994; Faggiano 2008; Figa‐Talamanca 1989 (F); Figa‐Talamanca 1989 (N.F); Gabrhelik 2012; Garcia 2005; Hort 1995; Howard 1996; Johnson 2009; Kellam 1998 (GBG); Kellam 1998 (ML); La Torre 2010 (A); La Torre 2010 (C); Nutbeam 1993 (FSE); Nutbeam 1993 (FSE+SAM); Nutbeam 1993 (SAM); Peterson 2000; Piper 2000 (HFL); Piper 2000 (HFL Age);Prokhorov 2008; Resnicow 2008 (Harm Min); Resnicow 2008 (LST); Ringwalt 2009a; Schulze 2006; Seal 2006; Simons‐Morton 2005; Spoth 2001 (ISFP); Spoth 2001 (PDFY); Spoth 2002 (LST); Spoth 2002 (LST + SFP); Storr 2002 (CC); Storr 2002 (FSP); Telch 1990 (No peers); Telch 1990 (Peers); Unger 2004 (CHIPS); Unger 2004 (FLAVOR); Valente 2007 (TND); Valente 2007 (TNDNetwork); Van Lier 2009; Walter 1986; Weichold 2011 (Peer); Weichold 2012 (Teacher); Wen 2010.
Group 1 studies Excluded from analysis	No control arm	Byrne 2005; Glanz 2007; Hamilton 2005; Murray 1984a
Group 1 studies Excluded from analysis	Data in format unable to use, no data provided or data unreliable	Abernathy 1992; Ary 1990; Crone 2003
Group 2 studies Included in analysis		Brown 2005; Clark 2010; Flay 1985; Forman 1990 (SI); Forman 1990 (SI ‐ NP); Forman 1990 (SI ‐ P); Kaufman 1994; Hecht 2003; Perry 2003 (Dare+ boys); Perry 2003 (Dare+ girls); Perry 2003 (Dare boys); Perry 2003 (Dare girls); Perry 2009; Reddy 2002 (School + F); Reddy 2002 (School only); Severson 1991 (High, F); Severson 1991(high, M); Severson 1991 (Middle, F); Severson 1991 (Middle, M); Shope 1996; Spoth 2001 (ISFP); Spoth 2001 (PDFY); St Pierre 2005 (Adult); St Pierre 2005 (Teen); Sun 2008 (Cognitive); Sun 2008 (Combined); Sussman 2007
Group 2 studies Excluded from analysis	Data in format unable to use	Spoth 2007
Group 3 studies Included in analysis		Botvin 1990a (Video); Botvin 1990a (Workshop); Botvin 2001; Campbell 2008; Chatrou 1999; Dijkstra 1999 (DM + B); Dijkstra 1999 (DM + no B); Dijkstra 1999 (SI + B); Dijkstra 1999 (SI + no B); Elder 1993; Gindre 1995; Hedman 2010 (Interview); Hedman 2010 (Lecture); Jøsendal 1998 (P); Jøsendal 1998 (P + T); Jøsendal 1998 (T); Laniado‐Laborín 1993; Lloyd 1983; Lotrean 2010; McCambridge 2011; Murray 1992 (MDEG); Murray 1992 (MSPP); Murray 1992 (SFG); Noland 1998; Perry 1996; Perry 2009; Rabinowitz 1974; Ringwalt 2009a; Schofield 2003; Scholz 2000 (G, female); Scholz 2000 (G, male); Scholz 2000 (R, female); Scholz 2000 (R, male); Sloboda 2009; Spoth 2002 (LST); Spoth 2002 (LST + SFP); Sussman 1995 TND1 CHS C; Sussman 1995 TND1 CHS SAC; Werch 2005
Group 3 studies Excluded from analysis	No control arm	Biglan 2000; Hansen 1991
	Data in format unable to use, no data provided or data unreliable	Biglan 1987b; Botvin 1990b; Bush 1989; Cameron 1999; Clarke 1986; Clayton 1996; Cohen 1989; Eisen 2003; Flay 1995; Focarile 1994; Gatta 1991; Gersick 1988; Gilchrist 1986; Gordon 2008; Hanewinkel 1994; Hansen 1988a; Hecht 2008; Hirschmann 1989; Longshore 2006; MacPherson 1980; Norman 2008; O'Donnell 1995; Rohrbach 2010a; Scheier 2001; Schinke 1984; Schinke 1985a; Schinke 1985b; Schinke 1985c; Schinke 1986a; Schinke 1986b; Schinke 1986c; Schinke 1988; Schinke 2000; Smith 2004; Sussman 1993; Vaughan 2007; Villalbí 1993; Walter 1985; Zheng 2005

Country	Number of studies	Study name
Austria	1	Faggiano 2008
Australia	1	Buller 2008 (Australia)
Belgium	1	Faggiano 2008
Canada	3	Armstrong 1990 (Peer); Armstrong 1990 (Teacher); Brown 2002; Denson 1981
China	2	Chou 2006; Wen 2010
Czech Republic	1	Gabrhelik 2012
Denmark	1	De Vries 2003 (Denmark)
Finland	1	De Vries 2003 (Finland)
Germany	4	Faggiano 2008; Hort 1995; Schulze 2006; Weichold 2011 (Peer); Weichold 2012 (Teacher)
Greece	1	Faggiano 2008
Italy	3	Faggiano 2008; Figa‐Talamanca 1989 (F); Figa‐Talamanca 1989 (N.F); La Torre 2010 (A); La Torre 2010 (C)
Netherlands	4	Ausems 2004 (Combined); Ausems 2004 (In school); Ausems 2004 (Out School); Crone 2011; De Vries 1994 (High); De Vries 1994 (Voc); Van Lier 2009
Portugal	1	De Vries 2003 (Portugal)
Spain	2	Faggiano 2008; Garcia 2005
South Africa	1	Resnicow 2008 (Harm Min); Resnicow 2008 (LST)
Sweden	1	Faggiano 2008
Thailand	1	Seal 2006
UK	4	Aveyard 1999; Conner 2010 (I); Conner 2010 (SE); De Vries 2003 (UK); Nutbeam 1993 (FSE); Nutbeam 1993 (FSE+SAM); Nutbeam 1993 (SAM)
USA	27	Ary 1990; Botvin 1980; Botvin 1982; Botvin 1983 (Intensive); Botvin 1983 (LST); Botvin 1999; Buller 2008 (USA); Coe 1982; Connell 2007; Elder 1996; Ellickson 1990 (HealthEd); Ellickson 1990 (Teen); Ellickson 2003; Ennett 1994; Howard 1996; Johnson 2009; Kellam 1998 (GBG); Kellam 1998 (ML); Peterson 2000; Piper 2000 (HFL); Piper 2000 (HFL Age); Prokhorov 2008; Ringwalt 2009a; Simons‐Morton 2005; Spoth 2001 (ISFP); Spoth 2001 (PDFY); Spoth 2002 (LST); Spoth 2002 (LST + SFP); Storr 2002 (CC); Storr 2002 (FSP); Telch 1990 (No peers); Telch 1990 (Peers); Unger 2004 (CHIPS); Unger 2004 (FLAVOR); Valente 2007 (TND); Valente 2007 (TNDNetwork); Walter 1986

Country	Number of studies	Study name
Canada	1	Flay 1985
India	2	Perry 2009; Reddy 2002 (School + F); Reddy 2002 (School only);
USA	12	Brown 2005; Clark 2010; Forman 1990 (SI); Forman 1990 (SI ‐ NP); Forman 1990 (SI ‐ P); Kaufman 1994; Hecht 2003; Perry 2003 (Dare+ boys); Perry 2003 (Dare+ girls); Perry 2003 (Dare boys); Perry 2003 (Dare girls); Severson 1991 (High, F); Severson 1991(high, M); Severson 1991 (Middle, F); Severson 1991 (Middle, M); Shope 1996; Spoth 2001 (ISFP); Spoth 2001 (PDFY); St Pierre 2005 (Adult); St Pierre 2005 (Teen); Sun 2008 (Cognitive); Sun 2008 (Combined); Sussman 2007

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Group 1: Pure Prevention cohort (adjusted) ‐ 1 year or less	40	32234	Odds Ratio (Fixed, 95% CI)	0.94 [0.85, 1.05]
1.1 Information giving curricula versus control	1	100	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
1.2 Social influences curricula versus control	25	20467	Odds Ratio (Fixed, 95% CI)	1.00 [0.88, 1.13]
1.3 Combined social competence and social influences curricula versus control	7	5370	Odds Ratio (Fixed, 95% CI)	0.49 [0.28, 0.87]
1.4 Multimodal programmes versus control	5	6000	Odds Ratio (Fixed, 95% CI)	0.89 [0.73, 1.08]
1.5 Other interventions	2	297	Odds Ratio (Fixed, 95% CI)	2.49 [0.10, 61.80]
2 Group 1: Pure Prevention cohort (adjusted) ‐ longest follow‐up	73		Odds Ratio (Random, 95% CI)	0.88 [0.82, 0.96]
2.1 Information giving curricula versus control	1		Odds Ratio (Random, 95% CI)	0.12 [0.00, 14.87]
2.2 Social competence curricula versus control	7		Odds Ratio (Random, 95% CI)	0.52 [0.30, 0.88]
2.3 Social influences curricula versus control	42		Odds Ratio (Random, 95% CI)	0.92 [0.84, 1.01]
2.4 Combined social competence and social influences versus control	10		Odds Ratio (Random, 95% CI)	0.50 [0.28, 0.87]
2.5 Multimodal programmes versus control	7		Odds Ratio (Random, 95% CI)	0.95 [0.64, 1.43]
2.6 Other interventions	6		Odds Ratio (Random, 95% CI)	0.91 [0.50, 1.66]
3 Group 2: Change in Smoking Behaviour over time ‐ 1 year or less	15	13137	Std. Mean Difference (Fixed, 95% CI)	0.04 [0.02, 0.06]
3.1 Information giving curricula versus control	1	1072	Std. Mean Difference (Fixed, 95% CI)	0.17 [‐0.04, 0.37]
3.2 Social competence curricula versus control	3	279	Std. Mean Difference (Fixed, 95% CI)	0.02 [‐1.19, 1.24]
3.3 Social influences curricula versus control	10	10689	Std. Mean Difference (Fixed, 95% CI)	0.04 [0.03, 0.06]
3.4 Combined social competence and social influences curricula versus control	1	1097	Std. Mean Difference (Fixed, 95% CI)	‐0.38 [‐0.59, ‐0.17]
4 Group 2: Change in Smoking Behaviour over time ‐ longest follow‐up	27		Std. Mean Difference (Fixed, 95% CI)	0.01 [‐0.00, 0.02]
4.1 Information giving curricula versus control	1		Std. Mean Difference (Fixed, 95% CI)	0.17 [‐0.04, 0.37]
4.2 Social competence versus control	5		Std. Mean Difference (Fixed, 95% CI)	‐0.04 [‐0.06, ‐0.01]
4.3 Social influences curricula versus control	16		Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
4.4 Combined social competence and social influences curricula versus control	3		Std. Mean Difference (Fixed, 95% CI)	‐0.02 [‐0.04, 0.00]
4.5 Multimodal programmes versus control	2		Std. Mean Difference (Fixed, 95% CI)	0.11 [‐0.01, 0.22]
5 Group 3: Point Prevalence of Smoking ‐ 1 year or less	21		Std. Mean Difference (Fixed, 95% CI)	Totals not selected
5.1 Information giving curricula versus control	1		Std. Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
5.2 Social influences curricula versus control	15		Std. Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
5.3 Combined social competence and social influence curricula versus control	3		Std. Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
5.4 Other interventions	2		Std. Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
6 Group 3: Point Prevalence of Smoking ‐ longest follow‐up	39		Odds Ratio (Fixed, 95% CI)	Totals not selected
6.1 Information giving curricula versus control	1		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.2 Social competence	1		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.3 Social influences curricula versus control	23		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.4 Combined social competence and social influence curricula versus control	10		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.5 Multimodal curricula versus control	2		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.6 Other interventions	2		Odds Ratio (Fixed, 95% CI)	0.0 [0.0, 0.0]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Low attrition ‐ 1 year or less	13	Odds Ratio (Fixed, 95% CI)	0.93 [0.75, 1.17]
1.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
1.2 Social influences curricula versus control	10	Odds Ratio (Fixed, 95% CI)	1.00 [0.79, 1.27]
1.3 Combined social competence and social influences curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.09]
1.4 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.73 [0.00, 119016.25]
2 Low attrition ‐ longest follow‐up	30	Odds Ratio (Fixed, 95% CI)	0.89 [0.78, 1.02]
2.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
2.2 Social competence curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.57 [0.32, 1.02]
2.3 Social influences curricula versus control	19	Odds Ratio (Fixed, 95% CI)	0.94 [0.81, 1.08]
2.4 Combined social competence and social influences	3	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.09]
2.5 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	1.03 [0.00, 433.58]
2.6 Other interventions	1	Odds Ratio (Fixed, 95% CI)	0.86 [0.44, 1.69]
3 Low & unclear attrition ‐ 1 year or less	31	Odds Ratio (Fixed, 95% CI)	0.92 [0.79, 1.07]
3.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
3.2 Social influences curricula versus control	20	Odds Ratio (Fixed, 95% CI)	0.96 [0.82, 1.13]
3.3 Combined social competence and social influences versus control	6	Odds Ratio (Fixed, 95% CI)	0.50 [0.28, 0.89]
3.4 Multimodal curricula versus control	2	Odds Ratio (Fixed, 95% CI)	0.72 [0.03, 19.98]
3.5 Other interventions	2	Odds Ratio (Fixed, 95% CI)	2.49 [0.10, 61.80]
4 Low & unclear attrition‐ longest follow‐up	58	Odds Ratio (Fixed, 95% CI)	0.90 [0.82, 1.00]
4.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
4.2 Social competence curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.57 [0.32, 1.02]
4.3 Social influences curricula versus control	35	Odds Ratio (Fixed, 95% CI)	0.94 [0.84, 1.04]
4.4 Combined social competence and social influences	8	Odds Ratio (Fixed, 95% CI)	0.50 [0.28, 0.89]
4.5 Multimodal curricula versus control	4	Odds Ratio (Fixed, 95% CI)	1.14 [0.21, 6.37]
4.6 Other interventions	5	Odds Ratio (Fixed, 95% CI)	0.89 [0.47, 1.69]
5 Low selection bias ‐ 1 year or less	16	Odds Ratio (Fixed, 95% CI)	0.97 [0.80, 1.17]
5.1 Social influences curricula versus control	12	Odds Ratio (Fixed, 95% CI)	1.02 [0.84, 1.24]
5.2 Combined social competence and social influences versus control	2	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.10]
5.3 Multimodal curricula versus control	2	Odds Ratio (Fixed, 95% CI)	0.72 [0.03, 19.98]
6 Low selection bias ‐ longest follow‐up	37	Odds Ratio (Fixed, 95% CI)	0.90 [0.80, 1.00]
6.1 Social competence curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.57 [0.32, 1.03]
6.2 Social influences curricula versus control	24	Odds Ratio (Fixed, 95% CI)	0.92 [0.82, 1.03]
6.3 Combined social competence and social influences versus control	2	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.10]
6.4 Multimodal curricula versus control	4	Odds Ratio (Fixed, 95% CI)	1.14 [0.21, 6.37]
6.5 Other interventions	2	Odds Ratio (Fixed, 95% CI)	0.88 [0.47, 1.66]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Female ‐ 1 year or less	7	Odds Ratio (Fixed, 95% CI)	0.69 [0.49, 0.96]
1.1 Social influences curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.69 [0.41, 1.14]
1.2 Combined social competence and social influences curricula versus control	2	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.09]
1.3 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.82 [0.45, 1.47]
2 Female ‐ longest follow‐up	9	Odds Ratio (Fixed, 95% CI)	0.82 [0.67, 1.00]
2.1 Social influences curricula versus control	6	Odds Ratio (Fixed, 95% CI)	0.83 [0.66, 1.04]
2.2 Combined social competence and social influences versus control	2	Odds Ratio (Fixed, 95% CI)	0.55 [0.28, 1.07]
2.3 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.99 [0.57, 1.73]
3 Male ‐ 1 year or less	6	Odds Ratio (Fixed, 95% CI)	0.66 [0.44, 0.98]
3.1 Social influences curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.92 [0.56, 1.52]
3.2 Combined social competence and social influences curricula versus control	1	Odds Ratio (Fixed, 95% CI)	1.15 [0.04, 37.54]
3.3 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.32 [0.16, 0.65]
4 Male ‐ longest follow‐up	8	Odds Ratio (Fixed, 95% CI)	0.96 [0.77, 1.20]
4.1 Social influences curricula verus control	6	Odds Ratio (Fixed, 95% CI)	0.97 [0.76, 1.23]
4.2 Combined social competence and social influences curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.77 [0.05, 11.85]
4.3 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.93 [0.54, 1.58]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 No Booster sessions ‐ 1 year or less	36	Odds Ratio (Fixed, 95% CI)	0.94 [0.85, 1.05]
1.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
1.2 Social influences curricula versus control	23	Odds Ratio (Fixed, 95% CI)	0.98 [0.86, 1.11]
1.3 Combined social competence and social influences curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.47 [0.14, 1.51]
1.4 Multimodal curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.89 [0.73, 1.08]
1.5 Other interventions	2	Odds Ratio (Fixed, 95% CI)	2.49 [0.10, 61.80]
2 No Booster sessions ‐ longest follow‐up	66	Odds Ratio (Fixed, 95% CI)	0.90 [0.83, 0.97]
2.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
2.2 Social competence curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.52 [0.30, 0.88]
2.3 Social influences curricula versus control	38	Odds Ratio (Fixed, 95% CI)	0.93 [0.85, 1.02]
2.4 Combined social competence and social influences curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.47 [0.15, 1.43]
2.5 Multimodal curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.83 [0.69, 1.01]
2.6 Other interventions	6	Odds Ratio (Fixed, 95% CI)	0.91 [0.50, 1.66]
3 Boosters sessions ‐ 1 year or less	4	Odds Ratio (Fixed, 95% CI)	0.70 [0.46, 1.07]
3.1 Social influences curricula versus control	2	Odds Ratio (Fixed, 95% CI)	0.90 [0.51, 1.56]
3.2 Combined social competence and social influences curricula versus control	2	Odds Ratio (Fixed, 95% CI)	0.50 [0.26, 0.96]
4 Booster sessions ‐ longest follow‐up	7	Odds Ratio (Fixed, 95% CI)	0.73 [0.55, 0.98]
4.1 Social influences curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.81 [0.58, 1.12]
4.2 Combined social competence and social influences curricula versus control	3	Odds Ratio (Fixed, 95% CI)	0.51 [0.27, 0.96]

Methods	Country: USA   Site: 22 middle/elementary and 15 high schools from 13 Oregon districts 'Project PATH' (Programs to Advance Teen Health)   Focus: Tobacco, alcohol and marijuana prevention   Design: Cluster RCT (Group1: never smoking prevention cohort, not included in analysis)
Participants	Baseline: 7837 (6263 completed pretest).   Age: 1943 6th graders (age 11 ‐ 12); 1890 7th graders; 698 8th graders; 1364 9th graders; 205 10th graders; 163 11th graders.   Gender: Not stated. Ethnicity : 89% W, 4.9% B, 2.2% A, 1.8% Latin American, 1.2% H Baseline smoking data: 9.9% weekly smoking.
Interventions	Category: Social Influences vs control. Programme deliverer: Science or health teachers (received 2 to 3 hrs training). Peer leaders presented some activities in 2 grades. Intervention: 5 classroom sessions in each of grades 6 through 10, typically taught over a one‐week period. "focused most heavily on cigarette smoking and smokeless tobacco use, it was designed to deter the use of marijuana and alcohol". At each grade level (a) awareness of social influences to engage in substance use (b) refusal skills training (c) health facts (d) contracting not to use cigarettes and other substances.   Programme different for each grade. Parent message group mailed 3 brochures.   Control: Groups typically received 10 classroom sessions of standard tobacco/drug use education.
Outcomes	Smoking: Pechacek 1984 self reported smoking index to yield an estimate of no. cigarettes smoked in last month (composite of no in last 6m, last month, last week, and last 24 hrs): Dichotomised on >1 cigarette in previous month. Expired air CO tested before survey completion.   Follow‐up: 9 ‐ 12m after pretest (Only results for grades 6 ‐ 9 given in Ary 1990)
Notes	Quality of intervention delivery: Surveys of teachers indicated that the control group received 10 sessions of standard tobacco and drug education (with 97% recognizing peer pressures, 97% short‐term effects on the body and brain, 96% long‐term health consequences, 84% decision‐making skills, 72% media pressures, and 67% refusal skills practice), and the experimental schools received a median of 5 sessions of other drug education in addition to PATH. Statistical quality: Was a power computation performed? No Was an intention‐to‐treat analysis performed? No Was a correction for clustering made? No Were appropriate statistical methods used? ANCOVA
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Schools were randomly assigned either to receive or not receive the intervention. The exception was one middle school assigned to the treatment condition because it had earlier served as a pilot school for program development. ...First, schools were blocked on urban/rural status. Second, schools were matched within blocks on characteristics such as level of tobacco and other drug use, ethnicity, and school size ...". In the 12 intervention schools, parents randomised to receive or not receive parent messages. No method of randomisation. Clusters: Schools. Cluster constraint: Blocked and matched. Baseline comparability: Not stated.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Low risk	"Questionnaire and biochemical data were provided by 7837 elementary‐, middle‐, and high‐school students and by 6263 students (80% of original subjects) at both initial assessment and approximately 1 year (9 ‐ 12 months) later". Attrition: 24.4 % in experimental and 24.6% in control schools; no differential attrition on pretest use by gender, grade, CO level, number of peers who smoked, offers of cigarettes, parental smoking.
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: Netherlands   Site: 8 local health departments were approached, 6 participated and 36 vocational schools participated.   Focus: Smoking prevention   Design: Cluster RCT (Group 1: never smoking prevention cohort, only arms 1 and 2 vs control included in the analysis))
Participants	Baseline: Intervention 1 (in‐school) = 525; Intervention 2 (out of school) = 513; Intervention 3 (combined in/out) = 829; control = 509.   Age: Average 13 yrs   Gender: 48% Male   Smoking status at baseline: 59.7% ever smoked; 19.5% current smokers.
Interventions	Category : Social influences (intervention 1/in school) vs social influences (intervention 2/out of school) vs control Programme deliverer: Teachers Intervention: In‐school: 3 lessons x 50 mins: ingredients of tobacco and physical and mental reactions of smoking; norms concerning smoking; pressures to smoke and skills to resist. Out‐of‐school: 3 letters mailed to students' homes, tailored to pretest attitudes, norms, self efficacy, smoking intentions and behaviour. Combined In‐school and out‐of‐school. Control: No statement.
Outcomes	Self reported never smoked even one puff; not in past month; smoked in past month. Follow‐up: 1yr, 18m.
Notes	Quality of intervention delivery: Process analysis for students was 15‐item questionnaire; and for teachers a 5‐item implementation questionnaire. Only 58% of schools returned the teacher process questionnaire; and only 65% of out‐of‐school students received and read the letters. Statistical quality: Was a power computation performed? Power computation to demonstrate an effect size with an OR = 2, with power = 80%, α 2‐tailed = 0.05, with 25 students per school, and between‐school variance = 0.30, implying an ICC = 0.08, required 36 schools, and sample size achieved. Was an intention‐to‐treat analysis performed? Yes. Missing data: replaced by previous observation; drop‐outs were treated as smokers. Was a correction for clustering made? Yes, using multilevel modelling Were appropriate statistical methods used? Multilevel regression modelling using MIXREG for continuous and MIXOR for dichotomous outcomes.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Nineteen schools that already participated in the in‐school program were randomly assigned to the in‐school or to the combined in‐school and out‐of‐school condition. The remaining 17 schools were randomly assigned to the out‐of‐school condition or to the control group". [i.e. randomisation of schools did not give all schools an equal chance of being assigned to the three groups]. No method of randomisation stated. Clusters: Schools Cluster constraint: Not stated Baseline comparability: Students in out‐of‐school condition older than control (OR 1.27, 95%CI 1.03 to 1.57)
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Numbers at 12m: Intervention 1 = 434 (83%); numbers at 18m: Intervention 2 = 265 (52%); Intervention 3 = 625 (75%);control = 317 (61%) "Attrition at student level was 17.3% at post‐test 1, 25.4% at post‐test 2, and 24.6% at post‐test 3". Attrition at post‐test 3 less likely if: living with both parents (OR = 0.53, 95%CI 0.37 to 0.77); with 2 Dutch parents (OR = 0.63, 95%CI 0.47 to 0.84), less 'diffusely' surrounded by smokers (OR = 0.87; 95% CI 0.84 to 0.90).
Selective reporting (reporting bias)	Low risk	Original goals of study met

Methods	Country: USA   Site: 2 suburban New York City schools.   Focus: Smoking prevention.   Design: Cluster RCT (Group 1: never smoking prevention cohort)
Participants	Baseline: 281   Age: 8th (age 13 ‐ 14), 9th and 10th graders.   Gender: Not stated Ethnicity: "predominantly white". Baseline smoking data: 70% nonsmokers.
Interventions	Category: Social influences and social competence vs control. Programme deliverer: Outside specialists. Intervention: 10 lessons over 12 weeks. Social influences and psychosocial skills; group discussion, modelling, behaviour rehearsal, and the application of special skills training to life situations, including the decision to smoke; homework; self improvement project. Control: No intervention.
Outcomes	Smoking: Self reported smoking (last month, and last week). Pretest smokers excluded from analysis.   Follow‐up: 6m from pretest.
Notes	Quality of intervention delivery: No process analysis of delivery of the intervention. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? No. Were appropriate statistical methods used? Yes, Χ², 2‐way ANOVA
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"After randomly assigning the two schools to an experimental (n = 121) and control (n = 160) condition...". Clusters: Schools. Cluster constraint: Not applicable as only 2 schools. Baseline comparability: "Both schools had approximately the same baseline smoking rates" (School A = 31%, School B = 29%).
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	"Unfortunately, follow‐up data (post‐test 2) were collected on only about 77% of the post‐test 1 sample (80% for the experimental group and 74% for the control group.". ["The first post‐test occurred at the completion of the smoking prevention program (12 weeks after the pretest), and the second post‐test occurred approximately three months later ..."] differential attrition from baseline.
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA   Site: 2 suburban New York City schools (all 7th grade classes)   Focus: Smoking prevention   Design: Cluster RCT (Group 1: never smoking prevention cohort)
Participants	Baseline: 426   Age: 7th graders (age 12 ‐ 12). Gender: Not stated   Ethnicity: W (school A 93%; school B 90%); B (2%,4%); A (3%,3%); H (2%,3%) Baseline smoking data: 74% of 374 analysable sample.
Interventions	Category: Social influences and social competence vs control. Programme deliverer: Peers (recruited from neighbouring high school, received 4 hr training workshop. Supervised by a teacher and project staff) Intervention: 12 1hr sessions over 12 weeks. Physiological effects; teenage smoking rates; LST smoking prevention programme skills (self image, self improvement, decision making, independent thinking, advertising techniques, coping with anxiety, communication skills, social skills, assertiveness); homework; a self improvement project. Control: No programme.   Note: See Botvin 1980 for similar programme delivered by outside specialists and Botvin 1983 for delivery by classroom teachers.
Outcomes	Smoking: Self reported smoking (last month, and last week). Pretest smokers excluded from analysis.   Saliva samples collected, 25% subsample analysed for thiocyanate.   Follow‐up: 1 yr after post‐test.
Notes	Quality of intervention delivery: No process analysis of delivery of the intervention. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? No. Were appropriate statistical methods used? X².
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Schools were randomly assigned to experimental and control conditions". Only two schools were randomised. Method of randomisation not stated. Clusters: Schools. Cluster constraint: Not applicable as only 2 schools. Baseline comparability: Not stated
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	"Complete pretest and post‐test data were obtained on 357 students. Of these, 264 (74%) were classified as nonsmokers at the time of the pretest. This group represents the nonsmoking cohort that was the focus of attention over the course of the intervention and follow‐up phases of the study". 426 baseline, 357 at 3m post‐test; no data on total numbers at 1 yr follow‐up; There were 264 nonsmokers at pretest, and of these 210 were reported present at 1 yr. Complete pre‐ and post‐test data on 84%, of whom 74% were nonsmokers at the pretest.
Selective reporting (reporting bias)	Low risk	No selective reporting

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Multi foci ‐ 1 year or less	14	Odds Ratio (Fixed, 95% CI)	0.92 [0.74, 1.16]
1.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
1.2 Social influences curricula versus control	9	Odds Ratio (Fixed, 95% CI)	0.99 [0.78, 1.27]
1.3 Combined social competence and social influences curricula versus control	3	Odds Ratio (Fixed, 95% CI)	0.54 [0.28, 1.06]
1.4 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.72 [0.02, 22.85]
2 Multi foci ‐ longest follow‐up	29	Odds Ratio (Fixed, 95% CI)	0.88 [0.77, 1.01]
2.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
2.2 Social competence curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.52 [0.30, 0.88]
2.3 Social influences curricula versus control	14	Odds Ratio (Fixed, 95% CI)	0.94 [0.82, 1.09]
2.4 Combined social competence and social influences curricula versus control	6	Odds Ratio (Fixed, 95% CI)	0.55 [0.29, 1.04]
2.5 Multimodal curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.82 [0.03, 25.09]
3 Tobacco focused ‐ 1 year or less	26	Odds Ratio (Fixed, 95% CI)	0.93 [0.83, 1.04]
3.1 Social influence curricula versus control	16	Odds Ratio (Fixed, 95% CI)	0.97 [0.84, 1.12]
3.2 Combined social competence and social influences curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.37 [0.12, 1.13]
3.3 Multimodal curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.89 [0.73, 1.08]
3.4 Other interventions	2	Odds Ratio (Fixed, 95% CI)	2.49 [0.10, 61.80]
4 Tobacco focused ‐ longest follow‐up	42	Odds Ratio (Fixed, 95% CI)	0.88 [0.80, 0.97]
4.1 Social influences curricula versus control	28	Odds Ratio (Fixed, 95% CI)	0.91 [0.81, 1.02]
4.2 Combined social competence and social influences curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.37 [0.12, 1.13]
4.3 Multimodal curricula versus control	4	Odds Ratio (Fixed, 95% CI)	0.83 [0.68, 1.00]
4.4 Other interventions	6	Odds Ratio (Fixed, 95% CI)	0.91 [0.50, 1.66]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Peer‐led ‐ 1 year or less	8	Odds Ratio (Fixed, 95% CI)	0.91 [0.56, 1.46]
1.1 Social influences curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.90 [0.56, 1.47]
1.2 Combined social competence and social influences curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.97 [0.11, 8.39]
2 Peer‐led ‐ longest follow‐up	11	Odds Ratio (Fixed, 95% CI)	0.94 [0.61, 1.47]
2.1 Social influences curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.93 [0.59, 1.47]
2.2 Combined social competence and social influences	2	Odds Ratio (Fixed, 95% CI)	0.99 [0.12, 8.00]
2.3 Multimodal curricula versus control	2	Odds Ratio (Fixed, 95% CI)	1.31 [0.16, 10.73]
3 Adult‐led ‐ 1 year or less	29	Odds Ratio (Fixed, 95% CI)	0.93 [0.83, 1.03]
3.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
3.2 Social influences curricula versus control	16	Odds Ratio (Fixed, 95% CI)	0.98 [0.86, 1.12]
3.3 Combined social competence and social influences curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.46 [0.26, 0.84]
3.4 Multimodal curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.89 [0.73, 1.08]
3.5 Other interventions	2	Odds Ratio (Fixed, 95% CI)	2.49 [0.10, 61.80]
4 Adult‐led ‐ longest follow‐up	56	Odds Ratio (Fixed, 95% CI)	0.88 [0.81, 0.96]
4.1 Information curricula versus control	1	Odds Ratio (Fixed, 95% CI)	0.12 [0.00, 14.87]
4.2 Social competence curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.52 [0.30, 0.88]
4.3 Social influences curricula versus control	30	Odds Ratio (Fixed, 95% CI)	0.92 [0.84, 1.01]
4.4 Combined social competence and social influences curricula versus control	7	Odds Ratio (Fixed, 95% CI)	0.47 [0.26, 0.84]
4.5 Multimodal curricula versus control	5	Odds Ratio (Fixed, 95% CI)	0.83 [0.68, 1.00]
4.6 Other interventions	6	Odds Ratio (Fixed, 95% CI)	0.91 [0.50, 1.66]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Low attrition ‐ 1 year or less	5	Std. Mean Difference (Fixed, 95% CI)	‐0.02 [‐3.01, 2.98]
1.1 Social influences curricula versus control	5	Std. Mean Difference (Fixed, 95% CI)	‐0.02 [‐3.01, 2.98]
2 Low attrition ‐ > 1 year, longest follow‐up	15	Std. Mean Difference (Fixed, 95% CI)	‐0.02 [‐0.05, 0.00]
2.1 Social competence curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	‐0.04 [‐0.06, ‐0.01]
2.2 Social influences curricula versus control	10	Std. Mean Difference (Fixed, 95% CI)	0.05 [‐0.06, 0.16]
2.3 Combined social competence and social influences curricula versus control	1	Std. Mean Difference (Fixed, 95% CI)	‐0.15 [‐0.36, 0.05]
2.4 Multimodal curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	0.11 [‐0.01, 0.22]
3 Low & unclear attrition ‐ 1 year or less	13	Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
3.1 Information curricula versus control	1	Std. Mean Difference (Fixed, 95% CI)	0.17 [‐0.04, 0.37]
3.2 Social competence curricula versus control	3	Std. Mean Difference (Fixed, 95% CI)	0.02 [‐1.19, 1.24]
3.3 Social influences curricula versus control	9	Std. Mean Difference (Fixed, 95% CI)	0.04 [0.03, 0.06]
3.4 Low & unclear attrition ‐ 1 year or less	0	Std. Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
4 Low & unclear attrition ‐ > 1 year, longest follow‐up	25	Std. Mean Difference (Fixed, 95% CI)	0.01 [‐0.00, 0.02]
4.1 Information	1	Std. Mean Difference (Fixed, 95% CI)	0.17 [‐0.04, 0.37]
4.2 Social competence curricula versus control	5	Std. Mean Difference (Fixed, 95% CI)	‐0.04 [‐0.06, ‐0.01]
4.3 Social influences curricula versus control	15	Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
4.4 Combined social competence and social influences curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	‐0.02 [‐0.04, 0.00]
4.5 Multimodal curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	0.11 [‐0.01, 0.22]
5 Low selection bias ‐ 1 year or less	2	Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
5.1 Social influences curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
6 Low selection bias ‐ > 1 year, longest follow‐up	11	Std. Mean Difference (Fixed, 95% CI)	0.02 [0.01, 0.04]
6.1 Social competence curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	‐0.04 [‐0.06, ‐0.01]
6.2 Social influences curricula versus control	7	Std. Mean Difference (Fixed, 95% CI)	0.05 [0.03, 0.06]
6.3 Multimodal curricula versus control	2	Std. Mean Difference (Fixed, 95% CI)	0.11 [‐0.01, 0.22]

Methods	Country: Canada   Site: All schools in Calgary, Alberta (94 intervention schools, 96 control schools) 'PAL Programme' (Peer Assisted Learning)   Focus: Smoking prevention   Design: Cluster RCT (Group 1: never smoking prevention cohort, not included in analysis)
Participants	Baseline: 7508   Age: Grade 6, age 11 ‐ 12 Gender: 49% F   Baseline smoking data: Never smoked 67% M 71% F
Interventions	Category: Social influences vs control Programme deliverer: Teachers and peers (received invitation to in‐service presentations about PAL programme) Intervention: 5 sessions over 3m. Information about the benefits of not smoking (with peer‐led component)   Control: No intervention
Outcomes	Smoking categories: Never smoked/ tried but no longer smoke/currently smoke (main analysis based on baseline never‐smokers)   Follow up: From start of programme: 1yr, 2yr, 3yr
Notes	Quality of intervention delivery: A telephone survey found that 5 teachers had not taught the programme; 40 had not taught the entire programme; and 49 had taught the complete programme. Statistical quality: Was a power computation performed? No Was an Intention‐to‐treat analysis performed? No Was a correction for clustering made? No Were appropriate statistical methods used? Analysis by X2 compared proportions smoking in the three groups
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	'Schools were classified into quintiles according to median neighbourhood income, and then were randomly assigned to either the test or control groups" Clusters: School Cluster constraint: Stratificaition Baseline comparability: Groups identical at pretest on smoking rates.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Questionnaires were anonymous then linked by a unique number. Unclear if students knew which arm of the study they were in.
Incomplete outcome data (attrition bias)   All outcomes	High risk	7207 (96%) after 12m; 6884 (92%) after 26m; and 6530 (87%) followed to the 9th grade. "of the students successfully matched across Grades 6, 7 and 8, matches were obtained for 3,567 (82.7%) Grade 9 students". The analysis sample is the 48% of the pretest sample who completed all four questionnaires; no analysis of differential attrition; In the evaluation, intervention classes were divided into those in which teachers reported teaching all lessons, and those where fewer were delivered
Selective reporting (reporting bias)	Unclear risk	Reported on primary outcome. But reporting not as expected because of changes during the study caused by incomplete teaching of the programme

Methods	Country: Australia   Site: 45 primary schools in Nedlands, Western Australia   Focus: smoking prevention   Design: Cluster RCT (Group 1: never smoking prevention cohort)
Participants	Baseline: (1981) 2366   Age: 7th grade (modal age 12 years) Gender: 49% F Ethnicity: Not stated   Baseline smoking data: Smoking prevalence 24 ‐ 37%
Interventions	Category: Social influences vs control Programme deliverer: Teachers and peers ("all leaders received appropriate previous training") Intervention: (6m duration) Peer‐led (selected by class), teacher facilitated; 5 sessions. Intervention based on Minnesota model. Components: estimating smokers in age group; negative consequences; why children smoke; physiological effects; information on % of smokers; listed situations where pressure to smoke; practised refusal techniques; students presented arguments for nonsmokers' rights; developed counter‐arguments to smokers' reasons for smoking; role of the family; advertising techniques; essay on reasons for remaining nonsmokers; public commitment Teacher‐led same programme Control: No planned intervention
Outcomes	Nonsmoking in previous 12m (not smoked a single cigarette, not even a few puffs). Saliva samples collected but not analysed.   Follow‐up: 12m, 24m, 7 yrs from end of programme.
Notes	Quality of intervention delivery: No process analysis of delivery of the intervention; the authors state "all leaders received appropriate previous training". Statistical quality: Was a power computation performed? No Was an intention‐to‐treat analysis performed? No Was a correction for clustering made? The data on schools were erased after 1yr, so that ICCs could not be computed, and the data were not corrected for the effects of clustering [the authors state: "Given the large number of original classes and the subsequent mixing of students that is described above, it is likely that any biases which arose in estimates of their effects or their precision because of the analysis of individuals rather than classes would be small". Were appropriate statistical methods used? Comparison of the proportions of students in the 3 groups who took up smoking was by Pearson's Χ2 (two‐sided); effects of other variables controlled in separate LRs (using EGRET) for boys and girls, and for each year of follow‐up, using only children present at baseline and both follow‐ups. Once the final models were chosen, the parameters were re‐estimated with an added risk model.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Each school was assigned at random to one of three interventional groups: control group (no planned intervention); peer‐led programme; and teacher‐led programme". No comment on method of randomisation. Clusters: Schools Cluster constraint: Stratified by class size and location Baseline comparability: No differences between groups at baseline, smoking prevalence higher for boys than girls
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	"At the time of the survey, the children, their teachers and those who conducted the survey did not know the interventional group to which the class was assigned"
Incomplete outcome data (attrition bias)   All outcomes	High risk	"Eighty‐two per cent and 64% of students were traced and re‐surveyed in the first and second follow‐up studies respectively". [after 1 and 2 years] "Seven years after the first survey of 2,366 Year 7 students in 1981 68% were traced through public records [Driver's Licences, electoral commissions and registries of births marriages and deaths]; 53 per cent of these responded to a new survey concerning smoking". [i.e. 37% of original sample] No differential attrition by treatment group at 12m follow up. Saliva samples were collected but not analysed. At the 7 yr follow up, non‐response was associated (P < .05) with being male, being in the control group, thinking most adults smoked, and mother and brother smoked.
Selective reporting (reporting bias)	Low risk	Abstract states: "How effective are peer‐led programmes in preventing the uptake of smoking by children?" This outcome is fully reported.

Methods	See Armstrong 1990 (Peer)
Participants
Interventions
Outcomes
Notes	This represents the 2nd intervention arm in Armstrong 1990 (Peer)

Methods	See Ausems 2004 (In school)
Participants
Interventions
Outcomes
Notes	This represents the 3rd intervention arm (combined in and out of school) within Ausems 2004 (In school)

Methods	Country: UK   Site: 53 West Midlands secondary schools.   Focus: Smoking prevention.   Design: Cluster RCT (Group 1: never smoking prevention cohort)
Participants	Baseline: (1997) 8352, 90% of potential participants.   Age: Year 9, 13 ‐ 14 yrs Gender: 50% Male   Ethnicity: 86% W, 5% Indian subcontinent, 4% Afro‐Caribbean.
Interventions	Category: Social influences vs control Programme deliverer: Teachers (received 2 day training course) Intervention: 6 hrs over 3 terms. 1 class lesson and 1 computer session per term for three terms based on Prochaska's transtheoretical model/ stages of change. Students used individual computers to answer questions about their smoking, and an expert system gave feedback on how their temptations compared to those of others in same stage, and their changes from previous sessions. The questionnaires were interspersed with video clips of young people talking about their thoughts about smoking that were relevant to the stage of change of the student concerned. Class lessons developed understanding of stages of change, and pros and cons of smoking at different stages. Students could be in one of nine stages (precontemplation to cessation maintenance) and were given advice appropriate to their stage, e.g., those in the acquisition preparation stage were told: "To be more like others who were thinking about trying it [smoking] but have chosen to stay smoke free, think more about the cons of smoking." Teachers delivered a one‐hr classroom 'transtheoretical model' intervention "how the pros and cons of smoking would vary in different stages, and lessons got young people to use these concepts" Control: Normal health education on tobacco. Teachers provided with lesson plans and handouts but were not required to use them, and received no training.
Outcomes	Self reported behaviour: Ex‐smoker/smoker/tried/never. Primary outcome was smoking one or more cigarettes a week. Questionnaires were confidential.   Follow‐up: 12m after start of intervention.
Notes	Quality of intervention delivery: 79% of baseline non‐regular smokers and 69% of baseline regular smokers received all three computer lessons; 70‐80% of sessions lasted long enough to read all the material; though baseline smokers were less likely to attend, and smokers were less likely to spend long enough to receive the individualised messages. Data on attendance and the students' reactions to the classroom lessons were not collected by the researchers. Half the teachers returned data, with a mean score of 4/5 for delivery of the lesson, and pupils' understanding and enjoyment. The researchers reported that: "All teachers reported that all intervention lessons were delivered, but we have no record of which individuals received the class‐based intervention. … Teachers were reluctant to return their questionnaires, despite prompting". Statistical quality: Was a power computation performed? Sample size of 8500 was calculated to achieve 90% power to detect a 4% difference in smoking with 5% Type I error (the ICC for smoking was calculated from a lifestyle survey as 0.008) Was an intention‐to‐treat analysis performed? Yes Was a correction for clustering made? Yes Were appropriate statistical methods used? Multilevel modelling to allow for clustering; sensitivity analysis for handling of losses to follow‐up; analyses performed by adjusting for baseline smoking status and other variables. Odds ratios used from Table 5 (Aveyard 2001)
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Schools sampled with probability proportional to size of year 9 enrolment; 89 schools approached,53 agreed to participate. Randomised in 5 strata based on year 9 size. "We randomly allocated schools, not individuals, to receive the intervention or be controls. We ensured that the arms were balanced by ordering schools into five groups based on numbers of students in year 9. We allocated each school a number between 1 and N (the maximum number in the group). A computer program generated n/2 random numbers between 1 and n, and these schools were allocated to intervention". Clusters: Schools Cluster constraint: 5 strata based on year 9 size. Baseline comparability: Equivalent.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	One school dropped out after randomisation leaving 52. 8352 13‐ and 14 yr olds enrolled; 7413 (90%) at year 1 and 6782 (82%) at year 2 follow‐ups. "For regular weekly smoking, the assumptions about those lost to follow‐up are as follows. We assumed that all those lost to follow‐up were smokers, those lost were not smokers, those lost had the same smoking status as at baseline (with unknown baseline smoking status counted as smokers), and those lost had the same smoking status as at baseline (with unknown baseline smoking status counted as nonsmokers). We then confined the analysis to all those who were followed up and all those and for whom smoking status could be calculated and all those followed up and who gave no inconsistent data on smoking status. Only the data for all those with known smoking status at follow‐up are presented in this report … In all these analyses there should be no reason why loss to follow‐up or unreliable data would be associated with the TTM or control group…".
Selective reporting (reporting bias)	Low risk	All study objectives met

Methods	Country: USA   Site: 3 high schools and 6 middle schools in Eugene, Oregon.   Focus: Smoking prevention (focus on effects of attrition)   Design: Cluster RCT (excluded from analysis)
Participants	Baseline: 1730 (873 7th; 588 9th; 262 10th graders)   Age: 7th (age 12 ‐ 13), 9th and 10th graders.   Gender: 49% F Ethnicity: "almost all white". Baseline smoking data: No data on baseline smoking rates
Interventions	Category: Social influences vs control Programme deliverer: Regular science or health teachers. Intervention: 3 consecutive days with a 4th session 2 weeks later. Social‐reinforcement short‐ and long‐term consequences of smoking; public commitment; teaching of refusal skills (film; practised role‐playing refusal skills; skits; teachers praised skills; class voted on best refusal). Control: No intervention.
Outcomes	Self reported smoking (Pechacek 1984 index) = a weighted average of the number of cigarettes smoked last week and the reported number smoked yesterday. Also categorised into 4 baseline groups: never‐smoked/triers/experimenters (1 ‐ 6 in previous week)/ regular. Expired air CO content. Refusal skills assessed for a sample (Hops 1986)   Follow‐up: 6m and 1yr
Notes	Quality of intervention delivery: No process analysis of delivery of the intervention. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? No. Were appropriate statistical methods used? X2 of proportions smoking in the two groups; ANCOVA of pretest smoking status, treatment condition, grade and gender (smoking rates log transformed to control skew).
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	9 schools from 2 school districts [no further statement on school selection]. "Within each school, the classes of teachers who had agreed to participate were randomly assigned either to the intervention or to a no‐program condition". No method of randomisation stated. Cluster: Classes. Cluster constraint: Not stated. Baseline comparability: No baseline difference between groups. Differences in baseline characteristics of drop‐outs: more likely to have been baseline smokers and have multiple risk factors for smoking.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Low risk	"Attrition rates were substantial at both 6 months (21.6%) and 1 year (31.8%)". Significant differences (P = 0.00) between those remaining and those missing both at 6 and 12m on cigarettes/week and for all family members and best friend smoking, but no differential attrition
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA   Site: 8 Oregon communities   Focus: Tobacco, alcohol, marijuana, and antisocial behaviours.   Design: Community‐ and school‐based RCT (excluded from analysis).
Participants	Baseline: 4438   Age: 6th grade (11‐12)   Gender: 52% M Ethnicity: 85% W, 7% H, 6% N‐A, 1% A‐A, 1% A, < 1% Other. Baseline smoking data: Smoking prevalence index for school‐based only intervention 8%, community intervention 10.5% (no actual numbers of nonsmokers/smokers, just index)
Interventions	Category: Social Influences vs social influences + multimodal. Programme deliverer: Teachers and community adults. Intervention 1: Schools Only PATH programme (Effects of smoking; refusal skills for smoking, drugs, antisocial behaviour; video assisted instruction in refusal skills; public commitment not to smoke; peer‐led education and skill practice; 35 sessions Grades 6 ‐ 9). Intervention 2: School PATH + Community Programme (4 modules: media advocacy, youth anti‐tobacco activities, family communication, and ACCESS module programmes to stores to reduce selling tobacco to minors). Control: No group.
Outcomes	Smoking defined as (1) level of smoking (never to pack+/day); (2) number of cigarettes (past month, week and day, with responses scaled to form Pechacek 1984 smoking index [monthly x 4.3 + weekly + daily/7] to form an index of the number of cigarettes smoked weekly), (3) net CO score (expired air minus classroom CO level). Similar measures were derived for smokeless tobacco. Follow‐up: Annually up to 5 yrs.
Notes	Quality of intervention delivery: Information was collected on adolescents' exposure to information about smoking cessation; awareness of efforts to reduce illegal tobacco sales to minors; and media activities; however, no process analysis for the school intervention component. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? No. Was a correction for clustering made? No. Were appropriate statistical methods used? Generalized estimating equations and MANOVA; individual students were nested within communities, and community means were the unit of analysis.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"It was a randomised controlled study in which small Oregon communities were assigned to one of two conditions." "Pairs of communities were matched on community socioeconomic status and population. One member of each pair was assigned at random (via the flip of a coin) to receive a school based tobacco and other substance use prevention programme (school based only (SBO) condition) in grades 6 through to 12. The other member received a community intervention in addition to the school based programme (CP condition)". Clusters: Communities Cluster constraint: Pair‐matched on community socioeconomic status and population. Baseline comparability: There were no differences at baseline between community pairs in size, per capita income, median household income, % below poverty level, % minority students, or % high school graduates.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Baseline = 4438; after 1 yr = 4515; after 2 yrs = 4395; after 4 yrs = 4708; after 5 yrs = 4165 [there is no explanation of the fluctuating numbers of over time, with more students after 1 yr compared to baseline and more students after 4 than 3 yrs; this is presumably due to in‐migration of students exceeding out‐migration]. Attrition was low at 6%; 13.5% of students were not assessed across all 5 yrs of the study; no assessment of differential attrition.
Selective reporting (reporting bias)	Low risk	No selective reporting

PERMALINK

School‐based programmes for preventing smoking

Roger E Thomas

Julie McLellan

Rafael Perera

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Background

The incidence and prevalence of smoking among children and adolescents

School‐based interventions

1. Information only curricula

2. Social competence curricula

3. Social influence curricula

4. Combined social competence and social influences curricula

5. Multimodal programmes

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Types of outcome measures

Search methods for identification of studies

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

Data synthesis

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Results

Description of studies

Included studies

1.

Excluded studies

Ongoing studies

Risk of bias in included studies

2.

Effects of interventions

1.1. Analysis.

1.2. Analysis.

3.

1.3. Analysis.

1.4. Analysis.

1.5. Analysis.

1.6. Analysis.

3.1. Analysis.

3.3. Analysis.

6.1. Analysis.

6.3. Analysis.

6.2. Analysis.

5.4. Analysis.

4.1. Analysis.

4.2. Analysis.

4.4. Analysis.

Discussion

Authors' conclusions

Implications for practice.

Implications for research.

What's new

History

Acknowledgements

Appendices

Appendix 1. MEDLINE search strategy

Appendix 2. CINAHL search strategy

Appendix 3. Raw data group 1 studies (included in analysis)

Appendix 4. Raw data group 2 studies (included in analysis)

Appendix 5. Raw data group 3 studies (included in analysis)

Appendix 6. Table to identify Group 1, 2 and 3 studies by analysis group

Appendix 7. Group 1 studies (included in analysis) by country

Appendix 8. Group 2 studies (included in analysis) by country

Appendix 9. Group 3 studies (included in analysis) by country

Methods	See Botvin 1983 (LST)
Participants
Interventions
Outcomes
Notes	This represents the 2nd intervention arm (LST intensive) within Botvin 1983 (LST)

Methods	Country: USA   Site: 7 schools in suburban New York (2 schools to intervention 1, 2 schools to intervention 2, 3 to control).   Focus: Smoking prevention   Design: Cluster RCT (Group 1: never smoking prevention cohort)
Participants	Baseline: 902.   Age: 7th grade (age 12 ‐13). Gender: Not stated. Ethnicity: 91% W.   Baseline smoking data: The numbers at pretest giving their smoking status ranged from 891 to 911 of whom nonsmokers were 92%.
Interventions	Category: Social influences and social competence vs control. Programme deliverer: Classroom teachers (received one‐day workshop training). Intervention: LST: immediate physiologic effects of smoking, self image, self improvement, decision making, advertising techniques, coping with anxiety, communication skills, social skills, assertiveness, techniques for resisting peer pressure to smoke. Direct comparison of long or short delivery format LST taught in 15 1hr sessions as part of science or health curriculum, over 15 weeks. LST in intensive mini‐course format, 15 sessions, consecutive days over approximately 1m. (One school also had 8 session booster between post‐test and 1yr follow‐up) Control: Standard smoking education mandated by NY State.
Outcomes	Self report of smoking (monthly recall; weekly recall; daily recall). Saliva samples collected but not analysed. Follow‐up: 1 yr from pre‐test.
Notes	Quality of intervention delivery: Process analysis performed but not reported. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? No. Were appropriate statistical methods used? X², ANCOVA.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"The seven schools in the study were randomly assigned to the following conditions: (1) LST Smoking Prevention Program, ...(2) LST Smoking Prevention Program, utilizing an intensive mini course ... and (3) control". Clusters: Schools. Cluster constraint: Not stated. Baseline comparability: Not stated.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	The number at the 1yr follow‐up ranged from 605 to 633 (67%); no attrition analysis. The largest number of participants at the pretest recorded in Table III was 876, and after one yr in Table V was 633 (72%).
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	See Botvin 1990a (Workshop)
Participants
Interventions
Outcomes
Notes	This represents the data from the second intervention within Botvin 1990a (Workshop)

Methods	Country: USA   Site: 56 schools in 3 regions of New York state.   Focus: Substance abuse prevention.   Design: Cluster RCT (Group 3: point prevalence).
Participants	Baseline: (1985) 5954.   Age: 7th graders (age 12 ‐ 13).   Gender: 48% F Ethnicity: 91% W, 2% B, 2% H, 1% N‐A. Baseline smoking data: Smoking based on 10 point scale. Intervention 1: mean (SE) = 1.10 (0.02), intervention 2: mean (SE) = 1.09 (0.01), control: mean (SE) = 1.10 (0.01).
Interventions	Category: Social influences and social competence vs control. Programme deliverer: Teachers. Intervention: 12 lessons over 15 class periods for 8 weeks in grade 7, 10 booster sessions in grade 8 and 5 in grade 9. LST (cognitive‐behavioural skills for building self esteem; resisting advertising pressure; managing anxiety; communicating effectively; developing personal relationships; asserting one's rights; developing specific skills to resist social influences to smoke, drink or use drugs). Formal (1 day) training/workshop and feedback on implementation. 2 hrs training by videotape, and no feedback. Control: No intervention.
Outcomes	Smoking: 10 point scale: 1 (never) ‐ 10 (more than a pack a day). Breath samples were collected, but not analysed.   Follow‐up: 3 yrs (9th grade, end of programme) and 5 ‐ 6 yrs (12th grade) from baseline.
Notes	Quality of intervention delivery: Average 68% implementation (ranging from 27% ‐ 97%), with only 75% of the students in the prevention conditions exposed to 60% or more of the prevention programmes. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? No. Was a correction for clustering made? No. Were appropriate statistical methods used? Appropriate analysis with GLM; MANOVA, and ANOVA; students who received at least 60% of the programme were included in the analysis.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	" ... schools were divided into tertiles consisting of schools with either high, medium or low cigarette smoking prevalence rates. From within groups of schools with similar levels of cigarette smoking, schools were randomly assigned within each of the geographic areas: (1) prevention program with a formal 1‐day training workshop and implementation feedback ...(2) prevention program with training provided by videotape ...and (3) a "treatment as usual" control group". Clusters: Schools. Cluster constraint: tertiles based on cigarette smoking prevalence rates, followed by geographical area. Baseline comparability: Botvin 1995: "No significant pretest differences were found ...".
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	5,954 7th graders participated in the pretest in the Fall of 1985‐86, 4,466 (75%) provided data at the end of the 9th grade, and 3597 (60%) in 1991 Pretest smokers more likely to be lost but no differential attrition across conditions
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA   Site: 10 suburban New York junior high schools (2 to each of 4 intervention groups, 2 to control).   Focus: Substance abuse prevention.   Design: Cluster RCT (excluded from analysis).
Participants	Baseline: 1311.   Age: 7th grade (age 12 ‐13).   Gender: 51% F (at 1yr follow‐up).   Ethnicity: 80% W, 13% B, 2% H, 2% A, 4% Other. Baseline smoking data: No data..
Interventions	Category: Social influences and social competence vs. control. Programme deliverer: Teachers and peers (received a 4 hr training workshop conducted by project staff). Intervention: All groups using LST approach. In 7th grade all experimental groups received a 20‐session multicomponent substance abuse prevention curriculum focusing on social, psychological, cognitive, and attitudinal factors ‐ facilitation of basic life skills and improvement of personal competence (teaching social resistance skills). In 8th grade selected groups received 10 booster sessions which were directed toward the consequences of smoking, decision making, resistance to advertising, anxiety coping skills, communication skills, social skills, assertiveness, and problem solving. Peer‐led. Peer‐led plus 8th grade booster. Teacher‐led. Teacher‐led plus 8th grade booster. Control: No intervention.
Outcomes	Smoking: monthly, weekly, and daily smoking dichotomous measures, and an index of smoking frequency (5‐point scale: never to everyday). Results presented as adjusted response proportions. Follow‐up: 1 yr.
Notes	Quality of intervention delivery: The field staff noted the low degree of fidelity of implementation by many teachers. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? No stated. Were appropriate statistical methods used? Attrition tested by ANOVA, treatment and control conditions compared using GLM ("One‐year follow‐up response frequencies were compared for each of the five conditions, with pretest response frequencies being used for covariates").
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"The 10 schools participating in the study had previously been randomly assigned to the following conditions during the first year of the study ...". No method of randomisation stated. Clusters: Schools. Cluster constraint: Not stated. Baseline comparability: Not stated.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Low risk	"Of the original sample of 1311 7th graders, 1185 (90%) were available for the initial post‐test and 998 (76%) were available for the one‐year follow‐up". No differential attrition between smokers and non‐smokers.
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA   Site: 29 New York junior high schools.   Focus: Reduction in tobacco and motivation to use substances by providing knowledge and skills to resist tobacco, alcohol and drugs.   Design: Cluster RCT (Group 1: never smoking prevention cohort).
Participants	Baseline: 2690   Age: 7th grade (11 ‐ 12).   Gender: 100% F Ethnicity: 60% A‐A, 23% H, 7% A, 3% W, 2% N‐A, 5% biracial or other. Baseline smoking data: 19% lifetime prevalence, 4% 30‐day prevalence; nonsmokers: intervention N = 1005, control N = 726.
Interventions	Category: Social Influences and social competence vs control. Programme deliverer: Teachers (received one‐day training workshop) Intervention: 15 session LST Programme, with cognitive‐behavioural skills to enhance assertiveness, resist advertising pressures, manage anxiety, communicate effectively, develop strong interpersonal relationships, and problem‐specific skills related to drug use influences, including assertiveness skills for use in situations in which students experience pressure from peers to smoke, drink or use drugs. The programme was modified for minority group use by changing the examples and the situations used for the behavioural exercises. They received 10 boosters the following year. Control: Received 10 sessions of an information‐only drug programme + 3 boosters the following year.
Outcomes	Smoking was defined as a 9‐point index from 1 (never) to 9 (more than 1 a day), and CO samples were collected at pre‐ and post‐test. Follow‐up: During grade 8 (approximately 1 yr).
Notes	Quality of intervention delivery: Project staff randomly monitored how much of the material was implemented by the teachers, and assigned an implementation score (material covered in full by 55%), which was used as a covariate in the ANCOVA. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? Not stated. Were appropriate statistical methods used? X² and GLM ANCOVA were used to compare the experimental and control groups.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Each of 29 participating junior high schools was randomly assigned to either receive the psychosocial prevention program or to serve as controls". E‐mail from Dr Botvin 29 January 2012: randomisation by computer. Clusters: Schools. Cluster constraint: Not stated. Baseline comparability: At baseline the intervention group differed from the control in higher % black (P < 0.001), higher % receiving free lunches (P < .0.001), lower grades (P < 0.02).
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Low risk	2,690 at baseline in 7th grade, 2209 (82%) in 8th grade; smokers had higher attrition rates (P < 0.0001), but there was no differential attrition across groups.
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA   Site: 29 inner city middle schools, New York.   Focus: Universal drug prevention.   Design: Cluster RCT (Group 3: point prevalence).
Participants	Baseline: Botvin 2001: 5222: 3621 intervention, 1477 control (Griffin 2003: 758 identified as at high risk of using drugs from Botvin 2001 study; 426 intervention; 332 control).   Age: Middle school students.   Gender: 53% F Ethnicity: 61% A‐A, 22% H, 6% A, 6% W, 5% mixed or Other. Baseline smoking data: Smoking: Intervention 1.36, Control 1.32 [? per week ? per month ? per year ?]
Interventions	Category: Social Influences and social competence vs control. Programme deliverer: Teachers. Intervention: LST taught drug resistance skills, norms against substance abuse, development of personal and social skills, improved self esteem, managing anxiety, communicating effectively, developing personal relationships, asserting one's rights, and resistance to advertising; main programme of 15 lessons in 7th grade, 10 boosters in 8th grade. Control: Substance abuse curriculum normally provided in NY schools.
Outcomes	Frequency of smoking from 1 (never) to 9 (more than once a day); quantity of smoking from 1 (none) to 8 ( > 2 packs a day); CO samples at pretest. Follow‐up: 3m (end of 8th grade),1 yr after first post‐test.
Notes	Quality of intervention delivery: Staff randomly monitored protocol adherence in classrooms (8 teachers monitored 167 times); average number of programme points covered = 48% (SD = 19.8), (compared to 68% in the Botvin 1990 implementation of the LST programme). Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Not stated. Was a correction for clustering made? Yes. Were appropriate statistical methods used? X², GLM ANOVA; and generalized estimated equations independent (PROC GENMOD in SAS); regression analyses.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Prior to randomisation, schools were surveyed and divided into high, medium, or low smoking prevalence. From within these groups, each of the 29 participating schools were randomised to either receive the intervention (16 schools) or be in the control group (13 schools)". Email from K Griffin 24 Jan. 2012: randomisation "was done by computer". Clusters: Schools. Cluster constraint: Grouped according to smoking prevalence. Baseline equivalence: No statistical differences on any substance use variables, but more Blacks in experimental (68%) than control (54%) group (P < 0.001), more Hispanic students in control (31%) than experimental (19%) (P < 0.001) and more students receiving free lunch in experimental (68%) than control (58)%) (P < 0.001) so all 3 included as covariates in regression analyses.
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	High risk	5222 7th graders, of whom 3621 (69%) received intervention ‐ 2144 completed pretest, intervention and provided data at the one‐yr follow‐up. "Significant differences were found in attrition rate according to pretest substance use, with smokers (F[1,5218] = 23.2, P < 0.0001), drinkers (F[1,5218] = 12.0, P < 0.0005), and those who use marijuana (F[1,5218] = 42.3, P < 0.0001), having higher attrition rates that that [sic] of non users".
Selective reporting (reporting bias)	Low risk	No selective publication

Methods	Country: Canada   Site: 6 school boards in SW Ontario; 30/35 schools participated.   Focus: Tobacco prevention.   Design: Cluster RCT (Group 1: never smoking prevention cohort).
Participants	Baseline: 2776 (out of 3028 students).   Age: Grade 8 (age 13 ‐ 14).   Gender: 50% F Ethnicity: Not stated Baseline smoking data: Smoking at end of Grade 8: intervention group 16%; control 18%.
Interventions	Category: Other interventions vs control. This intervention did not align with the main five categories; the programme intervenes by creating school anti‐smoking activities. Programme deliverer: Teachers and students. Intervention: A teacher in each school facilitated students and staff to participate in as many activities as possible inconsistent with smoking, build commitment to nonsmoking, and strengthen nonsmoking as a school norm. Co‐interventions not ascertained. Control: "usual care", not described further.
Outcomes	Outcomes: Intervention activities in each grade. Self reported never smoking, tried once, quit, experimental smoker [smoking < once a week]; and regular smoker [smoking weekly]. CO samples collected but not analysed. Follow‐up: Grade 10 (approximately 2 yrs).
Notes	Quality of intervention delivery: Adequate activities occurred: 3.8 intervention activities in Grade 9 and 3.5 in Grade 10. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Yes. Was a correction for clustering made? Yes. Were appropriate statistical methods used? Analysis of paired clusters using a variance term appropriate to the randomisation of schools.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The 30 schools were matched within school board (by size, number of elementary school cohort students projected to attend, and proportion of cohort students from the elementary school control condition), and then randomised within pairs to intervention or control conditions". Email from E Brown 18 Jan 2012: "one school from each matched pair was assigned to intervention condition via coin flip". Clusters: Schools Cluster constraint: Pair‐matched. Baseline comparability: No significant baseline differences in Grade 8 baseline smoking status, social models risk score or elementary school risk of smoking; but intervention schools included marginally higher proportion of children who had been in an elementary intervention group in Cameron 1999 study (P < 0.10).
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Collectors blinded to assignment.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	2776 at baseline in grade 8. "Data were provided by 2,643 students (95.2% of those who consented) at the end of Grade 10, with no differential attrition across conditions, and no difference between dropouts and the retained sample in gender ratio, Grade 8 smoking status, elementary study condition, or Grade 8 social models risk score".
Selective reporting (reporting bias)	Low risk	No selective reporting

Methods	Country: USA Sites: 10 elementary schools, north of Seattle (10/25 selected). 'Raising Healthy Children Project'. Focus: Reducing students’ likelihood to use alcohol, marijuana or cigarettes and altering the frequency at which students use alcohol, marijuana or cigarettes. Design: Cluster RCT (Group 2: change rates).
Participants	Baseline: Year 1 = 938 (1230 eligible), Year 2 = additional 102 from new intake (131 eligible); Age: Grades 1and 2; mean age 7.7 yrs. Gender: 54% M Ethnicity: 82% European American, 7% Asian/Pacific Islander, 4% A‐A, 4% H, 3% N‐A. Baseline smoking data: No data until Grade 7.
Interventions	Category: Social influences and social competence vs control (school and family). Programme deliverer: Study co‐ordinator, staff, student peers. Intervention: One intervention with 4 strategies to deter substance use: School intervention strategies ‐ Teacher and staff development workshops which provide proactive classroom management techniques: co‐operative learning, motivational strategies, participation, reading, interpersonal and problem‐solving skills. Plus monthly coaching sessions 1:1 to reinforce techniques. Workshops in year prior to recruitment – all staff to have at least 6 workshops. Annual booster sessions. ½ day observation of other staff. All staff observed 6 times during year by independent raters to ensure on track. Individual student intervention strategies: Volunteer student involvement in after‐school tutoring and study clubs in grades 4 ‐ 6. Peer Intervention strategies: Classroom instruction. Annual summer camps. Social skills booster retreats. Family intervention strategies: Multiple session parenting workshops or in‐home service for selected families during grades 1 – 8. Grade 8+ booster sessions delivered at home – tailored to needs of student and family (this was mailed to families who had moved out of the area). Control: No intervention stated.
Outcomes	Annual cigarette use: previous month and year; self reported; cigarette 6 point scale: 0 (no use) to 5 (more than 40 cigarettes a day). Follow‐up: Intervention from recruitment until grade 10. Data collection from grade 6 ‐ 10 (ages 11 ‐ 16). Grades 6 ‐ 9 complete group and 1:1 surveys during school hrs (those not in school completed by visit, mail or phone); Grades 9 ‐ 10 complete 1:1 interviews recorded directly on to computer.
Notes	Quality of intervention delivery: Over 94% of eligible teachers and staff in intervention schools attended the workshops with mean attendance of 5.7 sessions. 27% of intervention students attended study clubs. 40% of intervention students attended school retreats or workshops. 51% attended summer camps. 51% of intervention families attended at least one workshop (3 per year available). 35% of intervention families received individual contact. 77% of intervention families received at least one booster workshop. All intervention students and their families received at least one intervention component with overall mean of 28.3 contacts received by students and 12.6 by their families. Statistical quality: Was a power computation performed? No. Was an intention‐to‐treat analysis performed? Yes. Was a correction for clustering made? No. Were appropriate statistical methods used? Multiple regression latent growth models, with intervention status and background factors as covariates.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Of the 25 elementary schools in the district, the 10 schools that ranked the highest in an aggregate measures of risk (e.g., low income status, low standardized achievement test scores, high absenteeism, high mobility) were selected into the study. Schools were matched on these risk factors, and one school from each matched pair was assigned randomly to either an intervention (n = 5) or control (n = 5) condition". Exclusion of students who did not remain in the school for the first year of the intervention. No method of randomisation. Clusters: Schools Cluster constraint: Ranked and pair‐matched. Baseline comparability: Missing outcome data for N = 81; more female 9.8% than male (6.0%), P < 0.05 and logistic regression showed no difference in missing data between intervention and control. No comment on imbalances in smoking or smoking‐related factors. Email from E Brown 19 December 2011 "baseline for the Raising Healthy Children Project was when students were in Grades 1 and 2; therefore, technically we did have baseline equivalency (no smokers that young). However, to your point, analyses of these and other data indicated that students in intervention schools did not differ significantly between intervention and control schools on variables considered to be related to antisocial behaviours".
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	"To maintain confidentiality, students' parents, teachers, and other school personnel were not present and did not participate in any student data‐collection activities. All students were informed that their responses would not be shared with their parents or other school personnel". [we interpreted this as assuring confidentiality, but did not constitute blinding of participants or researchers to intervention status]
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Final pretest sample = 959 (92%) "77 excluded from analysis as missing substance use data for grades 6 ‐ 10, and 4 excluded as questionable validity of data". Retention rates in grade 6 ‐ 10 were all greater than 88%. No differential attrition between groups
Selective reporting (reporting bias)	Low risk	Two outcomes reported as stated.

Methods	Country: 1) Australia 2) America. Sites: 1) 25 secondary schools in Victoria and New South Wales (13 intervention, 12 control). 2) 21 middle schools in Colorado and New Mexico (10 intervention, 11 control). 'Consider This' Focus: Reduce 30‐day smoking prevalence. Design: Cluster RCT, internet‐based intervention (Group 1: never smoking prevention cohort).
Participants	Australia: Baseline: 2077 Age: Grades 7,8,9 (11 ‐ 14 yrs old). Gender: 48.3% M Ethnicity: Australian/European ancestry = 73.4%. Non‐European ancestry = 17%. Mixed ancestry = 7.4%. Baseline smoking data: 58.4% never smoked. America: Baseline: 1233 Age: Mostly years 6 and 7 (11 ‐ 13 yrs old). Gender: 48% M Ethnicity: W 55.8%; H 23.9%; A‐A 3.4%; N‐A 1.6%. A 3.9%. Native Hawaiian = 0.6%. Other 7.2% Baseline smoking data: 80.1% never smoked.
Interventions	Category: Social influences vs control. Programme deliverer: On‐line web‐based programme. Intervention: 73 online activities organised into 6 modules (Introduction, Media Literacy, Relationships, Mind and Body, Decision Making, and Resistance strategies). Programme aim to convince those who had not smoked not to start and persuade those who had already tried smoking to stop. Programme progression controlled by teachers. After pretest teachers ran 'Consider This' in computer lab classes, each session lasted 45 ‐ 60 minutes (first half of school year). Control: Standard health education.
Outcomes	Primary outcome: 30‐day smoking prevalence (number of days in the past month in which they smoked at least a whole cigarette). All students asked if they had ever smoked, even a puff. Those who had not were classified as nonsmokers and given a value of zero for the 30‐day prevalence. Remaining students classified as: Former smoker – not smoked in previous 30 days. Current experimenters – smoked in the previous 30 days. Secondary outcome: Completion of Pierce et al’s 3 susceptibility items and question on future likelihood of smoking. Follow‐up: For both trials at end of school yr.
Notes	Quality of intervention delivery: IT difficulties meant variation in time between testing and post‐test, loss of schools in the American trial, breaking of some matched pairs. In new Mexico the trial was delayed by one year. Australia: Children completed 43.2 out of 73 activities (59%) Only 26% of students completed at least 90% of activities. America: Children completed 46.6 out of 73 activities (64%). Only 24.8% of students completed at least 90% of activities. 83.1% of children completed module 1. Matching of schools failed because of teachers and IT problems. Statistical quality: Was a power computation performed? Yes, but not stated. Was an intention‐to‐treat analysis performed? No. Was a correction for clustering made? Yes by linear mixed models. Were appropriate statistical methods used? Descriptive statistics of participant characteristics. Linear mixed models. Bivariate linear mixed models to examine associations between outcome measures and potential covariates. Multivariate analysis to focus on significant predictors from the bivariate model.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Schools approached directly or via districts Schools paired on location, size, proportion of female, minority, and Hispanic students, and proportion of students who received free or reduced‐fee meals as an indicator of socioeconomic status of the catchment areas (American trial only). One school in each pair was assigned at random to the intervention group. Email from D Buller 19 December 2011 "Our project statistician used a computer program to randomise them after matching". Clusters: Schools. Cluster constraint: Paired matching based on number of factors. Baseline comparability: One Australian school enrolled without a match. Control group in US study had more children (78.3%) than experimental (83.2%) who had never smoked, but nonsmokers. (P = 0.92) "No significant differences in individual drop‐outs based on treatment group".
Allocation concealment (selection bias)	Unclear risk	No statement
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No statement
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Australia: Analysable sample = 1510 (intervention =754). Only 73% of original sample completed both pretest and post‐test. Large drop out: 207 due to classes withdrawing because of IT issues. Remainder largely due to timing of post‐test as number of students doing activities outside school. No significant differences in drop‐outs based on treatment group, gender, race/ethnicity or home language. America: Analysable sample = 1004 (intervention = 640). 82% of the original sample completed both pretest and post‐test. Half of drop out due to IT, remainder mostly absent. No differential attrition between groups.
Selective reporting (reporting bias)	Low risk	All outcomes clearly expressed.