Tobacco smoking is a major preventable cause of death in both developed and developing countries. Every day over 13,000 people die from tobacco-related diseases (WHO 2004). If current trends continue, by 2025 tobacco will contribute to the death of 10 million people worldwide each year, with seven million of these deaths occurring in developing countries (Mackay 2006). People who smoke are more prone to developing various types of cancer, such as those of the oral cavity, larynx, bladder and particularly lung cancer. Tobacco smokers are also at substantially increased risk of developing heart disease, stroke, emphysema and other fatal diseases (WHO 2004). Smoking also imposes a huge economic burden on society - currently up to 15% of the total healthcare costs in developed countries (Parrot 2004). Additionally, passive smoking is associated with serious morbidity (SCTH 1998). To reduce the growing global burden of tobacco-related mortality and morbidity, and the impact of tobacco use on economic indicators, tobacco control has become a world-wide public health imperative (WHO 2004).
Prevention and cessation are the two principal strategies in the battle against tobacco smoking. Nicotine is highly addictive (Surgeon General 1988). There is evidence that although 70% of US smokers say they want to quit, only five per cent are able to sustain cessation for one year (Schroeder 2002). The balance between the individual's motivation to stop smoking and his or her dependence on cigarettes influences smoking cessation success. Dependence in smokers and their motivation to stop smoking can be assessed by simple questions (West 2004). Maximizing the delivery of smoking cessation interventions can achieve more in terms of years of life saved and economic benefits than most medical interventions for smoking-related illnesses (Coleman 2004).
There is good evidence for the effectiveness of brief, therapist-delivered interventions, such as advice from a physician (Stead 2008b). There appears to be additional benefit from more intensive behavioural interventions, such as group therapy (Stead 2005), individual counselling (Lancaster 2005a) and telephone counselling (Stead 2006). However, these more intensive therapies are usually dependent on a trained professional delivering or facilitating the interventions. This is both expensive and time-consuming for the health providers, and often inconvenient to the patient, because of lengthy waiting times and the need to take time off work. Another major limitation of these more intensive types of interventions is that they reach only a small proportion of those who smoke.
It is estimated that in 2009, there were 1.73 billion Internet users worldwide (Pingdom 2010), and the number of Internet users is likely to increase rapidly over a relatively short time-frame (Modis 2005). The Internet has the potential to deliver behaviour change interventions (Japuntich 2006; Strecher 2006; Swartz 2006; Graham 2007). Internet-based material is an attractive intervention tool, because of relatively low costs per user, resulting in high cost-effectiveness (Swartz 2006). The Internet can be accessed in people's homes, in public libraries and through other public Internet access points, such as Internet cafes and information kiosks. The Internet is available 24 hours a day and 365 days a year, even in areas where there are not the resources for a smoking cessation clinic (such as some rural or deprived areas and low-income countries). Online treatment programmes are convenient from the users' perspective, because content can be accessed at any time, and they also offer a greater level of anonymity than in-person or phone-based counselling. They have the potential to reach audiences who might not otherwise seek support, because of limited health care provision or possible stigmatisation. Existing smoking cessation services such as advice from health professionals and NRT are under-utilized by young people (Rodgers 2005). Internet use by young people has grown exponentially and has a powerful role in influencing youth culture. It may therefore reach a target population of young people who smoke more effectively than the more traditional providers are able to do.
The Internet is a promising vehicle for delivering smoking cessation treatment either as a stand-alone programme or as an adjunct to pharmacotherapy (Swartz 2006; Graham 2007). According to the Pew Internet & American Life Project (Fox 2005), seven per cent of adult US Internet users, (approximately eight million people), reported having searched online for information on 'how to quit smoking'. In the USA, 18% of those with less than high school education searched the web for information on how to quit smoking, which represents a higher proportion than those with more education (Fox 2005).
Materials tailored for individual smokers are more effective than untailored ones, although the absolute size of the effect is still small (Lancaster 2005b). Internet programmes can be highly tailored to mimic the individualization of one-to-one counselling. A web-based programme that collects relevant information from users and tailors the intervention to their specific needs had significant advantages over a web-based, non-tailored cessation programme (Strecher 2005). A Dutch study explored existing self-help materials which are currently available in the Netherlands, and found them to be ineffective for smoking cessation. However, this study suggested that computer-tailored interventions could potentially be successfully designed, and may be a promising means of communicating information on smoking and cessation (Dijkstra 1999). Another study on the efficacy of web-based tailored behavioural smoking cessation materials for nicotine patch users showed that participants in the tailored programme reported significantly higher 12-week continuous abstinence rates (22.8%) than those in a non-tailored programme (18.1%) (Strecher 2005).
Using the Internet for smoking cessation programmes may also have limitations. There are a large number of smoking cessation web sites, but they do not all provide a direct intervention. Some studies of popular smoking cessation web sites and their quality (Bock 2004; Etter 2006) suggest that smokers seeking tobacco dependence treatment online may have difficulty discriminating between the numerous sites available (Etter 2006). In addition, web sites that provide direct treatment often do not fully implement treatment guidelines and do not take full advantage of the interactive and tailoring capabilities of the Internet (Bock 2004). Furthermore, a study on rates and determinants of repeat participation in a web-based health behaviour change programme suggested that such programmes may reach those who need them the least. For example, older individuals who had never smoked were more likely to participate repeatedly than those who currently smoke (Verheijden 2007). The Internet is also less likely to be used by people on lower income, who are more likely to smoke (Eysenbach 2007; Kontos 2007).
There have been two recent reviews of this area (Myung 2009, Shahab 2009). Myung 2009 evaluated both web-based and computer-based cessation programmes. The review included 22 trials in a single random effects meta-analysis that showed a significant effect on cessation; similar effects were also seen in a range of subgroups including one limited to nine trials that used a web based intervention. They concluded there was evidence to support the use of web and computer-based cessation programmes for adults who smoked, but not adolescent smokers. Shahab 2009 focused on interactive online interventions, and also sought to identify treatment effect moderators and mediators. The review included eleven randomized trials. The authors concluded that web based tailored and interactive interventions increased abstinence compared to booklet or email control, based on three trials. Both reviews identified large amounts of heterogeneity in both study designs and effect sizes.
To determine the effectiveness of Internet-based interventions for smoking cessation.
Criteria for considering studies for this review
Types of studies
Randomized or quasi-randomized controlled trials. Examples of quasi-random methods of assignment include alternation, date of birth, and medical record number.
Types of participants
Smokers who participated in Internet interventions for smoking cessation, with no exclusions on the basis of age, gender, ethnicity, language spoken or health status. Studies on adolescents and young adults were analysed separately from the general population studies, as both subgroups are important subgroups with particular needs, which warrant separate investigation.
Types of interventions
We included Internet studies in all settings and from all types of provider. There was no exclusion with respect to intervention method or duration. We included trials where the Internet intervention was evaluated as an adjunct to a pharmacotherapy such as nicotine replacement therapy (NRT), bupropion or varenicline, but only where the Internet component was the intervention being tested. The trials compared different types and combinations of intervention. The trials compared Internet-based programmes with no treatment or with other forms of treatment, such as self-help booklets. We included trials of interactive, personalised and non-interactive interventions, which focused on standard approaches to information delivery. Interactive interventions were not necessarily personalised.
Personalised interventions can vary considerably, from minimal personalisation to those which have been developed based on theoretical models which are relevant to desired treatment outcomes, such as self-efficacy. The interventions used in each study were fully described, as the heterogeneity of the interventions (for example, in relation to varying content, intensity, number of sessions, duration of contact time).
We excluded trials which used the Internet solely for recruitment and not for delivery of smoking cessation treatment. We also excluded trials where Internet-based programmes were used to remind participants of appointments for treatment that is not conducted online, e.g. face-to-face counselling, or pharmacotherapy. Text messaging interventions were covered in a Cochrane review of mobile phone interventions (Whittaker 2009) and are not covered in this review.
Types of outcome measures
The primary outcome was smoking cessation at least six months after the start of the intervention, and longer wherever the data were available. In order to assess short-term cessation we included trials with shorter follow-up period, but the gold-standard was six months' cessation. We excluded trials with fewer than four weeks follow up. We preferred sustained or prolonged cessation over point prevalence abstinence, but did not exclude studies which only reported the latter. We included studies which relied on self-reported cessation, as well as those which required biochemical validation of abstinence.
Where reported, we extracted data on user satisfaction rates for the Internet intervention compared to no or alternative interventions.
Search methods for identification of studies
We searched the specialised register of the Cochrane Tobacco Addiction Group, using the following terms: ('Internet', 'web$', 'www$', 'online', 'net', 'web-based', 'interventions') in the title, abstract or as keywords. The most recent search of the register was June 2010.This register has been developed from electronic searching of MEDLINE, EMBASE, Science Citation Index and PsycINFO, together with hand searching of specialist journals, conference proceedings, dissertations and reference lists of previous trials and overviews. We also performed ad hoc searches of a number of electronic databases, including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, PsycINFO and CINAHL, using the exploded MeSH terms which included: 'smoking cessation', 'tobacco cessation', 'Internet' , 'computer' and 'online'. We also searched Google Scholar, and checked the first 500 records retrieved.
We searched online trials registers, e.g. Controlled Clinical Trials (www.controlled-trials.com), the National Research Register (www.nrr.nhs.uk), US registries (clinicaltrials.gov), and WHO registries (www.who.int/trialsearch/) for ongoing and recently completed studies.
We searched through the reference lists of identified studies for other potentially relevant trials. We contacted authors and experts in this field for unpublished work.
There were no restrictions on language or date of publication.
Data collection and analysis
Selection of studies
Two authors independently assessed potentially relevant papers for inclusion. Disagreements were resolved through discussion between the two authors responsible for screening. This was done by each author writing their reasons for inclusion/exclusion until a consensus was reached. An arbiter was not used at this stage of initial screening selection. Reasons for exclusion were noted.
Data extraction and management
Two authors independently extracted data; the first author then checked data and compared the findings. This stage included an evaluation of study quality. An arbiter (JC) was used where there was disagreement amongst two selecting authors over the further exclusion of studies at this stage.
We extracted the following information from each trial:
- Country and setting
- Method of selection of participants
- Definition of smoker used
- Methods of randomization (sequence generation and allocation concealment), and blinding of trialists, participants and assessors
- Demographic characteristics of participants (e.g. average age, sex, average cigs/day)
- Proportion of site visitors who are actively trying to quit smoking vs information seeking
- Intervention and control description (provider, material delivered, control intervention if any, duration, level of interactivity etc)
- Outcomes including definition of abstinence used, and biochemical validation of cessation
- Proportion of participants with follow-up data
- Any harms or adverse effects
- Sources of funding
Assessment of risk of bias in included studies
Two independent authors assessed each study according to the presence and quality of the randomization process, whether or not trialists and assessors were 'blinded', whether the analysis was appropriate to the study design, and the description of withdrawals and drop-outs, whether the baseline measurements were comparable and outcome measures valid.
Measures of treatment effect
We calculated a risk ratio (RR) for the outcome for each trial, defined as (number who stopped smoking in the intervention group / total number randomized to the intervention group) / (number who stopped smoking in the control group / total number randomized to the control group). We aimed to conduct an intention-to-treat (ITT) analysis, i.e. including all those randomized to their original groups, whether or not they remained in the study. We treated drop-outs or those lost to follow up as continuing smoking. A risk ratio greater than one indicates that more people stopped smoking in the intervention group than the control group. We displayed risk ratios with 95% confidence intervals in forest plots.
Assessment of heterogeneity
We considered clinical, statistical and methodological heterogeneity. We assessed statistical heterogeneity using the I² statistic, which assesses the proportion of the variation between studies is due to heterogeneity rather than to chance (Higgins 2003). Values over 50% suggest substantial heterogeneity, but its significance also depends upon the magnitude and direction of the effect, and the strength of the evidence (as estimated by the confidence interval or p value).
Due to heterogeneity of design and variable quality overall and within prespecified subgroups, only very limited meta-analysis of the studies available for inclusion at this time, was possible. We have presented the results in a narrative form, and displayed the cessation outcomes for each study graphically. We separated trials in adolescents from those in young adults and older adults. We would have also considered studies in pregnant women separately. We also distinguished between tailored and non-tailored interventions. In the absence of heterogeneity we would have estimated a pooled weighted average of RRs, using the Mantel-Haenszel fixed-effect method, with a 95% confidence interval. We would also have used funnel plots to help identify possible biases.
Had it been appropriate to use meta-analysis we planned to use sensitivity analyses to investigate the impact on the estimated treatment effect of excluding from the meta-analysis any trials of questionable design, methodology or outcome measures.
Description of studies
We identified 71 potentially relevant records. One paper reported on four studies of an Internet-based intervention of which two were eligible RCTs (Munoz 2006 Study 3; Munoz 2006 Study 4). Twenty studies met the inclusion criteria.
Triallists were mostly based in the USA and recruited participants based there. One trial was conducted in Switzerland, one in Norway, one in the Netherlands, one in England and one in the Republic of Ireland. The studies by Munoz and colleagues recruited from multiple countries.
Recruitment was mainly web-based; participants found the sites through search engines and browsing. Several trials used press releases, billboards, television advertisements and flyers in addition to the web-based recruitment. Therefore, participants included in these trials were smokers motivated to quit smoking, who chose the Internet as a tool for smoking cessation support. Clark 2004 recruited people undergoing chest computerized tomography (CT) as a screening assessment for lung cancer at their first follow-up visits. Strecher 2008 recruited from members of two health management organizations (HMOs) participating in the National Cancer Institute's Cancer Research Network, Swan 2010 recruited participants from a large healthcare organization. Sixteen studies recruited a full adult age range. An 2008 recruited young adult college students and three studies recruited adolescents (Woodruff 2007; Mermelstein 2006; Patten 2006). Sample sizes ranged from less than 150 (Mermelstein 2006; Woodruff 2007) to nearly 12,000 (Etter 2005). In some studies, participants were offered financial compensation for completing assessment surveys and for biochemical analysis (An 2008, Munoz 2006 Study 3; Munoz 2006 Study 4; Mermelstein 2006; Te Poel 2009; Woodruff 2007).
Descriptions of the main features of each study intervention are provided in the Characteristics of included studies table.
A range of types of Internet interventions was tested in the included studies, from a very low intensity intervention, providing a list of web sites for smoking cessation (Clark 2004), to highly intensive interventions consisting of Internet, email and mobile phone delivered components (Brendryen 2008a; Brendryen 2008b). Tailored Internet interventions differed by the amount of tailoring, from a bulletin board facility (Stoddard 2008), a multimedia component (McKay 2008), tailored and personalised access (Strecher 2005; Rabius 2008) to very high-depth tailored stories and a highly personalized message sources (Strecher 2008 ). More details are given under comparisons section below.
Sixteen studies assessed smoking status at least six months after the start of the intervention (Clark 2004; Japuntich 2006; Mermelstein 2006; Patten 2006; Munoz 2006 Study 3; Munoz 2006 Study 4; Woodruff 2007; An 2008; Brendryen 2008a; Brendryen 2008b; McKay 2008; Rabius 2008; Strecher 2008; Munoz 2009, Swan 2010;Te Poel 2009). All of these except Strecher 2008 and Te Poel 2009 also assessed smoking at one or more intermediate follow-up points. Four studies followed participants for less than six months (Etter 2005; Strecher 2005; Swartz 2006; Stoddard 2008). Studies reported a range of definitions of abstinence at the time of follow up. Where studies reported abstinence rates for more than one definition we displayed the effect using the most conservative outcome [with the exception of An 2008, see below]. For most studies, seven-day smoking abstinence was the main outcome measure (Clark 2004; Etter 2005; Japuntich 2006;McKay 2008; Munoz 2006 Study 3; Munoz 2006 Study 4; Munoz 2009; Stoddard 2008;Strecher 2008; Swartz 2006; Te Poel 2009). A few studies reported 30-day self-reported smoking abstinence (Mermelstein 2006; Patten 2006; Swan 2010). One study assessed six-month prolonged abstinence from smoking (An 2008); this study also reported seven-day and 30-day prevalence abstinence. We used 30-day rates as our primary outcome because the programme did not involve setting a quit date, and the prolonged abstinence was based on self report of time since last cigarette rather than repeated assessments of abstinence. Three of the four short term studies assessed self-reported point prevalence abstinence at three-month follow up only (Etter 2005; Swartz 2006; Stoddard 2008) whilst Strecher 2005 assessed 28-day continuous abstinence rates at six-week follow up, and 10-week continuous abstinence rates at 12-week follow up. In one study, seven-day smoking abstinence was a secondary outcome, while time spent on the website, utilization of pages, cessation aids used in the past and during the study period were the main outcome measures (Stoddard 2008).
Due to the limited face-to-face contact and due to data collection through Internet or telephone interview, biochemical validation to confirm self-reported smoking abstinence was conducted in only five trials (Clark 2004; Japuntich 2006; Mermelstein 2006; Patten 2006; An 2008). All these measured carbon monoxide (CO) in expired air.
Utilization of the Internet site or programme use was a secondary outcome measure in 10 studies (Clark 2004; Japuntich 2006; Mermelstein 2006; Swartz 2006; Brendryen 2008b; McKay 2008; Rabius 2008; Strecher 2008; Munoz 2009; Swan 2010). Satisfaction with treatment condition was assessed in five trials (Strecher 2005; Woodruff 2007; Stoddard 2008; Munoz 2009; Te Poel 2009). Use of NRT or other pharmacotherapies was a secondary outcome measure in five trials (Patten 2006; Brendryen 2008b; McKay 2008; Strecher 2008; Swan 2010). Two of the studies in adolescents assessed reductions in the number of cigarettes or in smoking frequency as secondary outcomes (Patten 2006; Woodruff 2007). Munoz 2009 reported the impact of method of obtaining follow-up data on quit rates; comparing phone with online follow-up procedure.
We first grouped studies in subgroups according to whether (1) they compared an Internet intervention with no intervention or a non-Internet intervention, or (2) compared a more complex or tailored Internet intervention with a less complex one. At the suggestion of peer reviewers we added a third comparison that included studies in which the intervention offered access to an interactive website and the control could be either a static website, or a control without Internet access. In five trials, all participants were using, or were offered, pharmacotherapy (Strecher 2005; Japuntich 2006; Brendryen 2008b; Strecher 2008; Swan 2010) and the Internet component was thus being evaluated as an adjunct to pharmacotherapy. These were grouped in comparisons based on the nature of the Internet component and the control.
Internet intervention compared to no Internet intervention or no intervention at all
Ten trials compared an Internet intervention to a non-Internet based smoking cessation intervention or to a no intervention control (Clark 2004; Japuntich 2006; Mermelstein 2006; Patten 2006; Swartz 2006; Woodruff 2007; Brendryen 2008a; Brendryen 2008b; An 2008; Swan 2010). Six trials recruited adults, one targeted young adult university students (An 2008) and three were conducted among adolescents (Mermelstein 2006; Patten 2006; Woodruff 2007).
Of the six trials that targeted adults, two studies (Brendryen 2008a; Brendryen 2008b) evaluated 'Happy Endings', a 1-year programme delivered via the Internet and cell phone, consisting of more than 400 contacts by email, web pages, interactive voice response, and short message service (SMS) technology. Brendryen 2008a recruited people attempting to quit without NRT, whilst Brendryen 2008b offered a free supply of NRT to all participants. Follow up was after 12 months in both studies. Japuntich 2006 evaluated a web-based system incorporating information, support and problem-solving assistance; this was tested as an adjunct to bupropion and brief face-to-face counselling, with follow up after six months. Swartz 2006 recruited participants via worksites to test a video-based Internet site that presented strategies for smoking cessation and motivational materials tailored to the user’s race/ethnicity, sex and age. This study only reported short follow up (i.e. at 90 days); after this time the control group had access to the programme. Clark 2004 tested a very low intensity intervention for smokers having CT lung screening; a handout with a list of 10 Internet sites related to stopping smoking with a brief description of each site compared to printed self-help materials. Follow up was at 12 months. Swan 2010 was a three arm trial comparing an established proactive telephone counselling intervention, an interactive web site based on the same programme, and a combination of phone and Internet components, all providing behavioural support in conjunction with varenicline use.
An 2008 recruited college students who reported smoking in the past 30 days; the sample smoked an average of four cigarettes per day. Intervention group participants received $10 a week to visit an online college magazine that provided personalized smoking cessation messages and peer email support. The control group received only a confirmation email containing links to online health and academic resources. Both groups were informed about a campus-wide Quit&Win contest sponsored by University Health Service.
The three small studies in adolescents recruited populations of relatively light smokers. Mermelstein 2006 evaluated the effectiveness of enhancing the American Lung Association’s Not on Tobacco programme (NOT) with a web-based adjunct (NOT Plus), which included access to a specially designed web-site for teenagers, along with proactive phone calls from the group facilitator to the participant. Twenty-nine high schools were randomly assigned to either the NOT programme alone or to the NOT Plus one. Self-reported smoking behaviour was assessed at the end of the 10-week programme and three months later. Patten 2006 compared a home-based, Internet-delivered treatment for adolescent smoking cessation with a clinic-based brief office intervention (BOI) consisting of four individual counselling sessions. Adolescents assigned to the Internet condition had access to the web-site for 24 weeks and abstinence was assessed at the end of this period. Woodruff 2007 evaluated an Internet-based, virtual reality world combined with motivational interviewing, conducted in real-time by a smoking cessation counsellor. There was a measurement-only control condition involving four online surveys. Smoking status was assessed at baseline, post-intervention and at three and twelve-month follow up.
Comparisons between different Internet interventions
Ten trials compared two or more different Internet interventions (Etter 2005; Strecher 2005; Munoz 2006 Study 3; Munoz 2006 Study 4; Munoz 2009; McKay 2008; Rabius 2008; Stoddard 2008; Strecher 2008; Te Poel 2009)
A series of three studies by Munoz and colleagues evaluated adjuncts to an online resource, the 'Guia', a National Cancer Institute evidence-based intervention first developed for Spanish speaking smokers. In separate English language (Munoz 2006 Study 3) and Spanish (Munoz 2006 Study 4) studies the control condition was the provision of access to the Guia and 'Individually Timed Educational Messages' (ITEMs). The intervention tested was the addition of an online mood management course consisting of eight weekly lessons. Munoz 2009 also used the 'Guia' as the control condition, but in a four-arm design that evaluated the successive addition of ITEMs; the mood management condition used in the Munoz 2006 studies; and a 'virtual group' asynchronous bulletin board. The study recruited English and Spanish-speaking Internet users from 68 countries.
In two trials (Strecher 2005; Rabius 2008) the control condition provided access to a relatively static Internet site whilst one or more intervention conditions provided more tailored and personalised access. Rabius 2008 compared five tailored and interactive Internet services with the targeted, minimally interactive American Cancer Society site providing stage-based quitting advice and peer modelling. Follow-up surveys were conducted four and 13 months after randomization. Strecher 2005 assigned purchasers of a particular brand of nicotine patch to receive either web-based, tailored behavioural smoking cessation materials or web-based non-tailored materials. This study measured smoking behaviour at three-month follow up.
The remaining five studies were distinctive in their designs. Etter 2005 compared the efficacy of two versions of an Internet-based, computer tailored cessation programme; the control group received a shorter version modified for use by those smoking and buying NRT over the counter, although use of NRT was not a condition for enrolment. Follow up was at 2.5 months. Stoddard 2008 evaluated the impact of adding a bulletin board facility to the 'smokefree.gov' cessation site. This study measured smoking behaviour three months after enrolment. Strecher 2008 identified active psychosocial and communication components of a web-based smoking cessation intervention and examined the impact of increasing the tailoring depth on smoking cessation among nicotine patch users. Five components of the intervention were randomized using a fractional factorial design: high versus low depth tailored success story, outcome expectation versus efficacy expectation messages; high versus low personalized source; and multiple versus single exposure to the intervention component. Abstinence was assessed after six months. McKay 2008 compared the Quit Smoking Network (QSN), a web-based tailored cessation programme with a multimedia component, with Active Lives, a web-based programme providing tailored physical activity recommendations and goal setting in order to encourage smoking cessation. Abstinence was assessed after six months. Te Poel 2009 compared tailored to untailored cessation advice letters sent by email, after participants had completed on online survey. We included this study in the review because information was gathered via a website.
Risk of bias in included studies
The majority of studies did not explicitly describe the way in which the randomization sequence was generated or concealed until patient enrolment. In eleven studies, computer randomization was used to assign participants to intervention or control condition (Etter 2005; Strecher 2005; Munoz 2006 Study 3; Munoz 2006 Study 4; Swartz 2006; Brendryen 2008a; Brendryen 2008b; Stoddard 2008; Munoz 2009; ;Te Poel 2009; Swan 2010). Although there was little information about allocation concealment, when investigators used computerized randomization and had minimal interaction with participants we judged there to have been a low risk for selection bias. Munoz 2006 Study 3, Munoz 2006 Study 4, and Munoz 2009, used the baseline questionnaire to automatically implement stratified randomization by gender and major depressive episode (MDE) status (no MDE history, past MDE, current MDE) to the two conditions.
In the two studies (Mermelstein 2006; Woodruff 2007) that randomized schools to conditions there was the potential for bias due to the way in which individual students were recruited once their school was randomized. In both there were differences in the baseline smoking behaviour of intervention and control participants. The two studies also needed to take account of the non-independence of outcomes for students clustered within schools. Mermelstein 2006 used hierarchical linear modelling to allow for clustering. Woodruff 2007 assessed baseline variable intraclass correlations and average cluster sizes. Intraclass correlations were generally small (0.1 or less) and the magnitude of the effect sizes was below two, so analyses were conducted at the individual level without a school-level cluster term.
Incomplete outcome data
All studies included in this review used ITT analysis, reporting analyses based on the total number randomized, with drop-outs and participants lost to follow up classified as smoking. Wherever possible, we have noted the number of participants that completed the study in the Characteristics of Included studies table. In one study (Clark 2004) there were no drop-outs at one-year follow up, as all study participants attended their annual review at that point. Five studies (Japuntich 2006; An 2008; Brendryen 2008a; Brendryen 2008b; Strecher 2008) ascertained smoking status for over 80% of participants at follow up. Five studies ascertained smoking status for 50-80% of participants at follow up (Strecher 2005; Mermelstein 2006; Patten 2006; Swartz 2006; Woodruff 2007; Swan 2010). Eight studies had over 50% loss to follow up (Etter 2005; McKay 2008;Munoz 2006 Study 3 ; Munoz 2006 Study 4, Munoz 2009, Rabius 2008, Stoddard 2008;Te Poel 2009 ). All studies reported similar proportions loss to follow up in each group except in one study where survey non-response was higher among intervention participants then among controls (Woodruff 2007). We tested the sensitivity of our results to the assumption that lost participants were still smoking, by excluding those not reached at follow up from the denominator in the analyses, and report this below.
Effects of interventions
Internet intervention compared to no Internet interventions or no interventions at all
There were six trials in adult populations (Clark 2004; Japuntich 2006; Swartz 2006; Brendryen 2008a; Brendryen 2008b; Swan 2010). Both the Happy Endings trials detected a significant effect of the Internet intervention on sustained abstinence at 12 months whether it was compared to a self-help control (Brendryen 2008a, RR 2.94, 95% CI 1.49 to 5.81) or tested as an adjunct to NRT (Brendryen 2008b, RR 1.71, 95% CI 1.10 to 2.66). Whilst the relative benefit of the intervention was similar, prolonged abstinence at 12 months was higher, 13%, amongst control group participants in Brendryen 2008b, who were offered NRT, than amongst controls who were not given pharmacotherapy (seven per cent in Brendryen 2008a). The authors noted that the relative effects were smaller for the outcome of point prevalence abstinence at 12 months, rather than repeated abstinence, because abstinence increased over time in the control group in both trials. Japuntich 2006 did not detect a significant effect of the Internet component as an adjunct to counselling and bupropion (RR 1.27, 95% CI 0.70 to 2.31) whilst Swan 2010 did not detect an effect of an Internet component either compared to, or as an adjunct to, telephone counselling and varenicline (RR 0.94 95% CI 0.79 to 1.13, combining web only and web plus phone arms). All three conditions achieved similar quit rates, ranging from 27.4% to 30.6% for 30 day abstinence at six months. Clark 2004, which was only a minimal intervention, also did not detect an effect (RR 0.45, 95% CI 0.14 to1.40). Relative effects were similar in these trials at shorter follow-up points. One study with only short-term follow up (Swartz 2006) detected a significant effect of Internet intervention compared to no intervention at all (RR 2.46, 95% CI 1.16 to 5.21).
One study in a population of college students (An 2008) detected a significant effect on 30-day abstinence at 30-week follow up (RR 1.95, 95% CI 1.42 to 2.69) although rates of prolonged abstinence were only six per cent and did not differ between groups.
Patten 2006 compared a home-based Internet delivered intervention (SOS) to a brief office intervention (BOI) for adolescent smoking cessation, and did not detect a difference in abstinence. Rates at 24 and 36 weeks follow up were higher for BOI (RR 0.44, 95% CI 0.14 to 1.36 at 36 weeks). Mermelstein 2006 detected a significant effect of the web-based adjuncts to the group-based approaches for adolescent smoking cessation (crude RR 1.96, 95% CI 1.02 to 3.77; also reported as significant, (p < .05), using mixed model logistic regression to account for clustering within schools). Woodruff 2007 recruited eligible adolescents based on a report of smoking in the past month; at baseline some described themselves as 'former' smokers or had not smoked in the past week. Intervention participants had lower past week abstinence rates at baseline than controls (14% vs. 29%). At the post-assessment, they had significantly higher abstinence rates than controls (35% vs. 22%), but by the final 12-month follow up, the two groups had almost identical past-week abstinence rates (RR 0.93, 95% CI 0.60 to 1.44). The interaction term considering all four assessments was not significant. Intervention participants (68%, n = 52) completed a five item questionnaire assessing their satisfaction with the programme immediately after the post-test assessment; 89% of participants reported they would recommend the programme to another person who smoked.
Comparison of different Internet intervention
None of the three Munoz studies detected significant benefits of adding a mood management intervention, and pooling the comparable arms did not make a difference at 12 months (RR 0.90, 95% CI 0.70 to 1.15 ; I² = 51%; Analysis 2.1) or in the shorter term. In fact Munoz 2006 Study 3 found that the more complex intervention (GUIA + ITEMS + MM) yielded significantly lower quit rates at 12 months. All four arms of Munoz 2009 had similar long- and short-term quit rates, ranging from 19.1% to 22.7% at 12 months, with no evidence that the more tailored conditions had any incremental benefit over the static website control. Authors of this trial also assessed user satisfaction with the treatment condition, and reported that there were significant differences in satisfaction across conditions at all follow-up assessments, with groups 3 and 4 generally reporting greater satisfaction at each time point.
Strecher 2005 and Rabius 2008 also compared tailored to static sites but Strecher 2005 used pharmacotherapy and only reported short-term outcomes, and Rabius 2008 compared multiple sites. Rabius 2008 detected no significant differences between a static cessation site and any of the other Internet sites included in the evaluation. Approximately 10% of enrolled participants assigned to the static site reported abstinence after 13 months compared with 8% to 12% among those assigned to any of the five different interactive sites (RR 1.12, 95% CI 0.92 to 1.36 pooling any interactive website versus the control site). At baseline, 30% of participants reported an indicator of depression. Post-hoc analyses found that this subgroup had significantly lower 13-month quit rates than those who did not have signs of depression (8% vs. 12%, P < .001). Amongst the 70% of participants who did not report an indicator of depression at baseline, the more interactive, tailored sites, as a whole, were associated with higher quitting rates than the less interactive American Cancer Society site (13% vs. 10% P = .04). This exploratory analysis suggested that tailored, interactive web-sites might help smokers who do not report the indicator of depression at baseline to quit and maintain cessation. Strecher 2005 reported higher continuous abstinence rates in the tailored condition than the control, as an adjunct to NRT. At 12 weeks, continuous abstinence rates were 22.8% vs. 18.1% respectively (RR 1.26, 95% CI 1.10 to 1.44). Satisfaction with the programme was also significantly higher in the tailored than in the non-tailored program.
McKay 2008 detected no difference at three or six-month follow up between two cessation interventions, one with a focus on physical activity presenting very few strategies for quitting smoking.
Te Poel 2009 detected evidence of a benefit from a tailored email letter compared to a non tailored one, (RR 2.48, 95% CI 1.11 to 5.55). All participants in this study accessed a website once to complete a questionnaire about their smoking.
Stoddard 2008 compared the publicly available version of smokfree.gov, designated as usual care condition (UC), to an identical-looking website that included an asynchronous bulletin board (BB) and found that quit rates for participants in both conditions were similar after three months (RR 0.95, 95% CI 0.64 to 1.40). Satisfaction with the website was high and did not differ significantly between conditions (UC: 90.2%, BB: 84.9%, P= 0.08).
Strecher 2008 compared multiple conditions in a fractional factorial design so results are not displayed in the analyses. Participants could receive up to three high-depth components, addressing efficacy expectations, outcome expectations and success stories, as part of their tailored web based intervention. Tailoring depth was marginally related (p<0.08) to 6-month smoking cessation in ITT analyses and was significantly related to cessation using per-protocol analysis (excluding participants who used other cessation aids during the follow-up period). The adjusted 6-month cessation rates among participants receiving all the three high-depth tailored components was 27.7% in the ITT analysis. There was some evidence that high-depth tailored success stories had a particular influence on participants with lower levels of education although this interaction was not significant in the ITT analysis. Participants reported using an average of 5.1 weeks of their supply of nicotine patches, with 26.7% using the patch for the full 10 weeks.
Comparison of interactive Internet sites with static sites or non Internet controls
From the studies discussed above, eight studies with long term outcomes (Munoz 2006 Study 3; Munoz 2006 Study 4; An 2008; Brendryen 2008a; McKay 2008;Rabius 2008; Brendryen 2008b; Munoz 2009) and another three with only short-term data (Etter 2005; Strecher 2005; Swartz 2006) might be pooled to assess whether sites that give smokers content tailored to their needs and interests are more useful than a control. However we found that the level of statistical heterogeneity was too large for a pooled estimate to be valid (I² = 75% for long term outcomes, Analysis 3.1, I² = 76% for short term outcomes, Analysis 3.2).
The Internet, with all its richness of options and opportunities for communication and sharing information, has now become a regular part of daily life for the majority of people in many countries. Therefore it is appropriate to consider using it as a tool to increase choice and access to smoking cessation support. Online treatment is convenient in that it can be accessed anywhere, at any time; it also offers the option of anonymity. For healthcare providers it has the potential for being very cost-effective if provided as an automated service. Internet interventions for smoking cessation can be provided in conjunction with other cessation support such as individual or group counselling and NRT or other pharmacotherapy.
We identified 20 trials yielding data from nearly 40,000 participants. Despite this large volume of data, this remains a relatively new field of research, with all trials published since 2004. The studies included in this review assess ways to help people quit smoking using a variety of Internet cessation programmes. Trials varied in the interventions studied and in the duration of the interventions; they also varied in the timing of follow-up assessments and the way in which abstinence was defined.
There is only a small number of studies comparing the long-term effects of Internet to a no-Internet or no intervention control. The results of these studies have been mixed so there is as yet only limited evidence to support a long-term effect of programmes that use the Internet.
Two studies assessed whether an Internet intervention was more effective than a self-help booklet for smoking cessation (Brendryen 2008a; Brendryen 2008b). These have shown evidence of both short- and longer-term effect with up to one-year follow up, but the programme delivered to the treatment group was fully automated, relatively intensive and proactive, and was delivered via both the Internet and mobile phone. The three other studies reporting long-term results did not detect a benefit, but these were relatively less intensive programmes with fewer proactive elements to encourage programme use. One additional short-term study providing tailored online cessation videos did detect an increase in abstinence at three months (Swartz 2006).
When considering trials comparing different intensities of Internet support, there is some evidence of a short-term beneficial effect of individually tailored Internet programmes compared to static web-sites or non-tailored programmes. For example, Etter et al found that an original more tailored programme versus a modified and less tailored programme increased quit rates and helped maintain short-term abstinence (Etter 2005). Strecher et al found that continuous abstinence rates at six and 12 weeks were significantly higher in the tailored intervention than in the non-tailored intervention (Strecher 2005). In contrast, however, one study found there was no significant difference in abstinence rates between participants assigned to interactive sites and participants assigned to static sites, even though interactive sites led to higher utilization of the websites ( Rabius 2008). There are also reported benefits of interactive sites with regard to satisfaction of the users (Munoz 2009; Strecher 2005; Woodruff 2007). One form of tailoring is to encourage website use by sending tailored email messages. Three studies by Munoz and colleagues used these as a component of an Internet programme, but none of them evaluated the effect compared to a non-Internet control, and only Munoz 2009 used a static site as a control; this study failed to detect any additional benefit of this particular tailoring. All three studies evaluated the effect of adding a mood management component and none detected any evidence that this was helpful, and the trend in two studies was for a reduction in success. In a post hoc analysis we explored pooling studies that evaluated tailored, interactive Internet based interventions, whatever the nature of the control, but found there to be too much between study heterogeneity in effect sizes.
Pharmacotherapies such as NRT and bupropion can help people who make a quit attempt increase their chances of success (Hughes 2007; Stead 2008a). Face-to-face behavioural support has an independent benefit but many people are not willing to attend or cannot access group based programmes or multi-session counselling. Internet programmes can be used to deliver additional behavioural support to people using pharmacotherapy, but there is so far little evidence of an effect of Internet intervention in addition to pharmacotherapies. Both Japuntich 2006 and Swan 2010 failed to detect any benefit, although Strecher 2005 suggested a short term advantage of a tailored site over a static one. Strecher 2008 identified components of tailored sites that could assist cessation.
Studies conducted among adolescents and young adults vary not only in the Internet intervention used, but also in methodology and setting. All three trials in adolescents were relatively small. Two suggested that the Internet condition had less benefit than the comparison condition, while one reported a marginally significant benefit sustained after the end of the programme (Mermelstein 2006). In this study, participants in the web arm also had phone calls from a counsellor so the independent effect of the web component is uncertain. These studies show that Internet assistance is attractive and suggest that tailored web sites are more popular among young people (Mermelstein 2006; Woodruff 2007; An 2008).
To achieve success in smoking cessation programmes, interventions must be accessible, efficacious and cost-effective and transportable. Only two of the included studies in this review reported any information about cost-effectiveness of their intervention (Etter 2005, Rabius 2008). Etter et al estimated that the total cost of implementing the website, for a reach of 8000 participants in computer tailored programmes and for 600 000 visitors per year to the website, is comparable to the cost of running a small smoking cessation clinic which would treat about 50 smokers a month. Therefore, Internet services provide greater potential for cost-efficiency because they can provide assistance to many smokers at a very low cost. Internet interventions can also be delivered alongside other more traditional smoking cessation programmes, providing smokers who are motivated to quit smoking with different tools which increases their overall choice (Etter 2005). Rabius et al found Internet assistance for smoking cessation to be cost-effective, since four days of programming at a cost of less than US $2000 allowed approximately 5000 additional users for services from the five tailored interactive service providers, comparing with the much larger cost of serving 1000 new clients with telephone counselling (approximately US $100,000) (Rabius 2008).
Quality of the evidence and potential biases
The trials enrolling adults generally relied on self-reported data on smoking status. Biochemical validation of self-reported cessation was only attempted in two trials where participants had face-to-face contact during the follow-up visit (Clark 2004; Japuntich 2006). The Society for Research on Nicotine and Tobacco subcommittee on biochemical verification in clinical trials (SRNT 2002) considers that verification is not necessary when a trial includes a large population with limited face-to-face contact, and where the optimal data collection methods are through the mail, telephone, or Internet. There is a recommendation that biochemical verification be used in studies of smoking cessation in special populations, including adolescents (SRNT 2002). Only one of the four studies in adolescents and young people did not use biochemical verification of self-reported abstinence (Woodruff 2007). Four included studies followed participants for less than six months (Etter 2005; Strecher 2005; Swartz 2006; Stoddard 2008). It is hoped that reporting six-month outcomes will become routine (West 2005).
Conducting research via the Internet provides opportunities to generate very large sample sizes, but it is also methodologically challenging, because of threats to internal and external validity such as selection bias or differential drop-out (Feil 2003, Cobb 2005). Although there was limited detail about procedures for sequence generation and allocation concealment, we judged that the likelihood of selection bias was small in studies that recruited participants online. Rates of loss to follow up were varied and were high in some large online studies. In our primary analyses we followed the convention of assuming that all those lost to follow up continued to smoke. This can be argued to be a reasonable approach with a volunteer sample and low attrition but if attrition rates are high, or differential across conditions, the assumption may be wrong in some cases and introduce bias (Hall 2001). We undertook a sensitivity analysis ignoring losses to follow up by omitting them from the denominators, and did not find any important differences in relative effects. A range of alternative assumptions about those lost to follow up could be tested, but it is important to recognise that bias in the relative effect will only occur if there the proportion of quitters amongst those lost to follow up differs between intervention and control group.
Determining the contribution of a specific Web site presents a difficult challenge, since Internet users appear to access different sites when searching for information or support to a specific topic. For example, contamination in control groups may be difficult to prevent because of unrestricted access to the Internet, whilst on the other side we cannot be sure that the intervention group is using only the intended intervention (Eysenbach 2002, Feil 2003).
The two other recent reviews in this area drew somewhat less cautious conclusions about the strength of the evidence for the effectiveness of Internet interventions. One review (Myung 2009) pooled data from a number of studies of both web- and computer-based interventions, and concluded that there is now sufficient evidence to support the use of both categories of intervention for adult smokers. Their estimate for Internet interventions based on nine studies, using a random effects model, was RR 1.40, 95% CI 1.13 to 1.72. Shahab et al also suggest that interactive, web-based cessation can be effective in aiding cessation (Shahab 2009), based on 11 studies, all but one of which is included in our review (we were able to include longer term data for Pike 2007, as Rabius 2008). We excluded Prochaska 2008, because we were unable to confirm all data with the authors. Shahab 2009 pools the studies in a number of subgroups; the intervention (tailored/untailored); length of treatment; motivation to quit; and whether the intervention was fully automated or not. They estimated interactive web-based smoking cessation interventions to be effective compared to untailored booklet or e-mail interventions (random effects RR 1.8, 95% CI 1.4 to 2.3), but this was based on just three trials. they also estimated that tailored interventions increase six-month abstinence by 17%. They also suggest that only interventions aimed at smokers motivated to quit were effective (RR 1.3 95% CI 1.0 to 1.7). We consider that our more cautious conclusions are due to a conservative approach to subgroup analyses, and our preference not to pool, even with a random effects approach, when there is evidence of substantial heterogeneity.
Further studies of the long-term effects of Internet based cessation interventions are clearly needed, and there are several ongoing studies in this area (see Characteristics of ongoing studies). Future trials and reviews should include analyses of participants according to sociodemographic data, in order to be able to identify the types of smokers who seek Internet assistance in quitting smoking. Feil et al detected that a large proportion of women was recruited in their study (Feil 2003). According to our findings there is some evidence that females are more interested in smoking cessation programmes delivered via Internet; only three of the included trials reported that more males enrolled (Munoz 2006 Study 3; Munoz 2006 Study 4; Woodruff 2007), and three had equal number of male and female participants (Clark 2004; Patten 2006; Brendryen 2008b).
In the future there may be an interest from patients with depression seeking Internet assistance for quitting smoking. Although there is evidence that depression is an important factor in smoking cessation (Niaura 2002), and that depression inhibits quitting success by decreasing self-efficacy (Haukkala 2000), only a few studies evaluated the impact of Internet interventions among the subgroup of smokers reporting depression. Rabius et al found no overall difference in quit rate among smokers assigned to six experimental groups (five interactive and one static site), but they also found that those who reported an indicator of depression and were assigned to interactive site had lower cessation rates than those assigned to the static site, although this difference was not significant. The authors attribute these findings to the increased time investment required from participants of interactive sites (Rabius 2008).
Implications for practice
There is a small number of studies which provide very limited evidence of long-term benefits for programmes delivered only by the Internet compared to no-Internet controls. There is some evidence that tailored Internet interventions are more effective than non-tailored interventions.
Implications for research
More rigorous studies comparing the long-term effects of Internet interventions with non-Internet interventions or no intervention at all are needed in order to determine the true long-term effectiveness of the Internet as a tool for smoking cessation. These should, where possible, assess outcomes using objective measures, and also assess cost-effectiveness considerations. It is important that future trials also seek to describe the likely mechanisms through which these interventions may (or may not) be exerting their effects - they should therefore also report data on patient satisfaction, changes in knowledge, motivation, dependency, quit attempts and safety considerations.
Researchers should aim to assess smoking status after six months as a minimum, so that the longer-term benefit of programmes can be determined and meta-analyses of outcomes across studies be facilitated.
We thank Elizabeth Koshy and Matko Marlaias for their assistance in developing the protocol and selecting included studies for this review. We would like to thank Drs Jean-Francois Etter and Jennifer McClure for reading and commenting on earlier drafts of this protocol. We are also extremely grateful to Dr Etter for sharing with us his bibliography of Internet intervention trials. We also thank Shirley Manknell for consumer input.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Index terms
Last assessed as up-to-date: 14 July 2010.
Protocol first published: Issue 2, 2008
Review first published: Issue 9, 2010
Contributions of authors
Josip Car conceived the idea for this review and managed the review as contact author; he acts as guarantor for this paper. All authors contributed to the design. Marta Civljak lead the writing of the review. Lindsay Stead assisted with methodological support, data extraction and revising the draft review. Josip Car arbitrated the selection of studies and supervised the overall writing. Aziz Sheikh contributed to interpretation of the review findings and editing the text.
Declarations of interest
Sources of support
- Department of Primary Care and Social Medicine, Imperial College London, UK.
- Ministry of Science, Education and Sport, Croatia.
- NHS Connecting for Health Evaluation Programme (NHS CFHEP 001), UK.
Medical Subject Headings (MeSH)
MeSH check words
Adolescent; Adult; Female; Humans; Male
* Indicates the major publication for the study