Effectiveness and cost-effectiveness of computer and other electronic aids for smoking cessation: a systematic review and network meta-analysis

Introduction

The aim of the Health Technology Assessment (HTA) programme is to ensure that high-quality research information on the costs, effectiveness and broader impact of health technologies is produced in the most efficient way for those who use, manage, provide care in, or develop policy for, the NHS. Topics for research are identified and prioritised to meet the needs of the NHS. HTA forms a substantial portfolio of work within the National Institute for Health Research (NIHR) and each year about 50 new studies are commissioned to help answer questions of direct importance to the NHS. The studies include both primary research and evidence synthesis.

Research question

What is the effectiveness and cost-effectiveness of internet, PC and other electronic aids to help people stop smoking?

1. Technology Computer programs, internet sites, and other electronic media (such as mobile telephone texts) that help people stop smoking, or enhance the effectiveness of existing smoking cessation programmes by reducing relapse.

2. Patient group Adult smokers currently in NHS smoking cessation programmes.

3. Setting Those relevant to community-based programmes in the UK.

4. Control or comparator treatment No intervention or standard self-help material treatment.

5. Design An evidence synthesis of electronic aids that can be used in addition to standard NHS smoking cessation programmes to address the question that electronic media (and which aspects of the use of those media) have been shown to be effective in aiding smoking cessation and also to model the cost-effectiveness of using these media. Researchers should develop research recommendations for future HTA trials from their findings.

6. Primary outcomes Smoking cessation (ideally biochemically validated cessation rather than self-report).

7. Minimum duration of follow-up Six months.

Background to commissioning brief

Cessation services in the NHS achieve variable success rates with one-to-one and group sessions with smokers who want to quit, with wide variation across different age groups. Approaches to behaviour change can be supplemented with electronic aids, and this may significantly increase quit rates and prevent a proportion of cases that relapse.

In 2007, nearly 15 million households in Great Britain (61%) had internet access;1 many more people use mobile telephones. It is proposed that if computer and internet interventions to aid smoking cessation were shown to be effective, there would be potential for a large number of smokers to have access to this type of intervention at minimal cost to health services.

A recent systematic review showed some evidence that computer-based interventions changed smoking behaviour but identified a need for further research. 2 We are asking for further evidence synthesis because we want an estimate of cost-effectiveness and more precise specification of future research recommendations.

The reach of traditional smoking cessation interventions

Smoking cessation interventions are typically given face to face. An archetypal intervention is advocated in current English and Welsh National Institute for Health and Clinical Excellence (NICE) guidelines. 12 A smoker attends her or his general practitioner (GP) (family physician) and is advised to stop smoking (cessation induction). A smoker who agrees to this is offered referral to the NHS Stop Smoking Services (SSS). There, the smoker will be given regular behavioural support and will usually be prescribed pharmacotherapy to support the quit attempt (aid to cessation).

US national guidelines recommend that physicians behave in this way with every smoker at every consultation. 13 NICE guidance is less explicit on the frequency. 12 However, there is physician resistance to giving advice frequently and routinely, and rates of brief advice are much lower than suggested in guidelines. All UK primary care physicians record smoking status;14 about 30% of smokers receive advice to quit annually15 and about 3% receive pharmacotherapy to support quitting. 16 Perhaps, as a consequence, most people who try to quit do so without support from the NHS. 8 Only about 5% of smokers use the NHS SSS annually, out of > 40% who try to stop. 8

Self-help interventions to reach smokers

Despite their past experience, many smokers believe that they can and should quit smoking without formal treatment programmes. 17 Self-help interventions can support smoking cessation, by acting either as a cessation induction intervention or as an aid to cessation intervention, or both. The typical intervention is a leaflet, or perhaps an advertisement, in which a smoker is confronted with good reasons to stop smoking, as the GP might give in consultation. Other leaflets aim to take smokers through the quitting process, essentially by writing down the kind of advice that is typically given in a smoking cessation clinic. A Cochrane review showed that overall self-help interventions increased cessation rates, with a risk ratio (RR) of 1.21 [95% confidence interval (CI) 1.05 to 1.39]. 18 The self-help materials did not show evidence of additional efficacy when added to other interventions, such as physician advice or pharmacotherapy.

Static, targeted and tailored interventions, and collaborative filtering

In a static information intervention a typical leaflet is produced for all smokers and it is likely that some information therein will not be relevant for many readers. A targeted information intervention uses some broad-brush information to ensure that more of the content is relevant to its reader. For example, leaflets could be geared towards broad groups, such as people planning to stop smoking and looking for tips and advice on the best way to do this, or to others who are wondering whether or not they should. Such a self-help intervention could be delivered by leaflet, using the title of the leaflet to signal its content. In principle, electronic aids such as websites can function as static leaflets. Many websites feature information about the harms of smoking and ways to stop smoking that a person could choose to read. There is no obvious reason why smoking cessation leaflets should not work on the web, and these can be static or targeted.

A tailored intervention is more individualised. A typical behavioural support session will open with the therapist asking the patient to describe previous attempts to stop smoking and the difficulties encountered. The therapist will typically assess tobacco dependence. Advice is then tailored to the individual based on these data, formed around a basic quit plan on which to draw. A key competency for stop smoking advisors is ‘building rapport’ in order that the client feels that the practitioner ‘cares’ about their quit attempt and thus does not want to let them down; it also makes the client more receptive to information. For experienced practitioners this is where tailoring is most likely to take place, and therefore tailoring of intervention style as well as information that may be critical. Clearly, only the latter aspect translates to computer and electronic aids. Tailoring requires some individual assessment and hence uses either a person or an ‘expert system’ (EXP) to decode questionnaire responses and provide material tailored to an individual. The Cochrane review18 showed some evidence that tailored materials were more effective than standard self-help materials, with an RR of 1.31 (95% CI 1.20 to 1.42). Computerised interventions are nearly always needed to produce tailored self-help interventions.

Websites such as Amazon or iTunes use collaborative filtering to suggest products that one might like to purchase based on the choices of similar individuals. However, collaborative filtering has not been used extensively in smoking cessation interventions or more generally in behaviour change contexts.

Potential public health importance

Computerised interventions have considerable potential in public health for at least three reasons. The first relates to the evidence that many people are ambivalent about smoking,18,19 and a good number are prepared to make quit attempts with only modest prompting. 10,20 Electronic aids could provide such a prompt and, although most quit attempts end in early failure, a small proportion succeed. 21 Thus prompting more quit attempts will improve population cessation rates. The second reason relates to the established efficacy of behavioural support. Although the most effective support for cessation is the NHS SSS, only about 5% of smokers use this annually, whereas > 40% try to stop. 8 It is possible that the behavioural support provided by electronic aids could reach many of these smokers who otherwise use no support and thus might have much higher reach than the NHS SSS, even if it is somewhat less effective than behavioural support. Third, medication adherence is also often poor in studies without behavioural support,22,23 and better adherence is associated with a higher likelihood of quitting smoking. 24,25 By supporting adherence to medication, computerised interventions may improve the success of already effective interventions that are widely used.

The question of whether or not electronic aids are important to public health depends on their efficacy relative to the comparator, their reach, and their costs. Most smoking cessation interventions are only modestly effective in the medium term in comparison with other medical interventions. However, the benefits of cessation are very great and nearly all smoking cessation interventions are very cheap. Consequently, smoking cessation interventions are among the most cost-effective of all health-care interventions,26 according to NICE, meaning that the cost per quality-adjusted life-year (QALY) saved is lower than for many other health-care interventions. At first glance, electronic aids are likely to be cost-effective if they are effective. 27 Electronic aids might be costly to set up but their delivery is either free or very cheap as they require little in the way of labour costs to deliver. It is as cheap to deliver 5000 interventions as 50, as these can exploit economies of scale. This feature of electronic aids means that we need to extrapolate from the costs incurred in a trial to a population in which it might be delivered when considering the scale of the benefits that might accrue.

Search strategies

Searches were conducted in three phases: (1) the scoping searches to identify published and ongoing systematic reviews, which served as an additional source for identifying relevant primary studies and provided background information; (2) the main searches for primary studies to identify all relevant primary studies covering all of the three component reviews (effectiveness, supplementary and cost-effectiveness); and (3) the updated searches to identify relevant primary studies published during the preparation of this report. Detailed search strategies are described below.

Sifting of records retrieved from searches of electronic databases

Records retrieved from searches were imported into Reference Manager (version 11; Thomson ResearchSoft, San Francisco, CA, USA), which automatically detected and excluded duplicate records between electronic databases. Further duplicated records were identified and deleted manually. The titles and abstracts of the remaining records were examined for relevance by one of three reviewers (EA, DW, YFC). In order to improve the consistency of the sifting process, the reviewers independently screened a common set of the first 200 records, compared the results and resolved any disagreement by discussion before sifting through the remaining records. The initial sifting aimed to exclude obviously irrelevant records and focused on whether or not a paper possibly met the intervention criterion (out of the full set of study selection criteria, described in the next section).

Study selection

The study selection criteria and algorithm are described in Appendix 2. Full-text publications were ordered for all the records that passed through the initial sifting stage. At least two reviewers independently assessed the full publications against the inclusion/exclusion criteria (listed below). Disagreements between reviewers were resolved by discussion and by seeking further advice from additional members of the project team to reach a consensus. Where full publications could not be obtained, the records/studies were excluded. Details of these records (predominantly conference abstracts) are presented in the list of excluded studies (see Appendix 3).

Data extraction and quality assessment for effectiveness review

Data from included RCTs/quasi-RCTs were extracted on to a data extraction form by one of the reviewers (DW, IY, EA, OU). The data extraction form (see Appendix 3) was designed ad hoc for this review and included details of the citation, study design (population, interventions, comparators and co-interventions, outcome measures, and statistical methods) and results. The data extraction form also included a quality assessment checklist, which assesses the following domains:

1. methods of randomisation

2. allocation concealment

3. similarity in baseline characteristics between groups

4. similarity in care provided between groups other than the intervention/comparator being tested

5. biochemical validation

6. extent of dropout

7. presence of differential dropout between groups, and methods for adjustment in analysis

8. use of intention-to-treat (ITT) analysis.

The main outcome measure of interest is prolonged abstinence. Data on point prevalence abstinence and other measures of motivation to quit (e.g. movement in the stages of change) were also recorded.

All of the extracted data and results of quality assessment were independently checked by another reviewer to ensure accuracy and consistency. Disagreements were resolved by discussion and/or consulting a third reviewer.

Grouping of studies and study arms

In an attempt to guide the report structure and to facilitate quantitative analysis, the components of the care provided in each study arm (irrespective of whether they are considered as an intervention, a control or a co-intervention) with regard to smoking cessation were coded using a coding scheme shown in Table 4. Components were categorised as either ‘electronic’ or ‘non-electronic’ and these were coded separately. The coding scheme was developed, piloted, and revised during the data extraction phase. Coding was undertaken independently by two reviewers (IY, YFC). Disagreements were resolved by discussion and by seeking further advice/arbitration from another team member (PA). The coding was finalised before data analysis took place.

The rationale behind our categorisation in Table 4 is as follows. Given the focus of this review, interventions (or parts of interventions) were firstly classified as either electronic or non-electronic. Electronic (or electronic part of) interventions were then grouped into five different categories (e1–5) according to the number and nature of ‘components’ included within an intervention. Components were defined according to both the mode of delivery (given that this review aims to determine which electronic media may be effective) and whether the contents within each mode of delivery are generic or individually tailored. We think the latter is important because previous reviews18 suggest that individually tailored materials appear to be more effective compared with generic, non-tailored (NT) material. Our categorisation therefore attempts to differentiate electronic interventions primarily on the basis of whether their contents are generic or tailored. The categorisation also aims to explore whether or not inclusion of more than one component (i.e. more than one mode of delivery and/or inclusion of both generic and tailored contents) enhance the effectiveness of interventions.

With respect to non-electronic interventions, our thinking was also informed by the relevant Cochrane review. 18 A simple self-help leaflet is marginally more effective on its own than no intervention (RR = 1.21, 95% CI 1.05 to 1.39) and thus they were coded separately. However, there is no evidence that self-help material adds to brief advice from a physician. The RR for brief advice is larger than for a self-help intervention. Likewise, behavioural support or counselling has been shown to be more effective than brief advice and the evidence indicates that proactive telephone counselling might be similarly effective as face-to-face counselling. Pharmacotherapy is also effective in enhancing smoking cessation. There is a common consensus that behavioural support and pharmacotherapy have additive effects. The coding scheme therefore groups each of these non-electronic interventions (and the combination of counselling and pharmacotherapy) separately, with ascending code numbers corresponding to potentially more effective interventions.

In addition to the coding scheme to categorise individual study arms, each study was also classified by one reviewer (YFC) with respect to the study population, type of electronic media, and comparisons made within each study based on the framework shown in Figure 1. The mapping of studies to the framework provided further guidance on report structure and meta-analysis within each major section.

FIGURE 1.

Classification of included studies in relation to study population, mode of delivery of the electronic intervention and comparisons made within each study.

We have also attempted to apply a standardised coding with respect to the contents of each intervention using a taxonomy developed by Abraham and Michie. 126 However, our initial pilot indicated that information presented in published papers was often insufficient to allow accurate coding of each intervention/comparator. Detailed coding and analysis of the contents of the interventions were therefore not carried out.

Data handling and analysis

Quantity and quality of the evidence

In total, 3969 records remained in the Reference Manager database after duplicate records were removed. Of these, 270 were considered potentially relevant, and full papers were ordered for further examination.

The searches for primary studies were updated in December 2009. An additional 151 records were retrieved from the updated searches, of which 34 were considered potentially relevant and were ordered. Three additional potentially relevant papers were identified from a reference list of previous reviews and from contact with experts.

We could not obtain 14 of the 307 papers ordered – all of them conference abstracts. The remaining 293 papers were examined and 77 papers reporting results from 60 RCTs were selected for inclusion in the main effectiveness review. Papers potentially relevant for the cost-effectiveness review and supplementary review were forwarded to relevant team members for further considerations. A flow diagram for study selection process is shown in Figure 2 and a list of excluded studies and reasons for exclusion can be found in Appendix 4. Included studies are described in the following sections according to the categorisation of electronic interventions shown in Table 4.

FIGURE 2.

Flow diagram for study selection process. a, Some trials included multiple intervention arms with various levels of tailoring/numbers of components; b, meta-analysis only included comparisons of electronic interventions to minimal (generic self-help material) or no intervention control groups.

Overview of effectiveness

This section provides an overview of the effectiveness of computer and other electronic aids compared with no or minimal intervention (e.g. generic self-help material) across the different types of electronic interventions included in this review. Acknowledging the diverse nature of interventions, participants, methods and context across the included studies, the purpose of this section is to provide a panoramic view to answer the broad question of whether or not computer and other electronic aids are effective for smoking cessation. Quantitative findings from individual studies are presented in forest plots, along with pooled estimate of intervention effect across studies and measures of heterogeneity between studies. As stated above (see Data analysis), where the statistical heterogeneity between studies is high (e.g. I² ≥ 50%), the forest plots are mainly used to graphically present results from individual studies and to illustrate the differences between study findings. The pooled estimates shown in the plots, if not discarded, need to be interpreted with extreme caution in these cases.

In addition to providing an overall quantitative estimate of effectiveness (where appropriate), the meta-analyses also offered opportunities to identify major study-level factors that may influence the estimated size of effects. Three factors were explored by subgroup analyses: (1) aid to cessation studies compared with cessation induction studies; (2) stratification of studies by mode of delivery of the interventions; and (3) stratification of studies by concurrent non-electronic co-interventions. The subgroup (1) was prespecified in the review protocol. Subgroups (2) and (3) are exploratory. In addition, subgroup/sensitivity analyses were carried out to explore the impact of quality of method/reporting (generation of random sequence, allocation concealment) and biochemical validation.

For each analysis, a forest plot for point prevalence abstinence measured at 6 months is first presented, followed by forest plots for point prevalence abstinence and for prolonged abstinence at longest follow-up of each study. As only a small number of studies reported 6-month prolonged abstinence, no separate forest plot is generated for this outcome measure. Funnel plots examining potential publication bias or small study effects are presented at the end of this chapter. More detailed descriptions of individual studies and results of other comparisons (e.g. direct comparison between different electronic interventions) are presented in subsequent sections (see Interventions with single or multiple generic components, Interventions with single tailored component and Interventions with multiple tailored components).

Aid to cessation compared with cessation induction

Results are presented in Figures 3–5. For point prevalence abstinence at 6 months (see Figure 3), substantial heterogeneity exists within each subgroup of studies (aid to cessation vs cessation induction) and this prevents valid comparison between the subgroups and meaningful interpretation of the pooled estimates within and across the subgroups. Point prevalence abstinence and prolonged abstinence measured at the longest follow-up of each study (see Figures 4 and 5, respectively) suggest that overall the computer and electronic aids are more effective than control (no intervention or generic self-help material). The effect sizes are small (point prevalence abstinence, pooled RR = 1.14, 95% CI 1.07 to 1.22) to moderate (prolonged abstinence, pooled RR = 1.32, 95% CI 1.21 to 1.45) and do not appear to differ significantly between the subgroups.

FIGURE 3.

Point prevalence abstinence measured at 6 months: aid to cessation studies vs cessation induction studies.

FIGURE 4.

Point prevalence abstinence measured at longest follow-up of each study: aid to cessation studies vs cessation induction studies.

FIGURE 5.

Prolonged abstinence measured at longest follow-up of each study: aid to cessation studies vs cessation induction studies.

There is substantial heterogeneity (I² = 51%) among aid to cessation studies for the outcome of prolonged abstinence (see Figure 5). Further analysis reveals that the heterogeneity may partly be attributed to the difference between the five studies with single tailored component (RR = 1.22, 95% CI 1.10 to 1.37, I² = 6%) and the two studies with multiple tailored components (RR = 2.10, 95% CI 1.25 to 3.53, I² = 42%), but a large part of the heterogeneity is attributed to the difference between the two studies with multiple tailored components as exemplified by the I² value of 42%.

Mode of delivery

Stratification of studies according to the mode of delivery of electronic interventions does not reveal a clear pattern of effect among electronic interventions using different modes of delivery. For point prevalence abstinence measured at 6 months (Figure 6), significant heterogeneity was observed among cessation induction studies using computer-generated tailored printed materials, and among studies using multiple modes of delivery. For point prevalence abstinence measured at longest follow-up (Figure 7), substantial heterogeneity was also observed within the subgroups of e-mail-based interventions, web-based interventions and interventions using multiple modes of delivery. Results for prolonged abstinence (Figure 8) were relatively homogeneous and suggested that, overall, tailored printed materials (RR = 1.29, 95% CI 1.18 to 1.40, I² = 0%) and interventions utilising multiple channels of delivery (RR = 1.67, 95% CI 1.16 to 2.39, I² = 31.6%) are effective. The effectiveness of web-based intervention appears to vary between studies/contexts. The number of studies utilising stand-alone computers, mobile telephone text messages, interactive voice response (IVR) and e-mails was small. Their effectiveness particularly in terms of prolonged abstinence has not been demonstrated.

FIGURE 6.

Point prevalence abstinence measured at 6 months, stratified by mode of delivery of interventions.

FIGURE 7.

Point prevalence abstinence measured at longest follow-up of each study, stratified by mode of delivery of interventions.

FIGURE 8.

Prolonged abstinence measured at longest follow-up of each study, stratified by mode of delivery of interventions.

Non-electronic co-interventions

Non-electronic co-interventions that were used in conjunction with electronic interventions were classified into the following categories: none, self-help material, brief advice, counselling, pharmacotherapy, and pharmacotherapy plus counselling. Based on point prevalence abstinence (Figures 9 and 10), computer and electronic aids appear to be effective when used in conjunction with other non-electronic co-interventions, except when added to a combination of pharmacotherapy and counselling. Substantial heterogeneity in point prevalence abstinence measured at 6 months was observed when computer and electronic aids were used without any non-electronic co-interventions. Results for prolonged abstinence (Figure 11) are more homogeneous and are broadly consistent with the findings according to point prevalence abstinence.

FIGURE 9.

Point prevalence abstinence measured at 6 months, stratified by non-electronic co-intervention.

FIGURE 10.

Point prevalence abstinence measured at longest follow-up of each study, stratified by non-electronic co-intervention.

FIGURE 11.

Prolonged abstinence measured at longest follow-up of each study, stratified by non-electronic co-intervention.

Biochemical validation

Seventeen studies adopted some methods of biochemical validation of abstinence, such as measuring cotinine level in saliva or carbon monoxide level in exhaled air. Four studies39,62,98,120 presented data for both self-reported and biochemically validated abstinence, whereas six studies33,42,45,55,82,100 provided data on only biochemically validated abstinence. A further three studies34,36,73 conducted biochemical validation on a proportion of (usually randomly selected) participants and thus data for biochemically validated abstinence were not available for ITT analysis. The remaining four studies60,80,101,113 were not included in meta-analyses, as their control groups were neither ‘no intervention’ nor ‘non-electronic generic self-help material’.

This section explores the impact of biochemical validation in two subgroup/sensitivity analyses. First, a subgroup analysis was carried out comparing studies that reported biochemically validated abstinence to studies that only reported self-reported abstinence. Where both biochemically validated and self-reported data were available, the former was used in this analysis (in contrast with the main analysis, for which self-reported abstinence took preference). As mentioned above, in a few studies34,36,39,73 biochemical validation was conducted in a small proportion of participants. These studies were included in the ‘biochemically validated’ group in this analysis, although only self-reported abstinence was available and was included in the analysis. In these cases the limited biochemical validation was more akin to bogus pipeline and thus this sensitivity analysis explores, in part, the effect of biochemical validation and/or bogus pipeline compared with no validation at all. Second, a sensitivity analysis was conducted using the four studies39,62,98,120 in which both self-reported and biochemically validated abstinence was reported. This analysis allows comparison and contrast between these two types of data without being confounded by other study characteristics.

Figure 12 shows the results of studies with and without biochemical validation for point prevalence abstinence at 6 months. The pooled RR (1.09, 95% CI 0.94 to 1.27) for studies with biochemical validation suggests a much smaller and statistically insignificant effect than for the pooled RR for studies without biochemical validation. The statistical heterogeneity among the latter studies was very high (I² = 85.5%) and thus the pooled estimate is difficult to interpret and further investigation of source of heterogeneity is required.

FIGURE 12.

Findings from studies with biochemical validation vs studies without biochemical validation: point prevalence abstinence at 6 months.

Figure 13 shows the results for point prevalence abstinence at longest follow-up of each study. Again the pooled RR (1.07, 95% CI 0.95 to 1.20) for studies with biochemical validation suggest a smaller and statistically insignificant effect compared with that for studies without biochemical validation (RR 1.18, 95% CI 1.10 to 1.27), which shows a small but significant intervention effect. Test for interaction indicates the difference between subgroups is not statistically significant (p = 0.157), but the result of the test needs to be interpreted with caution, given that there is still moderate heterogeneity (I² = 38.0%) in the ‘no biochemical validation’ group.

FIGURE 13.

Findings from studies with biochemical validation vs studies without biochemical validation: point prevalence abstinence at longest follow-up.

Comparison of self-reported and biochemically validated abstinence using data from studies that reported both outcomes is shown in Figure 14. The pooled RR for biochemically validated abstinence suggests slightly larger effect than that for self-reported abstinence, but test for interaction indicates the difference is not statistically significant (p = 0.455).

FIGURE 14.

Comparison of self-reported abstinence and biochemically validated abstinence.

It is worth noting that in studies such as these, in which there is little or no therapeutic relationship between the patient and the individual offering the intervention, the scope for misreporting of smoking status is lower. In addition, collection of biochemical validation in such studies may be difficult, leading to a low response rate. If non-responders are rated as smoking, as is typical, this may serve to attenuate treatment effects disproportionately. This should be borne in mind when interpreting these results.

Methods of randomisation

This section explores whether or not methods of randomisation, including the generation of random sequence and concealment of allocation, had significant impact on the estimates of intervention effects.

Figure 15 shows prolonged abstinence at longest follow-up for each study, grouped according to the adequacy of methods for generating random sequence. Intervention effects estimated from studies with truly random methods appear to be similar to those estimated from studies with unclear description of the methods. Only one study109 clearly stated a quasi-randomised design. The results for point prevalence abstinence at 6 months and point prevalence abstinence at longest follow-up are generally similar to that of prolonged abstinence, although there is high level (I² > 70%) of heterogeneity within the subgroups (‘truly random’ and ‘unclear’) for point prevalence abstinence at 6 months (data not shown).

FIGURE 15.

Prolonged abstinence at longest follow-up according to whether or not the methods for generating sequence were truly random.

Figure 16 shows prolonged abstinence at longest follow-up, grouped according to the adequacy of allocation concealment. Allocation concealment was unclear in the majority of cases and was judged to be clearly adequate in only two trials with very heterogeneous results.

FIGURE 16.

Prolonged abstinence at longest follow-up according to whether or not allocation concealment was adequate.

Figure 17 shows the results for point prevalence abstinence at longest follow-up. Pooled RRs are very similar between the subgroups, with either adequate or unclear allocation concealment. There is moderate heterogeneity (I² = approximately 30%) within each of the two subgroups. Results for point prevalence abstinence at 6 months suggest that studies with adequate allocation concealment tended to produce smaller effect sizes (Figure 18). However, very high levels of heterogeneity (I² > 70%) were observed within each subgroups and this precludes a valid test for interaction to be conducted and any firm conclusions to be drawn.

FIGURE 17.

Point prevalence abstinence at longest follow-up according to whether or not allocation concealment was adequate.

FIGURE 18.

Point prevalence abstinence at 6 months according to whether or not allocation concealment was adequate.

Funnel plots

Funnel plots were generated for point prevalence abstinence at 6 months, point prevalence abstinence at longest follow-up, and prolonged abstinence at longest follow-up. They are shown in Figure 19. Consistent with the above findings, substantial heterogeneity is observed for point prevalence abstinence at 6 months. Egger’s test for all the three outcome measures does not suggest significant funnel plot asymmetry, although the possibility of some missing studies with negative outcomes cannot be ruled out by inspection of the funnel plots for point prevalence abstinence and prolonged abstinence at longest follow-up.

FIGURE 19.

Funnel plots showing all studies comparing computer and electronic aids to control of no or minimal intervention. (a) Point prevalence abstinence at 6 months, Egger’s test p = 0.15; (b) point prevalence abstinence at longest follow-up, Egger’s test p = 0.30; (c) prolonged abstinence at longest follow-up, Egger’s test p = 0.25. Ln, natural log.

Interventions with single or multiple generic components

The previous section (see Overview of effectiveness) provided an overview of quantitative findings from included studies in terms of the effectiveness of the interventions compared with no or minimal intervention. This section and the following two sections (see Interventions with single tailored component and Interventions with multiple tailored components) describe individual studies in more detail, including the assessment of their methodological quality. Studies were grouped according to the categorisation described previously (see Grouping of studies and study arms). Each of the sections includes descriptions of interventions and co-interventions, study design and characteristics of participants, quality assessment of included studies, and comparisons and findings. The latter present different types of comparison (see Figure 1) in the following order:

• electronic interventions compared with no intervention, usual care or untailored printed self-help material

• comparison of electronic interventions with other non-electronic interventions

• comparisons between electronic interventions

• other comparisons.

This section summarises two studies100,123 that evaluated interventions with single or multiple generic components; both were aid to cessation studies. The characteristics of these studies and their participants are presented in Table 5 and are briefly described below.

Interventions with single tailored component

The vast majority of included studies evaluated interventions with a single tailored component, with or without additional generic component(s). This section is therefore further subdivided according to the mode of delivery of the interventions.

Computer-generated, tailored printed materials

Thirty studies34,36–40,42–44,46,48,49,53,55–58,61,66,69,72,75,80,88,91,96,97,109,110 evaluated electronic interventions with computer-tailored printed materials, 1034,36,38,48,49,58,61,66,75,97 of which were aid to cessation studies and 2037–40,42–44,46,53,55–57,69,72,80,88,91,96,109,110 of which were cessation induction studies. The study characteristics, results of quality assessment and key findings are described below.

Interventions and co-interventions

Table 7 shows the characteristics of included studies and their participants. The interventions consisted of individualised feedback/printed letters generated by special software or ‘expert systems’. Most of the letters were tailored based on conceptual models on relevant theories of smoking cessation and behaviour change (e.g. transtheoretical model or social cognitive theory).

TABLE 7 - Characteristics of RCTs evaluating computer-generated, tailored printed materials

BO, both sorts of information; CO, no information control; CQP, Committed Quitters Programme; CQP + C, Committed Quitters Programme plus outbound telephone call; EMIS, Egton Medical Information System; MAN, stage-matched manual; MT, multiple tailored intervention; NR, not reported; OC, information on outcomes of quitting; SE, self-enhancing information; SHG, self-help guide; ST, single tailored letter; TEL, automated counselling; UG, only user’s guide plus audiotape.

Study and country	Co-intervention (non-electronic)	Population, key criteria related to smoking history and method of recruitment	Mean age (years); % female	Intervention, comparators and sample size (n)	Outcome measure and length of follow-up; comparison code [0–5] (see Figure 1 for the coding scheme)
Aid to cessation studies
Owen et al. 198934 Australia	None	Inclusion/exclusion criteria were not clearly stated. Smokers were recruited by public advertisements in newspapers and local radio and television	42; 45	Quit kit control (40) vs standard correspondence course (86) vs personalised correspondence (82)	7–10 days point prevalence (not clearly defined) Follow-up: 3 and 9 months [1]
Curry et al. 199136 USA	Self-help material	Smokers recruited through advertisement in View magazine	44.1; 65	Intrinsic group (304) vs extrinsic (304) vs both (304) vs control (305)	7-day point prevalence abstinence Follow-up: 3 and 12 months [0]
Strecher et al. 199438 – study 2 USA	None	Adult patients from 12 community-based group family practice physicians’ office, aged 18–75 years	36.7; 67	Tailored smoking letters (NR) vs no messages (NR) Total sample size = 296	7-day point prevalence Follow-up: 6 months [0]
Borland et al. 200466 Australia	Self-help material	Callers to the Victorian Quitline (a community service providing information and advice on all aspects of smoking cessation), who are adult smokers or recent quitters who requested written materials only. English speakers with no obvious psychiatry or neurological problems	32.3; 54	Tailored advice (521) vs control (537)	7-day point prevalence and 6-month prolonged abstinence Follow-up: 12 months [0]
Borland et al. 200361 Australia	Brief advice (plus self-help material)	Adult smokers who called to the Victorian Quitline telephone counselling and advice service	NR; 54	Computer-generated, tailored advice + call-back telephone counselling (528) vs computer-generated, tailored advice only (523) vs control ( 527)	Not smoked for 3 months or 9 months Follow-up: 3, 6 and 12 months [0][5]
Strecher et al. 200575 USA	Telephone counselling	Callers to National Cancer Institute’s Cancer Information Service enquiring about smoking cessation. English speakers, ≥ 18 years of age, who smoke at least five cigarettes per day, interested in quitting smoking, not involved in another cessation programmes	41; 70	Single untailored (NR) vs single tailored (NR) vs multiple tailored (NR) vs multiple retailored materials (NR) Total sample size = 1978	7-day abstinence Follow-up: 5 and 12 months [1][3]
Sutton and Gilbert 200797 UK	Telephone counselling (+ self-help material)	Current and recent ex-smokers/participants, who smoked on average 20.5 cigarettes per day/via callers to the Quitline	38.1; 66	Computer-generated, tailored advice letter + usual care (765) vs usual care (562)	3-month prolonged abstinence; 24-hour and 7-day point-prevalence abstinence Follow-up: 24 hours, 7 days, and 1 and 3 months [0]
Orleans et al. 200048 USA	Transdermal nicotine	Smokers aged ≥ 65 years, who filled a transdermal nicotine prescription	72; NR	Tailored treatment group = NR Control group = NR	7-day point prevalence Follow-up: 6 and 12 months [0]
Shiffman et al. 200049 USA	NRT (+ self-help material)	Adults/current cigarette smoker/smokers who purchased nicotine polacrilex gum and called the CQP programme through freephone	41.5; 54	CQP (1217) vs CQP + C (1207) vs UG (1203)	28-day abstinence at week 6 10-week continuous abstinence at week 12 Follow-up: 6 and 12 weeks [0]
Shiffman et al. 200158 USA	NRT (+ self-help material)	US purchasers of NicoDerm CQ/smoke almost 30 cigarettes per day/calling into a freephone number	40; 64	CQP (1865) vs UG (2344)	28-day and 10-week continuous abstinence Follow-up: 6 and 12 weeks [0]
Cessation induction studies
Prochaska et al. 199337 USA	None	Adult smokers recruited by newspaper advertisements. No criteria related to smoking history were used	43; 62	Counsellor calls + computer-generated, tailored reports + MAN vs tailored reports + MAN vs MAN vs standard self-help manual Total sample randomised = 756	24-hour point prevalence abstinence; not smoking at two consecutive follow-ups Follow-up: 6, 12 and 18 months [1][5]
Strecher et al. 199438 – study 1 USA	None	Adult patients of a large family practice, aged 40–65 years, with telephone, who do not share a household with someone in the same sample list	49.5; 68	Tailored smoking letters (NR) vs standardised smoking letters (NR) Total sample size = 72	7-day point prevalence abstinence Follow-up: 4 months [1]
Curry et al. 199539 USA	None	Enrolees in a health maintenance organisation/17.3 cigarettes/day/telephone survey	41.1; 52	Booklet (327) vs feedback (323) vs telephone (150) vs control (324)	7-day abstinence Follow-up: 3, 12 and 21 months [0][1][5]
Kreuter and Stretcher 199640 USA	None	Adult family practice patients	40; 65	Enhanced health risk appraisal (427) vs typical health risk appraisal (427) vs control (463)	Smoking prevalence Follow-up: 6 months [0][3]
Dijkstra et al. 199842 The Netherlands	None	Adult smokers/average of 20.3 cigarettes/day; smoked for an average of 21.8 years/advertisement in a local newspaper	39.7; 59	Only OC (386) vs Only SE (387) vs BO (387) vs no information at all (CO) (286)	7-day abstinence, not smoking for at least 24 hours, 12-month prolonged abstinence Follow-up: 3 and 14 months [0][3]
Dijkstra et al. 199843 The Netherlands	None	Adult smokers who were not planning to quit within the next 6 months. Recruited by advertisements in local newspapers	39.3; 60	A tailored letter three times (multiple) with SHG (MT-plus condition; n = 140) vs three times (multiple) a tailored letter only (MT-only condition; n = 156) vs a ST with a SHG (ST-plus condition; n = 157) vs a ST only (ST-only condition; n = 152) vs single NT letter once (NT condition; n = 147)	24-hour quit attempt, 7-day abstinence, intention to quit, stage transition, self-efficacy expectations, smoking behaviours Follow-up: 4 months [1][3]
Dijkstra et al. 199944 The Netherlands	None	Adult smokers who were not planning to quit within the next 6 months. Recruited by advertisements in local newspapers	41.7; 63	MT (214) vs ST (206) vs SHG (215) vs CO (208)	7-day point prevalence abstinence Follow-up: 7 months [0][1][3]
Velicer et al. 199946 USA	None	Adult smokers of a managed care system identified through mail and telephone surveys. No criteria relating to smoking history or readiness to quit were applied	38; 56	EXP+ MAN – one contact (357) vs EXP + MAN – two contacts (353) vs EXP+ MAN – three contacts (362) vs EXP+ MAN – six contacts (357) vs MAN – one contact (359) vs MAN – two contacts (368) vs MAN – three contacts (366) vs MAN – six contacts (360)	24-hour and 7-day point prevalence abstinence 30-days and 6-month prolonged abstinence Follow-up: 6, 12 and 18 months [1][3]
Etter and Perneger 200153,54 Switzerland	None	Residents aged 18–60 of the French-speaking part of Switzerland/daily cigarette smokers/randomly selected from a general population register	36.3; NR	Computer-generated tailored counselling letter (1467) vs no intervention (1467)	7-day point prevalence abstinence and 1-month abstinence Follow-up: 7 months [0]
Lennox et al. 200155 UK	None	Smokers aged 17–65 years registered at six general practices in Aberdeen	NR	Computer-tailored letter (870) vs standard letter (869) vs no intervention (871)	7-day point abstinence, stage of change Follow-up: 6 months [0][1]
Prochaska et al. 200156 USA	None	Rhode Island smokers/20.5 cigarettes per day/random digit-dialling telephone calls	41.3; 55	EXP – computer report (1358) vs assessment only (2786)	24-hour and 7-day abstinence 30-day and 6-month prolonged abstinence rates Follow-up: 6, 12, 18 and 24 months [0]
Prochaska et al. 200157 USA	None	Adults/criteria related to smoking history not reported/contacted via telephone and mail	38.1; 56	EXP (368) vs EXP + counselling (359) vs EXP + LifeSign computer (366) vs Assessment only (359)	24-hr and 7-day point prevalence abstinence 30 days and 6 months prolonged abstinence Follow-up: 6, 12 and 18 months [0]
Prochaska et al. 200469 USA	None	Parents/criteria related to smoking history not reported/list of parents who had a 9th grader	42.5; 75	Computer-generated tailored materials (1209) vs assessment only (1251)	24-hour and 7-day point prevalence abstinence 6-month prolonged abstinence Follow-up: 12 and 24 months [0]
Prochaska et al. 200572 USA	None	Primary care patients/17.2 cigarettes/day/recruited at home via telephone	44.7; 70	EXP intervention (2667) vs assessment only (2740)	7-day abstinence, not smoking for at least 24 hours, 6-month prolonged abstinence Follow-up: 12 and 24 months [0]
Hall et al. 200680,81 USA	None	Adult patients receiving outpatient treatment for depression; having smoked one or more cigarettes per day during the week before recruitment. Recruited by provider referral, invitation letters, and flyers in the participating clinics	41.8; 61	Computer-generated tailored materials + one-to-one face-to-face counselling + nicotine patches + bupropion (163) vs a folder containing a list of referral to smoking cessation programmes + a stop smoking guide (159)	7-day point prevalence abstinence Follow-up: 3, 6, 12 and 18 months [5]
Smeets et al. 200796 The Netherlands	None	Volunteers ages 18–65 years/15.6% smoked at baseline/contacted by telephone	47; 57	Computer-generated tailored materials (1417) vs printed untailored materials (1410)	Smoking behaviour was measured by asking smokers to state the number of cigarettes they smoked a day and whether or not they had made any quit attempts during the last 3 months Follow-up: 3 months [1]
Meyer et al. 2008109 Germany	None (usual practice)	Primary care patients aged 18–70 years who smoked cigarettes daily in the past 4 weeks. Patients who attended study practices were screened for smoking status and those who fulfilled the inclusion criteria were invited to participate	34.0; 48	Computer-tailored letters (488) vs practitioners’ brief advice (402) vs assessment only (609)	24-hour and 7-day point prevalence abstinence; 4-week and 6-month prolonged abstinence Follow-up: 6, 12, 18 and 24 months [0][2]
Schumann et al. 2008110,115,127 Germany	None	Adults/former and current smokers/from an existing general population health examination survey	44.5; 47	Computer-generated, tailored materials (302) vs no intervention (309)	7-day abstinence rate, prolonged abstinence of at least 6-months Follow-up: 6, 12, 18 and 24 months [0]
Gilbert et al. 200791 UK	Self-help material (+ standard care from GP)	Smokers between 18 and 65 years/93.6% smoked everyday/selected from GP practice using EMIS	42.1; 54	Computer-tailored feedback report (38) vs no intervention (40)	24-hour and 7-day point prevalence abstinence; 1- and 3-month prolonged abstinence Follow-up: 3 months [0]
Velicer et al. 200688 USA	NRT	Smokers from a New England Veterans Affairs Medical Center/smoked on average 24.5 cigarettes per day/via letter	50.5; 23	MAN (523) vs NRT plus MANs (522) vs EXP + NRT + MANs ( 509) vs TEL + EXP + NRT + MANs (500)	24-hour and 7-day point-prevalence abstinence 6-month prolonged abstinence Follow-up: 10, 20 and 30 months [0]

Four studies48,49,58,88 included pharmacotherapy as a co-intervention in each of the study arms; two studies75,97 included telephone counselling as a co-intervention, while another study61 included brief advice. Seven studies,36,46,49,58,61,66,97 including four mentioned above,49,58,61,97 had generic self-help material as a co-intervention in each of the study arms. Twenty studies34,37–40,42–44,53,55–57,69,72,80,91,96,109,110 do not have non-electronic co-interventions.

Study design and characteristics of participants

All the 30 studies, except Meyer et al. ,109 which was a quasi-RCT, were RCTs that randomised participants individually. 34,36–40,42–44,46,48,49,53,55–58,61,66,69,72,75,80,88,91,96,97,109,110 Follow-up ranged from 3 months to 30 months88 and was shorter than 12 months in 10 studies. All studies reported point prevalence, whereas 1639,42,46,49,56–58,61,66,69,72,88,91,97,109,110 reported prolonged abstinence.

Participants were recruited into trials as advertised through various media. Three studies66,97,115 included both current smokers and ex-smokers; only current smokers at study entry were included in the analysis in this section.

Mean age of study participants ranged from 32.366 to 72 years. 48 The proportion of females varies between 23%88 and 76%. 69

Quality assessment of included studies

Results of quality assessment of included studies are presented in Table 8. The methods of generation of random sequence, allocation concealment and blinding are unclear in the vast majority of the studies. Only seven studies34,36,39,42,55,62,80 included biochemical validation of smoking status in respondents. Loss to follow-up ranged from as low as 11% at 6 months44 to 56% at 12 months. 75 Six studies39,43,49,55,57,115 reported significant difference between intervention groups at baseline.

TABLE 8 - Quality assessment of RCTs/quasi-RCTs evaluating interventions using computer-generated, tailored printed material

CO, no information control; CQP, Committed Quitters Programme; CQP + C, Committed Quitters Programme plus outbound telephone call; MAN, stage-matched manual; NR, not reported; MT, multiple tailored intervention; SHG, self-help guide; ST, single tailored letter; TEL, automated counselling; UG, only user’s guide plus audiotape.

Study	Generation of random sequence	Allocation concealment	Blinding	Biochemical validation	Baseline characteristics similar between groups	Lost to follow-up	ITT analysis
Schumann et al. 2008110,115,128	Yes	Unclear	Unclear	No	No (difference in intention to quit within 6 months and past quit attempts)	Overall: 175/611 (28.6%) Computer-generated, tailored materials: 101/302 (33.4%) No intervention: 74/309 (23.9%)	Yes
Meyer et al. 2008109	No (based on the time of practice attendance)	Unclear	Outcome assessors	No	Yes	558/1499 (37.2%)	Yes
Gilbert et al. 200791	Yes	Unclear	Unclear	No	No	26/72 (36.1%)	Yes
Sutton and Gilbert 200797	Yes	Yes	Outcome assessors	No	Yes	Computer-generated, tailored advice + usual care: 159/765 (20.8%) Usual care: 181/743 (24.4%)	Yes
Smeets et al. 200796	Unclear	Unclear	Unclear	No	Yes	NR	Unclear
Velicer et al. 200688	Yes	Unclear	Participants	No	Yes	MAN: 126/523 (24.1%) NRT + MAN: 142/522 (27.2%) EXP + NRT + MAN: 153/509 (30.0%) TEL + EXP + NRT + MAN: 155/500 (31.0%)	Yes
Hall et al. 200680	Yes	Unclear	Outcome assessors	Yes	Yes	Computer-generated, tailored material + counselling +nicotine patched + bupropion: 41/163 (25.2%) List of referrals to smoking cessation programmes: 49/159 (30.8%)	Yes
Strecher et al. 200575	Unclear	Unclear	Unclear	No	Yes	34% at 5 months and 56% at 12 months	Yes
Prochaska et al. 200572	Unclear	Yes	Outcome assessors	No	Yes	EXP: 784/2667 (29.3%) Assessment only: 595/2740 (21.7%)	Unclear
Prochaska et al. 200469	Unclear	Unclear	Unclear	No	Yes	Computer-generated, tailored materials: 273/1209 (22.6%) Assessment only: 272/1251 (21.7%)	Yes
Borland et al. 200366	Unclear	Unclear	Unclear	No	Yes	Tailored advice = 170/521 Control = 146/537	Yes
Borland et al. 200361	Yes	Unclear	Participants	No	Yes	369/1578 (23.4%)	Yes
Etter and Perneger 200153,54	Yes	No	Unclear	No	Yes	Computer-generated, tailored counselling letter: 357/1467 (24.3%) No intervention: 121/1467 (8.2%)	Yes
Shiffman et al. 200158	Yes	Unclear	Unclear	No	Yes	Overall: 2194/4209 (52.1%) Computer-tailored program: 860/1865 (46.1%) UG: 1334/2344 (56.9%)	No
Prochaska et al. 200156	Unclear	Unclear	Unclear	No	No (proportions for ethnic and racial subgroups)	Overall: 1573/4144 (38.0%) EXP: 556/1358 (40.9%) Assessment only: 1017/2786 (36.5%)	No
Lennox et al. 200155	Yes	Unclear	Unclear	Yes	No (proportion of heavy smokers higher in tailored group)	Computer-tailored = 213/870 Standard letter = 236/869 Control = 166/871	Yes
Prochaska et al. 200157	Unclear	Unclear	Unclear	No	No (for number of cigarettes per day)	NR	Yes
Shiffman et al. 200049	Unclear	Unclear	Open label	No	Yes	CQP: 347/1217 (28.5%) CQP + C: 349/1207 (28.9%) UG: 303/1203 (25.2%)	Yes
Orleans et al. 200048	Unclear	Unclear	Unclear	No	Unclear	22% at 12 months	No
Dijkstra et al. 199944	Unclear	Unclear	Unclear	No	Yes	At 6 months: MT: 27/214 (12.6%) ST: 26/206 (12.2%) SHG: 15/215 (7.0%) CO: 26/208 (12.5%)	Yes
Velicer et al. 199946	Unclear	Unclear	Unclear	No	Yes	NR	Unclear
Dijkstra et al. 199843	Unclear	Unclear	Unclear	No	No (higher proportion classified as emotive in non-tailored group)	147/752 (19.5%) at 4 months	No
Dijkstra et al. 199842	Unclear	Unclear	Unclear	Yes	Yes	555/1546 (35.9%)	Unclear
Kreuter et al. 199640	Unclear	Unclear	Unclear	No	Yes	186/1317 (14.1%)	Unclear
Curry and Strecher 199539	Unclear	Unclear	Unclear	Yes	No (previous quit attempt)	272/1137 (23.9%)	Yes
Strecher et al. 199438 – study 1	Unclear	Unclear	Unclear	No	Yes	29.2%	No
Strecher et al. 199438 – study 2	Unclear	Unclear	Unclear	No	Yes	33%	No
Prochaska et al. 199337	Unclear	Unclear	Unclear	No	Unclear	13% at 6 months; 14% at 12 months; 20% at 18 months	No
Curry et al. 199136	Unclear	Unclear	Unclear	Yes	Yes	Unclear	Yes
Owen et al. 198934	Unclear	Unclear	Unclear	Yes	Yes	12%	Unclear

Comparisons and findings

Electronic interventions compared with no intervention, usual care or untailored printed self-help material

This comparison was addressed by 27 out of the 31 studies comparing computer-generated, tailored printed materials to either no intervention, usual care or untailored printed self-help materials, with or without concurrent co-interventions. Figure 20 shows the results of point prevalence abstinence at 6 months, which was reported in 13 studies. 37,40,44,46,48,53,55–57,61,97,109,110 Pooled result showed that computer-generated, tailored printed materials, when added to varied co-interventions, increase the likelihood of cessation in studies in which smokers are prepared to quit at the beginning of the studies (RR = 1.29, 95% CI 1.10 to 1.51). Tailored printed materials also appear to be effective in studies in which the smokers were not ready to quit at the start of the study, but there was significant heterogeneity in effect estimates (I² = 85%, p < 0.001). Overall, only five studies53,56,88,97,109 demonstrated a statistically significant increase in abstinence rates in the intervention groups compared with the control groups.

FIGURE 20.

Electronic interventions with computer-generated, tailored printed materials vs control: point prevalence abstinence at 6 months.

Figure 21 shows the results of point prevalence abstinence measured at the last follow-up of each study. Twenty-eight studies contributed to the overall summary estimates. There was some heterogeneity (I² = ∼30%) between the studies. Pooled results showed a similar increase in abstinence from the use of computer-generated, tailored printed materials compared with control group regardless of whether a study was a cessation induction study or aid to cessation study. Only six studies37,49,56,88,97,109 demonstrated a statistically significant increase in abstinence rates in the intervention groups compared with control groups. Of these studies, three56,88,109 reported smaller effect size at maximum follow-up compared with 6 months.

FIGURE 21.

Electronic interventions with computer-generated, tailored printed materials vs control: point prevalence abstinence at maximum follow-up.

Figure 22 presents the results for prolonged abstinence at 6 months and are represented by two studies,61,97 which are aid to cessation studies. The pooled result (RR = 1.31, 95% CI 1.00 to 1.71) just reaches statistical significance and is in line with point prevalence abstinence. Similarly, the results of prolonged abstinence at maximum follow-up are also in line with point abstinence at maximum follow-up and are shown in Figure 23.

FIGURE 22.

Electronic interventions with computer-generated, tailored printed materials vs control: prolonged abstinence at 6 months.

FIGURE 23.

Electronic interventions with computer-generated, tailored printed materials vs control: prolonged abstinence at maximum follow-up.

Comparison of electronic interventions with other non-electronic interventions

This comparison was addressed only in one cessation induction study. Meyer et al. 109 compared (up to three) computer-tailored letters with brief advice delivered by GPs, who received 2-hour on-site training in counselling techniques. Similar rates of point prevalence abstinence (tailored letters vs brief advice, 6.4% vs 4.0%, RR = 1.60, 95% CI 0.89 to 2.88 at 6 months; 15.4% vs 14.2%, RR = 1.08, 95% CI 0.79 to 1.49 at 24 months) and prolonged abstinence (tailored letters vs brief advice, 10.2% vs 9.7%, RR = 1.06, 95% CI 0.71 to 1.57) were observed between the groups. The abstinence rates in both intervention groups were significantly higher than an ‘assessment only’ control group within the same study (the ‘tailored letters’ vs ‘assessment only’ comparison has been described in the previous section).

Comparisons between electronic interventions

Six studies40,42–44,46,75 compared different electronic interventions with a single tailored component against each other. The comparisons and findings are summarised in Table 9 and each study is briefly described below. As settings and contents of the electronic interventions being assessed were different, it was not possible to include them in a meta-analysis and compare the electronic interventions with each other.

TABLE 9 - Summary of studies comparing different interventions with a single tailored component

Unless otherwise specified, the reported abstinence rates are 6-month prolonged abstinence and 7-day point prevalence abstinence measured at 6 months (or nearest time points).

Number of participants in individual arms not reported. 95% CIs were calculated assuming equal size in each of the intervention arms.

Study	Study type	Co-intervention	Comparisons	Key findingsa
Comparing different forms/contents/provision of interventions within each mode of delivery (comparison code 3, see Figure 1)
Strecher et al. 200575	Aid to cessation	One initial telephone counselling cessation	A series of four printed materials initially tailored to baseline data and subsequently retailored to follow-up data at 5 months vs a series of four printed materials tailored only to baseline data vs a single tailored smoking cessation guide (Total n for the trial = 1978)	Point prevalence abstinence (at 12 months): Multiple retailored 10.5% vs multiple tailored 10.3% vs single tailored 7.2% RR (multiple retailored + multiple tailored vs single tailored) = 1.43, 95% CI 0.99 to 2.06b Prolonged abstinence: NR
Kreuter and Strecher 199640	Cessation induction	None	‘Enhanced HRA’ providing tailored feedback on risk information + behaviour-change information (n = 94) vs ‘typical HRA’ providing feedback on risk information only (n = 102)	Point prevalence abstinence (at 6 months): Enhanced HRA vs typical HRA 7.4% vs 7.8% RR = 0.95, 95% CI 0.36 to 2.52 Prolonged abstinence: NR
Dijkstra et al. 199842	Cessation induction	None	2 × 2 factorial design. One-off tailored material containing: (information on outcomes of quitting vs no information) × (information on self-efficacy enhancement vs no information) Total n = 1546	Point prevalence abstinence (at 14 months): Outcomes information 6.3% vs no information 7.2% RR = 0.88, 95% CI 0.60 to 1.27 Self-efficacy information 7.2% vs no information 6.3% RR = 1.14, 95% CI 0.79 to 1.65 Prolonged abstinence (at 14 months): Outcomes information 2.3% vs no information 1.6% RR = 1.50, 95% CI 0.73 to 3.09 Self-efficacy information 2.6% vs no information 1.3% RR = 2.00, 95% CI 0.94 to 4.23
Dijkstra et al. 199843	Cessation induction	None	2 × 2 factorial design: (multiple tailored letters vs ST) × (self-help manual vs no manual) Total n = 605	Point prevalence abstinence (at 4 months): Multiple tailored letters 4.1% vs ST 0.6% RR = 6.26, 95% CI 1.41 to 27.75 Self-help manual 1.0% vs no manual 3.6% RR = 0.28, 0.08 to 1.00 Prolonged abstinence: NR
Dijkstra et al. 199944	Cessation induction	None	Three tailored letters (n = 214) vs single tailored letter (n = 206)	Point prevalence abstinence (at 6 months): Three tailored letters 2.8% vs ST 3.9% RR = 0.72, 95% CI 0.26 to 2.05 Prolonged abstinence: NR
Velicer et al. 199946	Cessation induction	None	Different number of mailings of tailored reports: one (n = 357) vs two (n = 353) vs three (n = 362) vs six (n = 357)	Point prevalence abstinence (at 6 months): One mailing 11.2% vs two mailings 9.1% vs three mailings 12.2% vs six mailings 9.0% Prolonged abstinence (at 12 months): One mailing 2.5% vs two mailings 3.4% vs three mailings 5.5% vs six mailings 3.4%

Strecher et al. 75 compared the effectiveness of the following four interventions when added to a brief cognitive–behavioural cessation telephone counselling: (a) a single, untailored smoking cessation guide; (b) a single, computer-tailored smoking cessation guide; (c) a series of four printed materials tailored only to baseline data; and (d) a series of four printed materials initially tailored to baseline data and subsequently retailored to follow-up data at 5 months. Point prevalence abstinence at 12 months was 8.1%, 7.2%, 10.3% and 10.5% for groups a, b, c and d, respectively. A global test for a trend of increasing effectiveness with increasing intensity and tailoring was not significant. In a possibly post hoc analysis, the authors found that the abstinence rates for the two multiple tailored material groups were significantly higher than the two single materials groups (either untailored or tailored). The reported odds ratio (groups c + d vs groups a + b) was 1.41 (95% CI 1.04 to 1.99).

Kreuter and Strecher40 compared two versions of health risk appraisal (HRA), which provided printed feedback to participants by post and covered seven health-related behaviours, including smoking, exercise, seat belt use, alcohol consumption, dietary fat intake, fruit and vegetable consumption, and safe gun storage. The ‘typical HRA’ provided personal feedback on risk information and the ‘enhanced HRA’ provided, additionally, behaviour-change information that was tailored individually. Similar point prevalence abstinence rates were observed at 6 months among adult family practice patients (enhanced HRA vs typical HRA using baseline smokers as denominators, 7.4% vs 7.8%, RR = 0.95, 95% CI 0.36 to 2.52). These abstinence rates were slightly lower than the 9.8% observed in the control group of no feedback within the same study (this comparison was covered in the previous section).

Dijkstra et al. 42 compared the effectiveness of three one-off tailored materials that contained different contents: (a) information on outcomes of quitting; (b) information on self-efficacy enhancement; and (c) information on both. A ‘no-information’ control group was also included. The study could be analysed as a factorial trial (outcomes information vs no information) × (self-efficacy information vs no information), as there appeared to be no significant interaction between the two factors. The inclusion of either information did not increase point prevalence abstinence at 14 months. Prolonged abstinence was increased slightly by inclusion of either information in the tailored material, but the increase was not statistically significant.

Dijkstra et al. 43 incorporated a factorial design of [multiple tailored letters vs single tailored letter (ST)] × (self-help manual vs no manual). Tests for interaction were not significant. The study specifically recruited smokers who were not planning to quit within the next 6 months. The point prevalence abstinence at 4 months was significantly higher in the two groups receiving multiple tailored letters (4.1%) than in the two groups receiving a ST (0.6%). The addition of a self-help manual appeared to have negative impact on point prevalence abstinence.

Dijkstra et al. 44 compared an intervention with three tailored letters to an intervention with a ST in a further study of smokers with low readiness to quit. Contrary to the findings of the previous study, no significant difference in point prevalence abstinence between groups receiving multiple tailored letters and a ST was found.

Velicer et al. 46 conducted a trial with a factorial design of (tailored report + manual vs manual only) × (one vs two vs three vs six contacts). They found no dose–response relationship between the number of contacts/mailings and abstinence rates in both the ‘tailored report + manual’ group and the ‘manual-only’ group.

Other comparisons

In an aid to cessation trial conducted by Borland et al. 61 among callers to a quitline service in Australia, the combination of computer-tailored letters plus call-back telephone counselling generally produced higher abstinence rates than computer-tailored letters alone, although the differences were not statistically significant when data were analysed on an ITT basis assuming that participants who were lost to follow-up did not quit. Point prevalence abstinence (not clearly defined) was 20.8% compared with 16.8% (RR = 1.24, 95% CI 0.96 to 1.59) at 6 months and 18.9% compared with 17.3% (RR = 1.09, 95% CI 0.84 to 1.41) at 12 months for the tailored letters + telephone counselling group compared with the tailored letters group. Three-month prolonged abstinence measured at 6 months (11.2% vs 8.0%, RR = 1.39, 95% CI 0.96 to 2.03) and 9-month prolonged abstinence measured at 12 months (6.1% vs 4.6%, RR = 1.32, 95% CI 0.79 to 2.21) showed a similar trend.

Prochaska et al. 37 compared a combined intervention incorporating four proactive counsellor calls, computer-tailored reports and stage-matched manuals (MANs) with a similar intervention without the counsellor calls. The intervention with added counsellor calls in this study produced lower abstinence rates compared with the intervention with tailored reports only. Reported point prevalence abstinence rates were 14% vs 16% at 6 months and 18% vs 25% at 18 months for the intervention with and without the addition of counsellor calls. Reported prolonged abstinence rates were also higher in the group without counsellor calls. The number of participants for individual arms was not reported in this study and the abstinence rates mentioned above appeared to have been calculated using available data (i.e. not ITT).

Curry et al. 39 investigated the effect of adding three telephone counselling calls to an intervention that consisted of computer-tailored feedback and a self-help manual among smokers in a large health maintenance organisation identified through a telephone survey. Higher abstinence rates were observed in the counselling + feedback group compared with the feedback-only group. The difference was statistically significant for point prevalence abstinence at 3 months (11% vs 4%, RR = 2.87, 1.43 to 5.75) but not at 12 months (11% vs 9%) and 21 months (15% vs 10%). No significant difference was found for continuous abstinence during 3–12 months (counselling + feedback 5% vs feedback alone 3%) and 3–21 months (4% vs 2%).

Hall et al. 80 conducted a trial comparing a staged care intervention that incorporated computer-tailored feedback, psychological counselling and pharmacotherapy to a brief contact control in which self-help materials and a list of local providers of smoking cessation programmes were provided. Participants were adult smoking patients receiving treatment for depression in an outpatient setting. Patients who received the staged care intervention reported higher 7-day point prevalence abstinence than the patients in the control group at 3 months (13.5% vs 9.4%, RR = 1.43, 95% CI 0.77 to 2.66), 12 months (14.1% vs 9.4%, RR = 1.50, 95% CI 0.81 to 2.76) and 18 months (18.4% vs 13.2%, RR = 1.39, 95% CI 0.83 to 2.33), but not at 6 months (14.1% vs 15.7%, RR = 0.90, 95% CI 0.53 to 1.51).

Summary and discussion

Overall, 30 studies34,36–40,42–44,46,48,49,53,55–58,61,66,69,72,75,80,88,91,96,97,109,110 that randomised between 72 and 3627 participants and followed up between 3 and 30 months evaluated the effect of tailored printed materials. In the meta-analysis, computer-generated, tailored printed materials were effective in increasing smoking cessation compared with no intervention or generic self-help materials. This finding was consistent irrespective of the study type (aid to cessation study or cessation induction study), type of outcome associated with smoking cessation measured (point prevalence abstinence or prolonged abstinence), the time at which follow-up was measured (6 months’ follow-up or maximum follow-up).

Six studies40,42–44,46,75 assessed comparisons between various electronic interventions. As settings and contents of the electronic interventions being assessed were different, it was not possible to include them in a meta-analysis and compare the electronic interventions with each other. However, four studies43,44,46,75 compared the effect of multiple tailored printed materials over a single tailored letter. There was no clear evidence to suggest multiple interventions was better than a single intervention as two studies43,75 favoured multiple interventions, whereas in the other two studies44,46 there was no significant difference between multiple tailored printed letters and a single letter.

Stand-alone computer program

Six studies (two aid to cessation studies,33,113 four cessation induction studies47,60,62,78) evaluated electronic interventions with stand-alone computer programs. The characteristics of the studies and participants, results of quality assessment and key findings are described below.

Interventions and co-interventions

Table 10 presents the characteristics of the included studies and their participants. In five studies33,60,62,78,113 the interventions contained stand-alone computer programs only. Of the other two studies, one had telephone hotline, stop-smoking contest and self-help material as a co-intervention in each of the study arms,33 whereas the other had brief advice by midwives and self-help manual as a co-intervention. 62

TABLE 10 - Characteristics of included studies which evaluated electronic interventions with stand-alone computer programs

NR, not reported.

Study	Co-intervention	Population, key criteria related to smoking history and method of recruitment	Mean age (years); % female	Intervention, comparators and sample size (n)	Outcome measure and length of follow-up; comparison code [0–5] (see Figure 1 for the coding scheme)
Aid to cessation studies
Prokhorov et al. 2008 USA113,114	None?	Young adult/participants began smoking regularly at 16 years of age and smoked 12.5 cigarettes per day/announcements by college instructors, public service announcements in student newsletters and newspapers, school marquee announcements, and flyers	22.8; 59	‘Look At Your Health’ LAYH computer-assisted, smoking cessation counselling (219) vs standard care (207)	7-day abstinence rate Follow-up: 10 months [5]
Burling et al. 1989 USA33	Telephone hotline + stop-smoking contest + self-help material	Veterans’ administration employees/recruitment?	43.6; 57	Computer + contest (29) vs contest only (29)	Abstinence (details NR) Follow-up: 3 and 6 months [0]
Cessation induction studies
Lawrence et al. 2003 UK62–64	Brief advice by midwives + self-help manual	Pregnant women/via antenatal clinics	NR; 100	Computer (324) vs manuals (305) vs control subjects (289)	Sustained abstinence assessed at 28–30 weeks and 10 days postnatally [0] [5]
Riley et al. 2002 USA60	None	Adult smokers/regular smokers (15–50 cigarettes per day for ≥ 1 year)/television advertisements	44.8; 44	Stand-alone computer program (44) vs selective elimination via manual instructions (49)	7-day point prevalence abstinence Follow-up: 6 and 12 months [5]
O’Neill et al. 2000 USA47	None	Young adult smokers/daily smokers/via telephone	19.7; 63	Computer-administered intervention on cigarette smoking (31) vs computer-administered intervention on other health behaviours (34)	1 ° outcome: stage of change 2 ° outcome: abstinence rate Follow-up: 7 months [3]
Wolfenden et al. 2005 Australia78,79	None	Surgical patients having non-cardiac elective surgery with pre-existing health problems, who are smokers aged 18 years and above, could read English, had booked day for surgery	43; 62	Interactive, tailored counselling delivered by computer + tailored self-help material + telephone counselling (124) vs usual care (86)	Abstinence (not defined) Follow-up: 3 months [5]

Study design and characteristics of participants

All of the six studies were RCTs that randomised participants individually. Length of follow-up ranged from 378 to 12 months. 60 Three studies47,60,113 evaluated point prevalence abstinence. Two studies33,78 did not define the abstinence. The study by Lawrence et al. 62 reported sustained abstinence assessed at 28–30 weeks of gestation and 10 days postnatally. The study by O’Neill et al. 47 reported stage of change and abstinence rate.

Participants included responders to invitations to participate in the trial advertised through various media: pregnant women via antenatal clinics, surgical patients having non-cardiac elective surgery with pre-existing health problems, and patients undergoing outpatient surgery or invasive screening in a large managed care organisation.

Mean age ranged from 19.7 years47 to just over 44.8 years. 60 Proportion of females varied from 44%60 to 100%. 62

Quality assessment of included studies

The results of quality assessment are presented in Table 11. None of the studies provided information on blinding. Four studies33,60,62,113 included biochemical validation of smoking status. Loss to follow-up varied significantly and ranged from 7% (at 6 months)33 to 45% 60 (at 12 months). Three studies60,62,78 included ITT analysis.

TABLE 11 - Quality assessment of included studies that evaluated electronic interventions with stand-alone computer programs

Study	Generation of random sequence	Allocation concealment	Blinding	Biochemical validation	Baseline characteristics similar between groups	Lost to follow-up	ITT analysis
Aid to cessation studies
Prokhorov et al. 2008113,114 USA	Yes	Unclear	Unclear	Yes	Yes	Overall: 100/426 (23.5%) Computer-assisted counselling: 61/219 (27.9%) Standard care: 39/207 (18.8%)	No
Burling et al. 1989 USA33	Unclear	Unclear	Unclear	Yes	Unclear	4/58 (6.9%)	Unclear
Cessation induction studies
Lawrence et al. 2003 UK62–64	Yes	Unclear	Unclear	Yes	Yes	Postnatal follow-up: Computer: 249/324 (77%) Manual: 219/305 (72%) Control: 185/289 (64%)	Yes
Riley et al. 2002 USA60	Unclear	Unclear	Unclear	Yes	Yes	Stand-alone computer program: 14/44 (31.8%) Selective elimination via manual instruction: 22/49 (44.9%)	Yes
O’Neill et al. 2000 USA47	Unclear	Unclear	Unclear	No	Unclear	Overall: 9/65 (13.8%) Computer-intervention on smoking cessation: 4/31 (12.9%) Computer-intervention on other health behaviours: 5/34 (14.7%)	No
Wolfenden et al. 2005 Australia78,79	Yes	Unclear	No	No	Yes	Smoking cessation programme = 16/124 Usual care = 10/86	Yes

Comparisons and findings

Electronic interventions compared with no intervention, usual care or untailored printed self-help material

One aid to cessation study (Burling et al. 33) and one cessation induction study (Lawrence et al. 62) compared stand-alone computer programs with either no intervention, usual care or untailored printed self-help material, with or without concurrent co-interventions.

Figure 24 shows the results of point prevalence abstinence measured at the last follow-up of each study. The results for the study (aid to cessation) by Burling et al. 33 at 6 months showed that the stand-alone computer program intervention increased cessation compared with control but the sample size was small (n = 58) and the difference was not statistically significant (RR = 2.00, 95% CI 0.55 to 7.24). The results for the study (cessation induction) by Lawrence et al. 62 showed that the stand-alone computer program intervention was not effective at 18 months (RR = 1.06, 95% CI 0.57 to 1.97).

FIGURE 24.

Stand-alone program vs control: point prevalence abstinence at maximum follow-up.

Comparisons between electronic interventions

One study by O’Neill et al. 47 reported point prevalence rate of 29% (9/31) for intervention and 21% (7/34) for control at 3 months, and 30% (9/31) for intervention and 32% (11/34) for control at 6 months. It also reported prolonged abstinence rate of 19% (6/31) for intervention and 15% (5/34) for control at 6 months. However, both point prevalence rate and prolonged abstinence rate were not defined and none of the differences reached statistical significance.

Other comparisons

Four studies60,62,78,113 estimated the effect of the combined electronic and non-electronic intervention compared with control. However, the effects of the electronic component cannot be separated out. The studies by Prokhorov et al. 113,114 compared computer-assisted motivational counselling and tailored material with brief counselling. The studies by Lawrence et al. 62–64 compared computer programs, printout by post and stage-based brief advice with non-stage-based brief. The study by Riley et al. 60 compared computerised scheduled gradual reduction (LifeSign) and programme guide with manual-based selective elimination. The studies by Wolfenden et al. 78,79 compared computer counselling program, telephone counselling and NRT with brief advice.

Summary and discussions

Six studies33,47,60,62,78,113 evaluated the use of a stand-alone computer program for smoking cessation. Only two of them33,62 compared the intervention with no intervention, usual care or untailored printed self-help material. Stand-alone computer program did not significantly increase cessation, but the number of trials and sample sizes were too small to allow a firm conclusion to be drawn.

Mobile telephone/interactive voice response-based intervention

Five studies (two aid to cessation studies,73,120 three cessation induction studies45,95,121) evaluated electronic interventions with mobile telephone or IVR. The characteristics of the studies and participants, results of quality assessment and key findings are described below.

Interventions and co-interventions

Table 12 presents the characteristics of the included studies and their participants. The interventions of three studies73,120,121 contained mobile telephone text messaging only. One study45 had an intervention of IVR and a co-intervention of self-help material. The other study95 had an intervention of IVR and a co-intervention of inhospital brief advice, NRT, counselling and self-help material.

TABLE 12 - Characteristics of included studies that evaluated mobile telephone/IVR-based intervention

SMS, short message service.

Study	Co-intervention	Population, key criteria related to smoking history and method of recruitment	Mean age (years); % female	Intervention, comparators and sample size (n)	Outcome measure and length of follow-up; comparison code [0–5] (see Figure 1 for the coding scheme)
Aid to cessation studies
Free et al. 2009120 UK	None	Aged ≥ 16 years, currently smoking cigarettes daily and interested in quitting, currently owns a mobile telephone, living within an hour of London and familiar with text messaging	36; 38	The txt2stop intervention (102) vs control (98)	7-day point prevalence and 28-day continuous abstinence Follow-up: 4 weeks and 6 months [1]
Rodgers et al. 200573 New Zealand	None	Aged ≥ 16 years, smokers (smoking cigarettes daily) who wanted to quit in next month (i.e. in contemplative stage of change) and owned a mobile telephone and able to send/receive texts	22; 59	Regular personalised text messages providing advice, support and distraction (852) vs one text message every 2 weeks, which only meant to thank them for being in the study, provide study contact details and inform them of the rewards for participants (853)	7-day point prevalence and 24-week continuous abstinence Follow-up: 6 months [1]
Cessation induction studies
Haug et al. 2009121 Germany	None	Young adults/daily smokers/at the university cafeteria	25.0; 57	Three SMS (60) vs one SMS (50) vs no SMS (64)	7-day point prevalence abstinence Follow-up: 3 months [0][3]
Reid et al. 200795 Canada	In-hospital brief advice, NRT, counselling and self-help material	Current smokers (five or more cigarettes per day) aged > 18 years, hospitalised for acute coronary syndrome, percutaneous coronary intervention or diagnostic catheterisation related to coronary heart disease	54; 32	IVR group (50) vs usual care (49)	7-day point prevalence abstinence Follow-up: 3 and 12 months [0]
Ershoff et al. 199945 USA	Self-help material	English-speaking pregnant women/smoked average 17 cigarettes/day/during antenatal booking	29.4; 100	Booklet only (131) vs motivational interview (126) vs IVR (133)	7-day abstinence (biochemically confirmed) Follow-up: 4 weeks [0][2]

Study design and characteristics of participants

All the five studies45,73,95,120,121 were RCTs that randomised participants individually. Length of follow-up ranged from 4 weeks45 to 12 months. 95 All the studies evaluated 7-day point prevalence abstinence. The study by Rodgers et al. 73 also measured 24 weeks’ continuous abstinence.

Participants included those aged ≥ 16 years: (1) currently smoking cigarettes daily and interested in quitting, currently owning a mobile telephone; (2) currently smokers who were hospitalised for acute coronary syndrome, percutaneous coronary intervention or diagnostic catheterisation related to coronary heart disease; or (3) presenting as English-speaking pregnant women smokers during antenatal booking.

The mean age ranged from 2273 to 54 years. 95 The proportion of females varied from 32%95 to 58–59%. 73

Quality assessment of included studies

The results of the quality assessment are presented in Table 13. Most of the studies were single-blind for care providers. 45,73,120 The study by Haug et al. 121 was blinded for participants, and the study by Reid et al. 95 did not report blinding information.

TABLE 13 - Quality assessment of included studies which evaluated mobile telephone/IVR-based intervention

SMS, short message service; UC, usual care.

Study	Generation of random sequence	Allocation concealment	Blinding	Biochemical validation	Baseline characteristics similar between groups	Lost to follow-up	ITT analysis
Aid to cessation studies
Free et al. 2009120	Unclear	Yes	Single blind	Yes	Unclear	Control = 8/98 Intensive group = 8/102	Yes
Rodgers et al. 200573	Yes	Yes	Care providers	Yes	Yes	Intervention 261/852 (30.6%) Control 179/853 (21%)	Yes
Cessation induction studies
Haug et al. 2009121	Yes	Unclear	Participants	No	Yes	Three SMS – 9/60 (15%) One SMS – 5/50 (10%) No SMS – 5/64 (7.8%)	Yes
Reid et al. 200795	Yes	Yes	Unclear	No	No (education level)	UC group = 83.7% IVR group = 86%	Yes
Ershoff et al. 199945	Unclear	Unclear	Care providers	Yes	Yes	58/390 (14.9%)	No

Three studies45,73,120 included biochemical validation of smoking status. Loss to follow-up varied significantly and ranged from 8% (at 4 weeks, 3 months)120,121 to 86–87%95 (at 12 months). Four studies73,95,120,121 included ITT analysis.

Comparisons and findings

Electronic interventions compared with no intervention, usual care or untailored printed self-help material

Two aid to cessation studies73,120 and three cessation induction studies45,95,121 compared mobile telephone text messaging or IVR with either no intervention, usual care or untailored printed self-help material, with or without concurrent co-interventions.

Figure 25 shows the results of point prevalence abstinence at 6 months. This outcome was reported in two aid to cessation studies,73,120 neither of which gave a significant result. The pooled results showed that the mobile text messaging intervention was not effective (RR = 1.03, 95% CI 0.83 to 1.28, I² = 10.4%).

FIGURE 25.

Mobile telephone text messaging vs control: point prevalence abstinence at 6 months.

Figure 26 shows the results of point prevalence abstinence measured at the last follow-up of each study. In addition to the two aid to cessation studies73,120 shown in Figure 25, three cessation induction studies reported this outcome at various durations of follow-up,45,95,121 two of which showed increased cessation in the intervention group but neither of which gave a significant result (see lower part of Figure 26). The pooled result did not reach statistical significance (RR = 1.08, 95% CI 0.83 to 1.40, I² = 4%).

FIGURE 26.

Mobile telephone text messaging vs control: point prevalence abstinence at maximum follow-up.

Comparison of electronic interventions with other non-electronic interventions

The study by Ershoff et al. 45 compared the effectiveness of IVR with motivational interview having self-help booklet as a co-intervention in each arm. The result showed no significant difference for IVR compared with motivational interview (RR = 0.81, 95% CI 0.39 to 1.69).

Comparisons between electronic interventions

The study by Haug et al. 121 compared three short message service (SMS) with one SMS. The result showed that three SMS did not perform significantly better than one SMS (RR = 1.25, 95% CI 0.82 to 1.91).

Summary and discussion

Five studies45,73,95,120,121 evaluated the use of mobile telephone text messaging or IVR for smoking cessation. No significant increase in abstinence rates was observed when these interventions were compared with no intervention, usual care or untailored printed self-help material, and to other non-electronic interventions. However, the number of RCTs is too small to allow a firm conclusion to be drawn.

Internet sites

Six studies (four aid to cessation studies,35,87,117,123 two cessation induction studies111,112) evaluated internet sites with/without co-interventions. The characteristics of the studies and participants, results of quality assessment and key findings are described below.

Interventions and co-interventions

Table 16 presents the characteristics of included studies and their participants. The interventions of three studies35,87,111 contained internet sites only. The studies by Oenema et al. 111 and Swartz et al. 87 evaluated the effectiveness of the internet sites compared with no intervention or delayed intervention. The study by Schneider et al. 35 evaluated the effectiveness of the internet sites with different contents. The other three studies had NRT,117 Health Risk Intervention,112 and automated e-mails plus online mood management course123 as a co-intervention in each of the study arms. The study by Strecher et al. 117 evaluated the effectiveness of interventions with a variation of each of the five two-level intervention factors related to the depth of tailoring of different components, personalisation of introductory messages and time schedule of the interventions. These factors will be further described later in this section under ‘comparison between electronic interventions’. The study by Prochaska et al. 112 evaluated the effectiveness of Online Transtheoretical Model and Motivational Interviewing with the adjunct of Health Risk Intervention. The study by Muñoz et al. 123 evaluated the effectiveness of Guía (static website) with or without individually timed educational messages (ITEM) compared with Guía with ITEM and an eight-lesson cognitive behaviour mood management course.

TABLE 16 - Characteristics of included studies which evaluated internet sites with a single tailored component

HMO, health maintenance organisation; HRI, health risk intervention; ITEM, individually timed educational message; MI, motivational interviewing; MM, mood management; TTM, transtheoretical model; VG, virtual group.

Study, country	Co-intervention	Population, key criteria related to smoking history and method of recruitment	Mean age (years); % female	Intervention, comparators and sample size (n)	Outcome measure and length of follow-up; comparison code [0–5] (see Figure 1 for the coding scheme)
Aid to cessation studies
Swartz et al. 200687 USA	None	Adults currently smoking cigarettes on daily basis/worksites advertisements and links to websites, electronic newsletter, e-mails	NR; 52	1–2–3 Smokefree website (171) vs delayed access (180)	7-day abstinence Follow-up: 90 days [0]
Schneider et al. 199035 USA	None	Adult subscribers to the CompuServe Information System who smoked cigarettes daily; recruited via articles in magazines and online newsletter	NR (≥ 18); 16	2 × 2 factorial design (1158) Interactive computer program vs non-interactive program (both delivered via internet) × internet forum vs no forum	Abstinence (not defined) Follow-up: 1, 3 and 6 months [3]
Strecher et al. 2008 USA117,118	NRT	Members of two HMOs; smoking at least 10 cigarettes per day and considering quitting within the next 30 days; invitation letter and HMO newsletter	46.3; 60	Study participants (1866) received a variation of each of the five two-level intervention factors: (1) depth of tailored outcome expectation feedback; (2) depth of efficacy expectations; (3) depth of success stories; (4) personalisation of source; and (5) exposure schedule	7-day point prevalence abstinence Follow-up: 6 months [3]
Muñoz et al. 2009123 Multicountry	None	Young adults/smoking at least five cigarettes/day/recruited online	37.9; 45	Guía: static website (247) vs Guía ITEMs (251) vs Guí + ITEMs + plus eight-lesson cognitive behaviour MM course (251) vs Guí + ITEM + MM + VG (251)	7-day point prevalence abstinence Follow-up: 1, 3, 6 and 12 months [4]
Cessation induction studies
Oenema et al. 2008111 The Netherlands	None	Adults ≥ 30 years of age from a pool of an online research panel, recruited by e-mail invitation	43.6; 54	A website with tailored information modules on saturated fat intake, physical activity and smoking cessation (1080) vs delayed intervention (1079)	Self-reported smoking cessation (not defined) and forward stage transition Follow-up: 1 month [0]
Prochaska et al. 2008112 USA	Online health risk assessment and HRI	Employees at a major medical university/9.7% current smoker/sent persuasive letter and e-mail, some given incentives to participate. Some received follow-up telephone call	41.6; 79	HRI (464) vs online: TTM + HRI (504) vs MI + HRI (433)	Stage of change, point prevalence abstinence Follow-up: 6 months [2][3]

Study design and characteristics of participants

All the six studies35,87,111,112,117,123 were RCTs that randomised participants individually. Length of follow-up ranged from 1111 to 12 months. 123 Four studies87,112,117,123 evaluated point prevalence abstinence. The study by Oenema et al. 111 reported the smoking prevalence of the current, ex-smokers or never smokers. The study by Schneider et al. 35 reported the outcome of abstinence, which was not defined. Participants were responders to invitations to participate in the trial advertised through various media. Mean age ranged from 37.9 years123 to just over 46.3 years. 117 One study focused specifically on young adults. 123 The proportion of female participants varied from 16%35 to 79%. 112

Quality assessment of included studies

The results of the quality assessment are presented in Table 17. Studies varied in the adequacy of methods of blinding. None of the studies included biochemical validation of smoking status. Loss to follow-up varied significantly and ranged from 16% (at 1 month)111 to 56% (at 6 months). 35 The study by Prochaska et al. 112 did not report loss to follow-up. Three studies87,111,117 included ITT analysis.

TABLE 17 - Quality assessment of included studies that evaluated internet site interventions

ITEM, individually timed educational message; NR, not reported; VG, virtual group.

Guía = static website.

Guía + ITEM + MM = static website plus ITEM plus eight-lesson cognitive behaviour mood management course.

Study	Generation of random sequence	Allocation concealment	Blinding	Biochemical validation	Significant difference between groups at baseline	Lost to follow-up	ITT analysis
Aid to cessation studies
Swartz et al. 200687	Yes	Unclear	Open label	No	No	Overall: 154/351 (43.9%) 1–2–3 Smokefree website: 84/171 (49.1%) Delayed access: 70/180 (38.9%)	Yes
Schneider et al. 199035	Yes	Unclear	Unclear	No	No	Overall: 56.3%	Unclear
Strecher et al. 2008117,118	Yes	Yes	Participants	No	Yes	451/1866 (24.2%)	Yes
Muñoz et al. 2009123	Yes	Yes	Outcome assessors	No	Unclear	Guía:a 72/247 (29.1%) Guía + ITEMs: 78/251 (31.1%) Guía + ITEMs + MM:b 78/251 (31.1%) Guía + ITEMs + MMb + VG: 80/251 (31.9%)	Unclear
Cessation induction studies
Oenema et al. 2008111	Yes	Yes	Outcome assessors	No	Yes	Static website: 231/1080 (21.4%) No intervention: 171/1079 (15.9%)	Yes
Prochaska et al. 2008112	Unclear	No	Unclear	No	No	NR	Unclear

Comparisons and findings

Electronic interventions compared with no intervention, usual care or untailored printed self-help material

One aid to cessation study (Swartz et al. 87 and one cessation induction study (Oenema et al. 111) compared internet site interventions with either no intervention, usual care or untailored printed self-help material, with or without concurrent co-interventions.

Figure 27 shows the results of point prevalence abstinence measured at the last follow-up of each study. The results for the study by Swartz et al. 87 at 3 months showed that the internet site intervention is significantly effective, with RR = 2.46, 95% CI 1.16 to 5.21. The results for the study by Oenema et al. 111 at 1 month showed that the internet site intervention is not effective, with RR = 1.28, 95% CI 0.52 to 3.13. Pooled estimate just failed to reach statistical significance (RR = 1.86, 95% CI 0.98 to 3.50).

FIGURE 27.

Internet site interventions vs control: point prevalence abstinence at maximum follow-up.

Comparison of electronic interventions with other non-electronic interventions

One cessation induction study (Prochaska et al. 112) provided information on the relative effectiveness of an internet site intervention (online transtheoretical model) and a non-electronic intervention (three sessions of motivational interviewing conducted face to face or over the telephone) when added to an online health risk intervention. The point prevalence abstinence at 6 months was 21% and 35% for online transtheoretical model and motivational interviewing, respectively. The difference was, however, not statistically significant owing to small sample size, as the study was conducted in a worksite population with a variety of health risk factors (i.e. not just smoking) and only 9.8% of the study participants were smoking at baseline and were included in this comparison.

Comparisons between electronic interventions

Two aid to cessation studies (Strecher et al. ,117 Schneider et al. 35) and two cessation induction studies (Prochaska et al. ,112 Muñoz et al. 2009123) made direct comparisons between different electronic interventions.

In a RCT with factorial design, Schneider et al. 35 evaluated two intervention components: (tailored vs untailored online programme) × (online forum vs no forum). The results of this study are described in the next section (see Interventions with multiple tailored components) (as the tailored programme-plus-online forum arm is considered as having multiple tailored components).

Using a fractional factorial design, Strecher et al. 117,118 conducted a RCT to examine the impact of five intervention components on 7-day point prevalence abstinence at 6 months. They found that two of the intervention components, namely depth of success stories (hypothetic stories of successful cessation created using information on individual characteristics and personal circumstances to address smokers’ motives and barriers to quitting) and personalisation of source (using photographs, signatures and friendly, personal-manner texts in the introduction to the programme) were associated with increased cessation. The other three components including depth of outcome expectations (tailored feedback and advice related to smokers’ motives for quitting), depth of efficacy expectations (tailored feedback and advice related to smokers’ barriers to quitting), and exposure schedule (single instalment vs a series of instalments over 5 weeks) were not significantly associated with cessation.

The aforementioned study by Prochaska et al. 112 also allowed comparison between different electronic interventions. It was found that the point prevalence abstinence at 6 months was 21% for the online transtheoretical model plus health risk intervention and was 17% for the online health risk intervention only. The difference was not statistically significant owing to the small sample size. Comparisons between generic components of the electronic interventions (static online cessation guide vs static guide + automated e-mails vs static guide + automated e-mails + an eight-lesson cognitive–behavioural mood management course) in the study by Muñoz et al. 124 have been described above (see Interventions with single or multiple generic components, Comparisons and findings, Comparisons between electronic interventions). The addition of an asynchronous bulletin board (considered as a tailored component) to these generic components did not significantly increase point prevalence abstinence (with and without bulletin board, 12.7% vs 14.3% at 6 months and 22.7% vs 20.7% at 12 months).

Summary and discussion

Six studies35,87,111,112,117,123 evaluated web-based interventions with a single tailored component. Two of them87,111 compared the interventions with a no-intervention control. Pooled estimate of point prevalence abstinence based on short-term follow-up (≤ 3 months) favours the web-based interventions. The difference reached statistical significance with a fixed-effects model but not a random-effects model.

Interventions with multiple tailored components

Fourteen studies (11 aid to cessation studies,35,70,76,82,92,104,105,107,116,119,124 three cessation induction studies57,88,101) evaluated electronic interventions with multiple tailored components. The characteristics of the studies and participants, results of quality assessment and key findings are described below.

Comparisons and findings

A variety of comparisons were made in the included studies. Relevant pair-wise comparisons were categorised according to the scheme shown in Figure 1 and the findings for each type of comparison are described below.

Electronic interventions compared with no intervention, usual care or untailored printed self-help material

Only six57,82,88,104,105,124 out of the 14 studies compared electronic interventions with multiple tailored components with either no intervention, usual care or untailored printed self-help material, with or without concurrent co-interventions.

Figure 28 shows the results of point prevalence abstinence at 6 months. This outcome was reported in five studies57,82,104,105,124 with varied interventions. Significant heterogeneity in effect estimates was observed. Only the two studies by Brendryen et al. 104,105 demonstrated a statistically significant increase in abstinence rates in the intervention groups compared with control groups. Both studies compared the multimedia Happy Ending programme (web + e-mails + IVR + text messages) with self-help manuals, one with concurrent NRT, the other without. Two of the studies82,124 assessing interactive websites incorporating some form of social interaction/support failed to show a significant increase in abstinence rates when these interventions were added to counselling plus pharmacotherapy. In the study by Prochaska et al. ,57 the combination of EXP reports and LifeSign program did not increase abstinence rates compared with the assessment only control.

FIGURE 28.

Electronic interventions with multiple tailored components vs control: point prevalence abstinence at 6 months.

Figure 29 shows the results of point prevalence abstinence measured at the last follow-up of each study. The effect size for the two studies by Brendryen et al. 104,105 at 12 months appears to become smaller compared with 6 months, mainly owing to increased abstinence in the control groups. One additional study88 showed that the combination of an EXP report and automated telephone counselling did not increase abstinence rates when they were added to NRT and stage-matched self-help manual.

FIGURE 29.

Electronic interventions with multiple tailored components vs control: point prevalence abstinence at maximum follow-up.

Results for prolonged abstinence are shown in Figure 30. The results are broadly in line with point prevalence abstinence.

FIGURE 30.

Electronic interventions with multiple tailored components vs control: 6-month prolonged abstinence at maximum follow-up.

Comparison of electronic interventions with other non-electronic interventions

Two studies (one aid to cessation study124 and one cessation induction study57) provided information on the relative effectiveness of an electronic intervention with multiple tailored components (or one component of the intervention) compared with telephone counselling.

Swan et al. 124 compared an interactive web-based programme (including a discussion forum) with multiple (up to five) proactive telephone counselling in treatment seeking smokers from a large health-care organisation in the USA. All of the participants received concurrent varenicline. Seven-day point prevalence abstinence was lower among smokers receiving the web-based programme than those receiving proactive telephone counselling. The difference between groups was statistically significant at 3 months but not at 6 months (30.7% vs 34.3%, RR = 0.89, 95% CI 0.73 to 1.09).

Prochaska et al. 57 evaluated the effect of adding either LifeSign program or counsellor calls to an EXP intervention that included three computer-tailored reports and a stage-matched self-help manual. Abstinence rates were similar or lower in the LifeSign + EXP group compared with counsellor calls + EXP group but the difference was not statistically significant (7-day point prevalence abstinence at 6 months 8.3% vs 11.6%, RR = 0.71, 95% CI 0.46 to 1.11; prolonged abstinence at 12 months 5.1% vs 5.3%, RR = 0.97, 95% CI 0.52 to 1.79). Neither study arms appeared to demonstrate additional benefit compared with a third arm of the trial in which smokers received the EXP alone. The comparison between LifeSign + EXP versus EXP alone is described in the next section.

Comparisons between electronic interventions

Ten studies35,57,70,76,88,92,101,107,116,119 allowed direct comparisons between different electronic interventions. These comparisons were further classified according to whether the studies compared different provision of contents/intensity/level of tailoring for interventions that were delivered through the same type of media (comparison code 3, see Figure 1), or whether the studies compared different combinations of electronic intervention delivered through different types of media (comparison code 4, see Figure 1).

Table 20 summarises the comparisons made and key findings from these studies. As settings and contents of the electronic interventions being assessed were different, it was not possible to include them in a meta-analysis and compare the electronic interventions with each other. A brief description of each study is provided below.

TABLE 20 - Studies which directly compared different contents, intensity and/or combinations of electronic interventions

BB, bulletin board; MP, modified programme; OP, original programme; QSN, Quit Smoking Network; TEL, automated telephone counselling.

Unless otherwise specified, the reported abstinence rates are 6-month prolonged abstinence and 7-day point prevalence abstinence measured at 6 months (or nearest time points).

Number of participants in individual arms not reported. 95% CIs were calculated assuming equal size in each of the four intervention arms.

Assuming participants who reported point prevalence abstinence at both 3 and 6 months also achieved 6-month prolonged abstinence.

Study	Study type	Co-intervention	Comparison	Key findingsa
Comparing different forms/contents/provision of interventions within each mode of delivery (comparison code 3; see Figure 1)
Schneider et al.199035	Aid to cessation	None	2 × 2 factorial design: (tailored online programme vs untailored online programme) × (forum vs no forum) (total n = 1158)	Point prevalence abstinence: Tailored programme 10.5% vs untailored programme 8.4% (RR = 1.24, 95% CI 0.87 to 1.78b) Forum 9.6% vs no forum 9.2% (RR = 1.06, 95% CI 0.74 to 1.51b) Prolonged abstinence: NR
Etter et al. 200570	Aid to cessation	Interactive smoking cessation website Stop-tabac.ch	Tailored web-based programme (n = 4346) vs modified web-based programme focusing more on NRT and nicotine dependence and less on health risks and coping strategies (n = 4336)	Point prevalence abstinence (at 2.5 months): OP 10.9% vs MP 8.9% (RR = 1.21, 95% CI 1.07 to 1.38) Prolonged abstinence: NR
Pike et al. 200792,93	Aid to cessation	None	Five interactive websites (n = 5404) vs one static website (n = 1047)	Point prevalence abstinence (at 3 months): Interactive websites 11.0% vs static website 10.9% (RR = 1.01, 95% CI 0.84 to 1.22) Prolonged abstinence: NR
Mckay et al. 2008107	Aid to cessation	None	Tailored web-based smoking cessation programme (QSN) (n = 1159) vs tailored web-based programme for promoting physical activity (Active Lives) (n = 1159)	Point prevalence abstinence: QSN 9.7% vs Active Lives 10.4% (RR = 0.93, 95% CI 0.73 to 1.19) Prolonged abstinence: c QSN 3.9% vs Active Lives 3.8% (RR = 1.02, 95% CI 0.68 to 1.54)
Stoddard et al. 2008116	Aid to cessation	Interactive smoking cessation website smokefree.gov	BB (n = 684) vs no BB (n = 691)	Point prevalence abstinence (at 3 months): With BB 6.6% vs without BB 6.9% (RR = 0.95, 95% CI 0.64 to 1.40) Prolonged abstinence: NR
Etter 2009119	Aid to cessation	Interactive smoking cessation website Stop-tabac.ch	One-off tailored report vs one-off untailored report	24-hour point prevalence abstinence (at 2 days): Tailored report 12.1% vs untailored report 12.1% (RR = 1.00, 95% CI 0.79 to 1.25) Prolonged abstinence: NR
Comparing different (combinations of) electronic interventions (comparison code 4; see Figure 1)
Strecher et al. 2005b76,77	Aid to cessation	NRT	Tailored programme (website + e-mails) vs untailored website	10-week prolonged abstinence (at 3 months): Tailored programme 20.1% vs untailored website 15.9% (RR = 1.26, 95% CI 1.11 to 1.44) Point prevalence abstinence: NR
An et al. 2008101	Cessation induction	Stop smoking contest	RealU (tailored website + multiple tailored e-mails by peer coaches) vs one-off untailored e-mail control	Prolonged abstinence (at 7 months): Stated no difference between groups (overall 6%) Point prevalence abstinence (at 7 months): RealU 59.1% vs one-off untailored e-mail 38.5% (RR = 1.54, 95% CI 1.28 to 1.85)
Prochaska et al. 200157	Cessation induction	Stage-matched self-help manual	Three EXP tailored reports + LifeSign (n = 374) vs EXP (n = 362)	Prolonged abstinence (at 12 months): ES + LifeSign 5.1% vs EXP 6.1% (RR = 0.84, 95% CI 0.46 to 1.52) Point prevalence abstinence: EXP + LifeSign 8.3% vs EXP 11.6% (RR = 0.71, 95% CI 0.46 to 1.11)
Velicer et al. 200688	Cessation induction	Stage-matched self-help manual + NRT	EXP report + TEL (n = 500) vs EXP (n = 509)	Prolonged abstinence (at 10 months): EXP + TEL 6.6% vs EXP 5.4% (RR = 1.24, 95% CI 0.76 to 2.04) Point prevalence abstinence (at 10 months): EXP + TEL 11.1% vs EXP 9.7% (RR = 1.16, 95% CI 0.81 to 1.67)

Schneider et al. 35 conducted a trial evaluating one of the earliest internet-based smoking cessation programmes in the late 1980s. The trial adopted a factorial design that allowed comparison in a full tailored version of the programme compared with an untailored version, and between the provision of an internet forum compared with no forum. Test for interaction between the two factors was not reported. Calculation assuming equal sample size between treatment arms suggested no significant interaction, and thus effect estimates assuming no interaction are presented in Table 20. Smokers who received the untailored version and had no access to the forum achieved the lowest abstinence, but the differences between groups and between the tested factors were not statistically significant at 6 months.

Etter70 compared an internet-based, tailored programme designed according to psychological and addiction theory with a modified programme (MP) that was designed for users of nicotine replacement products (but NRT was not part of the intervention within this trial). Both programmes were run within an interactive website (Stop-tabac.ch), from which the visitors were recruited to the study. The results suggested the original programme was more effective than the MP.

Pike et al. 92,93 compared five interactive websites (combined as a group) with a static website and found no significant difference in point prevalence abstinence between them. The contents, features and utilisation by smokers varied between the five interactive websites, and the authors suggested in a post hoc analysis that point prevalence abstinence was higher among two highly utilised sites compared with three less-utilised sites (12.2% vs 10.2%, RR = 1.18, 95% CI 1.01 to 1.38).

Mckay et al. 107 compared two web-based smoking cessation programmes: the Quit Smoking Network, which included tailored web pages, a peer-to-peer web forum and a professionally moderated forum; and the Active Lives programme, which was a web-based exercise enhancement programme that was ‘adapted somewhat to encourage smoking cessation’ and included tailored web pages and peer support forum. No significant difference in abstinence rates was observed between the groups.

Stoddard et al. 116 found that the addition of a bulletin board to an interactive website (www.smokefree.gov) did not further improve the effectiveness of the website, and the utilisation of the bulletin board was low. Only 12% (81/684) of smokers allocated to this arm actually viewed or posted individual messages.

Etter119 compared the immediate effect (2 days after enrolment) of a single tailored report with a single untailored report among visitors of an aforementioned interactive smoking cessation website (Stop-tabac.ch) and found no difference between the two groups.

Strecher et al. 76,77 compared a web-based computer-tailored programme with web-based non-tailored materials as a supplement to nicotine patch therapy. The tailored programme included a web-based cessation guide and tailored newsletters, behaviour support messages delivered via e-mails and tailored advice delivered by e-mail to a supportive person identified by the smoker. Results indicated that the tailored programme was more effective than the untailored web materials.

An et al. 101 assessed the RealU online cessation intervention at the University of Minnesota. The RealU intervention consisted of an interactive website, online magazine and peer e-mail support. Compared with a control group which received an one-off e-mail containing links to online cessation, health and academic support websites, smokers in the RealU group reported a significantly higher point prevalence abstinence at 30 weeks. However, there was no difference in prolonged abstinence between groups. The observed point prevalence abstinence rates in both RealU and control groups were higher than those observed in other studies. The authors attributed the finding to the high rates of occasional smoking at baseline. A campus-wide Quit and Win contest that took place around the 30-week assessment could also have had some impact. Given the nature of the intervention (cessation induction), the follow-up at 30 weeks was likely to be too short to capture any longer-term impact of the intervention.

The study by Prochaska et al. 57 mentioned in the previous section allowed the comparison of the combination of LifeSign program and an EXP intervention compared with the EXP intervention alone. The combined intervention achieved lower abstinence rates compared with the EXP intervention alone, but the differences between the two arms were not statistically significant.

In a study by Velicer et al. ,88 the addition of regular automated telephone counselling (which utilised a series of pre-recorded voice files to form a conversation that was tailored to the smoker’s responses to assessment questions entered using a telephone keypad) to one tailored report generated by an EXP produced slightly higher abstinence rates than the tailored report alone, but the differences between the two arms were not statistically significant. Smokers in both arms also received stage-based self-help manuals and NRT (when the smokers progressed to the appropriate stages).

Summary of findings

This effectiveness review includes 60 RCTs/quasi-RCTs reported in 77 publications that evaluate the use of computers and other electronic aids for smoking cessation. The results of meta-analyses show that computer and other electronic aids increase the likelihood of cessation compared with no intervention or generic self-help materials, but the effect is small. The effectiveness does not appear to vary with respect to mode of delivery and concurrent non-electronic co-interventions. Overall, similar sizes of effect are observed in both aid to cessation studies (in smokers who are ready to quit) and cessation induction studies (in smokers who are not yet ready to quit), but there is substantial heterogeneity among the latter (cessation induction studies), particularly when 6-month point prevalence abstinence is used as the outcome measure.

Our review has several points of strength:

• Compared with previously published reviews that have focused on specific types of computer and/or other electronic aids, this review is wider in its scope and encompasses all interventions that make use of automated features brought by the advance in information technology and telecommunication in the past couple of decades. The broader scope allows us to include a larger evidence base in this review and to examine the potential impact of different computer/electronic tools on the effectiveness of the interventions.

• Comprehensive searches were undertaken. Study selection, data extraction and coding followed standardised procedure and explicit criteria.

• Where sufficient data were available, meta-analyses were carried out following a stringent ITT principle, which minimises one of the major threats on the validity of effect estimates owing to loss to follow-up.

• The effectiveness review which focuses on evidence from RCTs/quasi-RCTs is complemented by a supplementary review and a cost-effectiveness review examining other types of relevant evidence.

This review has some limitations which should be considered:

• The review focuses on smoking cessation programmes in the adult population. It does not cover smoking cessation in adolescents, which, in our view, is a distinct population warranting a separate review.

• Although we have used explicit criteria based on evidence from previous reviews to categorise electronic interventions with respect to the nature of tailoring and the number of components, the bulk of included evidence falls within one specific category (interventions with a single tailored component) and this has somewhat limited the utility of the categorisation.

• Similar to the above point, although we are able to examine the effectiveness of electronic interventions utilising different mode of delivery, the volume of evidence for many of the delivery modes, for example e-mails and mobile telephone text messaging, remains limited. Therefore, the findings of lack of sufficient evidence for proving or refuting effectiveness should not be regarded as evidence of ineffectiveness.

• We have examined only a small number of factors that could potentially influence the effectiveness of the interventions. A comprehensive evaluation of potential effect modifiers at study level in a systematic review of complex interventions remains challenging. For example, in addition to the categorisation of interventions with respect to tailoring and numbers of components, we have also attempted to apply a standardised coding with respect to the contents of each intervention. However, our initial pilot indicated that information presented in published papers is often insufficient to allow accurate coding of each intervention/comparator.

Time-to-relapse models for continuous abstinence

A number of studies report continuous abstinence of some duration as an outcome measure. Some studies record this result at multiple durations. To synthesise this range of outcomes and allow our desired intervention effect (change in 12-month continuous abstinence) to be estimated, we base our model on the time from initiation of the quit attempt to relapse. This approach is possible, as some trials report continuous abstinence at multiple time points.

The model interprets data on continuous abstinence as samples from a binomial distribution, where success is defined as maintaining abstinence up to the observation time. The probability of avoiding relapse is derived from a survival model in which the time-to-relapse for participants in a given trial and arm are assumed to be independent and identically distributed. We explore two parametric models for this distribution – the Weibull and exponential distributions. The models are described fully in Appendix 6.

Incorporating intervention effects

Interventions were assumed to have an effect on the scale parameter of the time-to-relapse model but not (in the Weibull model) the shape – this is equivalent to fitting a proportional-hazard survival model. Given the considerable heterogeneity in the data, as described in Chapter 2, a random-effects treatment model was assumed, where intervention effects were assumed to vary between studies but be drawn from a common distribution of intervention effects. This model for study-level intervention effect is given by:

where:

• λ_i,j = the scale parameter in the time-to-relapse distribution for arm j of trial i

• µ_i = the log-scale in the control arm (j = 1) of trial i

• δ_i,j = the effect of the intervention given in arm j of trial i

• I(j,1) = 1 if j = 1, 0 otherwise to indicate the baseline arm.

The study-specific intervention effects on the log-scale δ _i,j were assumed to be normally distributed (random-effects assumption) with mean µ_Di,j, where:

• treat_i,j = the intervention used in arm j of trial i

•  = the mean intervention effect for studies of intervention treat against reference intervention [in this case no electronic (e0, see Table 21) or non-electronic components (c0, see Table 21)].

The assumption made is that the indirect evidence between two different interventions is consistent with the direct head-to-head evidence on those interventions. By imposing this assumption across all interventions in the evidence base, we ensure that the correct intervention comparison is made for the evidence from each trial, and a coherent set of estimates of intervention effects is generated from the evidence that respects the randomisation in the original trials, as described in Chapter 2.

As described in Chapter 2, the interventions given consist of an electronic component and/or a conventional intervention. Both types consisted of six classes, including nothing (listed in Table 21), and so this pairing gives 36 intervention combinations (a system for labelling these combinations, used in Table 21, is to define an intervention consisting of conventional intervention A and electronic intervention B as eAcB. The interventions then range from e0c0 (no support) to e5c5 (professional face-to-face counselling and pharmacotherapy combined with a complex electronic intervention, such as an interactive website with a bulletin board).

We could assume that the effects of these 36 interventions are completely independent from each other (Assumption A1). This would involve fitting 35 treatment parameters (as one of the 36 interventions is placebo), which is the most general model possible. However, this ignores the fact that the interventions share components, and also it may not be possible to fit this many treatment parameters given the available data. We also fitted a more restrictive model that assumed that the intervention effect of combining types was additive, so that the impact of an intervention with both active electronic and conventional interventions was equal to the sum of the effect of each intervention given in isolation (Assumption A2), giving 10 treatment parameters, five electronic and five conventional. We then explored the impact of assuming that the effect of each electronic intervention was the same irrespective of its category (Assumption A3), reducing the total number of treatment parameters to six (five conventional and one electronic). Finally, we fitted a more relaxed version of this model, still assuming that the effect of an electronic intervention did not depend on its category, but allowing for that effect to vary depending on the conventional co-therapy.

To illustrate the impact of these alternatives, consider four interventions: e0c3, e4c0, e4c3 and e2c3. If D(x) is defined as the effect of intervention x, then the three different intervention effect models imply respectively:

• Assumption A1 D(treat) is different for different interventions. So, for example, D(e0c3), D(e4c0) and D(e4c3) are independent.

• Assumption A2 D(eAcB) = D(eA) + D(cB). So, for example, D(e4c3) = D(e4) + D(c3), and D(e2c3) = D(e2c0) + D(e0c3), etc.

• Assumption A3 D(eAcB) = D(e) + D(cB), where D(e) = D(e0) = D(e1) = … = D(e6). So, for example, D(e4c3) = D(e2c3) = D(e) + D(c3), etc.

It can be seen that Assumption A1 allows for an interaction between the effect of an electronic intervention and the choice of conventional co-therapy, whereas Assumption A2 and Assumption A3 assume that there is no such interaction.

Incorporating readiness to quit

One hypothesis regarding the value of electronic aids to smoking cessation is that they might encourage those not actively seeking to quit to make an attempt to do so. A number of trials include such participants. If this effect exists, the impact of electronic interventions will be higher in this population. We explore this hypothesis within the model assuming a single intervention effect across electronic interventions (Assumption A3), by amending it to include two separate treatment parameters, one for studies restricted to those actively seeking to quit, and one for those not stating such a restriction, giving seven intervention effect parameters.

Model comparison

To explore the alternative assumptions described above, we compared the fit of models with the data:

• Model 1 35 intervention effects (Assumption A1), exponential model for time to relapse.

• Model 2 35 intervention effects (Assumption A1), Weibull model for time to relapse.

• Model 3 10 intervention effects (Assumption A2), Weibull model for time to relapse.

• Model 4 6 intervention effects (Assumption A3), Weibull model for time to relapse.

• Model 5 7 intervention effects (Assumption A3 incorporating readiness to quit), Weibull model for time to relapse.

The choice of time-to-relapse distribution was made by comparing models 1 and 2. Model choice was assessed using the deviance information criterion (DIC). 134 This statistic measures a compromise between model fit (measured by D¯) and model complexity (measured by pD), and is calculated as follows:

D¯ is the posterior mean of the deviance, a measure of model fit. A model is considered to provide an adequate fit to the data when D¯ is approximately equal to the number of data points. pD is interpreted as the effective number of parameters in the model. It reflects the flexibility with which the model can adjust to fit the data. The lower the value of DIC, the better, as this reflects improved model fit and/or a more parsimonious model.

Overview of evidence

Of the studies identified in the main effectiveness review, only 28 reported continuous abstinence. 36,37,39,43,46,47,49,53,56–58,62,63,66,69,73,76,88,91,97,100,101,104,105,109,110,120,124

Table 22 shows how many arms provided information on each of the types of electronic intervention. There are many studies evaluating class three and class five electronic interventions; far fewer evaluating interventions of class one, two or four. Figure 31 illustrates the network of evidence on continuous abstinence, by depicting each pair-wise comparison contained within the evidence base reporting this measure. It can be seen that only 19 of the 36 possible comparisons are evaluated. However, the comparisons do form a single connected network for those 19 interventions, so that an intervention effect can be estimated for any pair of comparisons of these interventions. For example, although no trials have compared e3c0 and e4c2 directly, trials exist comparing the former to e0c0 and the latter to e0c2. There is also a head-to-head trial of e0c0 and e0c2, so if the assumption of consistency is made, the intervention effect of e4c2 compared with e3c0 can be calculated.

Table 23 shows the variation between studies in terms of the readiness-to-quit of participants. Only two studies explicitly state that they excluded participants actively seeking to quit. 73,104

Comparison of fit for evaluated models

Table 24 gives the model fit statistics D¯, pD and DIC. The first conclusion to draw from comparing the models is that the exponential is an extremely poor fit with the data, and that the more flexible Weibull distribution improves model fit considerably. The model fitted assumes that the shape parameter α is treatment invariant, and its posterior mean was found to be around 0.14 in models 2–5. This suggests that the hazard of relapse decreases sharply over time. Although the fit was improved, none of the Weibull models achieved a value for that was close to 89 or less (based on the rule of thumb mentioned earlier, which compares with the number of data points), suggesting that there is still some lack of fit. A comparison of observed and fitted values found that the observations with the largest deviance were related to arms from two studies105,109 with repeated observations, where a very high relapse rate at the first observation was followed by a relapse rate in subsequent observations at or close to 0%.

Reducing the number of parameters from 35 to 10 (Model 3) did have an impact on model fit, but the DIC suggests that this impact is not strong enough to justify the model with full interactions (Model 2). Assuming that the intervention effect was identical across intervention classes had little further impact on model fit (although using fewer parameters), reflected in a reduction in the DIC statistic for Model 4. Adapting this model to include separate intervention effects according to readiness to quit, or interaction effects, led to a slightly worse model fit.

The conclusion from this comparative exercise is that, based on the DIC, the preferred model is one in which time to relapse has a Weibull distribution, and the intervention effect is assumed identical across electronic intervention classes, without interaction effects with the conventional co-intervention. The fit of such models with or without a differentiation of intervention effect between those ready to quit, and those not, is comparable, suggesting no evidence that there is a difference in intervention effect on continuous abstinence between these different populations.

Intervention effect estimates

The models give intervention effects that can be interpreted as hazard ratios (HRs) for the time to relapse. As described in section 5.1.2, the basic parameters of each model are the effects of each intervention relative to reference intervention e0c0. Using the consistency assumption we can then form the HR for any head-to-head comparison, calculated by dividing the HR of the active intervention (vs reference) by the HR of the control (vs reference). As the hazard is that of time to relapse, a lower HR represents an improvement.

Based on model fit alone, the preferred assumption is that of a single intervention effect across different classes of electronic intervention, with (Model 5) or without (Model 4) a modifier for readiness to quit. However, we were also interested in the impact of allowing separate intervention effects for each of the five classes of electronic intervention (Model 3) as providing the inputs required in the cost-effectiveness analysis. Table 25 gives the intervention effects for these three models. Although the inclusion criteria restricted trials to those with an electronic intervention arm, the network allows for estimates of intervention effects against reference intervention (e0c0) for each of the conventional interventions. These estimates are given in Table 25 for Model 3 – models 4 and 5 produce very similar results. They suggest that there is considerable uncertainty in the effectiveness of these interventions, and whether or not they are more effective than e0c0 at all. The result arises because of the limited and largely indirect nature of the evidence included in this review to base such comparisons – evaluations specifically concerned with established interventions have drawn on a larger evidence base to derive estimates of these intervention effects that are more precise and more positive. The estimates of the effectiveness of conventional interventions is not the focus of this review, are therefore essentially ‘nuisance parameters’ in our model, and we do not use them in the cost-effectiveness analysis described in Chapter 4 (see Decision-analytic model).

Table 25 gives intervention effect estimates for electronic interventions vs nothing. Model 4 gives a single intervention effect for all electronic interventions showing a benefit that is significant, although not strong (mean HR 0.87, 95% credible interval 0.83 to 0.92). Model 5 allows for the intervention effect to depend on whether or not the study explicitly restricts participation to smokers actively seeking to quit, but the resultant HRs are virtually identical. For Model 3, five separate intervention effects are generated. The mean HRs vary from 0.85 (e5) to 1.02 (e4). There is no discernable trend (either positive or negative) in mean effectiveness as intervention complexity increases. Reflecting the available evidence, there is considerable uncertainty in the intervention effect for e1 (mean HR 0.89, 95% credible interval 0.66 to 1.16), e2 (mean HR 0.98, 95% credible interval 0.78 to 1.21) and e4 (mean HR 1.02, 95% credible interval 0.78 to 1.32). Only interventions e3 (mean HR 0.88, 95% credible interval 0.83 to 0.93) and e5 (mean HR 0.85, 95% credible interval 0.75 to 0.96) have an intervention effect whose credible interval excludes 1, which is not surprising given the larger number of studies falling into this category.

Figure 32 illustrates the intervention effects (mean and 95% credible intervals) generated by the three models, so that the findings can be compared simultaneously. It illustrates the considerable uncertainty around the five separate intervention effects from Model 3, relative to the differences between the means, explaining why the DIC favoured Model 4. The resulting single estimate from that model, e_pooled, is clearly influenced most by those interventions with the most evidence (e3 and e5). It can also be seen that there is almost no difference in intervention effect between trials that restrict inclusion to those actively seeking to quit, and those that do not.

FIGURE 32.

Caterpillar plot of HRs (mean and 95% credible intervals) derived from synthesis of evidence using models 3–5. ‘e1–e5’ are the separate intervention effects for each class, estimated by Model 3. ‘e_pooled’ is the intervention effect from Model four, which assumes the intervention effect is independent of class. ‘e_NQ’ is the intervention effect for those not actively seeking to quit and ‘e_Q’ is the intervention effect for those who are seeking to quit (both estimated by Model 5).

Limitation of the analysis and recommendations for future research

A limitation of the system used to categorise electronic interventions was that it does not allow us to categorise the impact of mode of delivery. This was because there is a wide range of possible modes of delivery, with only limited data to compare them. In order to simplify the analysis and derive meaningful results, we developed a categorisation system based on expert advice which identified underlying factors (such as interaction and intensity) that might explain the relative efficacy of different electronic interventions. The additional impact of alternative modes of delivery should be explored, once data are available to allow this. We were unable to explore the impact of interventions on the quit–relapse cycle and the impact of factors such as age and motivational status on the optimal intervention due to a lack of evidence. There is a clear need for research designed to support these further analyses.

Summary of findings

• There is no evidence from the trials available that there is a differential intervention effect across categories of electronic intervention (see Tables 24 and 25 and Figure 32).

• When viewed as a single category, the addition of an electronic intervention of some sort to conventional smoking cessation aids improves continuous abstinence (mean HR 0.87, 95% credible interval 0.83 to 0.92).

• The available evidence does not show any difference in intervention effect for electronic interventions between trials restricted to smokers actively seeking to quit and those not imposing this inclusion criterion.

• There is considerable uncertainty in the estimates for e1 (single generic component, mean HR 0.89, 95% credible interval 0.66 to 1.16), e2 (multiple generic components, mean HR 0.98, 95% credible interval 0.78 to 1.21) and e4 (single generic component plus multiple tailored components, mean HR 1.02, 95% credible interval 0.66 to 1.16), reflecting the lack of trials assessing these categories of intervention.

Given these findings, we consider two sets of results for use in the cost-effectiveness analysis. When considering the cost-effectiveness of electronic interventions as a class, we will use the pooled estimate from Model 4. When exploring the question of which type of intervention to use, we will use the separate estimates from Model 3, as shown in Table 25 and Figure 32.

Search strategy

The search strategy was identical to the strategy used to identify effectiveness studies (described in Appendix 1), which was designed to also capture any studies reporting outcomes relevant to the cost-effectiveness of electronic aids for smoking cessation.

Inclusion criteria

All studies, of any design, that reported a full economic evaluation (i.e. cost-effectiveness analysis, cost-utility analysis, or cost–benefit analysis) were eligible for inclusion. The same criteria on the basis of the intervention were used as for inclusion in the effectiveness review (described in Appendix 2).

Data extraction

Data were extracted by one reviewer (JJM) and independently checked for accuracy by a second reviewer (NJW). Disagreements were resolved by discussion. Outcome measures extracted were intervention costs, incremental cost per quitter, incremental cost per life-year saved, incremental cost per QALY saved, for each intervention compared with the control scenario. Descriptions of the model used and economic data reported are given below.

Quality assessment

Quality assessment of cost-effectiveness studies was performed using standard criteria. 135 The results of this process are documented in Appendix 5.

Description of retrieved studies

Population

The review identified two distinct populations for whom these interventions have been used:

• Those making a committed quit attempt using the electronic aid as an adjunct to pharmacological interventions.

• Those at an earlier stage in the quitting process who are less motivated and not using pharmacological interventions. For these smokers, we would expect successful quit rates to be lower. However, the progression to making a committed quit attempt in the future is a relevant secondary outcome.

A simplified model of the quit process is illustrated in Figure 34.

FIGURE 34.

Decision model for electronic aid interventions for those who are about to undertake a quit attempt (committed quit attempt) and those who are not intending to make such an attempt (non-committed quit attempt) at the time the intervention is provided.

Interventions

The need for, and derivation of, a classification system for electronic interventions is presented in the main effectiveness review. The five-level classification system developed for this purpose is set out in Table 21. However, the evidence synthesis exercise of Chapter 3 suggested that there is considerable uncertainty around the comparative benefits of different classes of intervention. Following the results of that exercise, we also consider ‘lumping’ together all electronic aid interventions as a single intervention. This does raise issues in identifying a single cost for this ‘lumping’ of such heterogeneous interventions.

Control scenarios

The decision problem for our economic analysis is defined as whether or not some form of electronic aid should be added to conventional behavioural support. This raises the question of what exactly the conventional intervention used for the control arm should be. A variety of control interventions were used in the identified effectiveness studies. Current guidance provided by NICE on technology appraisals140 defines the relevant comparators as routine practice in the NHS and best practice (where this is different to routine care). Regarding non-electronic behavioural support for smoking cessation, current NICE guidance recommends pharmacotherapy in conjunction with brief opportunistic advice or professional counselling. 12 In line with this recommendation, we define the control scenario for the decision analysis as pharmacotherapy with opportunistic advice, and explore the control scenario of counselling plus NRT in sensitivity analysis. In each case, the intervention is defined as electronic interventions as well as, rather than instead of, the conventional control.

Decision questions

Given the description of the decision context, the decision problem consists of several separate questions:

1. Are electronic aids, as adjunct to current best conventional practice, cost-effective interventions for smoking cessation in committed quitters?

2. If so, what type of electronic aid is most cost-effective in this population (i.e. which components as defined above)?

3. Are electronic aids cost-effective for smoking cessation in smokers not yet attempting to quit?

4. If so, what type of electronic aid is most cost-effective in this population?

5. Is there value in carrying out further research into electronic aids as adjunct therapy for smoking cessation (either in committed quitters or those not actively seeking to quit)?

Our intention was to answer each of these questions. However, the evidence synthesis exercise found considerable uncertainty in differentiating between classes of intervention, and between populations, which will have an impact on the extent to which this aim can be achieved.

Model specification

We use a decision tree model to assess the cost-effectiveness of electronic aid interventions for smoking cessation (see Figure 34). Decision tree models are well established in HTA for smoking cessation;141,142 our model draws from this previous work. This avoids duplication of effort and also ensures consistency with previous HTAs in the area of smoking cessation.

The appropriate intervention to support behaviour change in smokers will depend on their intentions. Pharmacological interventions, such as bupropion and NRT, assist those who are actively trying to quit. Some interventions are targeted instead at smokers who feel they are unwilling or unable to quit, and aim to encourage them to make an attempt to quit that otherwise they would not consider. Electronic smoking cessation aids can potentially be effective in both populations, depending on the specific details of the intervention. Figure 34 illustrates the decision tree we have developed to represent this distinction. For those smokers actively trying to quit, we split the process into a first stage of success or failure. There are a number of ways to define success – along with previous HTAs141,142 we define this as continuous abstinence over a period of 12 months. For those who achieve this, we assume only some will remain ex-smokers indefinitely. This is also in line with previous evaluations. We assume only those who avoid relapse gain health benefits as a result of the intervention, and draw on the literature to estimate the size of these benefits (see Life-years saved, on long-term health benefits of quitting smoking). Again, this is in line with previous economic analyses of smoking cessation interventions. 55,81,141

For interventions aimed at those not yet actively attempting to quit, we extend the base of the decision tree so that an intermediate end point is the proportion who go on to make such an attempt. Those who do can then progress through the decision tree for those attempting to quit, if we assume that the treatment effect on quit rates in this group is similar to that for those actively attempting to quit prior to receiving an electronic intervention.

Review of additional evidence

Development of cost scenarios

Given the heterogeneous nature of the interventions and the lack of information on resource use, it is difficult to provide precise cost estimates for the five classes of electronic intervention. In response, we developed five scenarios for these costs. The primary source for this was the study by Meenan et al. ,146 corroborated by findings from Runge et al. ,148 Smith et al. 136 and Southard et al. 149 These scenarios are presented in Table 26 and their impact explored in sensitivity analyses.

Long-term medical costs

On the one hand it has been argued that smokers increase health-care costs because of increased incidence of smoking-related diseases. 152 On the other hand, it has also been argued that smokers have a lower life expectancy and so the health-care burden in the elderly is subsequently reduced in the long term. 153 Some have suggested that these two factors may in effect cancel out, so that the net change in long-term cost is small. 154 As there is no consensus on this issue, we make the assumption that there are no long-term cost implications as a result of quitting, only health benefits. This is in line with previous economic evaluations141 of aids to smoking cessation.

Twelve-month continuous cessation rates for baseline interventions

The evidence synthesis provides estimates of treatment effects as HRs from a Weibull time-to-relapse model. In that exercise we compared a model with independent treatment effects with a model where the effect of intervention, on the log-hazard scale, was additive (see Chapter 3, Overview of evidence and the mixed-treatment comparison approach). As the additive model was found to provide a better fit than either of the models allowing for interactions, we assume that the effectiveness of the electronic aid interventions is the same regardless of the control arm intervention. In other words, the effect of the electronic aid intervention is additive when used as an adjunct to pharmacological and/or counselling control interventions. We convert the HR to a RR of 12-month continuous abstinence using the Weibull model. If the probability of 12-month continuous abstinence for the control and the control plus intervention are defined as p_c(12) and p_i(12):

From the time-to-relapse model:

where h is the HR for the intervention vs placebo. This gives

We represent uncertainty around the baseline probability of 12-month continuous abstinence by assigning the parameter a normal distribution, truncated at 0, with mean 6% and standard deviation 1.6% (brief advice), or mean 12% and standard deviation 1.4% (counselling). The odds ratio for nicotine therapy is assumed to be normally distributed with mean 1.67 and standard deviation 0.06. These estimates were taken from previous economic evaluations of conventional therapies. 141,142

Lifetime cessation rates

The quit process is not always straightforward – some may relapse after a considerable period of abstinence, whereas others may achieve permanent cessation after several failed short-term attempts. A common approach in economic evaluations of smoking cessation interventions is to define the initial outcome in terms of the proportion achieving continuous abstinence for 12 months, but then to assume that a proportion will subsequently relapse. Furthermore, it is often assumed that long-term relapse rates are similar across intervention, and that health benefits only accrue to those that avoid relapse. We follow this approach in our base case.

Trials commonly follow participants and report results up to a fixed point in time, so that observational data are required to extrapolate their outcomes to permanent quit rates. Woolacott et al. 141 carried out a review of estimates of the proportion who maintain continuous abstinence for 12 months but subsequently relapse. The mean reported rate was 40%, and the estimates ranged from 30% to 50%. In an analysis of the use of genetic information to target pharmacotherapy, Welton et al. 155 capture this information by assigning to it a beta(38,57) distribution, which has a mean of 40% and a 95% CI of 30% to 50%. 155 We use this distribution in our cost-effectiveness analyses.

Life-years saved

To fully inform decision-making, an estimate of the long-term health impact of quitting is required. This is a complex parameter and difficult to assess. 141 At an individual level, the risk reduction achieved by quitting will depend on several factors. These include the age at which the smoker quits, the length of time they had been a smoker, the number of cigarettes smoked and the gender of the smoker. 156 The average benefit for the relevant population will further depend on its composition. The calculation is complicated by the wide range of conditions for which smoking is a risk factor.

Woolacott et al. 141 identified 17 economic evaluations of smoking cessation therapies, of which 12 use a value for life-years or QALYs gained by quitting. The average discounted QALY gain reported was 2.7, with a range of 1.35–4.07 QALYs gained per quitter. Welton et al. 155 capture this range of estimates for the discounted QALY gain by assigning it a normal distribution, truncated at 0, with mean 2.7 and variance 0.47, and we use this distribution for the parameter in our model. This approach allows us to represent the uncertainty around this parameter, although it does have limitations which are discussed below (see Limitations of the analysis and recommendations for further research).

Discounting and time horizon

We take a lifetime perspective for the economic analysis. However, for reasons given above (see Long-term medical costs), we limit our costs to the immediate resource requirements of the intervention, which are short term. The only long-term component of the costs and benefits arising in the model relate to the long-term health benefits of quitting. As described above (see Life-years saved), the distribution used for this benefit is based on a range of estimates from the literature, all of which discount health benefits. The distribution therefore reflects a discounting of these future health gains. The source studies for this distribution do not use a consistent rate for discounting health benefits – an attempt to adjust each estimate for the rate used was considered, but given the considerable uncertainty around this measure, such an adjustment would be unlikely to materially alter the distribution we use to represent this uncertainty.

Summary of analyses

As outlined above (see Decision questions), there are several questions that arise in determining the cost-effectiveness of electronic interventions. We began by considering the population of those actively seeking to quit. We first explored whether or not electronic interventions were cost-effective in this group when considered as a single class of intervention. To analyse this, we used the results from the evidence synthesis arising when all classes are assumed to be equally effective (Model 4 of Table 25). Given the heterogeneity of these interventions, we carried out a cost threshold analysis. This involved deriving the distribution for QALYs gained, and determining the maximum cost at which the intervention was cost-effective at commonly quoted thresholds (£20,000–30,000 per QALY). As described above (see Control scenarios), we consider electronic intervention as an adjunct to two possible conventional interventions: (1) pharmacotherapy plus brief advice and (2) pharmacotherapy plus professional counselling.

We then explore the question of which class of intervention should be chosen, based on cost-effectiveness, if a decision is made to implement an electronic intervention of some type. For this, we draw on the results of the evidence synthesis arising when separate effects are assumed for the different classes of electronic intervention (see Model 3 of Table 25). As with the single intervention case, we carry out separate analyses for the two possible conventional best practice interventions. We also carry out separate analyses exploring the five cost scenarios developed above (see Derivation of cost data for electronic interventions).

Having explored cost-effectiveness in those actively seeking to quit, we consider those who are not committed to such an attempt. Our approach to this question is based on the result found in the evidence synthesis, that there is no evidence to suggest the relative effect of electronic interventions in this group. We allow for the difference by adjusting the conventional intervention used in the control arm. As these are participants who are not seeking to quit, we assume that they will not turn to the two control interventions described above. Instead, we reperform our analysis, based on spontaneous quit rates, which are estimated at 1% over 12 months. 141 We explore the cost-effectiveness of electronic interventions as a single class in this population, as there is insufficient evidence to comment on what type of electronic intervention might be particularly suitable for this population.

Results for those actively seeking to quit

Cost-effectiveness of electronic intervention compared with no electronic intervention

Table 27 gives an analysis of the economic impact of implementing an electronic intervention alongside a conventional control. The mean QALY gain depends on the choice of conventional therapy (0.053 for brief advice, 0.069 for counselling). The lower bound of the 95% credible interval only just includes negative values, reflecting a small possibility that the addition would actually reduce long-term quit rates. As we are considering a broad class of interventions with a wide range of plausible costs, we carry out a threshold analysis to explore cost-effectiveness. This involves multiplying any desired willingness-to-pay (WTP) threshold by the mean QALY gain to determine the cost above which the therapy is no longer cost-effective. Table 27 gives the results of these calculations at WTP thresholds of £20,000 and £30,000 per QALY – the results in every case are well above the costs explored in the scenarios of Table 26.

TABLE 27 - Cost-effectiveness of electronic interventions (mean and 95% credible interval) when given adjunct to pharmacotherapy and brief advice or individual counsellinga

The results shown include the absolute QALY gain, and the cost above which electronic interventions are no longer cost-effective (at commonly quoted thresholds).

Control	QALY gain from electronic intervention		Maximum acceptable cost (£)
	Mean	Credible interval	£30,000/QALY	£20,000/QALY
Pharmacotherapy + brief advice	0.053	–0.016 to 0.152	1579	1053
Pharmacotherapy + counselling	0.069	–0.023 to 0.190	2081	1387

To illustrate the uncertainty around the cost-effectiveness results, we construct cost-effectiveness acceptability curves (CEACs) for the two control scenarios. This requires some assumption to be made around the cost of the intervention. We present CEACs based on assuming this cost to be either £10, £50 or £90 per user in Figure 35, as this reflects the range of values included in our base case for the costs of the electronic intervention types. In all cases considered, the probability of cost-effectiveness reaches 90% before the WTP reaches £20,000/QALY, and remains below 95%, reflecting the probability that the electronic intervention reduces long-term quit rates.

FIGURE 35.

(a) CEACs for electronic intervention vs NRT plus brief advice. (b) CEACs for electronic intervention vs NRT plus counselling. Three CEACs are shown based on different assumptions for the cost of the intervention.

Comparative cost-effectiveness of electronic interventions

We also carried out an analysis of the comparative cost-effectiveness of the five electronic interventions. The results of this exercise are given in Table 28 (where the conventional intervention is pharmacotherapy plus brief advice) and Table 29 (where the conventional intervention is pharmacotherapy plus counselling). Results are given for each of the five cost scenarios listed in Table 26. They show that for all scenarios except high cost to tailoring, e3 interventions (single tailored component) dominate (provide more QALYs at less cost) e4 (multiple components including a single tailored component) and e2 (multiple generic components) interventions. Even in the exceptional scenario, the ICERs for e3 (single tailored component) interventions compared with e2 (multiple generic components) interventions are very favourable (£55/QALY as an adjunct to brief advice, £40/QALY as an adjunct to counselling). The e3 (single tailored component) interventions compare unfavourably with e1 (single generic component) interventions – the latter either dominates or provides additional benefits at an ICER of ≤ £4400/QALY). In turn, the e5 (multiple tailored components) interventions appear cost-effective compared with e1 (single generic component) interventions in most scenarios when added to either conventional intervention. The least favourable scenario for e5 (multiple tailored components) is the intensive support scenario, under which the ICER for e5 compared with e1 is £28,000/QALY when added to brief support, and £18,000 when added to counselling.

TABLE 28a - Cost-effectiveness under the base-case cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	44.80	0.0004	112,000	Dominated
e2	30.10	0.0145	2076	Dominated
e3	14.70	0.0525	280	Dominated
e1	7.70	0.0576	134	134
e5	77.00	0.0669	1151	7452

TABLE 28b - Cost-effectiveness under the high cost to tailoring cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	62.30	0.0004	155,750	Dominated
e2	30.10	0.0145	2076	Dominated
e3	32.20	0.0525	613	Dominated
e1	7.70	0.0576	134	134
e5	77.00	0.0669	1151	7452

TABLE 28c - Cost-effectiveness under the limited saving cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	59.50	0.0004	148,750	Dominated
e2	44.80	0.0145	3090	Dominated
e3	14.70	0.0525	280	280
e1	30.10	0.0576	523	3020
e5	77.00	0.0669	1151	5043

TABLE 28d - Cost-effectiveness under the intensive support cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	44.80	0.0004	112,000	Dominated
e2	30.10	0.0145	2076	Dominated
e3	14.70	0.0525	280	Dominated
e1	7.70	0.0576	134	134
e5	270.20	0.0669	4039	28,226

TABLE 28e - Cost-effectiveness under the mass uptake cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	0.98	0.0004	2450	Dominated
e2	0.70	0.0145	48	Dominated
e3	0.28	0.0525	5	5
e1	0.70	0.0576	12	82
e5	2.10	0.0669	31	151

TABLE 29a - Cost-effectiveness under the base-case cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	44.80	–0.0034	Dominated	Dominated
e2	30.10	0.0171	1760	Dominated
e3	14.70	0.0693	212	Dominated
e1	7.70	0.0728	106	106
e5	77.00	0.0874	881	4756

TABLE 29b - Cost-effectiveness under the high cost to tailoring cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	62.30	–0.0034	Dominated	Dominated
e2	30.10	0.0171	1760	Dominated
e3	32.20	0.0693	465	Dominated
e1	7.70	0.0728	106	106
e5	77.00	0.0874	881	4756

TABLE 29c - Cost-effectiveness under the limited saving cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	59.50	–0.0034	Dominated	Dominated
e2	44.80	0.0171	2620	Dominated
e3	14.70	0.0693	212	212
e1	30.10	0.0728	413	Dominated (extended)
e5	77.00	0.0874	881	3442

TABLE 29d - Cost-effectiveness under the intensive support cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	44.80	–0.0034	Dominated	Dominated
e2	30.10	0.0171	1760	Dominated
e3	14.70	0.0693	212	Dominated
e1	7.70	0.0728	106	106
e5	270.20	0.0874	3093	18,016

TABLE 29e - Cost-effectiveness under the mass uptake cost scenario

Class of electronic intervention	Cost of intervention (£)	Mean QALYs gained	Cost (£)/QALY vs conventional only	ICER (£ per QALY)
e4	0.98	–0.0034	Dominated	Dominated
e2	0.70	0.0171	41	Dominated
e3	0.28	0.0693	4	4
e1	0.70	0.0728	10	Dominated (extended)
e5	2.10	0.0874	24	101

These results relate to the mean estimates of cost-effectiveness. As described in the main effectiveness review (see Chapter 2) and in the mixed-treatment comparison (see Chapter 3), there is considerable uncertainty around the comparative efficacy of the different classes of electronic intervention, and this translates into uncertainty around the mean estimates given in Tables 28 and 29.

Table 28 shows the cost-effectiveness results (under five cost scenarios) for electronic interventions as an adjuvant to pharmacotherapy plus brief advice.

Table 29 shows cost-effectiveness results (under five cost scenarios) for electronic interventions as an adjuvant to pharmacotherapy plus counselling.

Figure 36 gives the CEACs illustrating the likelihood that each intervention class is the most effective, depending on the value chosen for the WTP threshold. CEACs are shown for each scenario, assuming that the therapy is added to pharmacotherapy and brief advice (curves for the alternative control scenario are almost identical).

FIGURE 36.

Cost-effectiveness acceptability curves illustrating the probability that each class of electronic intervention is the most cost-effective, given different assumed WTP thresholds.

The CEACs illustrate the considerable uncertainty around which of the five classes of intervention should be preferred on cost-effectiveness grounds. This uncertainty is reflected across the cost scenarios, suggesting that it is the lack of information on efficacy that drives the uncertainty. Although class five interventions had the most favourable mean cost-effectiveness, the probability that they are indeed the most cost-effective class is around 30–35%. The probability that class one interventions are the most cost-effective tends to be similar, if not higher, and is favoured by lower WTP thresholds and the intensive support scenario. The probability that class three interventions are the most cost-effective tends to be lower, but still considerable, at 20–30%. Even class two and class four interventions, which are dominated by the other classes in almost every scenario at the mean, have a probability of cost-effectiveness of between 5% and 10%.

Results for those not actively seeking to quit

The evidence synthesis found that the relative treatment effect of electronic interventions did not differ between trials in those actively seeking to quit and trials in other populations. To explore the cost-effectiveness of electronic aids in this group, we repeated the analysis described in Results of the cost-effectiveness analysis, Summary of analyses. As in that section, we use the relative treatment effect arising from the assumption that electronic aids form a single class of intervention. The only difference is that we assume that smokers not actively seeking to quit would not seek out conventional interventions. Therefore, we apply the same treatment effect, but use a different baseline quit rate, reflecting spontaneous unsupported quitting. The 12-month quit rate in this situation has been estimated at 1%,141 and we apply the HR for electronic intervention to this rate as described in the section Twelve-month continuous cessation rates for baseline interventions.

We found the mean QALY gain from implementing an electronic intervention in this population to be 0.014 (95% credible interval 0 to 0.04). This is considerably less than we predict in the population of smokers actively seeking to quit. We carried out a cost-threshold analysis as described in the section Results for those actively seeing to quit, above, and found that the electronic intervention would not be cost-effective if it cost more than £271 (assuming a WTP of £30,000/QALY) or £406 (assuming a WTP of £20,000/QALY). This is considerably less than the threshold for smokers willing to quit but still well above the values in the various cost scenarios.

As in Results for those actively seeing to quit, above, we developed CEACs to represent the uncertainty around this finding. The CEAC for this population is presented in Figure 37. The probability of cost-effectiveness at a WTP threshold of £20,000/QALY is 77% if the cost per user is £90, rising to 91% if this cost is £10.

FIGURE 37.

Cost-effective acceptability curve for electronic intervention vs placebo in a population of smokers not actively seeking to quit. A 1% spontaneous quit rate is assumed for the comparator. CEACs are presented assuming costs of the electronic intervention of £10, £50 and £90.

Estimated value of additional information

The results shown above suggest that adding an electronic aid of some sort to current conventional support for smokers attempting to quit would be cost-effective, and that the intensive class five intervention would be expected to maximise cost-effectiveness. We have shown that there is considerable uncertainty around this finding, particularly regarding the choice of class of intervention. One response to this uncertainty would be to delay the choice until further information was available. Value-of-information (VOI) methods provide a framework for determining whether or not this is an advisable course of action. 157

Value-of-information methods assess the likelihood that additional information will lead to a change in the decision, and provide an estimate of the expected loss from acting without that information. The expected value of perfect information (EVPI) is this estimate when the additional information provides exact values for all model parameters and eliminates all uncertainty. We calculate the EVPI first assuming that the treatment effect is identical across electronic intervention classes, and used a uniform (0,200) distribution to represent our uncertainty about cost. The resulting estimate, in terms of benefit per person, depends on the threshold uses to assign a monetary value to health benefits. At a value of £20,000 per QALY, the EVPI was £31 per person if the comparator includes brief advice, or £41 per person if the comparator included counselling.

We repeated the exercise based on the evidence synthesis model allowing separate treatment effects across electronic intervention classes. We explored the impact of different conventional comparators and cost scenarios, and also compared the EVPI assuming the current decision would be in favour of class one interventions or class five interventions. We found that the EVPI was much higher in this case, as we would expect given the additional uncertainty around the comparative efficacy of intervention classes. The strongest factor affecting the EVPI was the choice of comparator: when it included brief advice, estimates (given a threshold of £20,000 per QALY) ranged from £2047 to £2132 per person; when it included counselling, this rose to £2643–2802. As can be seen in these ranges, the other factors had comparatively little impact on EVPI.

The EVPI provides an upper bound on the benefit of conducting future research, as such research is unlikely to reduce uncertainty completely. The benefit is also proportional to the relevant population affected by the decision, and the choice of timescale to assess the decision. Given the large numbers that are eligible for smoking cessation support, this suggests that even a large study would be cost-effective. In 2009–10, over 750,000 smokers in England used NHS SSS (reference statistics on NHS SSS England 2009–10, The NHS Information Centre; lifestyle statistics available at www.ic.nhs.uk/webfiles/publications/Health%20and%20Lifestyles/SSS_2009_10_revised.pdf). This indicates the potential reach of electronic intervention, although it is not possible to predict how extensive their uptake would be. Assuming that electronic interventions would reach 20% of these users, that the decision time horizon is 5 years, and applying a discount rate of 3.5%, the population EVPI would be approximately 700,000 times higher than the EVPI per person results quoted above. Therefore, population EVPI would lie between £1.4B and £2B. Even if it is assumed that all electronic interventions are equally effective, the population EVPI would be approximately £20–30M. Table 30 further illustrates the sensitivity of population EVPI to assumptions regarding uptake and the time horizon. These are EVPI estimates, so the expected benefit of a specific study would be lower. Also, these estimates are sensitive to assumptions around the decision population and time horizon that must be based on limited robust evidence. Nevertheless, they illustrate the potential value in conducting further research, particularly on the relationship between the design of electronic interventions and their efficacy.

Limitations of the analysis and recommendations for further research

A limitation of the cost-effectiveness analysis, shared with several previous cost-effectiveness analyses of smoking cessation interventions,104,105 is that intervention benefit is restricted to the first quit attempt. Thus even if those who fail at the first attempt go on to quit permanently at some future attempt, such attempts do not need to be accounted for, as we are assuming that the probability of success in those attempts is unchanged by exposure to the intervention. This assumption simplifies the model considerably, as it allows us to ignore subsequent quit attempts following relapse. However, the numbers of those who sustain their first quit attempt to permanent abstinence is small. If some of those who fail their first quit attempt go on to make a future quit attempt that is successful and interventions increase the success rate of these subsequent attempts then we will have underestimated the benefits of effective treatments. We would have liked to explore this issue using a more realistic model that included the impact of interventions on subsequent attempts. Unfortunately, such models require more detailed information on patient event histories than is available from the evidence base currently available.

To translate the efficacy estimates of Chapter 3 into estimates of cost-effectiveness, the model requires information that cannot be obtained from the trials identified in our systematic review. This includes information on the long-term relapse rate, and the QALY gain associated with permanently quitting smoking. A detailed investigation of the evidence around these parameters was beyond the scope of this work. Instead, we have based our estimates on evidence from previous evaluations of smoking cessation interventions, and used distributions to capture the strength (or weakness) of evidence in that work on the parameters of interest. There are limitations to this approach. A more extensive analysis might have allowed us to uncover new evidence, or identify biases and other factors driving divergent estimates in the existing literature. The studies underpinning estimates of the QALY gain from quitting, for example, vary in terms of the data used and the choice of discount rate for health benefits (which range from 0–6%). It may well be, therefore, that our approach overestimates the uncertainty around these parameters. However, our results already show that electronic interventions of some type are so likely to be cost-effective that it is unlikely that better estimates of long-term relapse and QALY gains would change that conclusion.

A further limitation of our analysis is that we have not included reductions in smoking-related health costs in our analysis. This is in line with previous evaluations (Woolacott et al. 141) found that most of the cost-effectiveness studies identified in their review of the literature on smoking cessation interventions excluded such costs. The justification for this was that there is uncertainty as to whether or not such costs might be offset by the increased costs of providing other health services, pension costs and reduced tax revenue. However, it is not appropriate to include these offsetting costs – the costs of unrelated diseases are generally excluded in economic evaluations, and tax revenues are merely transfer payments. Our main reason for excluding any reductions in smoking-related health-care costs for quitters was the lack of robust estimates of this parameter in the literature, and the fact that the present value of such savings will be reduced substantially by discounting. We may therefore have underestimated the cost-effectiveness of effective interventions. However, the threshold analysis presented above (see Cost-effectiveness of electronic intervention compared with no electronic intervention) suggests that the costs of electronic interventions are already likely to be well below the level at which such interventions are cost-effective. The inclusion of additional cost savings will only strengthen this conclusion. Once further effectiveness data are available to allow the analysis of the incremental cost-effectiveness of alternative types of electronic intervention then the analysis may become more sensitive to the values used for these additional parameters (smoking related health-care costs, relapse rates, health benefits of quitting).

Although our analysis has focused on the health benefits accruing directly to the ex-smoker, it is important to note that there are others who may gain. Passive smoking, smoking during pregnancy and the influence of behaviour on other smokers are all means by which a decision to quit may impact on others. These effects are difficult to quantify, but if they were to be included, effective interventions would become even more cost-effective than our results suggest.

The EVPI analysis gives some guidance on the benefits of conducting further research to inform parameters in the model. The estimates of EVPI quoted above (see Estimated value of additional information) should be interpreted by considering the number of people affected by, and the time horizon of, the decision being informed by the cost-effectiveness analysis. Estimates of EVPI are several orders of magnitude lower when it is assumed that all electronic interventions are equally effective, compared with estimated EVPI without this assumption. This suggests that there is substantial value in research on the comparative short-term efficacy of alternative classes of intervention. However, given the numbers who are potentially affected by this decision, research into other model parameters (such as the rate of long-term relapse and the benefits of quitting) is also likely to be worthwhile.

Expected value of perfect information methods can estimate only the value of information related to parameters included in the economic model. Further comparative research on alternative methods for electronic interventions would allow analysis of the impact on cost-effectiveness of aspects of interventions not evaluated in the current analysis (e.g. mode of delivery). Research on issues such as the influence of the age and motivational state of the smoker would allow the issue of the most cost-effective design of electronic intervention for specific groups to be explored.

Summary of findings

• If we assume that the treatment effect is identical across classes of electronic intervention then some sort of electronic intervention is likely to be cost-effective (when added to non-electronic behavioural support).

• Based on mean cost-effectiveness, e5 interventions (multiple tailored components, such as an interactive website and chat room) have the highest cost-effectiveness in most scenarios. Estimates of incremental cost-effectiveness of e5 [multiple tailored components + generic component(s)] vs e1 (single generic component), which is usually the next best option, range from £100–28,000 per QALY. However, probabilistic sensitivity analysis suggests that there is substantial uncertainty around this estimate, and the probability that e5 is indeed the most cost-effective type of intervention is 30–35%.

• Given this uncertainty, further effectiveness research is likely to be cost-effective, particularly around the most effective type of electronic intervention. EVPI calculations suggest the upper limit for the benefit of this research is around £2000–3000 per person. Given the large numbers that are eligible for smoking cessation support, this suggests that even a large study would be cost-effective.

• There is some support in the evidence for the cost-effectiveness of electronic interventions in smokers not actively seeking to quit but there is very little evidence available in this population on which to base any conclusions.

Profile of users

Interventions delivered via electronic aids, particularly the internet, have the potential to reach large numbers of smokers. 29,104 This review aimed to examine whether or not electronic interventions were able to reach smokers who were unwilling or unable to attend face-to-face interventions, or to contact a telephone quit line. One limitation of controlled studies of electronic aids (such as those included in the effectiveness review component of this report) is that the inclusion criteria they apply may mean that the smokers recruited to the trial are not necessarily representative of those who would choose to use an intervention in practice. Alternative study designs, particularly cross-sectional studies that examine the profile of users of an existing intervention, can therefore shed some light on who might use electronic aids and how the reach and acceptability of such aids can be enhanced.

This supplementary review included four studies that described the characteristics of existing users of already available internet sites (with or without a tailored component) for smoking cessation. 158–161 It also included three pilot studies that examined the profile of those who volunteered to test electronic aids. 162–164

Acceptability of interventions to groups of smokers

Previous reviews of electronic aids for behaviour change have suggested that receptivity to such aids varies significantly between patient groups or populations. 183 The supplementary review therefore examined the extent to which electronic aids were acceptable to different groups of smokers, namely:

• young adults

• lower-income groups

• black and minority ethnic (BME) communities.

Acceptability was not always directly measured but was assessed by uptake or continued use, for example return visits to a website.

Feasibility of delivery in different settings

Three cross-sectional studies focus on the delivery of electronic aids for smoking cessation in different settings. One study176 examines the uptake of and outcomes from a web-based intervention offered from the workplace. Two related studies177,178 describe the development, piloting and outcomes from a tailored computerised cessation programme delivered in a presurgical secondary care setting.

Workplace interventions to support behaviour change can provide a useful way of improving the health of adults of working age. Reviews of workplace interventions for smoking cessation have concluded that, although they are likely to yield only a modest impact on smoking prevalence owing to low participation rates, they can be effective in supporting individual smokers to quit. 185,179 Graham et al. 176 report results from an internet-based workplace smoking cessation intervention delivered to IBM employees in the USA in 2003. During online enrolment for the company’s health-care benefits programme, employees were asked about smoking status. Those willing to identify themselves as current smokers (just 6.6%; n = 8688) were given a choice of a self-help printed cessation intervention or an internet-based intervention with tailored and interactive elements. Those who agreed to take up either option were given a benefits premium discount (worth US$132 per year). Although 72% of smokers took up one of the two options (n = 6235), just 28.5% of these (1776) chose to use the website and 1746 registered on the site.

Most of those that registered were male (65%; average age 44 years) and, as in the studies described above under ‘profile of users’, a proportion who chose to use the site (11.5%) had recently stopped smoking. Around one-third (32%) of those who registered were successfully followed up at 12 months, and it was found that the 7-day point prevalence abstinence rate was still 12.8% using an ITT approach. 176

Haile et al. 177 initially piloted a computerised interactive cessation programme in a pre-surgical admission clinic in a hospital in Australia in 1999. They reported results from this pilot in 2002 including assessing the acceptability of this electronic aid with patients. The intervention was brief, involving smokers attending the clinic (n = 56) completing a baseline questionnaire on a computer at the clinic and then receiving a short (completion time 12–43 minutes) computer-delivered cessation intervention during the same visit. Nine months later these smokers were followed up by telephone to examine cessation outcomes. Patients reported that they found the programme acceptable and had used the information learned in the programme. The authors also examined the cost of the programme and concluded that its principal advantage was to convey ‘expert’ smoking cessation counselling and feedback that clinic staff did not have the time or capacity to provide. 177

Members of the same research team then went on to refine and expand the intervention in the same clinic setting with results reported in 2009. 178 The expanded intervention integrated the computer-based programme into both preoperative and postoperative care. In this study the team also reviewed relevant literature and interviewed preoperative clinic staff (number of interviews not reported) regarding the barriers to the provision of effective smoking cessation care. They identified the following barriers:

• lack of organisational support

• perceived patient objection

• lack of systems to identify smokers

• lack of staff time and skill

• perceived inability to change care practices

• perceived lack of efficacy of cessation care

• cost of providing care.

The expanded programme involved completion of the same preoperative computerised questionnaire for smokers and delivery of tailored counselling via a computer in the clinic. Additions included brief advice from clinic nursing and anaesthetist staff that was guided by computer-generated prompts, followed by preoperative and postoperative provision of NRT. Just before admission, patients were telephoned and a computer-administered telephone interview providing further cessation counselling was delivered. Postdischarge patients were referred to a telephone quitline.

The authors concluded that the computer-based intervention provided a way to systematically and accurately identify smokers, did not require much clinical staff time or skill, was viewed by staff and patients as an acceptable form of care, and was inexpensive to deliver compared with other surgical costs. They also noted that the programme continued to be offered in the preoperative clinic after the research had ended. At the time the authors wrote the paper, the programme had been in place for 3 years since the end of the study and continued to be offered despite caseload expansion and restructuring of services in the hospital where the study took place.

Usability of interventions

A number of studies describe the features of electronic aids that can enhance user acceptability and potentially efficacy. These observations are not usually the main focus of articles but are reported in the description of the intervention and then commented on as part of the results. A number of themes relating to usability were identified from studies in the supplementary review. These themes were identified following review and data extraction (into narrative tables) of the main findings of each included study:

• involving users in the development of electronic aids can enhance usability

• enrolment in programmes involving electronic aids should be as easy as possible in order to retain smokers

• aids that include interactive tools or social support elements may be more acceptable and/or effective

• more frequent use of electronic aids may enhance efficacy.