Types of studies

All trials were individually randomised controlled trials. Five trials involved randomisation into two groups which were both included in our analysis (Marteau 2012; Mooney 2007; Nollen 2011; Schmitz 2005; Tucker 2017), and three trials involved randomisation into three groups, where only two of these groups were eligible for this review (Chan 2010; Chan 2011; Mooney 2005). One trial involved a 2 × 2 × 2 factorial design with eight randomised groups, but these groups were collapsed into a two‐group comparison, relevant to this review, by the study authors (Smith 2013). One trial involved a 2 × 2 × 2 × 2 × 2 factorial design with 32 randomised groups, with these groups collapsed into three two‐group comparisons relevant to this review (Schlam 2018).

Types of participants and settings

Eight studies included a general population of smokers. Two studies included only participants with a specific clinical condition, namely erectile dysfunction (Chan 2010) and HIV/AIDS (Tucker 2017). The mean ages of participants in trials ranged from 34.6 years (Mooney 2005) to 49 years (Schmitz 2005). In two trials, all participants were women (Mooney 2007; Schmitz 2005). In one trial, all participants were men (Chan 2010). In the remaining trials, percentage women ranged from 7.5% (Tucker 2017) to 62.5% (Nollen 2011). Seven trials took place in the USA (Mooney 2005; Mooney 2007; Nollen 2011; Schlam 2018; Schmitz 2005; Smith 2013; Tucker 2017), two in Hong Kong, China (Chan 2010; Chan 2011), and one in the UK (Marteau 2012). Regarding setting, all but one of the included studies featured interventions that were at least in part delivered in‐person, with the other delivering the intervention by telephone (Smith 2013). The interventions were delivered in clinic (e.g. smoking cessation or outpatient clinics) or social service settings, apart from one that was delivered by telephone (Smith 2013), one where one of the three adherence interventions was delivered by automated telephone call (Schlam 2018), and two where the setting was unclear (Chan 2010; Chan 2011). Those delivering the intervention were trained counsellors or project staff (Chan 2010; Chan 2011; Mooney 2005; Nollen 2011; Schlam 2018; Smith 2013; Tucker 2017), nurses (Marteau 2012; Schmitz 2005), or cognitive behavioural therapy (CBT) practitioners (Mooney 2007).

Types of interventions

The trials all offered pharmacological treatment and some behavioural support, comprising a form of smoking cessation counselling with no particular emphasis on adherence (e.g. providing dosing instructions and weekly checks of adverse effects; Schmitz 2005), to participants in the control arm. Support for the control arm varied from a single support session of 16 minutes (Tucker 2017) or 20 minutes (Mooney 2005) to seven weekly sessions (Marteau 2012; Mooney 2007; Schmitz 2005). In the main, the intervention consisted of an additional component to the standard behavioural support, with eight studies providing additional contact time for those in the intervention arm (Chan 2010; Chan 2011; Mooney 2007; Nollen 2011; Schlam 2018; Schmitz 2005; Smith 2013; Tucker 2017). In the other two studies, the nature of the contact changed but its duration did not significantly differ (Marteau 2012; Mooney 2005). The interventions typically provided information on the rationale for, and emphasised the importance of, adherence to medication, and aided participants in developing strategies to overcome problems and barriers to maintaining adherence. As such, they included a combination of two intervention strategies outlined within a taxonomy of interventions to increase adherence (Haynes 2008), that is included in Appendix 2, namely 1. instruction for participants on medication use or 2. counselling about smoking, and the value of medication in overcoming addiction. Two studies included interventions that involved personalised feedback of medication taking, monitored electronically (Mooney 2007; Schmitz 2005); one study elicited participants' beliefs about medication taking and then provided personalised counselling relating to those beliefs (Mooney 2005); one study tailored medication dose to either genotype or degree of tobacco dependence and explained the rationale for this to participants (Marteau 2012); and five studies added additional counselling contact time to standard behavioural support, with content focusing on medication adherence, including the use of motivational interviewing techniques and the 4/5R approach to increasing motivation (counselling addressing risks, rewards, roadblocks, and repetition, and relevance in the case of the 5Rs; Chan 2010; Chan 2011), a focus on motivation to use the medication and behavioural skills for achieving this (Nollen 2011; Tucker 2017), and targeting medication beliefs with monitoring and feedback on adherence (Smith 2013). One study examined multiple adherence interventions concerning personalised feedback of adherence behaviour, automated medication reminder calls and additional behavioural support content focused on adherence (Schlam 2018).

Seven studies prescribed NRT (Chan 2010; Chan 2011; Marteau 2012; Mooney 2005; Schlam 2018; Smith 2013; Tucker 2017), two studies prescribed bupropion (Mooney 2007; Schmitz 2005), and one study prescribed varenicline (Nollen 2011).

We categorised the content of each intervention by reference to PAPA. Of 12 comparisons included in the review, three comparisons assessed the impact of changing perceptions (Marteau 2012; Mooney 2005; Schlam 2018 (medication adherence counselling intervention)), and five comparisons assessed the impact of interventions aiming to improve the practicalities of medication‐taking (Chan 2010; Chan 2011; Mooney 2007; Schlam 2018 (electronic monitoring feedback intervention); Schmitz 2005). Four comparisons assessed Interventions of perceptions and practicalities (Nollen 2011; Schlam 2018 (automated calls intervention); Smith 2013; Tucker 2017).

We also assessed whether interventions aimed at changing perceptions or practicalities assessed participants' particular beliefs or difficulties (patient‐centred) or provided a standardised intervention (clinician‐centred). Nine comparisons were patient‐centred (Chan 2010; Chan 2011; Mooney 2005; Mooney 2007; Schlam 2018 (medication adherence counselling intervention); Schlam 2018 (electronic monitoring feedback intervention); Schmitz 2005; Smith 2013; Tucker 2017), and three comparisons were clinician‐centred (Marteau 2012; Nollen 2011; Schlam 2018 (automated calls intervention)).

Types of outcome measures

Measures of adherence varied across studies. Five studies reported at least one continuous outcome, measured as the percentage or amount of prescribed medication that was consumed (Marteau 2012; Mooney 2005; Nollen 2011), number of days on which it was used (Smith 2013), or percentage of days on which a person was adherent (Schlam 2018). Five studies used a dichotomous outcome, meaning people were classified as either achieving or not achieving a specified degree of adherence that was deemed adequate (Chan 2010; Chan 2011; Mooney 2007; Schmitz 2005; Tucker 2017). The definitions of adequate adherence naturally varied by medication type and because there may not be agreed standards for what constitutes desirable levels of adherence. Furthermore, the operationalisation of this was not always clear. In assessing adherence, seven studies at least partly used tablet counts (Marteau 2012; Mooney 2005; Nollen 2011; Tucker 2017), or electronic monitoring systems (Mooney 2007; Schlam 2018; Schmitz 2005). One study used self‐report (Smith 2013), while the means of assessing adherence was unclear in two studies (Chan 2010; Chan 2011). The period for which the primary adherence outcome was being assessed ranged from approximately two weeks (Mooney 2005; Smith 2013), to three months (Nollen 2011).

To assess abstinence seven studies used biochemically validated outcomes (Chan 2010; Chan 2011; Marteau 2012; Mooney 2005; Mooney 2007; Nollen 2011; Tucker 2017), but only six of these provided useable data in study reports (Chan 2010; Chan 2011; Marteau 2012; Mooney 2005; Nollen 2011; Tucker 2017). Two studies provided self‐reported abstinence data (Schlam 2018; Smith 2013), and one study did not report abstinence (Schmitz 2005). Time of abstinence outcome measurement ranged from two weeks (Mooney 2005), to six months (Chan 2010; Chan 2011; Marteau 2012; Schlam 2018; Smith 2013), to one year (Schlam 2018).

Adherence to medication for tobacco dependence

Five studies reported dichotomous adherence measures (Chan 2010; Chan 2011; Mooney 2007; Schmitz 2005; Tucker 2017). Chan 2010 and Chan 2011 assessed whether or not there had been continuous use of NRT, for four weeks (Chan 2010) and eight weeks (Chan 2011). Mooney 2007 and Schmitz 2005 both assessed whether or not participants had taken two daily doses of bupropion as prescribed over the seven‐week treatment period. Tucker 2017 assessed whether participants had used six or more nicotine patches per week, for those participants who provided complete data at baseline and follow‐up. Five studies used continuous adherence measures (Marteau 2012; Mooney 2005; Nollen 2011; Schlam 2018; Smith 2013). Marteau 2012 assessed the proportion of prescribed NRT consumed over the four‐week treatment period and reported the group mean. Mooney 2005 reported the mean pieces of nicotine gum used during the first 15 days of a quit attempt in those who completed the treatment period only. Nollen 2011 assessed the proportion of prescribed varenicline doses taken over three months, for those who remained engaged. Schlam 2018 measured the percentage of days in the first six weeks of the quit attempt where participants were adherent (i.e. where participants used both a nicotine patch and four or more pieces of gum), in those participants who completed the treatment phase. Smith 2013 assessed self‐reported number of days of nicotine patch use in the first two weeks, for those remaining engaged.

Pooled analysis of these data, comprising 12 comparisons from 10 studies, showed that adherence interventions produced a small improvement in adherence, with no significant statistical heterogeneity being observed (SMD 0.10, 95% CI 0.03 to 0.18; I² = 6%; n = 3655; Figure 3). Re‐expressing this effect size produced by the primary random‐effects meta‐analysis in a more familiar metric (see Data synthesis) suggested that interventions to increase adherence could have an effect equivalent to a 3.9% increase (95% CI 1.2% to 7.0%) in the mean proportion of prescribed medications consumed over 28 days.

Figure 3

---

---

Forest plot of comparison: 1 Medication adherence intervention plus standard care versus standard care alone, outcome: 1.1 Adherence (combined dichotomous and continuous).

GRADE assessment indicated that the evidence for this outcome was of moderate certainty, meaning that the true effect is probably close to the estimated effect. This judgement was reached through consideration of the following criteria. The current evidence was downgraded once due to risk of bias, because the majority of studies were judged to be at high or unclear risk of bias. We did not downgrade the evidence further based on other GRADE considerations. For imprecision, the number of participants (sample size) incorporated into this meta‐analysis exceeded the optimal information size (i.e. a sufficient sample size for a single adequately powered trial), and the 95% CI ranged from a very small to a small benefit. For inconsistency, there was minimal heterogeneity and considerable overlapping of CIs. There was no clear reason to downgrade certainty of evidence for indirectness (providing it was emphasised that moderate evidence related only to those receiving an adherence intervention in addition to behavioural support for smoking cessation, compared to behavioural support alone). Finally, for other considerations, including publication bias, the certainty of the evidence was not downgraded. Although a funnel plot of the primary outcome data suggested possible asymmetry (Figure 4), only one of 12 included comparisons was statistically significant, and there was not a clearly consistent pattern of smaller studies resulting in greater intervention effect estimates than larger studies. This limited the plausibility of publication bias as an explanation for asymmetry (Sterne 2011).

Figure 4

---

---

Funnel plot of comparison: 1 Medication adherence intervention plus standard care versus standard care alone, outcome: 1.1 Adherence (combined dichotomous and continuous).

Subgroup analyses

We conducted three subgroup analyses of the primary analysis in order to examine the relative impact of specific intervention types in terms of their focus on 'perceptions'; 'practicalities'; or 'both' (Analysis 1.2); and whether the intervention was participant‐centred or clinician‐centred (Analysis 1.3). The third analysis considered these two factors in combination (Analysis 1.4).

There was no strong evidence that the effect of interventions that focused on perceptions, practicalities, or on both differed in their effect on adherence (I² = 13%, P = 0.32; Analysis 1.2). That said, the effect of interventions focused on practicalities appeared slightly larger than the other two groups, with an SMD of 0.21 (95% CI 0.03 to 0.38; I² = 39%; n = 1752), compared with perceptions (SMD 0.10, 95% CI –0.03 to 0.24; I² = 0%; n = 839), or a combination of both perceptions and practicalities (SMD 0.04, 95% CI –0.08 to 0.16; I² = 0%; n = 1064).

There was no clear evidence that participant‐centred interventions differed in effectiveness from clinician‐centred interventions (I² = 0%, P = 0.71; Analysis 1.3). The SMD for participant‐centred interventions was 0.12 (95% CI 0.02 to 0.23; I² = 20%; n = 2791) and for clinician‐centred interventions was 0.09 (95% CI –0.05 to 0.23; I² = 0%; n = 864).

There was also no strong evidence that combining these two classification systems led to subgroup differences in the effect of interventions on medication adherence (I² = 0%, P = 0.65; Analysis 1.4).

We conducted a further subgroup analysis to examine whether there were differential effects of the intervention depending on which medication was prescribed (Analysis 1.5). In this analysis, there was stronger evidence of subgroup differences (I² = 68%, P = 0.04). The effect of interventions to increase adherence to bupropion (SMD 0.58, 95% CI 0.14 to 1.01; I² = 0%; n = 152) was much larger than that for NRT (SMD 0.09, 95% CI 0.02 to 0.17; I² = 0%; n = 3442) or varenicline (SMD from only one study was –0.22, 95% CI –0.73 to 0.29; n = 61).

Short‐term abstinence (less than six months)

Five studies contributed data to the analysis of short‐term abstinence (Marteau 2012; Mooney 2005; Nollen 2011; Smith 2013; Tucker 2017). Marteau 2012 assessed biochemically validated prolonged abstinence at 28 days, Mooney 2005 assessed biochemically validated point‐prevalent abstinence at two weeks and Nollen 2011 assessed biochemically validated point‐prevalent abstinence at three months. Smith 2013 measured self‐reported 30‐day point‐prevalent abstinence at six weeks, while Tucker 2017 assessed biochemically validated continuous abstinence over 90 days.

Random‐effects meta‐analysis pooling these data produced an RR of 1.08 (95% CI 0.96 to 1.21; I² = 0%; n = 1795; Analysis 1.8). This suggested a potential small effect of adherence interventions on short‐term abstinence from smoking but with considerable uncertainty due to CIs overlapping no effect and including the possibility of a very small negative effect on abstinence.

GRADE assessment indicated that the evidence for this outcome was of low certainty, meaning that the true effect might be markedly different from the estimated effect. This judgement was reached through consideration of the following criteria. The current evidence was downgraded by one level due to risk of bias, because the majority of studies were judged to be at high or unclear risk of bias. It was also downgraded by one level due to imprecision, because while the number of participants (sample size) incorporated into this meta‐analysis exceeded the optimal information size (i.e. a sufficient sample size for a single adequately powered trial), the 95% CI overlapped no effect and ranged from a very small harm to a small benefit. We did not downgrade further due to other considerations, namely inconsistency (because there was negligible heterogeneity), indirectness or publication bias (with insufficient studies for formal assessment).

Long‐term abstinence (six months or longer)

Five studies (seven comparisons) contributed data to long‐term abstinence (Chan 2010; Chan 2011; Marteau 2012; Schlam 2018; Smith 2013). All five studies assessed abstinence at six months, which was biochemically validated in three studies (Chan 2010; Chan 2011; Marteau 2012), and based on self‐report in two studies (Schlam 2018; Smith 2013). Random‐effects meta‐analysis pooling these data produced an RR of 1.16 (95% CI 0.96 to 1.40; I² = 48%; n = 3593; Analysis 1.9). This suggested a potential small effect of adherence interventions on long‐term abstinence; however, considerable uncertainty arose due to the lower CI including the possibility of a very small negative effect on abstinence. Participants subject to interventions to improve adherence were between 4% less likely and 16% more likely to be abstinent at six months than those given standard behavioural support.

GRADE assessment indicated that the evidence for this outcome was of low certainty, meaning that the true effect might be markedly different from the estimated effect. This judgement was reached through consideration of the following criteria. The current evidence was downgraded by one level due to risk of bias because the majority of studies were at high or unclear risk of bias. It was also downgraded by one level for imprecision, because while the number of participants (sample size) incorporated into this meta‐analysis exceeded the optimal information size (i.e. a sufficient sample size for a single adequately powered trial), the 95% CI overlapped no effect and ranged from a very small harm to a small benefit. We did not downgrade further for other considerations, namely inconsistency (because heterogeneity was not classed as substantial), indirectness or publication bias (there were insufficient studies for formal assessment).

Factors plausibly associated with increases in adherence

No studies reported any relevant outcomes (i.e. factors plausibly associated with increases in adherence, such as intention or motivation, or attitudes towards treatment).

Adverse events

Four studies explicitly reported adverse events (Marteau 2012; Mooney 2005; Schlam 2018; Smith 2013). In Marteau 2012, there were no adverse events that were plausibly related to the intervention or its effect on participants' exposure to medication. There were also no differences between groups in levels of anxiety at either one‐week or six‐month assessment times. In Mooney 2005, there was no difference in adverse events between groups and in Schlam 2018 and Smith 2013 there were no serious adverse events during the study.