For the 2015 version of the review (Hollands 2015b), the criteria for eligible interventions was refined between the protocol and the review. The original primary intention of the review was to examine the effect of interventions to increase adherence where this was the clearly intended focus of those intervening. However, this primary intention was not adequately reflected in the original criteria. As such, a large number of studies of interventions that could in theory alter adherence but where this was not the researchers' intention would have been relevant for inclusion. Furthermore, this lack of clarity meant that most extant studies that featured any intervention in smokers would have to be examined at the full‐text screening stage because a clear focus on increasing adherence (which can typically be derived from the title and abstract screening process) was not necessary for consideration for inclusion.

For the current update, we made additional changes to the original protocol.

Rather than including separate meta‐analyses combining each of continuous outcome data and dichotomous outcome data for a given outcome, we produced a single meta‐analysis that combined these using generic inverse variance methods. If a study reported both continuous and dichotomous outcomes that were similar in meeting other criteria, continuous data were used. In the previous version of the review, we had applied fixed‐effect models to our pooling of the data. This was in line with the protocol, in which we stated that we intended to group substantially similar studies. In practice, there was considerable methodological and clinical variance between studies, reflecting different characteristics of settings, participant groups, interventions and measures. As such, a random‐effects model was considered more appropriate.

We added formal coding of the focus of the content of the intervention, which while stated as an objective, had not been conducted for the previous version of this review. We coded studies by reference to two key factors: 1. focus on perceptions, practicalities, or both; 2. participant‐centred or clinician‐centred. This was then used as the basis for subgroup analyses to determine which types of interventions were most effective.

In the previous version of this review we proposed to assess the impact of using intention‐to‐treat data for adherence outcomes instead of using data for only those participants who remained engaged with treatment (the latter being our specified preferred approach). This was no longer considered necessary given justification for using the latter data, reflected in the two newly identified studies which both report adherence data for participants who remained engaged with treatment.

For risk of bias assessment, following guidance from the Cochrane Tobacco Addiction Group, we removed three domains (blinding of participants and personnel (performance bias); validity and reliability of outcome measures (other sources of bias); comparability of baseline characteristics (other sources of bias)). We followed guidance in Section 8.7 (Table 8.7a) of the Cochrane Handbook for Systematic Reviews of Interventions for deriving a summary risk of bias judgement from the domains that were assessed (Higgins 2011).