Smoking cessation for people with severe mental illness: systematic review and network meta-analysis

Abstract

Objective

To evaluate the comparative effectiveness and safety of pharmacological and non-pharmacological smoking cessation interventions in people with severe mental illness

Design

Systematic review and network meta-analysis.

Data sources

Medline, Embase, Scopus, Cochrane Central Register of Controlled Trials (CENTRAL), and Web of Science, from inception to 19 September 2024.

Eligibility criteria for selecting studies

Trials enrolling adults who had established diagnoses of schizophrenia, bipolar disorder, recurrent or current severe major depressive disorder, or post-traumatic stress disorder, randomised to a smoking cessation intervention versus another active treatment, placebo, standard care, or no treatment.

Results

74 randomised controlled trials (11 023 participants) evaluating nine smoking cessation interventions were included in the study. Compared with placebo or minimal care, varenicline (10 more per 100 achieving long term smoking abstinence, 95% confidence interval (CI) 5 to 16; high certainty evidence) and bupropion (5 more per 100, 1 to 10; moderate certainty evidence) increased long term abstinence. Effects on short term smoking abstinence were similar. Nicotine replacement therapy improved short term abstinence but with little or no long term abstinence benefit (moderate certainty evidence). Combination interventions (pharmacological with non-pharmacological interventions) may increase long term abstinence (6 more per 100, 95% CI 3 to 11; low certainty evidence). The certainty of evidence for other interventions was very low. Serious adverse event data were highly uncertain. Dropout from a trial because of harms was possibly no different for varenicline, bupropion, and nicotine replacement therapy compared with placebo or minimal care.

Conclusions

Varenicline, bupropion, and nicotine replacement therapy likely improved smoking abstinence in people with severe mental illness compared with placebo or minimal care (moderate to high certainty evidence). Combined pharmacological and non-pharmacological approaches may offer more benefit (low certainty evidence), but the risk of serious adverse events for all interventions was very uncertain.

Systematic review registration

PROSPERO CRD42022349498.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• People with serious mental illness are less likely to receive support for smoking cessation, even though tobacco use greatly harms their health

• Clinicians are often reluctant to offer these interventions because of concerns about side effects and the belief that individuals with severe mental illness need different treatment approaches than the general population

WHAT THIS STUDY ADDS

• Varenicline, bupropion, and nicotine replacement therapy probably improve smoking abstinence in people with severe mental illness; compared with placebo or minimal care 2 to 25 more people per 100 probably archieve short term abstinence and 1 to 16 more people probably archieve long term abstinence (with varenicline or bupropion)

• Combined pharmacological and non-pharmacological approaches may provide additional benefit; between 3 to 11 people more may archieve long term abstinence, but certainty of evidence is low

• Estimates of risk of serious adverse events for all interventions are very uncertain

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY

• By quantifying the extent that specific interventions can improve abstinence rates, the study can help manage these concerns and encourage clinicians to implement evidence based strategies in practice

Introduction

Severe mental illness, characterised by the three dimensions diagnosis, duration, and disability,¹ and including schizophrenia, bipolar disorder, and major depressive disorder affect 20 million, 45.5 million, and 163 million people, respectively, worldwide.²,⁵ Compared with the general population, adults with severe mental illness have a twofold to threefold higher risk of mortality⁶,⁹ and 10-20 year reduction in life expectancy.^{8 10 11} Physical diseases, particularly cardiovascular sequelae, contribute substantially to the increased risk of mortality.^{12 13} The World Health Organization has recently called for accelerated action for those with severe mental illness, and through its Special Initiative for Mental Health has committed to looking at gaps in mental health service and treatment internationally.¹⁴

About half of the deaths in people with severe mental illness are from preventable physical illnesses, particularly cardiovascular and respiratory diseases, which in turn are strongly linked to the use of tobacco.¹⁵,¹⁹ By targeting tobacco smoking as a modifiable lifestyle factor, many of these conditions could be prevented. Whereas tobacco smoking rates have substantially decreased among the general population (in the US from 20.9% in 2005 to 11.5% in 2021),²⁰ rates seem persistently raised among adults with severe mental illness.^{17 21} People with severe mental illness are 2-6 times more likely to smoke,¹⁹²²,²⁴ are heavier smokers (ie, smoke ≥20 cigarettes a day), and have a greater severity of nicotine dependence and withdrawal symptoms when attempting to quit smoking.¹⁷²³,²⁸

Pharmacological (varenicline, bupropion, and nicotine replacement therapy) and behavioural or psychosocial (eg, motivational interviewing and acceptance and commitment therapy) interventions may reduce smoking in people with severe mental illness.¹⁷²⁹,³⁵ So far, no recent comprehensive systematic review and meta-analysis has been conducted to assess the effectiveness and safety of these interventions for people with severe mental illness. A network meta-analysis,³⁶ rather than a pairwise meta-analysis, evaluates the comparative effectiveness and safety of multiple treatment alternatives, including those that have not been tested directly against each other. Network meta-analyses also combine direct evidence (between treatments that have been tested against each other) and indirect evidence (between treatments that have not) to strengthen inferences made about treatment comparisons.

We therefore conducted a systematic review and network meta-analysis of randomised trials to assess the comparative effectiveness and safety of pharmacological and non-pharmacological interventions for smoking cessation in people with severe mental illness. This review is part of the BMJ Rapid Recommendations project, a collaborative effort from the MAGIC (Making GRADE the Irresistible Choice) Evidence Ecosystem Foundation (www.magicevidence.org) and The BMJ to produce trustworthy recommendations in response to practice changing evidence. This systematic review will inform a parallel Rapid Recommendations guideline.

Methods

Our systematic review and network meta-analysis was registered in PROSPERO (CRD42022349498). Our results are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Network Meta-Analysis Extension checklist,^{37 38} and the study used the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) method.^{39 40}

The parallel BMJ Rapid Recommendations guideline panel provided general oversight of the project and defined the scope of this systematic review. The panel identified and prioritised patient important outcomes and provided outcome specific minimal important difference thresholds. The panel consisted of unconflicted patient partners, methodologists, general practitioners, internists, psychiatrists, psychologists, nicotine addiction specialists, and a pulmonary physician.

Search strategy

We conducted systematic searches of Medline, Embase, Web of Science, Scopus, and Cochrane Central Register of Controlled Trials (CENTRAL), from inception to 19 September 2024, without language restriction. One reviewer (JD) and an experienced information specialist (TV) developed all of the database specific search strategies (online supplemental appendix 1). We also reviewed the reference lists of eligible studies and pertinent reviews to identify other relevant studies.

Eligibility criteria

We included randomised controlled trials that enrolled adults aged ≥18 years who had an established diagnosis of schizophrenia, bipolar disorder, recurrent or current severe major depressive disorder, or post-traumatic stress disorder, who smoked tobacco, and who agreed to smoking cessation treatment. The randomised trials included adults receiving one or more interventions aimed at reducing or quitting smoking versus an alternate active intervention, placebo, standard care, or no treatment. All included trials assessed at least one prespecified patient important outcome. We also included studies that examined smoking cessation in a broad population but performed a subgroup analysis (eg, in people with a psychiatric diagnosis of major depressive disorder) and studies that included people with any psychiatric diagnosis but where at least 80% had schizophrenia, bipolar disorder, recurrent or current major depressive disorder, or post-traumatic stress disorder.

Eligible patient important outcomes were short term smoking abstinence (≥2 weeks), long term abstinence (≥6 months), mental health, quality of life, weight gain, serious adverse events, and dropout from the trial because of adverse events or harm. For studies reporting on abstinence with different measures, we prioritised biochemically verified results over self-report, sustained abstinence data over point prevalence, and biochemically validated point prevalence over self-reported sustained abstinence data, as outlined in a previous Cochrane review on smoking cessation.⁴¹ Online supplemental appendix 2 has the full eligibility criteria.

Study selection

Duplicates were removed by a stepwise semiautomatic procedure in Endnote,⁴² followed by automatic and manual removal with Covidence Software.⁴³ Then, paired reviewers (MDH, BA, GEB, JD, A-CV, and HC) used the prespecified eligibility criteria and the Covidence Software to screen the titles and abstracts of identified citations, followed by the full texts of potentially eligible studies. Disagreements were resolved by consensus.

Data extraction

For each eligible study, one reviewer (GEB, JD, HC, A-CV, or AA) extracted data by using standardised, pre-tested forms in Covidence and in an Excel spreadsheet, and a second reviewer crosschecked the data for accuracy. Data extracted from each study were: general information (trial registration number and title); study characteristics (country and location (hospital or city, or both), length of intervention, length of follow-up, start and end date of recruitment, and setting); participant characteristics (total number of randomised participants, mean age, sex, severe mental illness diagnosis, diagnostic classification system, illness stage (first episode v chronic condition), severity of illness, somatic comorbidities, drug treatments and co-treatments for severe mental illness, smoking history, previous attempts at quitting smoking, and whether the participant had an intention to quit smoking); details of interventions and comparators (including dose, route, intensity, and duration, where applicable); and results for prespecified outcomes.

Masked to the study results, a group of content experts and methodologists grouped similar interventions for the included studies. We categorised study arms into three groups: pharmacological interventions, non-pharmacological interventions, and combinations of pharmacological and non-pharmacological interventions. Although we explored more detailed networks (specifically separating single v dual nicotine replacement therapy components, psychological theory based v non-theory based non-pharmacological interventions), we chose to use broader networks because of lack of data.

Brief counselling, defined pragmatically as <20 minutes of education, advice, or support for each session, was allowed for all interventions. Counselling for >20 minutes for each session was considered non-pharmacological treatment. All studies assessing pharmacological interventions included brief counselling.

To ensure that all studies were included in one network, and given the reasonable similarities in anticipated effects, we combined placebo and minimal care into a common comparator node. Minimal care was defined as access or referral to usual smoking cessation care, including referrals to specialised settings.

We grouped treatments into categories or treatment nodes: placebo or minimal care (reference), varenicline, bupropion, nicotine replacement therapy, any pharmacological treatment (participants were given the opportunity to take smoking cessation drug treatments but not all participants actually took these drugs, which resembles clinical practice), and non-pharmacological treatments. We also considered any combination of pharmacological and non-pharmacological interventions as separate categories, such as nicotine replacement therapy with non-pharmacological interventions.

Risk of bias in individual studies

Pairs of reviewers (GEB, HC, AA, MV, and A-CV) independently assessed the risk of bias with the Cochrane risk-of-bias tool version 1, with answering options similar to version 2 of the tool.^{44 45} Disagreements were resolved by consensus. The tool’s items are: random sequence generation; allocation concealment (of random sequence); masking of patients, caregivers, outcome assessors, and analysts; incomplete outcome data (≥20% missing data was judged to be high risk of bias); and selective reporting. Risk of bias for each criterion was evaluated as low, probably low, probably high, or high. The two judgments, low and probably low, were categorised as low risk of bias, and the judgments probably high and high as high risk of bias. For random sequence generation and allocation concealment, if no information was provided by the trial authors, reviewers judged that the risk of bias was probably high.

Data synthesis

For our network meta-analysis, we first reviewed the direct evidence for each pairwise comparison with forest plots, and constructed network plots to visualise the data structure for each outcome. We then performed a network meta-analysis with a bayesian framework. For binary outcomes, we used a binomial likelihood with a log link to estimate the posterior mean log odds ratio with 95% confidence intervals. For continuous outcomes, we applied a normal likelihood with an identity link, with standardised mean difference as the effect measure to account for variability in outcome scales used across the included studies.

We assessed model convergence with Gelman plots, and examined the potential for inconsistency with node splitting. All analyses were conducted with MetaInsight,⁴⁶ and the gemtc,⁴⁷ BUGSnet,⁴⁸ and netmeta⁴⁹ packages in R Core team.⁵⁰ We performed pairwise random effects meta-analyses for outcomes with sparse data: dropout from the trial because of harm, quality of life, and weight gain. For this analysis, we used Revman,⁵¹ based on the DerSimonian and Laird method to estimate the average relative risk or mean difference across trials and the corresponding confidence intervals.

Interpretation of results

For dichotomous outcomes, we inferred baseline risks based on pooled event rates for the reference group (placebo or minimal care), and calculated absolute treatment effects incorporating relative effects and estimated baseline risks. For two continuous outcomes (depression and general psychopathology), we extrapolated to a commonly used scale with the standard deviation from well performed large studies.^{52 53}

To differentiate important from unimportant (ie, little or no difference, or trivial) effects, we used minimal important difference thresholds informed by the literature and an informal survey of the linked guideline panel.⁵⁴ We used a threshold of 5 per 100 adults for short term and long term abstinence, 5 on a scale from 0 to 52 for depression (Hamilton Depression Rating Scale),⁵⁵ and 15 on a scale from 30 to 210 for general psychopathology (Positive and Negative Syndrome Scale).⁵⁶ We used a null threshold for serious adverse events and dropout from the trial because of harm.

Categorisation of interventions

For each outcome in the network meta-analysis, we categorised interventions from most to least effective by using a minimally contextualised approach.⁵⁷ Treatments were categorised from most to least effective or harmful, based on treatment effect estimates, and simultaneously on the associated certainty of the evidence compared with the reference standard. We chose placebo or minimal care as the reference intervention. Next, we classified interventions based on whether they were better than the reference standard (group 1), and subsequently based on whether any interventions from group 1 were better than at least one other intervention in group 1 (group 2). Within the two groups, we categorised the interventions based on whether their pairwise comparisons relative to the reference standard were informed by high or moderate certainty evidence, or by low or very low certainty evidence. Because at most three interventions in group 1 had moderate or high certainty evidence, and only one intervention in group 2 had moderate or high certainty evidence, we have presented the results of all group 1 interventions, regardless of the level of certainty. With this approach, we ultimately provided an overview of all interventions that were better than the reference standard, divided into four certainty of evidence groups.

Subgroup and sensitivity analyses

We planned subgroup analyses based on severe mental illness diagnoses and on treatment setting, but these were not conducted because of lack of data in our nodes. During the review, we noted that some studies examined known smoking cessation interventions for different purposes, such as cognitive functioning. In these studies, participants may not have been aware of the smoking cessation intervention, which we judged had the potential to bias the results, based on input from clinical experts on the linked guideline panel. We therefore performed a sensitivity analysis excluding studies where smoking cessation interventions were evaluated for a primary indication other than smoking cessation. These studies only reported short term abstinence.

Assessing certainty of the evidence

We used the GRADE approach for network meta-analysis^{58 59} to assess the certainty of direct, indirect, and network estimates for all pooled outcomes. Certainty was judged as being high, moderate, low, or very low. Certainty ratings for randomised trials started as high. Certainty in direct evidence could be rated down for concerns about the risk of bias, inconsistency, indirectness, or small study effects. Certainty in indirect estimates started at the lowest GRADE rating among the direct comparisons contributing the most weight to the dominant first order loop. Network estimate certainty ratings accounted for the proportional contribution of direct and indirect evidence, and for imprecision of the network estimate. We rated down for imprecision if associated 95% confidence intervals included the prespecified minimal important difference for abstinence, for continuous outcomes, or the null threshold for serious adverse events.⁶⁰

When we identified significant incoherence between direct and indirect evidence, we rated down the certainty of the network estimate. We also assessed for transitivity, a requirement for indirect comparisons, which means that effect modifiers for the different sets of comparisons are similar.⁶¹ We generally chose a high threshold for rating down⁶² for intransitivity, given that we used strict population related eligibility criteria when identifying studies. For outcomes that were analysed with a pairwise meta-analysis, we used the standard GRADE approach to assess the certainty of the evidence.^{39 40} GRADE certainty ratings were performed by one reviewer (GEB) and checked for accuracy by a second reviewer (ND). Discrepancies were resolved by discussion. Absolute effect estimates across outcomes were presented with GRADE summary of findings tables for network meta-analyses.

Patient and public involvement

Four patients with severe mental illness and lived experience of smoking cessation interventions were involved in outcome selection and prioritisation, identification of thresholds for important versus unimportant effects, interpretation of results, and the generation of recommendations as part of the BMJ Rapid Recommendations initiative. Results of this study will be disseminated to patients as part of that recommendation.

Results

Of 12 005 unique hits, 74 studies of 11 023 participants were considered eligible for inclusion in our study (figure 1). We excluded 146 studies because of a wrong population, mostly because patients did not have a psychiatric confirmed diagnosis but might have had depressive symptoms, for example. Sixty six studies were excluded because of a wrong intervention. Typical reasons for exclusion were that interventions did not aim to reduce or quit smoking but instead aimed to reduce craving, or lifestyle interventions had multiple aims (eg, smoking cessation and improving physical fitness).

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart of studies included in the systematic review.

Three large studies (n>900)⁶³,⁶⁵ accounted for more than one third of the total patient sample. Results of the Evaluating Adverse Events in a Global Smoking Cessation Study (EAGLES) were reported as four separate subgroups based on diagnosis,^{30 52 63 66} and therefore our 74 studies reported on 77 unique populations. In 35 studies, participants had schizophrenic spectrum disorders (n=2361), in 14 studies participants had post-traumatic stress disorder (n=2278), in five studies participants had bipolar disorder (n=406), in two studies participants had recurrent or current severe major depressive disorder (n=1324), and 21 studies had a mixed population of people with severe mental illness (n=4654). All studies were conducted in high income countries. Mean age was 46 years (range 24-57) and 65% of participants were men. At baseline, participants smoked a mean of 20 cigarettes/day (range 7-34). Mean score on the Fagerström test for nicotine dependence was 5.8 (standard deviation 1.4). In 35 trials, the intention to quit smoking was explicitly reported as one of the inclusion criteria.

Most study interventions were multimodal (n=45), providing a combination of pharmacological treatment together with psychosocial approaches. The average length of the interventions was 12.5 weeks; the average follow-up time beyond the intervention was 21.8 weeks. Online supplemental appendix 3 reports all of the characteristics of the studies.

Risk of bias

Of the 74 studies, most trials (n=72, 94%) were at risk of bias for at least one domain. One study (1%) had an inadequate generated sequence, five (7%) had inadequately concealed allocation, with 54% and 58% with unclear reporting of the respective procedures for randomisation. In 39 (53%) studies, participants were not masked, in 42 (57%) studies healthcare providers were not masked, and in 26 (35%) studies outcome assessors were not masked. Seventeen trials (23%) reported frequent loss to follow-up and 29 (39%) had selective reporting (online supplemental appendix 4).

Main findings for each outcome

Table 1 presents the relative and absolute effect estimates for all treatments and prioritised outcomes (table 2 explains the colour scheme). Figure 2 shows the network plot for long term abstinence.

Table 1. Relative and absolute effect estimates for single and multimodal intervention treatments for prioritised outcomes.

Intervention v placebo* or minimal care	Absolute effect estimates (95% CI) and odds ratio (95% CI)
	Short term abstinence† (4-20 weeks)	Long term abstinence‡ (6-18 months)	Serious adverse events§	Depression¶	General psychopathology**
Single interventions
Varenicline*	17 more per 100 (11 more to 25 more) OR 3.89 (2.65 to 5.81)	10 more per 100 (5 more to 16 more) OR 2.84 (1.86 to 4.27)	0 more per 100 (0 more to 0 more) OR 1.12 (0.41 to 3.19)	0.27 lower (0.95 lower to 0.38 higher)	3.1 higher (2.5 lower to 8.7 higher)
Bupropion*	7 more per 100 (2 more to 13 more) OR 2.01 (1.30 to 3.11)	5 more per 100 (1 more to 10 more) OR 1.89 (1.17 to 2.96)	0 more per 100 (0 more to 0 more) OR 1.11 (0.28 to 4.31)	0.38 lower (1.10 lower to 0.38 higher)	21.7 lower (39.2 lower to 4.3 higher)
Nicotine replacement therapy*	8 more per 100 (4 more to 15 more) OR 2.30 (1.54 to 3.54)	4 more per 100 (1 more to 9 more) OR 1.71 (1.07 to 2.63)	0 more per 100 (0 more to 0 more) OR 1.28 (0.33 to 4.91)	0.11 higher (0.68 lower to 0.91 higher)
Non-pharmacological treatments	18 more per 100 (0 more to 55 more) OR 4.17 (0.98 to 20.2)	5 more per 100 (1 more to 10 more) OR 1.82 (1.21 to 2.95)
Multimodal interventions
Pharmacological with non-pharmacological treatments	8 more per 100 (4 more to 15 more) OR 2.23 (1.52 to 3.56)	6 more per 100 (3 more to 11 more) OR 2.09 (1.49 to 3.00)	0 more per 100 (0 more to 0 more) OR 1.34 (0.70 to 2.99)	0.27 higher (0.15 lower to 0.65 higher)
Nicotine replacement therapy with non-pharmacological treatments	10 more per 100 (2 more to 22 more) OR 2.60 (1.35 to 4.96)	4 more per 100 (1 less to 12 more) OR 1.71 (0.87 to 3.24)	0 more per 100 (0 more to 1 more) OR 1.21 (0.23 to 6.5)	0.72 lower (1.60 lower to 0.19 higher)	2.5 lower (19.8 lower to 3.7 higher)
Bupropion with non-pharmacological treatments	53 more per 100 (11 more to 83 more) OR 17.9 (2.72 to 117)	27 more per 100 (7 more to 61 more) OR 7.48 (2.23 to 30.7)
Bupropion with nicotine replacement therapy and non-pharmacological treatments	29 more per 100 (3 more to 64 more) OR 6.77 (1.48 to 30.2)	19 more per 100 (4 more to 48 more) OR 5.16 (1.59 to 17.7)		0.11 lower (2.36 lower to 2.17 higher)	8.1 lower (19.8 lower to 3.7 higher)

^*All study arms with pharmacological interventions or placebo also provided brief counselling.

^†Baseline effects in placebo or minimal care group, 8 per 100.

^‡Baseline effects in placebo or minimal care group, 6 per 100.

^§Baseline effects in placebo or minimal care group, 0 per 100 but for calculation purposes set to 0.1 per 100.

^¶Results for depression converted to Hamilton Depression Rating Scale (range 0-52, higher score reflects more symptoms, minimal important difference 3-5).⁵⁵

^**Results for general psychopathology converted to Positive and Negative Syndrome Scale (range 30-210, higher score reflects more symptoms, minimal important difference 15).⁵⁶

CI, confidence interval; OR, odds ratio.

Table 2. Colour scheme for level of certainly used in table 1.

Level of certainty	Category
	More effective than placebo or minimal care	Worse than placebo or minimal care	No different than placebo or minimal care
High certainty	Definitely more effective than placebo or minimal care	Definitely worse than placebo or minimal care	Definitely no different than placebo or minimal care
Moderate certainty	Probably more effective than placebo or minimal care	Probably worse than placebo or minimal care	Probably no different than placebo or minimal care
Low certainty	Maybe more effective than placebo or minimal care	Maybe worse than placebo or minimal care	Maybe no different than placebo or minimal care
Very low certainty	We are very uncertain

Figure 2. Network plot for long term abstinence. Lines between interventions are direct comparisons. Nodes represent specific interventions being compared in the network. The thickness of lines represents the number of studies for each direct comparison. The node placebo also includes minimal care interventions. Non-pharmacological=any non-pharmacological intervention; pharmacological-non-pharmacological=combined intervention consisting of any flexible, personalised pharmacological intervention with any non-pharmacological intervention.

Short term smoking abstinence

Fifty five randomised controlled trials reported on short term smoking abstinence and of these, 27 trials (n=5014) were included in the network meta-analysis. Online supplemental appendix 5 has details of the network meta-analysis and results of other studies.

Informed by moderate or high certainty evidence, three interventions were better than placebo or minimal care: bupropion (odds ratio 2.01, 95% confidence interval (CI) 1.30 to 3.11; 7 more per 100 adults, 95% CI 2 more to 13 more), nicotine replacement therapy (odds ratio 2.30, 1.54 to 3.54; 8 more per 100, 4 more to 15 more), and varenicline (odds ratio 3.89, 2.65 to 5.81; 17 more per 100, 11 more to 25 more) (table 1). Varenicline was better than nicotine replacement therapy (odds ratio 1.69, 95% CI 1.14 to 2.44) and bupropion (odds ratio 1.92, 1.32 to 2.93).

Post hoc sensitivity analysis, excluding two studies^{67 68} for short term abstinence, did not change the estimates of our network meta-analysis: bupropion (odds ratio 2.0, 95% CI 1.30 to 3.11), nicotine replacement therapy (odds ratio 2.25, 1.53 to 3.44), and varenicline (odds ratio 3.83, 2.69 to 5.58) (online supplemental figure S5).

Long term smoking abstinence

Forty two randomised controlled trials reported long term smoking abstinence outcomes and of these, 28 trials (n=6883) were included in the network meta-analysis. Online supplemental appendix 6 has details of the results. Two interventions supported by moderate or high certainty evidence and one supported by low certainty evidence were better than placebo or minimal care: bupropion (odds ratio 1.89, 95% CI 1.17 to 2.96; 5 more per 100, 95% CI 1 more to 10 more) and varenicline (odds ratio 2.84, 1.86 to 4.27; 10 more per 100, 5 more to 16 more). Low certainty evidence supported a combined pharmacological and non-pharmacological intervention (odds ratio 2.09, 1.49 to 3.00; 6 more per 100, 3 more to 11 more) (table 1). In these studies, 48-94% used any pharmacological intervention. Non-pharmacological interventions consisted of behavioural or motivational enhancing support, or both (online supplemental table S12). Varenicline was better than bupropion for long term abstinence outcomes (odds ratio 1.49, 1.02 to 2.27).

Serious adverse events

Forty two randomised controlled trials reported serious adverse events. Overall, 857 serious adverse events were reported in a total sample of 7184 participants, with large variation between studies. Eleven studies assessing six interventions and including 4221 participants were combined in the network meta-analysis. Relative effect estimates for all interventions showed a small increase in the risk of serious adverse events, but the certainty of the evidence was very low across comparisons (table 1). Eighteen trials (n=1295) reported no serious adverse events in any of the groups. Online supplemental appendix 7 has details on all studies, including those not included in the network meta-analysis

Psychiatric symptom ratings

Thirty three randomised controlled trials reported psychiatric symptoms based on symptom rating scales (online supplemental appendix 8). Twenty three trials reported depressive symptoms, and 12 of these (n=2574), assessing seven interventions, were included in the network meta-analysis. We found probably little or no difference in depressive symptoms with varenicline, bupropion, and nicotine replacement therapy compared with placebo or minimal care (moderate certainty evidence), and possibly little or no difference with a combination of pharmacological and non-pharmacological treatments (focusing on behavioural support, online supplemental table S25; low certainty evidence).

Of 14 studies reporting on global psychopathology, five (n=502) were included in the network meta-analysis. Varenicline probably resulted in little or no difference in global psychopathology scores compared with placebo or minimal care (moderate certainty). Similar results were found for nicotine replacement therapy and non-pharmacological interventions (cognitive behavioural therapy with or without motivational interviewing, online supplemental table S30), but the certainty of the evidence was low. Bupropion possibly resulted in lower scores (low certainty evidence).

Dropout from trial because of harm

Twenty four randomised controlled trials (n=3261) reported 75 participants dropping out of a trial because of harm. We found possibly little to no difference in dropout rates because of harm for nicotine replacement therapy (relative risk 1.26, 95% CI 0.50 to 3.14; 0.5 more per 100, 95% CI 0.9 fewer to 4.0 more), varenicline (relative risk 1.24, 0.57 to 2.71; 0.5 more per 100, 0.8 fewer to 3.3 more), and bupropion (relative risk 0.85, 0.25 to 2.91; 0.3 fewer per 100, 1.4 fewer to 3.6 more) compared with placebo or minimal care (low certainty evidence, online supplemental appendix 9).

Quality of life

Seven randomised controlled trials (n=1110) reported on quality of life (online supplemental appendix 10). Four studies compared a combined pharmacological and non-pharmacological intervention (behavioural support, motivational interviewing, or counselling) with placebo or minimal care, two studies compared two types of non-pharmacological interventions (specific v more general intervention), and one study compared nicotine replacement therapy with nicotine replacement therapy and a non-pharmacological intervention (motivational interviewing and support, online supplemental table S42). Relative to placebo or minimal care, low certainty evidence showed possibly little or no difference in quality of life for all treatment comparisons.

Weight gain

Ten randomised controlled trials with varying comparisons and measures reported body weight outcomes (online supplemental appendix 11). Three trials compared a combined pharmacological and non-pharmacological intervention (behavioural support with or without motivational support, online supplemental table S45) versus minimal care. Pooled results showed possibly little to no difference in body mass index at 12 months (0.42 body mass index units; 95% CI −0.78 to 1.53; low certainty evidence). For the remaining comparisons, the certainty of the evidence was very low.

Discussion

Principal findings

In our extensive network meta-analysis, we found high certainty evidence that compared with placebo or minimal care, varenicline increased both short term and long term smoking abstinence in people with severe mental illness seeking treatment for smoking cessation. We found probably little to no effect on depression and general psychopathology. For bupropion, moderate certainty evidence showed benefits for both short term and long term abstinence, with probably little to no effect on depression, and possibly improvement of general psychopathology. Nicotine replacement therapy was found to probably improve short term abstinence with little to no effect on long term abstinence and depression. Combined pharmacological and non-pharmacological treatments might also increase the likelihood of achieving long term smoking abstinence, although the effects on short term abstinence were very uncertain. The certainty of the evidence for serious adverse events was very low for all treatment comparisons. Effect estimates for other treatment comparisons relative to placebo or minimal care were mostly very low certainty. For quality of life and body weight, we found mostly little or no effects of treatments, but this finding was based on low or very low certainty evidence. Lastly, dropout from a trial because of harms was possibly no different for varenicline, bupropion, and nicotine replacement therapy compared with placebo or minimal care.

Comparison with other studies

Only a few previous systematic reviews have evaluated both pharmacological and non-pharmacological interventions in people with severe mental illness.^{17 31} These reviews, however, may not have prioritised patient relevant outcomes, performed a network meta-analysis, assessed the certainty of the evidence, or expressed results in absolute numbers. By applying these methods, our results are useful for discussing the benefits and harms of smoking cessation interventions with patients and for shared decision making.

Although current practice guidelines^{23 69} recommend combining pharmacological and non-pharmacological interventions, our network meta-analysis found mostly uncertainty about the degree of added benefit of combining non-pharmacological interventions with pharmacological interventions. Despite this result, we also found that the use of these interventions may (with low certainty evidence) be more effective than placebo or minimal care for long term smoking abstinence in patients with severe mental illness.

Rates of sustained smoking abstinence in people with severe mental illness are lower than in the general population,²⁹ although a recent Cochrane review⁷⁰ in healthy adults reported similar absolute long term abstinence rates. Direct comparison of these rates, however, may be challenging, because these rates were calculated in populations with different patient and treatment characteristics. This considerable variety in characteristics can have a profound effect on reported smoking cessation rates.⁷¹ Rates for abstinence in people with severe mental illness^{27 33 72} are worsened by higher severity of nicotine dependence, psychiatric symptom severity, and earlier age when starting smoking.⁷² Other barriers to abstinence are the idea of smoking as a coping mechanism and smoking to self‐medicate for symptoms of severe mental illness.⁷³ Despite these findings, smokers with severe mental illness should continue to have access to smoking cessation aids to decrease the risk of mortality and morbidity.^{18 74} Compared with the general population, specific smoking cessation treatment adaptations, perhaps more intense or longer treatments, may be needed to attain sustained abstinence in those with severe mental illness.³²

Strengths and limitations of this study

Key strengths of our meta-analysis were the comprehensive search strategies, incorporating several databases and broadly covering all available smoking cessation interventions, and the application of strict eligibility criteria, resulting in a homogeneous population and no serious concerns regarding intransitivity. Furthermore, by adhering to GRADE methods, we prioritised patient important outcomes.

Our analysis also had several limitations. Firstly, the included non-pharmacological interventions were diverse in nature and intensity (ranging from minimal care, psychoeducation, and brief advice to more complex and intense interventions), frequency, and duration. The interventions were delivered by a variety of individuals with different expertise levels, and were delivered variably in individual or group settings. The variability in interventions included in this treatment node decreased the certainty of the evidence in related treatment effects. Secondly, most studies were generally small and had a high risk of bias for at least one domain. Most analyses, however, were based on data from the robust EAGLES trials (although we recognise that a few trials dominating our analyses may also be perceived as a limitation). Thirdly, although e-cigarettes have become a popular cessation aid for smokers, we identified only one study on the effectiveness of e-cigarettes on smoking cessation in people with severe mental illness for inclusion in our analysis. Fourthly, whereas weight was identified as substantially important to patients, and treatments may exacerbate existing health problems with excess weight gain and affect adherence, few studies reported on the effect of smoking cessation on this outcome and the overall certainty of the evidence was low.

Also, whereas serious adverse events may preclude people with severe mental illness from starting or maintaining smoking cessation treatments, treatment effect estimates had very low certainty evidence for all interventions. Clinicians are often reluctant to recommend some treatments because of concerns about the lack of efficacy²⁷ or the potential adverse effects on psychiatric symptoms of these interventions.^{17 27} Despite the low certainty of evidence, several large observational studies did not find an increased risk of serious neuropsychiatric adverse events caused by smoking cessation for varenicline or bupropion compared with nicotine replacement therapy in those with and without psychiatric diseases.^{75 76} The European Medicines Agency and US Food and Drug Administration have removed their black box warnings from these drug labels.^{77 78} Finally, despite exclusively including trials evaluating smoking cessation interventions, only 45% of the included studies explicitly reported intention to quit smoking as an eligibility criterion. Although we assume that people entering a smoking cessation trial have an interest in either reducing or quitting smoking, inclusion of studies not explicitly reporting this outcome as an eligibility criterion may raise concerns about the generalisability of our findings. Inclusion of studies where reducing or quitting smoking was not the primary intention of participants may underestimate the effect of the study intervention, but this concern was not confirmed in another meta-analysis.⁷⁹ Moreover, in two included studies comparing varenicline with placebo^{67 68} with a focus on improving cognition, people may not have been aware that they received a smoking cessation intervention. Given concerns that including these studies could result in underestimation of the treatment effects of varenicline, we conducted a post hoc sensitivity analysis excluding the two studies, and found results consistent with the primary analysis. We believe inclusion of the two trials in our network meta-analysis broadened the included populations to people who were willing to try an intervention and therefore increased the generalisability of our findings.

In this specialty area, several questions warrant further research. Although the simultaneous use of multiple forms of nicotine replacement therapy is increasingly considered standard clinical care,⁸⁰ with some evidence that dual nicotine replacement therapy may be more effective than single form nicotine replacement therapy, and even as effective as varenicline,⁸¹,⁸⁶ this hypothesis could not be examined in our network meta-analysis because of lack of data. Also, knowledge of the factors predicting sustained abstinence in people with severe mental illness is limited.²⁷

Conclusions

In our systematic review and network meta-analysis, we found that varenicline, bupropion, and nicotine replacement therapy were more effective than placebo or minimal care in achieving short term or long term smoking abstinence in people with severe mental illness. Among the three pharmacotherapies, varenicline was most effective and had treatment effects supported by high certainty evidence. A combination of pharmacological and non-pharmacological interventions may also be more effective than placebo or minimal care. Effects on serious adverse events across treatments were very uncertain. Clinicians and people living with severe mental illness who smoke should consider starting smoking cessation treatment and engage in shared decision making, accounting for individual risk profiles, patient preferences, and the anticipated benefits and harms of candidate treatments.

Data availability statement

Data are available upon reasonable request.