Prevalence of tobacco smoking in people at clinical high-risk for psychosis: Systematic review and meta-analysis

Several hypotheses have been proposed to explain why individuals with psychosis consume more tobacco compared with the general population, but the reasons remain unclear. The phases predating the onset of psychosis could provide an interesting framework to clarify this association. The aim of this systematic review and meta-analysis is to provide an updated and comprehensive synthesis of the association between tobacco smoking and Clinical High Risk for Psychosis (CHR – P) status. We performed a multistep systematic PRISMA/MOOSE-compliant electronic search for articles published from inception until October 1st

Several hypotheses have been proposed to explain why individuals with psychosis consume more tobacco compared with the general population, but the reasons remain unclear. The phases predating the onset of psychosis could provide an interesting framework to clarify this association. The aim of this systematic review and meta-analysis is to provide an updated and comprehensive synthesis of the association between tobacco smoking and Clinical High Risk for Psychosis (CHR -P) status.
We performed a multistep systematic PRISMA/MOOSE-compliant electronic search for articles published from inception until October 1st, 2021. Web of Science was searched, complemented by a manual search of original articles reporting the outcome of tobacco consumption (defined as the number of individuals which were smoking tobacco at baseline) in a group of CHR-P patients versus healthy controls (HC). We employed quality assessment of the included studies with Newcastle Ottawa Scale (NOS). The effect size for the primary outcome was the odds ratio (OR) of smoking tobacco in CHR-P samples vs HC. We performed a random-effects model meta-analysis, assessment of heterogeneity with I 2 index, sensitivity analyses excluding one study at a time for primary outcome, meta-regressions with four independent moderators (mean age, female ratio, sample size, NOS) and assessment of publication bias with funnel plot and Egger's test.
We included 21 independent articles, totalling 2018 CHR-P individuals (mean age of 21.35 ± 2.91 years and average female ratio of 41 ± 7 %) and 1160 HC (mean age of 22.42 ± 3.70 years and average female ratio of 45 ± 11 %). The NOS score was 6.52 ± 1.25 (range from 0 to 9). The OR of smoking status was 2.22 (95%CI 1.74-2.84, p < 0.01). Heterogeneity (I 2 ) was 24.09 (p = 0.16). Sensitivity analyses, removing one study at a time, revealed the robustness of our main finding. Meta-regressions did not reveal any significant association between the moderators and the main outcome. Visual inspection of the funnel plot and Egger's test did not reveal evident publication bias.
Our main finding of an increased OR of smokers in the CHR-P individuals compared to healthy controls corroborates the accumulation of unhealthy lifestyles in this vulnerable group. This does not demonstrate any causal association between tobacco smoking and incidence of psychosis, which should be investigated in future prospective cohorts. In conclusion, the window of opportunity represented by CHR-P status should involve more efficient physical health screening and better investigating the aetiological impact of smoking in the development of psychosis.

Background
Prevalence of tobacco smoking in individuals with mental disorders is higher than healthy controls (36.1 % compared with 21.4 % (Gfroerer et al., 2013). The ratio rises significantly in individuals with schizophrenia, where smokers are four to five times higher than the healthy population and over 60 % of the patients are smokers (Hartz et al., 2014). Indeed, the life expectancy in patients with schizophrenia is substantially reduced because of natural causes such as cardiovascular diseases -partially resulting from unfavourable lifestyle -which probably account for more than two-thirds of deaths among this population (Correll et al., 2017;Hjorthøj et al., 2017;Walker et al., 2015).
Several hypotheses (which are not mutually exclusive) have been proposed to explain this association (Gogos et al., 2019), but the reasons why individuals with psychosis consume more tobacco are still unclear (de Leon and Diaz, 2005;Gurillo et al., 2015).
1) The self-medication hypothesis suggests that patients might smoke tobacco due to its direct effect on the affective, negative and cognitive symptoms of the disease (Parikh et al., 2016) or to counteract antipsychotic side effects (Desai et al., 2001).
2) Nicotine consumption may relieve anxiety (Anderson and Brunzell, 2015) and/or other symptoms of mental disorders in comorbidity.
3) An alternative hypothesis of a shared genetic liability (biological pleiotropy) between psychosis and nicotine dependence, supported by findings which underline shared genetic risk factors in calcium signaling, long-term potentiation and neuroactive ligand-receptor interaction pathways (Chen et al., 2016). 4) Tobacco might be a causal risk factor for psychosis (Gage et al., 2017), supported by the interference between nicotine and the cholinergic system (Scarr et al., 2013). In line with this evidence, higher levels of daily tobacco use are associated with higher risk of developing psychosis (Kendler et al., 2015), and nicotine dependence is associated with poorer outcomes and more severe symptoms in schizophrenia (Parikh et al., 2016).
The relationship between tobacco and psychosis is complicated by the high correlation between cannabis and tobacco consumption (MacCabe, 2018), correlation with other substance use (Hindocha et al., 2015), and combination of tobacco and cannabis in joints by many cannabis users (Gage et al., 2014).
In 77 % of cases, smoking becomes a habit before the onset of schizophrenia, defined as the first contact with psychiatric services (de Leon and Diaz, 2005), with an average anticipation of 11 years (Kelly and McCreadie, 1999) and could therefore coincide with the period in which the first symptoms of psychosis appear (Beratis et al., 2001;Mark Weiser et al., 2004). Thus, the CHR-P state provides an interesting but under-investigated framework to examine this association (Gogos et al., 2019) constituting a phase in which smoking habit and psychotic symptoms might change over time and influence each other. Also, this phase could be seen as a window of opportunity to investigate specific reasons for initiating tobacco use and provide effective smoking cessation support (van der Heijden et al., 2021a).
A meta-analysis published in 2016 (Carney et al., 2016) pooled data from 11 studies reporting odds ratios (OR) about tobacco smoking in young people at Clinical High Risk for Psychosis (CHR-P), showed a higher rate of tobacco smoking compared to healthy controls (OR = 2.3, 95%CI = 1.48-3.48) in line with previous research from our research team (Catalan et al., 2021a;Fusar-Poli et al., 2020a;Salazar de Pablo et al., 2021), reporting similar results (OR = 3.043, 95%CI 1.204-7.692) (Fusar-Poli et al., 2017). Since these earlier reports, numerous other primary studies have been published, requiring an update of the evidence. The aim of this systematic review and meta-analysis is to provide an updated and comprehensive synthesis of the association between tobacco smoking and CHR-P status, with the hypothesis that there will be an increased OR of smokers in the CHR-P individuals compared to healthy controls.

Methods
This study was conducted accordingly to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) (Moher et al., 2009) and the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) (Stroup et al., 2000)(eTables 1, 2). The study protocol is registered on https://osf.io/uvf2q.

Search and selection strategies
We performed a systematic PRISMA-compliant electronic search for articles published from inception until October 1st, 2021. We searched Web of Knowledge (which includes Web of Science Core Collection, BIOSIS Citation Index, KCI -Korean Journal Database, MEDLINE, Russian Science Citation Index and SciELO Citation Index) using the following search string: "psychosis risk AND smoking". Our lab has previously validated the keyword "psychosis risk" as the most comprehensive for evidence synthesis studies in CHR-P populations (Salazar de Pablo et al., 2021). References of systematic reviews and meta-analyses identified during this phase were manually searched. Articles were first screened as abstracts, then the remaining articles were assessed against the inclusion and exclusion criteria on a full-text basis and decisions were taken regarding their inclusion in the meta-analysis.
The inclusion criteria were: a) original studies, published in English; b) conducted in individuals meeting CHR-P criteria according to Structured Interview for Psychosis-Risk Syndromes (SIPS any version) (McGlashan et al., 2010), Comprehensive Assessment of At-Risk Mental States (CAARMS any version) (Yung AR, 2006), ERIraos (Rausch et al., 2013), Bonn Scale for the Assessment of Basic Symptoms (BSABS) (Gross, 1987), Schizophrenia Proneness Instrument, Adult Version (SPI-A) and Child and Youth (SPI-CY) version (Schultze-Lutter et al., 2007); Basel Screening Instrument for Psychosis (BSIP) (Riecher-Rössler et al., 2008); c) retrospective, cross-sectional and longitudinal studies investigating CHR-P subjects compared with healthy controls (HC). Intervention studies were also included when tobacco smoking at baseline was reported; d) reporting the meta-analytic outcome of tobacco consumption (defined as the number of individuals which were smoking tobacco at baseline). CHR-P status typically requires that patients included were 15-35 years old (Fusar-Poli et al., 2013).
The exclusion criteria were: a) reviews, conference proceedings, study protocols, case series or reports; b) unpublished data; c) studies evaluating patients with risk of psychosis, but using criteria different from those established above (e.g. genetic risk alone), or studies with samples of individuals with other persisting psychotic disorders (e.g. schizophrenia spectrum) diagnosed with DSM (any version) or ICD (any version); d) studies which not include a HC control group; e) studies not reporting data about tobacco consumption in CHR-P and HC; f) overlapping data sets.
In case of studies with overlapping datasets, we selected the largest and most recent data set. Disagreements in selection criteria were resolved through discussion and consensus with a senior researcher.

Outcome measure and data extraction
The primary outcome was defined as the odds ratio (OR) between smokers in CHR-P subjects and HC, meta-regressions represented instead the secondary outcomes of the study. Two independent researchers extracted data. Any discrepancies arising in extraction criteria were resolved through discussion with a senior researcher. The variables extracted and recorded in the main database were: author and year of the study, assessment tool, sample size of CHR-P groups, mean age of CHR-P groups, female ratio of CHR-P groups, number of CHR-P smokers, number of CHR-P non-smokers, sample size of HC groups, mean age of HC groups, female ratio of HC groups, number of smokers in HC groups, number of non-smokers in HC groups, continent.

Quality assessment
The quality of the studies included in the meta-analysis was evaluated using the Newcastle-Ottawa Scale (NOS) for case-control studies, which has been repeatedly used (Catalan et al., 2021b) to assess studies in this field. Studies were awarded a minimum of zero and a maximum of nine points on items related to the selection and definition of CHR-P patients and controls, representativeness, comparability, exposure.

Strategy for data synthesis
The effect size for the primary outcome was defined as the ratio between the OR of smoking tobacco in CHR-P samples vs HC. Values >1 indicated that tobacco smoking was more prevalent in the CHR-P sample, values lower than 1 indicated that it was more prevalent in the HC sample. Random effects models were selected to account for heterogeneity between studies. Q statistic was used to assess heterogeneity among study point estimates, while the proportion of total variability in the ORs was evaluated with the I 2 index (Lipsey and Wilson, 2001) which is not influenced by the number of studies included. We performed sensitivity analyses excluding one study at a time for the main outcome, to confirm the robustness of the findings. We performed metaregression analyses for four independent moderators: mean age, female ratio, sample size, NOS score. Meta-regressions were completed when at least 10 studies were available. We assessed the risk of publication bias by visual inspection of funnel plots and Egger's regression intercept test. The significance level was set at 0.05 (two-tailed). Meta-analysis was performed using Comprehensive Meta-Analysis Software, Version 3.

Database
Our systematic literature search identified 1798 articles through the electronic search (see eResults) and 14 additional articles were added through the manual search of relevant reviews. After the two-phase screening, we included 21 independent articles (PRISMA flow-chart, Fig. 1), each including one independent sample. The total database comprised 2018 CHR-P patients with a mean age of 21.35 years (range from 15.70 to 26.78) and an average female ratio of 41 % (range from 28 % to 57 %) and 1160 HC with a mean age of 22.41 years (range from 15.50 to 30.50) and an average female ratio of 45 % (range from 21 % to 61 %). Studies included an average of 96 CHR-P patients (range from 12 to 678) and 55 HC (range from 11 to 261). 11 studies included individuals recruited in Europe, 7 in North America, 3 in Australia, 2 in Asia and 2 in South America (3 studies recruited individuals in more than one continent). The mean quality of the studies included, evaluated with the Newcastle Ottawa Scale for case-control studies, was 6.52 ± 1.25 (ranging from 4 to 8). The characteristics of the studies included in the meta-analysis are illustrated in eTable 3.

Meta analysis of prevalence of smoking status in CHR-P individuals vs HC
The meta-analytical OR of smoking status was 2.22 (95%CI 1.74-2.84, p < 0.01) (Fig. 2) in CHR-P vs HC, indicating that CHR-P individuals were more likely to use tobacco compared to healthy controls. Although we were expecting high heterogeneity, I 2 resulted 24.09 with p = 0.16, therefore non-significant. We therefore appended a supplementary analysis employing fixed effects meta-analyses. The results were as follows (eFig. 2): meta-analytical OR = 2.20 (95%CI 1.81-2.68, p < 0.01); I 2 = 24.09 with p = 0.16. Excluding one study at a time for primary outcome confirmed the robustness of the finding (sensitivity plot is presented in eFig. 1). We performed all the planned meta-regressions (mean age, female ratio, sample size, Newcastle Ottawa Scale score), but none of them did reveal any significant association (eTable 4). Visual inspection of the funnel plot (Fig. 3) did not reveal evident publication bias and Egger's test for funnel plot asymmetry was not significant (p = 0.83).

Discussion
To our knowledge, this is the most up-to-date and comprehensive review and meta-analysis of the existing literature about the association between tobacco smoking and CHR -P. Compared to the previous metaanalysis on this topic, the current review included 10 new studies, relating to additional 1080 CHR-P patients and 427 controls. This represents a 115 % increase of the previous dataset. Furthermore, the current dataset includes additional continents (e.g., South America) and we performed new meta-regressions, sensitivity analyses and quality Fig. 1. Prisma 2020 flow diagram. A. De Micheli et al. assessments that were not included in the previous publication.
In line with our research hypothesis, the current paper demonstrates that CHR-P individuals are more likely to smoke tobacco than HC, without being able to assess the potential causative role of tobacco in psychosis. The OR of 2.22 (95%CI 1.74-2.84 p < 0.01) puts the CHR-P clinical cohort in an intermediate position between the general adult population that has described a decreasing trend in smoking rates across the last decades (14.4 % at the NHS Statistics of Smoking England 2019) and individuals with schizophrenia, with a meta-analytical prevalence of smokers of 65 % (Fornaro et al., 2021). The sensitivity analyses corroborate the robustness of our finding, that was based on a fair average quality of the included studies and low level of heterogeneity. Overall, this finding confirms data from the literature that presents high proportions of smokers (43 %, unpublished data) in CHR-P subjects in  specialized clinical services, such as the OASIS (Outreach and Support in South London) (Fusar-Poli et al., 2020b) and in multicentric studies (van der Heijden et al., 2021b;Ward et al., 2019).
Meta-regressions revealed that the association between the smoking and independent moderators (mean age, female ratio, sample size, NOS score) was not significant. However, we acknowledge the small number of studies included in this database and, thus, the likelihood that our analyses are under-powered. Given these limitations, we cannot determine whether age, sex, sample size, and study quality might constitute possible moderating factors. Due to the limited availability of data, we could not investigate further relevant moderators, such as substance use or CHR-P subgroups.
One of the most relevant limitations related to the granularity of the data was the variety of manners the smoking status was operationalized in the included studies.
Indeed, smoking status is largely collected as a self-reported measure and rarely by standardized measures (like the Fagerström test). Smoking status was expressed with the number of cigarettes per day or, more frequently, by dividing categorically the population into smokers and non-smokers subgroups. Thus, it was not possible to formulate any precise DSM/ICD diagnoses, as Tobacco Use Disorder (TUD), or to include a minimum threshold of cigarettes per day in the inclusion criteria.
As mentioned in the background, there are different hypotheses that may explain the impact of cigarette smoking in CHR-P individuals: 1) The self-medication hypothesis according to which individuals would consume tobacco to alleviate their symptoms (Parikh et al., 2016) and medications' side effects. In CHR-P populations, antipsychotics are currently not recommended by NICE guidelines (Taylor and Perera, 2015) or just considered in exceptional circumstances by EPA guidelines (Schmidt et al., 2015), so these young subjects may use tobacco to mitigate against their presenting symptoms. Unfortunately, quantitative data about daily tobacco consumption and symptoms intensity were available in a limited number of included studies, so we could not provide evidence on this potential association. Nevertheless, a recent finding of a dose-response relationship between the number of daily cigarettes and a higher level of general CHR-P symptoms severity (van der Heijden et al., 2021a) might be interpreted in line with selfmedication.
2) As the CHR-P state is frequently associated with other mental disorders in comorbidity, like anxiety or depression (Fusar-Poli et al., 2014;Rutigliano et al., 2016), smoking could be secondary to those disorders (Fluharty et al., 2017) rather than to attenuated psychotic symptoms. In a recent study, Ward et al. (2019) failed to demonstrate a significant association between smoking and CHR-P status when depression and anxiety were entered in the regression models. As nicotine has a well-known anxiolytic effect, possibly through β2* nicotinic acetylcholinergic receptors (Anderson and Brunzell, 2015), it can explain both the association between tobacco and non-psychotic disorders such as anxiety as well as the self-medication hypothesis (see eDiscussion). The lack of reported comorbid diagnosis did not allow us to investigate their potential role of confounders from a meta-analytical perspective.
3) Different studies found a shared genetic liability between tobacco use and schizophrenia (Gage et al., 2017;Wootton et al., 2020). The onset of smoking habits usually precedes the first episode of psychosis (Myles et al., 2012), suggesting a shared vulnerability between tobacco smoking and psychosis (Liu et al., 2019). Despite genetic data on this correlation are currently not available in the literature for CHR -P, the young subjects included in our sample already present a significative association with tobacco, even before the onset of clear-cut psychotic symptoms.
4) The impact of tobacco as direct risk factor for the onset of psychosis (Gage et al., 2017), also summarised by Gurillo and colleagues in their meta-analysis (Gurillo et al., 2015) including 5 longitudinal studies and showing a higher incidence of new psychotic disorders in daily smokers, compared with non-smokers (RR: 2.18) and reporting that daily smokers developed a psychotic illness at an earlier age than nonsmokers. Cigarette smoking as a risk factor was also investigated in a Finnish prospective cohort (Mustonen et al., 2018) that found an adjusted risk of psychosis (HR = 2.87) for heavy-smoker adolescents.
There is already meta-analytical evidence that tobacco smoking is associated with CHR-P status (OR = 3.03) (Fusar-Poli et al., 2017), with the first episode of psychosis (OR = 3.22) (Gurillo et al., 2015), and with psychotic-like experiences (OR = 1.47) (Bhavsar et al., 2018). The relationship between psychosis and tobacco is less recognised than other risk factors like cannabis (MacCabe, 2018), and the gap is even more evident when we talk about the CHR-P phase. Due to the limited proportion of prospective studies investigating tobacco smoking in CHR-P samples (<30 % of the studies included in this meta-analysis), we could not investigate the impact of tobacco on the incidence (i.e., new cases of psychosis in this patient group). Most included studies evaluate the association between smoking and the CHR-P phase cross-sectionally through psychometric interviews or self-reported questionnaires (Mustonen et al., 2018).
There are even fewer studies with a longitudinal design that assess the role of cigarette smoking in transition to psychosis from a CHR-P state. In the EU-GEI cohort, van der Heijden et al. (2021a) found a dose-response relationship between the number of daily cigarettes consumed and a higher level of general symptoms severity but no association was noticed between smoking status and transition risks. The NAPLS-2 study confirmed similar results, failing to find an association between smoking and transition to psychosis (Buchy et al., 2015;Ward et al., 2019). An older study (Kristensen and Cadenhead, 2007) found the nicotine to be significantly associated with later conversion to psychosis, but some individuals transitioning were smoking cannabis along with tobacco.

Limitations
The main limitation is that our study focuses on prevalence and not on prospective associations. Therefore, we cannot not rule out reverse causation and could not exclude that CHR-P psychopathology may lead to an increased use of tobacco. To dissect the direction of this association, future and well-powered longitudinal cohort studies should measure exposure to tobacco at baseline and the incidence of psychosis prospectively. These longitudinal studies are rare in the current literature and therefore a prospective meta-analysis is currently not feasible. As already mentioned in previous paragraphs, several limitations were due to the granularity of the data, including the lack of standardized measures as the Fagerström test or the scarce report of comorbid mental disorders. Of note, only one study out of 21 mentioned alternative ways of smoking tobacco in the discussion and no one reported this data for the participants. Regarding the matching between CHR-P subjects and HC, most studies included in the meta-analysis used age and gender as parameters, but these criteria were not entirely homogeneous.

Conclusions
Our main finding of a higher proportion of smokers in the CHR-P population compared to HC supports existing evidence that unhealthy lifestyles are prevalent in this clinical cohort of vulnerable subjects. The window of opportunity represented by CHR-P status should involve more efficient means to improve the long-term physical health outcomes of these patients and to better investigate the aetiological role of smoking in the development of psychosis.
Considering the alarming proportion of smokers in CHR-P population, we call for a more systematic assessment of their smoking status, implementation of ad-hoc preventative strategies to promote smoking cessation, currently still limited (van der Heijden et al., 2021b) (Ward et al., 2019) and a higher percentage of primary research with longitudinal designs to better investigate the interface between tobacco and psychosis.

Vailability of data and materials
The authors give no permission to share raw data.

Declaration of competing interest
PFP received honoraria or grant fees from Lundbeck, Angelini and Menarini in the past 36 months outside the current work.