Introduction
The search for environmental causes for psychosis (Dean & Murray, Reference Dean and Murray2005) in the past three decades has included factors that can be experienced after childhood, for example, migration and the use of substances (Morgan et al. Reference Morgan, Charalambides, Hutchinson and Murray2010). Investigations into the possible causal effects of cannabis have featured prominently in research into substances, with a meta-analysis estimating that cannabis users experienced nearly three times the odds of having psychosis compared with non-users [odds ratio (OR) 2.93, 95% confidence interval (CI) 2.36–3.64] (Semple et al. Reference Semple, Mcintosh and Lawrie2005). More recently, other drugs have been examined, most notably tobacco.
There is a strong positive association between smoking cigarettes and psychotic disorders (de Leon & Diaz, Reference De Leon and Diaz2005). The most recent meta-analysis of smoking as a risk factor for psychosis estimated the OR for daily smoking to be around 3, based on 11 case–control studies, and the relative risk, from five prospective studies, to be approximately 2 (Gurillo et al. Reference Gurillo, Jauhar, Murray and Maccabe2015). The positive association between tobacco smoking and psychotic illnesses has a number of candidate explanations. These include:
-
(a) Self-medication (Kumari & Postma, Reference Kumari and Postma2005), for example, of psychiatric symptoms (Smith et al. Reference Smith, Singh, Infante, Khandat and Kloos2002), cognitive deficits (George et al. Reference George, Vessicchio, Termine, Sahady, Head, Pepper, Kosten and Wexler2002; Sacco et al. Reference Sacco, Termine, Seyal, Dudas, Vessicchio, Krishnan-Sarin, Jatlow, Wexler and George2005; Barr et al. Reference Barr, Culhane, Jubelt, Mufti, Dyer, Weiss, Deckersbach, Kelly, Freudenreich, Goff and Evins2008) or adverse effects of psychiatric drugs (Goff et al. Reference Goff, Henderson and Amico1992),
-
(b) Shared genetic liability to both smoking and psychoses (Lyons et al. Reference Lyons, Bar, Kremen, Toomey, Eisen, Goldberg, Faraone and Tsuang2002; Smith et al. Reference Smith, Barch, Wolf, Mamah and Csernansky2008; Chen et al. Reference Chen, Bacanu, Yu, Zhao, Jia, Kendler, Kranzler, Gelernter, Farrer, Minica, Pool, Milaneschi, Boomsma, Penninx, Tyndale, Ware, Vink, Kaprio, Munafò and Chen2016),
-
(c) A causal effect of smoking on schizophrenia (Weiser et al. Reference Weiser, Reichenberg, Grotto, Yasvitzky, Rabinowitz, Lubin, Nahon, Knobler and Davidson2004; Kendler et al. Reference Kendler, Lönn, Sundquist and Sundquist2015),
-
(d) A reverse causal effect – mental health problems could result in people who smoke being less likely to quit, for example, because of more severe nicotine dependence or more limited access to smoking cessation treatment (Szatkowski & McNeill, Reference Szatkowski and Mcneill2014), and
-
(e) Confounding by other drug use – people who smoke are more likely to take other drugs, including cannabis and stimulants (Regier et al. Reference Regier, Farmer, Rae, Locke, Keith, Judd and Goodwin1990; Morral et al. Reference Morral, Mccaffrey and Paddock2002), which may be causally associated with psychosis (Semple et al. Reference Semple, Mcintosh and Lawrie2005; Large et al. Reference Large, Sharma, Compton, Slade and Nielssen2011).
In a recent prospective study, Kendler et al. found that smoking was associated with later schizophrenia in two Swedish cohorts, after accounting in the design for shared familial factors between people who developed schizophrenia and those who did not (Kendler et al. Reference Kendler, Lönn, Sundquist and Sundquist2015). Heavy smokers in discordant monozygotic twin pairs were around 1.7 times more likely to develop psychosis compared with the non-smoking twin, suggesting that genetic factors do not completely explain the relationship between smoking and later psychosis. Strengths of association were not affected by specifying different buffer periods between smoking assessment and first diagnosis, implying that the association did not arise as a result of people smoking as part of the psychosis prodrome.
It is increasingly argued that psychotic disorders represent the extreme end of a phenomenological continuum of psychotic experiences (PEs), which extend into the general, non-clinical population (Johns & Van Os, Reference Johns and Van Os2001; Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; Linscott & van Os, Reference Linscott and Van Os2013). However, although observational data from a number of sources have indicated high smoking prevalence in people with mental disorders [Royal College of Psychiatrists (RCPSYCH, 2013)], few studies have addressed the question of whether tobacco smoking is associated with PEs in the general population (Van Gastel et al. Reference Van Gastel, Maccabe, Schubart, Vreeker, Tempelaar, Kahn and Boks2013; Gage et al. Reference Gage, Hickman, Heron, Munafò, Lewis, Macleod and Zammit2014). Furthermore, the extent to which any association is explained by confounding cannabis or socioeconomic factors is unclear.
This study examined the association between cigarette smoking and PEs in a representative population-based sample of South London.
Our objectives were to:
-
(1) Estimate the association between smoking and PEs and between smoking and the number of PEs reported, taking into account possible confounding by cannabis, stimulant use and ethnicity, and
-
(2) Test for a linear trend in the odds of reporting PEs with quantity of cigarettes smoked per day.
Methods
Sample
The South East London Community Health study [SELCoH (Hatch et al. Reference Hatch, Frissa, Verdecchia, Stewart, Fear, Reichenberg, Morgan, Kankulu, Clark and Gazard2011)] is a representative household survey of South East London residents collected in 2008–2010. The analytic sample consisted of 1698 people, residing in 1075 households, collected through random sampling of a postcode address file, who were interviewed by lay researchers. Respondents were between 16 and 90 years of age. Of 2359 people eligible within participating households, 1698 (71.9%) participated.
Psychotic experiences
The rating scale used for the assessment of PEs was the Psychosis Screening Questionnaire [PSQ (Bebbington & Nayani, Reference Bebbington and Nayani1995)]. The PSQ is a self-report questionnaire designed to be administered by lay interviewers for use in large-scale epidemiological studies, for the purpose of screening respondents for possible psychotic disorder. It is a five-item questionnaire that assesses different positive psychotic symptom domains experienced in the previous year. These comprise: hypomania, strange experiences, persecutory experiences, auditory hallucinations, and thought interferences. Each domain contains an initial ‘probe’ item, which is followed by secondary questions. Because the present study was focused on non-affective psychosis, responses to the hypomania item were not examined. Endorsement of PEs was defined as positive response to items in the four remaining domains. This approach was consistent with a previous analysis of PEs originating from these data (Morgan et al. Reference Morgan, Reininghaus, Reichenberg, Frissa, Hotopf and Hatch2014). Information on the number of domains endorsed was also available. The PSQ displays good correspondence with psychosis items on the Schedules for Clinical Assessment in Neuropsychiatry (Bebbington & Nayani, Reference Bebbington and Nayani1995) and has seen frequent use in population studies (Johns et al. Reference Johns, Nazroo, Bebbington and Kuipers2002; Bebbington et al. Reference Bebbington, Bhugra, Brugha, Singleton, Farrell, Jenkins, Lewis and Meltzer2004; Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004).
Sociodemographic and clinical measures
Data on age, gender, employment status (employed, unemployed, student, other), ethnicity (White, Black Caribbean, Black African, Asian, and other), marital status (single, married/cohabiting, divorced/separated, and widowed), social class (measured by the National Statistics Socio-Economic Classification), a composite score of general cognitive ability (details available in Mollon et al. (Reference Mollon, David, Morgan, Frissa, Glahn, Pilecka, Hatch, Hotopf and Reichenberg2016), and highest educational attainment (no qualifications, General Certificate of Secondary Education, A level, and degree level or above) were available. The presence of symptoms of a common mental disorder in the previous 2 weeks was defined based on responses to the CISR [Clinical Interview Schedule, Revised (Lewis & Pelosi, Reference Lewis and Pelosi1990)], with a cut-off score of 12 (Lewis et al. Reference Lewis, Pelosi, Araya and Dunn1992).
Measurement of cigarette smoking:
Information on cigarette smoking analysed in this study was collected from SELCoH participants at four levels: the category of ‘never smoked’ was based on answering ‘no’ to the question: ‘Have you ever smoked a cigarette?’. Ex-smokers were defined as those answering ‘yes’ to the question: ‘Have you ever smoked a cigarette?’ And then answering ‘no’ to the question: ‘Do you smoke cigarettes at all nowadays?’. Sporadic smoking was based on answering ‘yes’ to the question: ‘Have you ever smoked a cigarette?’, then ‘yes’ to the question: ‘Do you smoke cigarettes at all nowadays?’, and then reporting a zero daily cigarette intake when asked: ‘About how many cigarettes a day do you usually smoke?’. Finally, daily smokers were defined by answering positively to both prior questions and providing an estimate of the number of daily cigarettes smoked. All participants defined as daily smokers were therefore current smokers.
Ascertainment of cannabis use
Participants were asked about cannabis use frequency and categorised into the following groups: never used, use less frequently than once a week, use more than once a week but less than daily, and use daily.
Evaluation of stimulant substance use
Participants reported use of amphetamines, ecstasy, cocaine, and crack use; all were combined into a single variable with three levels – never used, use but not in the previous year, and use in the previous year. All models which adjusted for substance misuse included this three-level variable.
Analysis
All analyses were carried out in STATA 14 (StataCorp, 2014) and took account of non-response weights and clustering of responses by household. Inverse probability weights (Pickles et al. Reference Pickles, Dunn and Vázquez-Barquero1995) were calculated from logistic regression models for non-response of an eligible individual within households. Predictor variables for non-response were selected for inclusion in the final weights model based on strength of statistical evidence (p values of <0.05) and whether the selected weighting scheme reproduced the means and prevalences of participants with complete data. The final prediction model contained effects of age and gender. Categorical descriptions of the sample by PEs were inspected. Univariate associations between PE status and cigarette smoking, stimulants, and sociodemographic variables (age, gender, and ethnicity) were evaluated and presented. Multivariate models were used to assess and account for confounding. Age and gender were included in all models. Covariates whose inclusion in the model did not deviate the association between PEs and daily cigarette smoking by more than 10% of the unadjusted OR were discarded (Greenland et al. Reference Greenland, Daniel and Pearce2016). This left age, gender, and ethnic group as covariates in modelling, alongside stimulant and cannabis use as potential confounders of primary interest. In particular, neither the inclusion of general cognitive ability, marital status, employment status, social class, nor educational attainment altered estimates sufficient for their inclusion. Having identified evidence of strong negative confounding by ethnicity, we explored the association between ethnic group and smoking, presented in online Supplementary Table S7. Descriptive data on the overlap between cigarette smoking and use of cannabis and stimulants are also presented as supplementary material. Modification of the association between current smoking and reporting any PEs by age, cannabis use, and common mental disorder was tested by fitting multiplicative interaction terms for smoking status by age, cannabis use, and common mental disorder in fully adjusted models. Ordinal logistic regressions were used to assess the association between smoking status and number of PEs (range from 0 to 4). Finally, we examined the possibility of a dose–response relationship by assessing linear trends in the association between the number of cigarettes smoked and (a) the odds of reporting any PEs (from logistic regression models), and (b) the odds of reporting one further PE (from ordinal logistic regression models).
Results
After excluding participants with missing data on the modelled variables, 1680 survey participants remained for analysis. Sociodemographic and substance use associations with PEs are shown in Table 1. PEs were more frequently reported by younger participants, and those with Black Caribbean and Black African ethnic status. Cannabis, ecstasy, cocaine, and other stimulants were associated with PEs. The estimate for crack cocaine, while indicating a possible strong association, was imprecise and not statistically significant, as its use was seldom reported. Cannabis use frequency was strongly associated with use of stimulant drugs (see online Supplementary Table S6). There was an association between PEs and daily, but not sporadic or past, cigarette smoking. Multivariate models for the odds of reporting any PEs are show in Table 2: when sociodemographic variables were included in the model, the estimate increased, indicating positive confounding by age, gender, and ethnicity. Further adjustment for cannabis frequency attenuated the OR for daily smoking on PEs. Finally, adjustment for stimulant use (recent and in the lifetime) modestly reduced the association. No statistical evidence was found for differences in the association between current smoking and the odds of reporting PE within different age groups, or at different levels of cannabis use.
Table 1. Counts and survey-weighted univariate associations between PEs and each variable used in this study, based on the analytic sample of 1680
PE, psychotic experience; CI, confidence interval.
Table 2. OR estimates for smoking pattern on PEs from survey weighted logistic regression
All models are based on 1680 participants. Age was adjusted for as a continuous variable.
OR, odds ratio; CI, confidence interval.
We found strong statistical evidence for a dose–response relationship between the number of cigarettes smoked and the odds of reporting any PEs, and the reporting of a greater number of PEs, in adjusted models. On average, an increase in daily cigarette consumption from 0 to 1–9, from 1–9 to 10–19, or 10–19 to 20 or more was accompanied by a 1.04 increase in the overall relative odds of reporting any PEs (95% CI 1.02–1.07; Table 3) and a 1.58-fold increase in the relative odds of reporting one further PE (95% CI 1.32–1.90; Table 3).
Table 3. OR estimates for the association between (a) any PEs (upper panel) and (b) the number of PEs (lower panel, reflecting the increase in relative odds for one more PE) with quantity of cigarettes smoked per day
Based on overall analytic sample of 1680. Test statistics (T) are from survey-weighted logistic regression models.
PE, psychotic experience; OR, odds ratio; CI, confidence interval.
Daily smoking was associated not only with an increased odds of reporting PEs, but also with increasing number of PEs, although this estimate lost precision after adjusting for stimulant use (fully adjusted OR 1.55, 95% CI 0.98–2.47, Table 4). The most common PE was strange experiences (6.05%), followed by auditory hallucinations (3.87%), then persecutory experiences (3.27%), with thought interferences the least common PE (1.32%). Individual types of PE were associated with daily smoking, with precise estimates for strange experiences, but not for the other symptoms. In fully adjusted models, associations remained for each symptom, but lost precision. On account of the association between PEs and other symptoms of mental disorder, we estimated associations of PEs with smoking pattern by common mental disorder, as shown in online Supplementary Table S5. No statistical evidence was found for variation in effect estimates by common mental disorder, although this test lacked power. Because of the association between ethnicity and PEs, and the attenuation in estimates observed when it was included in regression models, we described the association between ethnicity and smoking, as reported in online Supplementary Table S7: all non-White ethnic groups had lower proportions of reported daily, ex-, and sporadic smoking compared with the White reference group (p < 0.001).
Table 4. Models comparing daily smokers to never-smokers for an increase in number of PEs, and for separate types of psychotic experience
All models are based on 1680 participants. Estimates for ex-smokers and sporadic smokers are not presented.
PE, psychotic experience; OR, odds ratio; CI, confidence interval.
Discussion
Summary of findings
We found evidence of a cross-sectional association between daily cigarette smoking, but not ex-smoking, and PEs in a sample of household residents in South East London. The association was not explained completely by cannabis use frequency, or by use of stimulant drugs, or by ethnicity (ethnicity was strongly associated with daily smoking, see online Supplementary Table S7). There was an increasing strength of association observed by number of cigarettes smoked, and increased cigarette consumption predicted a greater number of PEs. We did not find statistical evidence for interaction of smoking with age, cannabis use, or with symptoms of common mental disorder.
Previous literature
Smoking is a crucial, potent, and modifiable cause of morbidity and mortality in the UK (Matcham et al. Reference Matcham, Carroll, Chung, Crawford, Galloway, Hames, Jackson, Jacobson, Manawadu, Mccracken, Moxham, Rayner, Robson, Simpson, Wilson and Hotopf2017). Although the number of people who smoke in the UK is falling (Action on Smoking and Health, 2015), this decline is not reflected in people with mental illness (McManus et al. Reference McManus, Meltzer and Campion2010); and data from the Health Survey for England suggest that smoking may be declining more slowly in people with mental health problems compared with those without (Szatkowski & McNeill, Reference Szatkowski and Mcneill2014). Therefore, identifying the mechanisms by which smoking and mental illness are associated could be beneficial for public health.
Our findings that PEs and daily smoking are associated, are consistent with a small body of literature suggesting that smoking is more common in people with sub-clinical PEs than the rest of the general population. Firstly, van Gastel et al. (Reference Van Gastel, Maccabe, Schubart, Vreeker, Tempelaar, Kahn and Boks2013) reported an analysis of an internet survey, finding that the cross-sectional association between scores on the community assessment of PEs and daily smoking for the past month remained apparent despite accounting for cannabis use and for a group of other confounders. Secondly, smokers were 1.3 times more likely to report PEs in the World Health Surveys compared with non-smokers, after adjustments, suggesting the association is consistent across national settings (Koyanagi et al. Reference Koyanagi, Stickley and Haro2016). Thirdly, Wiles et al. (Reference Wiles, Zammit, Bebbington, Singleton, Meltzer and Lewis2006) reported association between smoking and PEs in the 2007 UK Adult Psychiatric Morbidity Surveys, but found that the crude association was strongly attenuated by adjustment for cannabis, general cognitive ability and marital status. Fourthly, Saha et al. (Reference Saha, Scott, Varghese, Degenhardt, Slade and Mcgrath2011) found that daily smoking was associated with reporting delusion-like experiences in an Australian household survey (2011), after adjusting for a broad range of confounders. Fifth, in an analysis of prospective data from the Avon Longitudinal Study of Parents and Children, Gage et al. (Reference Gage, Hickman, Heron, Munafò, Lewis, Macleod and Zammit2014) reported that smoking at age 16 was predictive of PEs at 18, after accounting for cannabis use frequency and a range of early and mid-life confounders. Overall, few previous studies have assessed dose–response relationship with number of cigarettes smoked or by number of PEs reported, and few studies have adjusted for cannabis use in detail, for example, by including cannabis use frequency in statistical models.
How our results fit in
Our results, from a highly socioeconomically and ethnically diverse sample, are consistent with the previous literature suggesting the cross-sectional association between cigarette smoking and PEs is not fully explained by cannabis use, the use of stimulant drugs, or confounding by demographic or socioeconomic status, particularly by ethnic group. Furthermore, we present evidence that the relationship between odds of reporting any PEs, and a greater number of PEs is related to the number of cigarettes smoked per day. Finally, we extend previous literature by presenting evidence that daily smoking predicts the reporting of more PEs on a continuous scale. We also found no evidence of association between PEs and being an ex-smoker, implying that our analysis did not suffer from confounding by non-time-varying characteristics, such as unadjusted sociodemographic factors. The finding of no association between PEs and ex-smoking is consistent with other literature suggesting that mental health improves following smoking cessation (Taylor et al. Reference Taylor, Mcneill, Girling, Farley, Lindson-Hawley and Aveyard2014) raising the possibility that the increase in PEs associated with smoking may be reversible. Our results are also consistent with smoking being a more persistent behaviour in people with PEs compared with those without, and fit with some evidence that people with psychosis who smoke tend to have more severe positive symptoms and more limited social adjustment (Barnes et al. Reference Barnes, Lawford, Burton, Heslop, Noble, Hausdorf and Young2006; Krishnadas et al. Reference Krishnadas, Jauhar, Telfer, Shivashankar and Mccreadie2012).
Strengths and limitations
This was a cross-sectional study, and these associations could be explained by smoking occurring after PEs. Measurement of smoking, PEs, and cannabis use were by self-report in the same survey, and some way of confirming this information in independent data would have improved the validity of the study. Strong collinearity between exposure and a confounder limits the ability of regression methods to correctly adjust for confounding – in this study, the close overlap between cannabis use and cigarette smoking (Amos et al. Reference Amos, Wiltshire, Bostock, Haw and Mcneill2004) might not have fully allowed for the identification of separate effects for these two factors (Gage et al. Reference Gage, Hickman, Heron, Munafò, Lewis, Macleod and Zammit2014; Greenland et al. Reference Greenland, Daniel and Pearce2016). There were no data on the persistence or timeframe of PEs, further limiting inference. Although we were able to adjust estimates for stimulant use, this was in the form of an aggregated variable across four different stimulants, leaving open the possibility of residual confounding by the use of individual stimulants. However, despite these limitations to the data, the study did allow the assessment of this association in an urban, diverse population with relatively high levels of cannabis and stimulant use, in contrast to previous studies. The generalisability of our results to the rest of the UK population could be limited. However, a previous study based on these data suggested similarity in the distributions of age, gender, economic activity, and ethnicity to the overall English population recorded in the UK Census (Hatch et al. Reference Hatch, Frissa, Verdecchia, Stewart, Fear, Reichenberg, Morgan, Kankulu, Clark and Gazard2011). We found evidence that ethnic group was strongly related to the probability of daily smoking, in accordance with other studies (Best et al. Reference Best, Rawaf, Rowley, Floyd, Manning and Strang2001; Wanigaratne et al. Reference Wanigaratne, Dar, Abdulrahim and Strang2003; McCambridge & Strang, Reference Mccambridge and Strang2005), and adjusted for it as a possible confounder (see online Supplementary Table S7).
Conclusions
The association between PEs and smoking is apparent in a highly diverse population with relatively prevalent use of cannabis and stimulant drugs. The linear relationship between cigarette consumption and odds of reporting PEs requires urgent explanation in longitudinal studies and diverse populations.
Supplementary Material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291717001556.
Acknowledgements
VB is supported by a Wellcome Clinical Research Training Fellowship (101681/Z/13/Z). This paper represents independent research funded by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. The authors acknowledge the assistance of Josephine Mollon and Shirlee MacCrimmon in the preparation of this paper.
