There were 47 studies meeting inclusion criteria. The studies were broken down by type and direction of analysis (prospective, quasiprospective and cross-sectional) and are represented in Additional files 1, 2, 3, 4, 5.
A total of 13 studies were identified that utilized population-based samples to assess a prospective relationship between ADs, smoking behavior, and ND. These studies comprised random population samples drawn from the United States, New Zealand, Germany and The Netherlands, with some samples used in multiple studies. Discussion of the included studies has been grouped below based on the sample used.
Influence of anxiety disorders on risk of later smoking or nicotine dependence
(A summary of this section is as follows (see also Additional file 1, Table S1): the best available evidence is equivocal, but suggests that certain baseline ADs are risk factors for onset of smoking and nicotine dependence, although results vary across studies and across different disorders. The best available studies failed to control for comorbid substance use disorders.)
The first published prospective data were drawn from the Oregon Adolescent Project Depression project, which randomly recruited adolescents (aged 14 years to 18 years) from high schools in Western Oregon, USA in 1987 to 1989 and followed them at two time points (1 year later, and on their 24th birthday). Assessment of ADs was undertaken utilizing the Schedule for Affective Disorders and Schizophrenia for School-Age Children (K-SADS) for DSM-III revision (DSM-III-R) , and smokers were defined as those smoking ≥ 3 times per week. Brown et al.  demonstrated no difference in the odds for incident smoking at 1-year follow-up in those with versus without any ADs at baseline, both unadjusted and after controlling for a variety of demographic and other risk factors. This finding of no association was replicated in the 24-year-old follow-up  although only PD, and not grouped ADs, was used as the exposure variable.
Breslau et al.  utilized a random population sample of 1,007 young adults (21 years to 30 years), drawn from the Detroit Epidemiologic Study, to examine PD as a predictor of cigarette smoking. Baseline assessment occurred in 1989 and three follow-ups (1990, 1992 and 1994) were conducted. Diagnosis of PD was made through use of the National Institute of Mental Health Diagnostic Interview Schedule  for DSM-III-R criteria, and 'daily smoking' was defined as smoking daily for ≥ 1 month. Risk of smoking onset was increased in those with PD at baseline (HR = 2.20 (95% CI 1.10 to 4.42)), but this significance was lost when controlled for the presence of major depressive disorder (MDD). Breslau et al.  extended their investigation, incorporating two further follow-ups (1999 and 2001) to assess the interaction between PTSD and ND over 10 years. After adjusting for gender, race and education, the odds of incident ND at follow-up was 4.03 (95% CI 2.10 to 7.72) in those with baseline PTSD versus those without a trauma history.
Johnson et al.  followed a random sample of adolescents from New York state, assessed at baseline in 1983 (mean age 14 years) and at two subsequent follow-ups in 1985 to 1986 (mean age 16 years) and 1991 to 1993 (mean age 22 years). Diagnosis of ADs (grouped and individual) at baseline was made using the Diagnostic Interview Schedule for Children (DIS-C)  to DSM-III criteria. Smoking assessment was by self-report, and categorized into smoking > 1 pack (20+ cigarettes) versus smoking < 1 pack (1 to 19 cigarettes) per day (not non-smokers). After adjusting for a variety of demographic and other risk factors no association was detected between adolescent anxiety disorder status and onset of smoking in adulthood.
The Early Developmental Stages of Psychopathology (EDSP) study has been utilized to assess the association between ADs, ND and smoking behavior. In this study, a random community cohort of 3,021 adolescents and young adults (age 14 years to 24 years) was sampled from metropolitan Munich. The cohort was assessed at baseline (1995) and two follow-ups, the first between 1996 and 1997 and the second between 1998 and 1999 (a third follow-up was subsequently conducted in 2005 to 2006) . AD and ND diagnoses were made utilizing an updated version  of the Composite International Diagnostic Interview  for DSM-IV criteria and smoking was assessed by self-report. Isensee et al.  categorized participants from the EDSP into non-smokers, occasional smokers, non-dependent regular smokers and dependent regular smokers (see Additional file 1, Table S1 for definitions) and calculated odds ratios for incident smoking by baseline AD status. No associations were found between baseline AD status and odds of incident smoking. Sonntag et al.  extended this study in SP but once again found no association, although a positive association was found between those with social fears symptoms and later development of ND.
Woodward et al.  utilized a New Zealand birth cohort to compare the risk of DSM-IV ND between the ages of 18 and 21, dependent upon diagnosis of ADs between the ages of 14 to 16. A linear association was found between increasing number of ADs at age 14 to 16 (0 to 3+) and subsequent ND diagnosis, although this was not significant after controlling for childhood sexual abuse, alcohol abuse, parental changes and deviant peer affiliations.
Chou et al.  utilized the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) to investigate the association between ADs and ND in adults age 60 years and older (n = 8,012). Diagnoses were made utilizing the Alcohol Use Disorders and Associated Disabilities Interview Schedule for DSM-IV criteria (AUDADIS-IV)  and the risk of incident ND was assessed between baseline (2000 to 2001) and follow-up (2004 to 2005). No associations were found between baseline AD status and subsequent ND.
Cuijpers et al.  utilized data from the Netherlands Mental Health Survey and Incident Study (NEMESIS) to investigate the relationship between incident ADs (expressed as incident rate ratios) and previous smoking status. The NEMESIS study randomly recruited adults (18 years to 64 years) from 90 municipalities in The Netherlands, undertaking baseline assessments and 2 follow-ups at 1 year and 3 years (n = 4,796). The CIDI  was used for DSM-III-R diagnoses, and smoking was assessed by self-report, with participants placed into 4 categories (non-smokers, 1 to 9 cigarettes daily, 10 to 19 cigarettes daily and 20+ cigarettes daily). In a follow-up analysis, having 12-month (incidence rate ratio (IRR) 4.46 (P < 0.05)) and lifetime GAD (IRR 4.46 (P < 0.05)) was associated with increased risk of smoking onset at follow-up.
Johnson et al.  utilized the prospective follow-up of the full NESARC database (n = 34,653) to assess the impact of AD on smoking onset and persistence. Grouped ADs were associated with reduced daily smoking onset (OR 0.62 (95% CI 0.39 to 0.99)) when adjusting for demographics and socioeconomic status, but not with smoking persistence. Interestingly, comorbid substance use was an effect modifier; respondents with a comorbid substance use disorder (for example, alcohol, marijuana, amphetamines, opioids, sedatives, tranquilizers, cocaine, inhalants, hallucinogens, heroin, and other drugs) and AD demonstrated an increased risk of daily smoking onset (OR 2.22 (1.01 to 4.91)), whereas those without comorbid substance use disorder had a decreased risk of smoking onset (OR 0.43 (0.23 to 0.78)).
Most recently, Swendsen et al.  utilized a 10-year follow-up of 5,001 participants drawn from the National Co-morbidity Survey (NCS) to investigate mental disorders as a risk factor for onset of daily smoking or ND. The NCS, conducted in the US between 1990 and 1992, was a stratified multistage probability sample of 8,098 non-institutionalized residents (age 15 years to 54 years), utilizing the CIDI (V1.1), that assessed the interaction between smoking, ND and DSM-III-R mental disorders. The NCS2, performed in 2001 to 2002, was a 10-year re-interview of 5,001 participants from the NCS, but which utilized an updated version CIDI (V3.0) for DSM-IV criteria. After adjusting for sociodemographic characteristics, the odds of commencing daily smoking was increased for respondents with baseline PD, SP, GAD and specific phobia, but not for those with baseline PTSD or agoraphobia. In contrast, in those respondents with baseline daily smoking the odds of ND onset were raised in PTSD, agoraphobia and specific phobia, but not PD, SP or GAD. When considering the whole population odds of ND onset, having baseline PTSD, SP and specific phobia conferred an increased risk (see Additional file 1, Table S1).
Smoking and nicotine dependence as risk factors for later anxiety disorders
(A summary of this section is as follows (see also Additional file 2, Table S2): the available prospective evidence associating smoking and nicotine dependence as risk factors for incident anxiety disorders is limited and heterogeneous. However, smoking has been demonstrated as a risk factor for grouped anxiety disorders, panic disorder and generalized anxiety disorder in a number of studies, although these findings are not replicated in all studies.)
Data from the Oregon Adolescent Depression Project were utilized to assess the relationship between baseline smoking status and incident ADs. Goodwin et al.  demonstrated an association between increased odds of PD diagnosis at age 24 in those with daily smoking at baseline versus those not smoking daily (OR 5.1 (2.4 to 10.5)), which remained significant after controlling for other ADs and parental risk factors. No other associations were found.
Utilizing data from the Detroit Epidemiologic Study, Breslau et al.  found increased risk of subsequent PD onset in individuals with prior daily smoking even when controlling for gender and MDD (HR 13.13 (4.41 to 39.10)). In addition, prior daily smokers who continued to smoke were more likely to experience incident PD (HR 14.46 (4.81 to 43.5)) when controlled for gender and MDD.
In the New York Adolescent Cohort, relationships were discovered between odds of adult ADs when grouped (OR 10.78 (1.48 to 78.55)), GAD (OR 5.53 (1.84 to 16.66) and PD (OR 15.58 (2.31 to 105.14) when comparing baseline > 1 pack per day smokers versus < 1 pack per day smokers . Data from the EDSP studies  demonstrated relationships between increased incident PD, agoraphobia, SP and PTSD when comparing baseline ND smokers versus non-users, however all associations became non-significant when controlled for comorbid conditions at baseline (depressive disorders, panic attacks, other ADs, alcohol and drug disorders, and eating disorders). In the NESARC study, Chou et al.  assessed the relationship between ND at baseline and subsequent ADs, finding no associations.
Cuijpers et al.  utilized data from the NEMESIS to investigate the relationship between incident ADs (expressed as incident rate ratios) and past smoking status. Smoking at 1-year follow-up was associated with increased incidence of grouped ADs (IRR 1.77 (1.10 to 2.86)) and GAD (IRR 3.80 (1.09 to 13.21)) after controlling for demographics and other risk factors. No other relationships were found (see Additional file 2, Table S2).
(A summary of this section is as follows: a small number of studies have utilized a single time point analysis and retrospective self-report patient data to draw quasiprospective associations between smoking, ND and ADs. These studies generally indicate increased smoking behaviors or nicotine dependence in individuals with pre-existing anxiety disorders, and vice versa, although studies are limited by the retrospective nature of data.)
Baseline anxiety disorders and risk of smoking or nicotine dependence
See also Additional file 3, Table S3. Breslau et al.  examined the 4,411 participants who completed the tobacco supplement of the NCS to assess the interaction between smoking, ND and DSM-III-R ADs, utilizing discrete time survival models with ADs as time dependent variables and controlling for race, gender, education and age. The onset of daily smoking was the age at which respondents first smoked daily for ≥ 1 month. Increased odds for daily smoking were found in patients with pre-existing (OR 1.9 (1.05 to 3.7)) or currently active GAD (OR 2.1 (1.1 to 3.9)), pre-existing (OR 1.6 (1.3 to 1.8)) or currently active specific phobia (OR 1.5 (1.3 to 1.8)), pre-existing (OR 1.5 (1.2 to 1.7)) or currently active SP (OR 1.3 (1.1 to 1.6)), and pre-existing (OR 2.1 (1.6 to 2.9)) or currently active PTSD (OR 2.0 (1.4 to 2.9)).
Breslau et al.  further investigated the odds of smoking persistence and the transitioning from daily smoking to ND based upon pre-existing and currently active ADs. Increased odds of transitioning from daily smoking to ND were found in individuals with pre-existing and currently active agoraphobia, specific phobia, SP and PTSD. Interestingly, having pre-existing, but not currently active, PD was also strongly associated with increased odds of daily smoking to ND transition (OR 5.8 (3.0 to 11.6)). No associations existed between odds of smoking persistence and any pre-existing AD.
Koenen et al.  utilized data from the national Vietnam Era Twin (VET) registry and retrospective self-report of age of onset to test prospective onset associations between DSM-III-R ADs and ND. Associations were controlled for various demographic and other risk factors and time-dependent covariates (conduct disorder, MDD, alcohol and drug abuse or dependence) were entered into models. The results demonstrated that pre-existing PTSD was associated with increased odds of subsequent ND (OR 1.73 (1.38 to 2.17)). Utilizing the Greater Smoky Mountain Study (GSMS), a longitudinal representative study of 4,500 children (aged 9 years, 11 years and 13 years) from western North Carolina, USA, Costello et al.  found children, both boys and girls, with any AD were more likely to commence smoking than those without an AD (see Additional file 3, Table S3).
Smoking and nicotine dependence and risk of incident anxiety disorders
See also Additional file 4, Table S4. Breslau et al.  utilized the NCS, including respondent recall about their age of smoking and ND onset, to assess the effect of these parameters on developing ADs. Adjusting for demographic characteristics, pre-existing daily smoking (defined as onset > 1 year prior to disorder onset) was associated with increased odds of PD (OR 2.6 (1.2 to 5.4)) and agoraphobia (OR 4.4 (2.3 to 8.2)). The role of ND was assessed across all ADs. In this analysis, ND smokers and non-ND smokers maintained increased odds of PD and agoraphobia, but no other ADs. The only other associations were found in relation to past smokers (without ND) who exhibited decreased odds of PTSD (OR 0.2 (0.1 to 0.5)) when controlled for demographics and other pre-existing psychiatric disorders. Breslau et al.  extended their study by comparing the age of smoking onset (early vs not early; see Additional file 4, Table S4 for definitions), standardized pack years of smoking and time since quitting against odds of AD diagnosis. No association was found between early onset smoking and ADs, but increased years since quitting was associated with decreased odds of subsequent PD (OR 0.5 (0.4 to 0.7)), agoraphobia (OR 0.5 (0.5 to 0.8)) and SP (OR 0.6 (0.4 to 0.8)). The associations between standardized pack years of smoking were not significant in all ADs except PD, where increased pack years of smoking appeared protective in current smokers but a risk factor in past smokers, and GAD where increased pack years was associated with increased odds of GAD in both current and past smokers.
In a separate analysis utilizing a subsample of NCS data, Breslau et al.  investigated the interaction between smoking characteristics and subsequent onset of PD. Significant relationships were discovered between prior daily smoking (HR 2.93 (1.84 to 4.66)) and smoking persistence in prior daily smokers (HR 3.18 (1.99 to 5.10)) and subsequent onset of PD. In addition, pre-existing ND was associated with increased odds of subsequent PTSD onset (OR 2.24 (1.78 to 2.83)) in the aforementioned study drawn from the VET registry .
(A summary of this section is as follows (see also Additional file 5, Table S5): a large number of studies have reported cross-sectional relationships between cigarette smoking, nicotine dependence and anxiety disorders. Many demonstrate higher rates of smoking and nicotine dependence in those with anxiety disorders, and vice versa. However, their utility is limited due to their inherent inability to provide insight into direction of causality.)
Almost all studies included in this review reported cross-sectional associations between smoking and/or ND and ADs. Studies providing cross-sectional information [6, 7, 17, 19, 23, 27, 28, 32, 39–67] are listed in Additional file 5, Table S5. Descriptions of some selected larger studies utilizing population-based data are detailed below.
Smoking or nicotine dependence by anxiety disorder status
Lasser et al.  utilized the NCS to demonstrate increased rates of current and lifetime smoking in respondents with current and lifetime SP (39.5% and 54%), agoraphobia (38.4% and 58.9%), PD (35.9% and 61.3%), specific phobia (40.3% and 57.8%), PTSD (45.3% and 63.3%) and GAD (46% and 68.4%) when controlling for gender, age and geographical region. Utilizing the NCS-R data, Cougle et al.  explored the role of comorbidity in the association between ADs and smoking behavior. After controlling for demographics, depression and drug abuse/dependence, associations between increased odds of lifetime and 12-month daily smoking were observed with PTSD (Lifetime: OR 1.58 (1.21 to 2.06); 12-month: OR 1.46 (1.08 to 1.97)), 12-month daily smoking with PD (OR 1.42 (1.04 to 1.94)), and Lifetime daily smoking with GAD (OR 1.23 (1.05 to 1.61)).
In a nationally representative sample of the New Zealand population, individuals with ADs (grouped) had a smoking prevalence of 30.4% (27.7 to 33.0), and consumed approximately 16% of all cigarettes in New Zealand . In Australia, data from the nationally representative National Survey of Mental Health and Wellbeing 2007 reported rates of current and daily smoking in those with individual ADs (range for current smoker: 33% to 45%; range for daily smoker: 27% to 42%) well above the rates in respondents not reporting a mental disorder (current: 13.6%; daily: 10.8%) .
Cougle et al.  demonstrated increased odds of ND in patients with SP (OR 1.31 (1.01 to 1.71)), GAD (OR 1.59 (1.21 to 1.98)) and PTSD (OR 1.47 (1.01 to 2.16)) when adjusting for demographics, depression and drug abuse/dependence. Utilizing the NESARC for adults aged 18 to 25 years , the odds of 12-month ND were significantly increased for respondents with lifetime specific phobia (OR 1.8 (1.16 to 2.88)) after controlling for other psychiatric disorders, smoking and demographic characteristics.
Anxiety disorder by smoking or nicotine dependence status
Utilizing the German Transitions in Alcohol Consumption and Smoking study, Schumann et al.  calculated unadjusted odds ratios for individual ADs based on smoking and ND status. When comparing ND ever smokers (respondents with ND who had smoked at least one cigarette daily for ≥ 4 weeks at some point in their life) to non-ND ever smokers, increased odds were found for PD (OR 2.92 (1.78 to 2.73)), SP (OR 3.07 (1.70 to 5.57)), specific phobia (OR 2.09 (1.63 to 2.68)), GAD (OR 4.26 (1.85 to 9.84)) and PTSD (OR 2.08 (1.13 to 3.83)). Grant et al.  used data from NESARC to compare the odds of ADs on ND status. Unadjusted odds ratios with individual ADs as dependent variables were greater across all assessed ADs. Point estimates for odds ratios ranged from 2.6 for SP to 4.6 for PD with agoraphobia for respondents with versus without ND. Data from the UK National Households Survey  demonstrated increased rates of GAD (4.1% vs 2.4%), specific phobia (1.5% vs 0.8%) and PD (1.5% vs 0.5%) in respondents with ND versus those without ND. Degenhardt et al.  utilized the 1997 National Survey of Mental Health and Wellbeing in Australia to demonstrate increased odds of ADs (grouped) in current smokers versus never smokers (OR 1.50 (1.21 to 1.87)) when adjusted for demographic status, other drug use and neuroticism.