|Home | About | Journals | Submit | Contact Us | Français|
Background: Cigarette smoking was consistently found to be more prevalent in individuals with schizophrenia than in other psychiatric groups and the general population. These findings have been interpreted as evidence of a specific association between schizophrenia and smoking. However, the supporting data come primarily from cross-sectional studies, which are susceptible to confounding. Our aim was to test specificity of this link longitudinally in an epidemiologic sample. Methods: A cohort of 542 inpatients with psychosis was followed for 10 years after first hospitalization, completing 5 face-to-face interviews. Assessments included ratings of specific symptoms (psychotic, negative, disorganized, and depressive), Global Assessment of Functioning, and a categorical measure of cigarette consumption. All participants were assigned longitudinal consensus diagnoses by study psychiatrists, and 229 were diagnosed with schizophrenia spectrum disorders (SZ). Results: At baseline, 52.4% of participants were current smokers and 69.3% were lifetime smokers. Smoking rates did not differ among the diagnostic groups (schizophrenia spectrum, major depressive, bipolar, or other psychotic disorder) at any assessment point. Smokers were more severely ill than nonsmokers but did not differ in specific symptoms either cross-sectionally or longitudinally. Among smokers, changes in cigarette consumption were linked only with changes in depression (β=.16, P<.001). Conclusions: Rates of smoking were elevated in subjects with schizophrenia but were just as high with other psychotic disorders. Smoking was not associated with psychotic symptoms, but cigarette consumption covaried with depression over time. Given the devastating health consequences of cigarette use, smoking cessation interventions are urgently needed in this population and should specifically address depression.
Smoking is the leading cause of preventable mortality in developed countries, and it remains highly prevalent among patients with schizophrenia despite a steep decline in the general US population.1–5 A quantitative review of 42 studies from 20 countries reported a 62% pooled rate of smoking in patients with schizophrenia, which was significantly elevated relative to general population and psychiatric comparison groups.2 These data have been interpreted as evidence of a specific link between smoking and schizophrenia. It was hypothesized that schizophrenia leads to smoking, presumably, because nicotine reduces symptoms of the illness (ie, self-medication),6 that smoking may contribute to development of the disorder by altering neurochemical systems in the brain,7 and that both conditions could arise from a common genetic vulnerability.2
However, these hypotheses are based primarily on cross-sectional comparisons of smoking rates in schizophrenia and various other psychiatric disorders.2 Elevated rates found in such studies may be due to confounders, such as illness severity or impoverished environment.4,8,9 Several investigations attempted to control for potential confounders statistically, but the selection of variables varied widely. Typically, adjustments were made for age and gender, but some studies also controlled for socioeconomic status (SES), low IQ, education, marital status, comorbid diagnoses, and several other characteristics.4,10–13 The effect almost invariably persisted after corrections. However, no study controlled for illness severity. Prospective investigations can provide more direct evidence of the connection, but 2 studies conducted to date produced contradictory results. Weiser et al12 found that smokers were at an increased risk for developing the illness, whereas Zammit et al13 reported that smoking had a protective effect. Moreover, investigations of associations between smoking and severity of schizophrenia symptoms have also produced inconsistent results. One study reported a link to both negative and positive symptoms,14 4 found associations with one type of symptoms but not the others,11,15–17 and another 4 did not find any significant relations.18–21 In contrast, the link between smoking and depression is well established both cross-sectionally22 and prospectively.23–27 In fact, it was suggested that depression may contribute to the high rates of smoking in patients with schizophrenia,14,17 but this hypothesis has not been directly tested.
The objective of this study was to test the belief that schizophrenia has a specific connection to smoking. First, we examined the specificity of elevated smoking rates to schizophrenia by comparing it to disorders of similar severity (eg, mood disorders with psychosis). Second, we tested associations of smoking with schizophrenia and depression symptoms both cross-sectionally and longitudinally. Third, we evaluated links between changes in these symptoms and fluctuations in cigarette consumption over time among smokers.
Data for this study came from a first-admission cohort described in detail elsewhere.28–31 The patients were recruited from all 12 psychiatric inpatient units of Suffolk County, NY, between 1989 and 1995. The units included state, community, and university facilities. At the beginning of the study, all facilities permitted smoking, but in late 1991 and early 1992, policies changed to prohibit smoking on the units. Instead, patients were regularly taken outside to smoke. They were also offered nicotine replacement (gum or patch). Despite these changes, the rate of current smoking remained stable across the 6-year enrollment period (β=−.01, P=.83).
The purpose of the parent study was to investigate the course of schizophrenia longitudinally in an epidemiologic sample of first-admission cases. To obtain a representative sample, all new patients with psychosis were enrolled, and thus, a variety of psychotic disorders are present in this cohort. The inclusion criteria were aged 15–60, first admission either concurrent or during the 6 months prior to the index admission, clinical evidence of psychosis, ability to understanding the assessment procedures in English, and capacity to provide written informed consent. The procedures for obtaining informed consent were approved annually by the Institutional Review Boards at Stony Brook University and all hospitals where respondents were recruited. For participants aged 15–17, written consent of parents was also required. The response rate was 72%. Overall, 675 individuals met the inclusion criteria and were interviewed at the baseline. Of those, 31 were later found to be ineligible, primarily because it became clear that they never suffered from psychosis. Another 55 participants were excluded because of insufficient follow-up information to permit a longitudinal diagnosis (15 of them died in the first 24 months of the study), and 47 were removed because they had a substance-induced or organic psychosis. Hence, the analysis sample is composed of 542 participants (figure 1).
In addition to baseline, face-to-face interviews were performed at month 6 and years 2, 4, and 10 by trained master's level mental health professionals. Study psychiatrists formulated longitudinal consensus Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, research diagnoses after the year 2 assessment. The ratings were made independently by at least 2 psychiatrists and were discussed by the whole team. The raters based diagnoses on all available information from structured clinical interviews, medical records, and significant others collected at baseline, month 6, and year 2.30 The diagnoses were grouped as schizophrenia spectrum (SZ; schizoaffective disorder, and schizophreniform disorder; N=229), bipolar disorder with psychosis (N=148), major depressive disorder with psychosis (N=104), and other psychotic disorders (delusional disorder, brief psychotic reaction, and psychotic disorder not otherwise specified; N=61). The SZ group is the focus of the present study.
With extensive efforts to track the sample and keep respondents invested in the study, we were able to maintain contact with the vast majority of participants. By the 10-year point, 28 respondents had died, 29 were lost to the study, 47 refused to be interviewed, and 8 were too ill to give informed consent. Half of the 28 participants died of medical illnesses (most commonly cardiovascular disease, various forms of cancer, and liver failure), but only one of them died of lung cancer. Due to relatively low mortality and absence of a clear connection to tobacco use, we do not expect that mortality biased our findings. Of 430 contacted respondents, 73 provided limited data and could not be analyzed. Thus, all necessary information at year 10 was available for 357 respondents (completers), and 147 of them had an SZ diagnosis. We compared excluded participants (N=185) with completers on all variables used in this study and found no significant differences (all P>.10), which suggests that attrition had little systematic impact on the results. For instance, the prevalence of current smoking at baseline was 51.6% among noncompleters, as compared with 52.8% among completers (χ2(1)=.07, P=.80). Missing data were handled with maximum likelihood methods, and thus, all available data were utilized without deleting any records.
Smoking was assessed with items adapted from the National Household Survey on Drug Abuse interview.32 All participants were asked how much they smoked per day in the past 30 days (past 6 months at waves 2, 3, and 4) on the following scale: none, <1 cigarette, 1–5 cigarettes, ½ a pack, 1 pack, 1½ packs, and 2 or more packs. A participant reporting any usage was considered a smoker at that wave. At baseline, participants were also asked whether they ever smoked on a regular basis. Those who responded yes or reported cigarette usage at baseline were considered lifetime smokers. Responses were recoded as the number of cigarettes consumed on an average day using the mid-point of the corresponding interval except for the highest level, which was coded as 40.
Symptoms of schizophrenia in the 4-week period preceding the interview were measured with the Scale for the Assessment of Positive Symptoms (SAPS)33 and the Scale for the Assessment of Negative Symptoms (SANS).34 The SAPS consists of 31 items tapping 4 symptom domains and a global rating for each domain. The SANS is composed of 19 items measuring 5 domains as well as the global ratings. SAPS and SANS items are rated on a 6-point scale (0=none, 5=severe). The validity of these measures is well established.35–37 Specifically, the SANS is unidimensional, whereas the SAPS taps 2 factors: psychotic symptoms (hallucinations and delusions) and disorganized symptoms (bizarre behavior and formal thought disorder). Consistent with prior studies, we used individual items but not the global items.38–40 Common factor analysis supported the established structure in the present data. We removed 3 items that had consistently poor loadings in these analyses: SANS item 13 (inattentiveness during mental status testing), SAPS item 9 (persecutory delusions), and SAPS item 10 (delusions of jealousy). Thus, the final SANS composite consisted of 18 items (internal consistency reliability ranged α=.88–.90 over the 5 waves), SAPS psychotic of 16 items (α=.79–.85), and SAPS disorganized of 13 items (α=.71–.76). The interrater reliability of these ratings was very good (average interrater r=.72).41
Symptoms of depression in the month preceding the interview were assessed with the Structured Clinical Interview for DSM-III-R (SCID)42 depression module administered without skip-outs. Depression was operationalized as a composite of the 9 DSM criterion symptoms. Items were rated on a 3-point scale: 1=not present, 2=questionable, and 3=definite. Internal consistency reliability of the composite ranged from .83 to .86, and the interrater reliability was very good.28 The composite correlated .74–.80 with the Hamilton Rating Scale for Depression (HAM-D)43 across different waves. The HAM-D was not administered at year 4, and hence, depression was assessed with the SCID-based scale. However, we repeated all analyses with the HAM-D (without the year 4 data), and none of our findings changed.
Illness severity was operationalized as psychiatrists’ consensus Global Assessment of Functioning (GAF) rating for the worst week in the past month. This score had little variability at baseline because all participants were hospitalized. Thus, we also report on baseline GAF for the best month (GAF Best) in the year prior to hospitalization in some analyses.
GAF correlated strongly with the SANS (average concurrent r=−.61) but had only moderate associations with the other symptom scales (average r=−.33). The concurrent correlations among the symptom scales were low to moderate (average r=.19), which is consistent with the literature.35–37
We examined 3 demographic variables: sex, age at baseline, and SES of the head of the household. SES was assessed at baseline using the 7-point occupational scale of the Hollingshead Four Factor Index of Social Position.44 It refers to the participant unless he or she is supported by someone else.
Cross-sectional comparisons were made with t tests and chi-square tests. Adjustment for covariates was performed with logistic regression. Stability of smoking status was evaluated with tetrachoric correlations, estimated in Mplus version 4.1.45 Tetrachoric correlations quantify associations between dichotomous variables on the metric of Pearson r while correcting for biasing effects of low base rates. Longitudinal associations between smoking status and symptom/illness severity were examined with repeated-measures analysis of variance in SAS PROC MIXED.46 Changes in cigarette consumption over time and their correlates were evaluated with multilevel growth curve analysis,47 estimated in PROC MIXED. Specifically, cigarette consumption was modeled on 2 levels as repeated observations (level 1) nested within individuals (level 2). In level 1 (within subjects), cigarette consumption was evaluated as a function of time, symptoms, and illness severity, whereas level 2 (between subjects) modeled the mean and variance of the level 1 intercept as a function of demographic variables. Longitudinal trends in cigarette consumption were evaluated with a modified model. We removed all clinical and demographic covariates, so that level 1 described cigarette consumption as a linear function of time alone, and level 2 included only mean and variance of the intercept. All variables except time were standardized with respect to their grand means and SDs (across all subjects and waves) to facilitate interpretation. In both PROC MIXED analyses, the random-effects covariance structure was specified as an unstructured covariance matrix because it imposed the fewest assumptions and provided a better fit than various constrained covariance matrices. Although the SZ group was the focus of the study, all analyses were also repeated for the total sample.
In the total sample, the lifetime prevalence of smoking was 69.3%. It did not differ by diagnosis, although the SZ group had the lowest rate (67.4%; table 1). Current prevalence of smoking at baseline was 52.4% and was similar across the diagnoses. In fact, there were no diagnostic differences in rates of current smoking at any of the follow-up assessments. Adjusting these comparisons for age, sex, and SES did not change the findings, except that at year 4 the SZ group had a significantly lower rate of smoking that the “other psychoses” group (adjusted odds ratio=.32, confidence interval=.12–.86). We also stratified the sample by age (15–18, 19–25, 26–44, and 45–58 years) and repeated the analyses but observed no diagnostic differences in either of these strata (table 1).
Smoking status was very stable over the 10 years with tetrachoric correlations between waves ranging .80–.97 (table 2; top half). Indeed, 80% of baseline smokers, as compared with 21% of baseline nonsmokers, smoked at year 10. Of the 229 participants with SZ diagnoses, 147 smoked at one or more waves and constitute the smokers subsample.
Lifetime smokers did not differ from nonsmokers with respect to sex, age, and SES (table 1). Moreover, smoking status was not related to symptom and illness severity at any assessments point, except for one significant finding for the GAF and one for the SANS (table 3). We also examined these relationships longitudinally. Given the high stability of smoking status, participants were classified according to their pattern of smoking across all available assessments as abstaining (never smoked during the study, N=82), fluctuating (had a change in smoking status, N=51), or persisting (always smoked, N=96). The groups did not differ on diagnosis, gender, or SES, but the fluctuating group was 5.0 years younger than the rest of the sample (P<.001). The groups also did not differ in their trajectories of symptom and illness severity. Controlling for sex, age, and SES did not change the findings.
In the smokers subsample (N=147), individual cigarette consumption varied considerably over time, with Pearson correlations across waves ranging .33–.64 (table 2; bottom half). These results indicate that some participants increased their use while other decreased it. However, the group's average daily cigarette consumption did not change. It was 16.1 cigarettes a day at baseline and showed no significant upward or downward trend over the 10-year follow-up.
Cross-sectional analyses did not reveal any consistent associations between cigarette consumption and clinical variables (Appendix), including GAF Best at baseline (r=−.04). In the longitudinal analysis, we related changes in cigarette consumption to changes in symptom and illness severity over time. Each association was adjusted for demographic characteristics and influence of the other variables. Only depression severity showed a significant effect (β=.16, P<.001), which corresponds to consuming an extra cigarette a day for each additional DSM depression symptom (table 4).
Given that the rates of smoking did not differ among the diagnostic groups, we repeated all analyses in the total sample. We thus increased power and enhanced generalizability of the analyses. Smoking status was very stable in this sample (table 2), and 351 of the 542 participants smoked at least at one point during the study. In this group, cigarette consumption varied considerably over time (table 2). Average consumption at baseline was 18.1 cigarettes a day and declined at the rate of 0.36 cigarettes a year (P<.001) over the 10-year follow-up.
Lifetime smokers did not differ from nonsmokers with respect to sex, age, and SES (table 1). On the other hand, smokers had consistently lower GAF scores, with comparisons being significant at each wave except lifetime and baseline (table 3). However, lifetime and baseline smokers had significantly lower GAF Best scores (d=−0.23, P<.05 for both). Depression scores were consistently elevated in smokers and the difference reached significance at 3 time points. The SANS was also significantly elevated at years 4 and 10 (table 3). Controlling for sex, age, and SES did not change the findings, except for the effect of year 2 GAF in the total sample, whose significance changed from P=.030 to .081. Although more associations were statistically significant in the total sample than in the SZ subgroup, the effect sizes remained essentially the same (see mean effects in table 3). Thus, the apparent difference in findings likely reflects greater statistical power of the total sample. To appreciate the magnitude of association between smoking and GAF, we compared prevalence of severe illness (GAF≤50) at year 10 between smokers and nonsmokers. Many more smokers were severely ill (64.5% vs 45.1%, P<.001).
Longitudinal analyses also confirmed the association between smoking status and illness severity. Smoking pattern (abstaining, fluctuating, or persisting) had no association with symptom trajectories, but the persisting group had consistently lower GAF scores than the abstaining (P<.01), with the fluctuating falling in between (figure 1). Controlling for sex, age, and SES did not change the findings.
Results for cigarette consumption paralleled findings in the SZ group. Cross-sectional analyses did not reveal any consistent associations between cigarette consumption and clinical variables because no links were replicated across more than 2 waves (Appendix). In the longitudinal analysis, only depression severity had a significant association with the amount of smoking while adjusting for other clinical variables and demographic characteristics (table 4).
All analyses were repeated in the completers subsample. The results were unchanged, except that a few cross-sectional comparisons became nonsignificant due to reduced power and without an appreciable change in magnitude.
Consistent with previous reports, the point prevalence of smoking in the schizophrenia group of our epidemiologic first-admission sample (51.0% at year 10) was 2.5-fold greater than in the US population (20.9% in 2005).1 However, rates of smoking were just as high in the other diagnostic groups. The present findings underscore the significance of smoking as a public health problem that cuts across psychotic disorders and possibly all severe mental illnesses. There was little change in smoking status among participants with schizophrenia over the 10-year follow-up, and the average cigarette consumption remained stable, although for individual smokers the amount of consumption fluctuated notably over time. Neither smoking status nor amount of consumption was associated with schizophrenia symptoms. However, changes in the letter were linked with changes in depression symptoms.
We did not find evidence of specificity of smoking to schizophrenia relative to other psychotic disorders at any point during the follow-up, although our study had the power of .93 to detect the level of difference reported in a recent meta-analysis that compared schizophrenia with other psychiatric disorders.2 Further examination of smokers suggested that they were not different from nonsmokers in severity of specific symptoms but were more severely ill overall. For instance, at the end of the follow-up, smokers were 43% more likely to be severely ill than nonsmokers. Thus, it appears that reports of elevated smoking rates in schizophrenia may have confounded diagnosis with illness severity. In support of this interpretation, smoking did not show specificity to nonaffective psychosis in a nationally representative study of mental illness in the United States.48 Furthermore, as mentioned in the Introduction, prospective investigations of relations between smoking and schizophrenia produced contradictory results, and cross-sectional studies failed to find consistent associations between smoking and severity of schizophrenia symptoms. On the other hand, smoking status has been linked to illness severity in general psychiatric samples, albeit assessed with proxy measures.4,8,49 For instance, Aguilar et al49 found no clear links between nicotine dependence and schizophrenia symptoms but observed an association between smoking and frequent hospitalizations, an indirect measure of illness severity.
Examination of cigarette consumption among smokers did not support the self-medication hypothesis with regard to schizophrenia symptoms. However, we found that increases in depression were associated with heavier smoking. This observation is consistent with the extensive literature documenting cross-sectional and longitudinal correlations between smoking and depression in other populations.22 It appears that depression may have an etiologic connection with smoking behavior, and mechanisms for this association have been proposed.22,50–52 These data highlight the importance of addressing depression in smoking cessation interventions.
We did not find differences between smokers and nonsmokers in SES and gender composition, although in the general population smoking is more prevalent among men and in lower socioeconomic strata.1 The gender difference was also observed in schizophrenia samples.2 However, the patients in these studies have typically been ill for a long time. In contrast, our cohort was assessed at the time of the first admission and was followed through the early stages of the illness. We expect that gender and SES differences will emerge in our sample with time.
A notable caveat of this study is that the findings should not be generalized to patients who have never suffered from psychosis. With regard to our diagnostic comparisons, it is important to note that the majority of mood disorder patients did not have any psychotic symptoms at the follow-up assessments, yet no diagnostic differences were observed. Furthermore, severity of psychotic symptoms was associated neither with smoking status nor with cigarette consumption. Nevertheless, our conclusions should also be tested in nonpsychotic severely ill patients. The present findings do show that smoking is highly prevalent across psychotic disorders rather than being specific to schizophrenia. Another limitation of the study is that anxiety was not assessed, despite growing evidence of its links to smoking.5,48,53 Cigarette consumption is also associated with substance use. Evaluation of this relationship was beyond the scope of the present study, but it would be very informative to examine the links between substance abuse and smoking over time. Finally, smoking was measured with a single item and was rated on an ordinal scale rather that continuously. This likely limited reliability of the score and probably led to underestimation of the effect sizes.
Despite these limitations, our findings suggest that smoking is especially prevalent among those with severe mental illness, and special intervention efforts should be directed toward this group. It appears that factors contributing to high rates of smoking are likely to be consistent across the diagnostic groups studied here. One factor with potential etiological significance is depression, and it should be addressed specifically in smoking cessation protocols. Importantly, emerging evidence suggests that smoking cessation interventions can be successfully employed in psychiatric populations,54,55 and quitting does not seem to impede recovery from mental illness.56 Hence, provision of these services to severely mentally ill is an imperative.
National Institutes of Health (MH-44801 to E.B.).
The authors thank the mental health professionals in Suffolk County, the mental health project psychiatrists and staff, and most of all, the study participants and their families and friends. Special thanks to Elizabeth Persons for help in preparation of the manuscript.
|Baseline||Month 6||Year 2||Year 4||Year 10||Meana|
|Schizophrenia spectrum group|
Note: GAF, Global Assessment of Functioning; SAPS, Scale for the Assessment of Positive Symptoms; SANS, Scale for the Assessment of Negative Symptoms.
*P<.05, **P<.01, ***P<.001.