Predictors of Tobacco Use Among Persons With Mental Illnesses in a Statewide Population

The 2001-2002 National Epidemiologic Survey on Alcohol and Related Conditions suggested that the approximately 7.1 percent of the U.S. population with a psychiatric illness consumes more than 34.2 percent of all cigarettes ( 1 ). Forty-one percent of persons in the United States who reported having recent mental illnesses also smoked cigarettes, on the basis of National Comorbidity Survey data from the early 1990s ( 2 ). This is approximately twice the prevalence of cigarette smoking among adults in the general population ( 3 ). Among persons with mental illnesses, smoking varies greatly by psychiatric diagnosis ( 2 ) and is associated with higher rates of alcohol and drug use ( 4 ). If persons with mental illnesses have multiple psychiatric diagnoses, their tobacco use is further increased ( 2 ). This vulnerable population clearly carries a disproportionate share of the burden related to tobacco use and warrants ongoing attention.

Prior study of tobacco use among persons with mental illnesses has focused on the high prevalence of use among persons with specific diagnoses, such as schizophrenia, bipolar disorder, and anxiety ( 5 , 6 , 7 , 8 ), and as noted, elevated rates of smoking have been found in nationally representative samples of persons with mental illness ( 1 , 2 ). To our knowledge, prior research has not utilized administrative data sets to investigate tobacco use rates in this population.

As an alternative to past studies with small samples or epidemiological studies, there is growing interest in utilizing large administrative databases to inform resource allocation and treatment guidelines ( 2 ). Findings from large statewide databases might complement existing findings regarding rates of tobacco use among persons with mental illnesses. Statewide estimates of tobacco use by individuals served by the public mental health system might support planning for primary and secondary intervention approaches, suggest potential barriers, and serve as a benchmark for future outcomes and research.

The objectives of this study were to utilize Colorado Client Assessment Records (CCARs) to estimate the prevalence of tobacco use among persons with mental illnesses accessing public sector mental health care in Colorado and to determine the relationships between tobacco use and primary diagnosis (that is, schizophrenia, bipolar disorder, schizoaffective disorder, depression or dysthymia, and anxiety, compared with all other diagnoses), and alcohol and drug use (amphetamines, barbiturates, caffeine, cocaine, marijuana, heroin, inhalants, and all other substances). We tested the hypotheses that primary diagnoses of schizophrenia, schizoaffective disorder, and bipolar disorder would be associated with the highest rates of smoking, followed by depression or dysthymia and anxiety and then all other diagnoses grouped together. We also explored a secondary hypothesis that substance use other than tobacco would be associated with increased tobacco use, independent of the primary diagnosis.

Methods

The study population included 111,984 unduplicated individuals aged 12 years and older whom the Colorado public mental health system served during the 2003 and 2004 fiscal years and who had complete data on key variables. This represents 99.6 percent of all cases (406 cases were excluded because of missing substance use data for one or more drug categories). Data are inclusive of individuals receiving a continuum of inpatient and outpatient services.

Data were obtained from the CCAR and were maintained by the Colorado Division of Mental Health, which provided patient-level data stripped of individual identifiers. CCARs are required to be completed for all persons served by the public mental health system at admission, discharge, and annually. Individuals were considered tobacco users if they endorsed current tobacco use on any CCAR during the study interval. Clinicians completed the CCAR data fields on the basis of interviews, treatment records, and available history. Public mental health system clinicians who collected data had a spectrum of educational backgrounds, but all were required to complete standardized CCAR training.

For the analysis we used SPSS version 12.0. We classified individuals by primary diagnosis; thus diagnostic categories are mutually exclusive and indicator variables are coded 1, if yes (that is, primary diagnosis), and 0, if no, with "other" as the reference category. Substance use variables were coded as 1, if present on the CCAR, and as 0, if not, but these do not constitute mutually exclusive categories, and individuals can be positive for multiple substances. Categorical variables included gender (reference was male with categories for female and unknown), race or ethnicity (reference was white), and age group (reference was the 18-to-59-year age group). Data on primary diagnosis and tobacco use were available for all individuals. Categories for unknown gender and unknown race or ethnicity were included, but cases with missing data on substance use were excluded from the analysis.

We inspected frequency distributions for all variables and used chi square tests to examine bivariate relationships between tobacco use and diagnostic category, substance use, and sociodemographic characteristics. Variables that did not meet the assumptions of the chi square test because of small cell sizes were not considered for multivariate analysis. Because the large sample provided sufficient power to detect associations that might not be clinically significant, variables were included in the final model only if they were significantly associated with the outcome at p<.01 in the chi square tests.

To test the hypotheses that primary diagnosis and substance use other than tobacco would be independently associated with tobacco use, the analysis controlled for gender, age, and race or ethnicity and we used multiple logistic regression analysis, with tobacco use as the dependent variable and all independent variables entered simultaneously. Adjusted odds ratios (ORs) with 95 percent confidence intervals (CIs) were computed for all independent variables.

This study was exempt from institutional review board approval, because publicly available data with no personal identifiers were used.

Results

The overall prevalence of tobacco use in the sample was 38.7 percent (N=43,508). Table 1 presents the results of the multivariate logistic regression analysis. The regression model correctly classified 71.2 percent of cases (Nagelkerke R ² =.257). Males were more likely to report smoking tobacco than females (p<.001). There were significant differences in smoking prevalence between age groups. Compared with adults aged 18 to 59 years, adolescents were less likely to report using tobacco (OR=.37, p<.001) as were older adults (OR=.50, p<.001). Compared with the race or ethnicity reference group of whites, the American Indian and Alaska Native group was the only group more likely to use tobacco (OR=1.29, p<.001), whereas Asians, Native Hawaiians and Pacific Islanders, and Hispanics were less likely to use tobacco (p<.001 for all). At the extreme, Asians were half as likely as whites to smoke (OR=.50, p<.001). There were no differences between African-American or multiracial groups compared with the white group.

Enlarge table

Compared with a reference category of all other primary diagnoses, persons with diagnoses of schizophrenia, schizoaffective disorder, bipolar disorder (p<.001 for all), and depression or dysthymia (p<.01) were more likely to use tobacco, whereas individuals with anxiety did not differ from the "other" category. At the extreme, individuals with schizophrenia were 2.61 times and those with schizoaffective disorder were 2.14 times as likely as individuals in the "other" diagnostic category to use tobacco.

We also examined co-occurrence of tobacco use with alcohol or other substance use for the study sample. Individuals who used alcohol, amphetamines, caffeine, cocaine, or marijuana were significantly more likely to use tobacco (p<.001 for all).

Discussion

This study represents a statewide exploration of tobacco use for almost 112,000 persons with serious mental illnesses who received services from the public mental health system over the course of two years. The study's finding that 38.9 percent of the sample used tobacco approximates other population-based research examining the prevalence of smoking among adults with psychiatric disorders ( 2 ). This is about double the smoking rate of 20.4 percent for Colorado's adult population ( 9 ).

Smoking rates for specific diagnoses were also similar to those found in previous research, although rates for bipolar disorder (50.7 percent) and anxiety (32.0 percent) were lower than rates previously found, which were 60.6 and 46.0 percent, respectively ( 2 ). The data support our hypothesis that primary diagnoses of schizophrenia, schizoaffective disorder, and bipolar disorder are associated with the highest rates of smoking, compared with other diagnoses. Consistent with other research, the data support our hypothesis that substance use is also associated with higher rates of smoking ( 4 ). At a tobacco use rate of 30 percent ( Table 1 ), users of the public mental health system who have diagnoses other than schizophrenia, schizoaffective disorder, bipolar disorder, anxiety, or depression also smoked at a greater rate than the general population, which is indicative of these individuals' generally higher psychiatric symptom burden.

This study has several limitations. The CCAR is an instrument completed in the field by mental health clinicians. Thus the diagnostic categories and other ratings are based on data available to community-based clinicians. Much of the data are based on self-report. We also did not capture data on frequency, intensity, or duration of tobacco use behaviors or mental health treatment.

Conclusions

There are few prior reports of tobacco use rates and predictors of tobacco use for statewide populations accessing the public mental health system. This study furthers the work of existing small-sample and nationally representative investigations by providing a cross-sectional analysis of the clinical and demographic factors associated with smoking behavior in a large community-based sample. We found that this large statewide administrative data set presents an impressive data source that has broad relevance to the mental health community ( 10 ). By describing the predictors of tobacco use among persons with mental illnesses, study findings have assisted the state to identify priority subgroups of tobacco users, such as persons with schizophrenia and persons with co-occurring mental illnesses and substance use. We found that CCAR data are a low-burden basis for service planning, policy initiatives, and ongoing tobacco use research.

Acknowledgment

The authors gratefully acknowledge the assistance provided by Debra Kupfer, M.S.