This study has several main findings: 1) Psychiatric disorders were associated with higher prevalence of psychoactive substance use, regardless of type of disorder or substance; 2) CA, CD and CA+ CD prevalence rates were generally higher than unconditional prevalence rates among respondents with and without psychiatric disorders; 3) respondents with multiple disorders (mainly mood and anxiety disorders) had higher rates of CA+CD on most, but not all, psychoactive substances (e.g., not heroin), while schizophrenia was associated only with higher rates of tranquilizer CA+ CD, 4) psychiatric disorders had few associations with CA only and CD only. This pattern of findings supports the existence of common factors relating mood and anxiety disorders and substance abuse and dependence beyond just increased substance use, e.g., possible influence on the transition from use to abuse and dependence. In contrast, schizophrenia, while also associated with increased rates of substance use, was associated only with increased tranquilizer CA+ CD and amphetamine CA only, suggesting it may have little influence on the transition from use to abuse and dependence for most substances. However, the observed associations do not necessarily reflect direct causality. This question must be addressed by longitudinal studies and genetic epidemiology studies.
Respondents with mood or anxiety disorders or schizophrenia had higher lifetime prevalence rates of substance use than found in the overall NESARC sample of the US general population using an unconditional approach (Grucza et al., 2007
), but rates comparable to those in the general population from the 2002 NSDUH study, also using an unconditional approach (Grucza et al., 2007
; Substance Abuse and Mental Health Services Administration, 2003
). Respondents with psychiatric disorders also had higher CD rates (when considering CA+CD) than in other studies of population samples (which include respondents with psychiatric disorders) that used conditional approaches (Anthony et al., 1994
; Perkonigg et al., 1998
). This suggests that CD rates in the general population may be inflated due to the strong association between psychiatric disorders and substance use and substance use disorders, and that the presence of psychiatric comorbidity should be taken into account when measuring CD.
Respondents with no psychiatric disorder had generally lower CD and CA+CD rates than in other studies that used conditional approaches in the general population without stratifying by presence of psychiatric disorder (Anthony et al., 1994
; Perkonigg et al., 1998
) and lower lifetime prevalence rates of substance use (except for alcohol) than found in the overall NESARC sample of the US general population using unconditional approaches (Grucza et al., 2007
) and much lower rates than those in the general population from the 2002 NSDUH study also using unconditional approaches (Grucza et al., 2007
; Substance Abuse and Mental Health Services Administration, 2003
). Because the NESARC and NSDUH general population samples include respondents with psychiatric disorders, these differences suggest that estimates of substance use in the general population might be inflated due to higher rates of substance use among people with psychiatric disorders.
The proportion of respondents with mood or anxiety disorders who had CA+ CD was much higher than in studies of unconditional substance dependence (Conway et al., 2006
), suggesting possible influence of these disorders on the transition from use to abuse and dependence. The strength of associations between psychiatric disorders and CA+ CD in the present study somewhat resembles that found in other studies of lifetime substance use disorders (Compton et al., 2007
; Conway et al., 2006
; Conway et al., 2007
). In the present study, associations were stronger for CA+CD than for CA only and CD only (possibly due to small sample size of respondents with only substance abuse or dependence), and for those with multiple psychiatric disorders (mainly combinations of mood and anxiety disorders), rather than a single disorder (e.g., only a mood or anxiety disorder or schizophrenia). Compton and colleagues (2007)
found both past-year and lifetime unconditional drug abuse and dependence significantly associated with mood and anxiety disorders, even when adjusted for the presence of other psychiatric disorders.
Our results are only partially consistent with a recent analysis of 10-year prospective longitudinal data from the NCS (Swendsen et al., 2010
). That study found that pre-existing (baseline) mood disorder was associated with increased risk of developing tobacco CD over the following 10 years, but no increased risk of alcohol or illegal drug CD. Anxiety disorders were associated only with increased risk of alcohol CD. In contrast, the present study, which evaluated lifetime diagnoses, found that mood disorders were associated with increased prevalence of both tobacco and alcohol CD only, as well as CD and CA+ CD on various illegal drugs. In addition, anxiety disorders were associated with increased prevalence of tobacco CD only, alcohol CA only and CA+CD. Moreover, combined mood and anxiety disorders (multiple disorders group) were associated with tobacco, alcohol and several illegal drugs CD only and CA+CD.
These differences can probably be explained by differences in methodology between the NESARC and the NCS studies. Of note, substance dependence in the NCS was assessed only among individuals meeting criteria for DSM-IV abuse- known as a ‘gated approach’ (Swendsen et al., 2010
). Thus, the NCS study might have missed cases of substance dependence because this approach not only can reduce the number of identified cases of substance dependence but also can introduce bias in estimates of associations linked with substance dependence (Degenhardt et al., 2007
). In addition, the longitudinal design of the Swendsen et al. (2010)
study might have contributed to differences in findings if it takes more than 10 years (the period of the Swendsen study) for associations between co-morbidity and substance use transitions to become evident.
Our findings shed some light on the suggested link between marijuana use and psychosis (Moore et al., 2007
). We found a high rate (45.4%) of marijuana use among respondents with schizophrenia, comparable to that found in other epidemiological studies (Gregg et al., 2007
). This use rate was higher than among respondents with mood or anxiety disorders or with no disorder. However, respondents with schizophrenia were at no greater risk of marijuana CD than those with no psychiatric disorder. These findings suggest that the association between schizophrenia and marijuana is more related to the transition from abstinence to use than from use to dependence.
Our pattern of findings is not consistent with the self-medication hypothesis, which holds that individuals use a substance in order to alleviate particular psychiatric symptoms, e.g., depressed mood or anxiety (Khantzian, 2003
). As a consequence, there should be a considerable degree of psychopharmacologic specificity in an individual’s preferred drug (Khantzian, 2003
), e.g., those with depressed mood should be more likely to use and abuse stimulants, while those with anxiety should be more likely to use and abuse CNS depressants. We did not find any such clear-cut distinctions. Respondents with mood disorders only and anxiety disorders only had similar lifetime prevalence rates for use and similar adjusted odds ratios of developing lifetime CA+CD on both CNS depressants (alcohol, sedatives, marijuana) and stimulants (cocaine, amphetamine). Respondents with multiple disorders also had generally similar adjusted odds ratios of developing lifetime CA+CD on the various substances.
This study has several implications for clinical practice. Clinicians should keep in mind the increased risk for development of substance use disorders in patients with substance use and a comorbid psychiatric disorder. Screening all such patients for substance use disorders appears to be clinically worthwhile. Given the lack of support for the self-medication hypothesis, a narrow focus on specific substance co-morbidity in specific psychiatric disorders appears unjustified.
This study has several limitations. First, it is a cross-sectional study; thus, causal relationships cannot be inferred. Second, the diagnosis of schizophrenia was based on respondents’ self-report of receiving this diagnosis from a doctor, not on a structured diagnostic interview. Thus, the diagnosis of schizophrenia may have less validity than other diagnoses. However, the prevalence of schizophrenia in the NESARC sample (<1%) is consistent with that in other studies using structured diagnostic interviews (Eaton et al., 2008
), making it unlikely that schizophrenia was significantly misdiagnosed in the NESARC sample. Third, large epidemiologic studies may overestimate comorbidity associations, especially when the focus is on lifetime diagnoses in mixed-age samples (Kraemer et al., 2006
). Fourth, there might be some misdiagnosis of substance use disorders among respondents who are polydrug users, e.g., they might misattribute a dependence symptom to one drug that was actually related to another drug. Fifth, sample sizes were very small for some psychiatric disorder-substance combinations, resulting in very wide 95% CIs and precluding calculation of odds ratios in some cases. Therefore, this study may have been underpowered to detect significant differences in prevalence rates between some subgroups. Sixth, all information is based on self-report, as in all large-scale epidemiologic surveys. As such, the validity of these results is predicated on the accuracy of the information provided by respondents. However, the AUDADIS-IV has shown good reliability and validity (Grant et al., 2003
; Grant et al., 2005
). Overall, these limitations do not seriously weaken the validity of the observed differences. Small sample sizes and misdiagnosis or underdiagnosis would tend to blur distinctions and associations among subgroups, rather than generate spuriously significant differences. In addition, we believe that the conditional approach provides a stronger basis for any observed findings, instead of simply testing for associations using an unconditional approach.
In summary, this study is the first, to our knowledge, to evaluate the rates of substance-specific CA only, CD only, and CA+CD among respondents with psychiatric disorders that includes respondents with schizophrenia. This study identified different population subgroups at risk of developing CA and CD on different substances. In particular, it confirmed previous findings, using unconditional prevalence rates, that psychiatric comorbidity is associated with increased psychoactive substance use. The pattern of findings suggests that mood and anxiety disorders may influence the transition from substance use to abuse/dependence rather than from abstinence to use, while schizophrenia may influence the transition from abstinence to use (especially for marijuana), but does not support the self-medication hypothesis of substance use disorders.