The recommendations of Meta-analysis of Observational Studies in Epidemiology (MOOSE)11
were used as guidelines when conducting this meta-analysis. In order to find articles reporting the rate of CUDs in schizophrenia published from 1996 to 2008, we conducted searches of 3 electronic databases (PsycINFO, PubMed, and Web of Science), the most recent being in January 2009. The key words used were “schizophreni*,” “psychosis,” “psychoses,” and “psychotic” to locate studies on schizophrenic psychoses, as well as “cannabis abuse,” “cannabis dependence,” “cannabis use disorder,” “substance use disorder,” “substance abuse,” “substance dependence,” and “dual diagnosis.” A similar search was carried out at the same time for alcohol use disorders. Altogether 3323 articles were retrieved, and their abstracts and titles were analyzed by both J.K. and J.M. Of these, 611 were identified as possibly relevant, and their full texts were analyzed in detail by J.K. and J.M. The inclusion of each article was independently evaluated and agreed upon.
In addition, a manual literature search was performed for the same time period from the journals Acta Psychiatrica Scandinavica, American Journal of Psychiatry, Archives of General Psychiatry, British Journal of Psychiatry, Journal of Clinical Psychiatry, Psychiatry Research, Schizophrenia Bulletin, Schizophrenia Research, and Social Psychiatry and Psychiatric Epidemiology. These journals were selected because they were available, and each had published a considerable proportion of the articles (approximately 40%) in our systematic database search. We also contacted approximately 30 authors to obtain unpublished information.
The inclusion criteria published in studies in our collection were that (1) at least 80% of the participants were individuals with a schizophrenia-spectrum diagnosis (schizophrenia, schizophreniform disorder, schizoaffective disorder, delusional disorder). Results from studies reporting findings from several psychiatric classifications were also included if they had determined the rate of CUDs in schizophrenia-spectrum patients alone. Other inclusion criteria were the following: (2) the study reported on the rate of cannabis abuse or dependence, (3) the subjects were older than 16 years, and (4) the study sample included more than 15 participants. Only articles (5) reporting schizophrenia and CUD diagnoses according to the DSM or ICD criteria and (6) written in English were included. We excluded studies with samples that might have biased the presented rates of CUDs in the study (eg, samples recruited from prisons, forensic psychiatry units, or homeless shelters). Trials and intervention studies were also excluded.
Information was collected on the classification system used. The terminology for CUDs in this article is adopted from the DSM classification system (abuse and dependence). We examined whether the classification system used (ICD-10, DSM-III-R, or DSM-IV) affected the presented CUD rates. For schizophrenia, the classification criteria mainly differ in terms of the duration of psychotic symptoms: In the ICD, the symptoms should last 1 month and in the DSM 6 months before making the diagnosis. The diagnostic criteria for CUDs in ICD and DSM classification systems are presented earlier.
CUD rates were compared between first-episode and long-term schizophrenia patient samples. The average duration of illness was determined from the studies. In samples other than first-episode patients, the minimum reported average duration was 9 years, and all these studies were categorized as having long-term patient samples. In addition, we determined the study location and whether the sample consisted of inpatients or outpatients. Information on the gender distribution, the distribution of schizophrenia diagnoses, mean age, and age range were collected where reported. CUD rates were for patients with abuse, and dependence diagnoses were combined in studies in which the diagnoses were not overlapping.
In this article, we present the number of studies as well as the mean, SD, median, interquartile range (IQR), and range for estimates of the rate of CUDs in each of the categories of interest. Due to the significant heterogeneity in rate estimates12
< .001), we present random mean estimates, which is a conservative weighting method giving the same weight to all studies. When evidence is found of heterogeneity among studies in rate estimates, linear regression analysis applying the bootstrap resampling technique (or meta-regression12
) with the z
test can be used to analyze associations between rates and study characteristics. Bootstrap methods were used because they make fewer assumptions about the distribution of the rates.13
We created 1000 bootstrap samples by randomly resampling with replacement from the original data. Regression analysis was used to compare the effect of classification systems (DSM-III-R
), study setting (first-episode vs long-term sample, inpatients vs outpatients), and location (North America vs Europe) on the rate estimates. Both the gender distribution (proportion of males) and mean age were analyzed as continuous variables. For reasons of presentation, these were categorized into 2 groups. The results of regression analysis were adjusted for the method of diagnosing cannabis use, so that variables for abuse (no/yes), dependence (no/yes), and the time period (lifetime/current) were included in the regression models. We also applied more traditional meta-regression to check whether the results remained statistically significant. The data were analyzed with Stata 9.0.14