Association of cannabis use-related predictor variables and self-reported psychotic disorders: US adults, 2001-2002 and 2012-2013

Abstract

Objective:

To determine the association of cannabis indicators with self-reported psychotic disorders in the U.S. general population.

Methods:

Participants were from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC, 2001-2002; N=43,093) and NESARC-III (2012-2013; N=36,309). Logistic regression was used to estimate standardized prevalences of past-year self-reported psychotic disorders within each survey, and evaluate the association of past-year self-reported psychotic disorders with indicators of non-medical cannabis (any use; frequent use [≥3-times/week], daily/near-daily use, and DSM-IV cannabis use disorders [CUD]) compared to those with no past-year non-medical cannabis use. Whether associations changed over time was indicated by difference-in-difference tests (between-survey contrasts) and ratios of odds ratios between surveys.

Results:

Self-reported psychotic disorders were significantly more prevalent in those with than without any non-medical cannabis use (2001-2002: 1.65% vs 0.27%; 2012-2013: 1.89% vs. 0.68%). In 2001-2002, self-reported psychotic disorders were unrelated to either frequent use, or daily/near-daily use, but in 2012- 2013, self-reported psychotic disorders were more common both in those with frequent use and daily/near non-medical cannabis use, each compared to non-users (2012-2013:2.79% vs. 0.68%; 2.52% vs. 0.68%, respectively). Self-reported psychotic disorders were significantly more prevalent in participants with CUD than in non-users in both surveys (2001-2002: 2.55% vs. 0.27%; 2012- 2013: 3.38% vs. 0.68%). These associations did not differ significantly over time.

Conclusions:

Data from the U.S. general population, especially more recent data, suggest associations between frequent non-medical cannabis use indicators and self-reported psychotic disorder. Clinicians and policymakers should consider these relationships when monitoring patients and formulating programs.

INTRODUCTION

Schizophrenia spectrum and other psychotic disorders are a heterogenous group of serious mental health disorders that involve impairment in thinking, perception and emotion(1, 2). Despite being relatively uncommon in the general population, psychotic disorders result in substantial social, economic, and health-related burdens(1, 3–5), are leading causes of disability-adjusted life-years in the US and worldwide(6–8), and increase the risk of suicide and early mortality(9–11). Information on change over time in the prevalence of psychotic disorders can help gauge need for services and identify changes in potentially modifiable risk factors. However, many methodological issues make determining time trends in the prevalence of psychotic disorders challenging. While meta-analyses(12, 13) have not found evidence of change in the incidence or prevalence of psychotic disorders over time, considerable heterogeneity in study designs and resulting prevalence estimates could have obscured changes in prevalence. Further, most of the meta-analyzed data originated outside the US. Studies utilizing large-scale national data are needed to begin to understand time trends in rates of psychotic disorders in the US, and factors that may be associated with change.

One such factor may be cannabis use. Cannabis is one of the most widely used psychoactive substances in the US and worldwide(14). The prevalence of adult non-medical cannabis use, frequent use and cannabis use disorder (CUD) has increased in the US general population and in large-sample studies of patient populations(15–18). In addition, THC potency of illicit cannabis increased more than 3-fold since 1995, and THC potency of legal cannabis and other cannabis-derived products (e.g., wax and shatter) is often substantially higher(19–21). Findings from several prospective and cross-sectional studies indicate a dose-response relationship between frequency of cannabis use and the risk for psychosis, as illustrated in a recent meta-analysis(22). Further, there is increasing evidence of strong associations of high potency cannabis use and psychosis(23, 24).

While fewer studies addressed the relationship of cannabis use disorders to psychosis, some prospective findings suggest that cannabis use disorders are prospectively associated with increased risk for development of psychotic disorders(25, 26). While the nature of the relationship of cannabis to psychosis has been debated, i.e., whether the relationship is causal or due to shared genetic risk factors(27, 28), a prudent conclusion appears to be that some part of the relationship is causal(27, 28), and therefore that further study of the relationship is warranted.

Lengthy, detailed symptom-based measures of psychotic disorders have not been feasible in recent US national surveys, leading to a gap in knowledge about psychosis and potential risk factors among US adults. An alternative survey approach is to ask respondents to self-report on schizophrenia or psychotic illness that was diagnosed by a doctor or other health professional (self-reported psychosis [SRP]). This approach was used in the 2001-2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC)(29). One study of NESARC data showed associations between lifetime SRP and a combined substance use disorder category, but provided little information specific to cannabis(30). Other NESARC studies showed associations of lifetime SRP with cannabis use and CUD(31, 32), but these studies addressed lifetime, not current (past-year) disorders, and also reported on data collected before the substantial increases in adult cannabis use, cannabis potency, and CUD since the mid-2000s(15, 33). These changes in the US cannabis landscape warrant examination of whether the prevalence of SRP and its association with cannabis use or CUD has changed over time.

Accordingly, we used data from two US nationally representative adult surveys, the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) and the 2012-2013 NESARC-III, to examine the following three questions. 1) Did the prevalence of current SRP (self-reported psychotic episode in the past year) change over time? 2) Were cannabis use indicators (any non-medical use, frequent non-medical use, daily/near-daily non-medical use, or CUD) associated with current SRP in either survey? 3) Did the relationships of cannabis and current SRP change between 2001-2002 to 2012-2013?

METHODS

Samples and procedures

The NESARC(29) and NESARC-III surveys(34) used multistage designs to sample adults (age ≥18) in households and group quarters. Sample weights adjusted for nonresponse and probability of selection. The total sample analyzed was 79,402 (43,093 in NESARC, 36,309 in NESARC-III). Across surveys, rigorous field procedures were similar(16, 35), including structured in-class training and home-study for interviewers, random callbacks to verify interview data, and expert supervision. The examination of trends over time in important health outcomes was possible due to the methodological similarity between the two surveys(16, 36–38). For the 2001- 2002 NESARC, the US Bureau of the Census and Office of Management and Budget institutional review boards (IRB) approved the protocol and written consent procedures. Response rate for NESARC (2001-2002) was 81.0%. For the NESARC-III, the IRB at the National Institutes of Health and Westat approved the protocol and verbal (recorded electronically) consent procedures. Response rate for NESARC-III was 60.1%, similar to other US representative surveys conducted in similar years(39, 40).

Measures

In both surveys, substance use and substance use disorders were assessed using a structured computer-assisted interview, the Alcohol Use Disorder and Associated Disabilities Interview Schedule (AUDADIS).

The outcome, self-reported psychotic disorders in the past year, was measured in NESARC and NESARC-III with a nearly identical question, asking if a doctor or other health professional told the respondent that they had schizophrenia or psychotic illness or episode. This brief survey measure has been previously validated (41).

Predictors included four past-year cannabis use-related variables: any non-medical use, frequent non-medical use, daily/near-daily non-medical use, and DSM-IV CUD. In both surveys, identical questions were used to assess non-medical cannabis use, defined as use without a prescription or other than prescribed, e.g., to get high(35). Any use was assessed using a two-level variable, “yes” for those who used ≥1 time in the past year and “no” otherwise. Frequent use was assessed using a three-level variable: those who used ≥3 days per week; those with any use but less than three days per week; and those with no past year use. Similarly, daily/near-daily use was assessed using a three-level variable: those who used 5-7 days per week; any use but less than 5-7 days per week; and no past year use. CUD was assessed using a three-level variable: those with DSM-IV abuse or dependence, i.e., ≥3 of 6 DSM-IV dependence criteria (cannabis withdrawal was not included in DSM-IV) or ≥1 of 4 DSM-IV abuse criteria; any use but no CUD; and no past-year use. Abuse and dependence were combined because an extensive review of studies conducted in preparation for the publication of DSM-5 showed that the criteria for cannabis abuse and dependence were unidimensional, reflecting a single CUD diagnosis(42). Consistent with this, such a DSM-IV CUD variable remains widely used in large-scale studies(35, 43–47). In both surveys, the 22 CUD symptom items were mostly identical; the large differences in CUD prevalence across the two surveys could not be accounted for by the few slight differences in item wording(16, 35). For sensitivity analyses, we re-defined CUD, adding a cannabis withdrawal criterion and requiring 3 of 7 dependence criteria to be positive for CD. Cannabis withdrawal was assessed identically in NESARC and NESARC-III, as 3 or more of 5 withdrawal symptoms: nervousness/anxiety, sleep difficulty, depressed mood, restlessness, or physical symptoms (one or more of headache, shakiness, sweating, abdominal pain, and fever); or use to avoid or relieve withdrawal symptoms. This was done because DSM-5 includes withdrawal as a CUD criterion, due to evidence showing its validity and relatedness to the other CUD criteria(42). For consistency across predictors and clarity of interpretation, the reference group for all four predictors was no past-year non-medical use.

Control covariates included gender; age (18-29; 30-44; 45-64; ≥65 years); race/ethnicity (Hispanic; Non-Hispanic: White; Black; and Other [Native American, Asian, Pacific Islander]); education (<high school; high school graduate or GED; ≥some college); and urbanicity (urban, rural). Dichotomous variables were constructed for alcohol, tobacco, and stimulant use, indicating past-year use (yes/no), as these substances are potential confounders of the examined associations(48, 49). In sensitivity analyses, we included a covariate indicating if respondents’ states of residence had medical cannabis laws (MCL), as determined by economic and legal experts as in previous papers(35, 44, 50). The MCL variable was defined with three levels: never-MCL; MCL enacted by 2001 (NESARC); and MCL enacted between 2002 and 2012 (NESARC-III). Seven states (California, Colorado, Hawaii, Maine, Nevada, Oregon, Washington) had MCL by 2001. Nine more states (Arizona, Connecticut, Maryland, Massachusetts, Michigan, Montana, New Jersey, New Mexico, Vermont) had MCL by 2012.

Statistical Analysis

As in other studies evaluating trends between the two surveys(16, 35–37), the NESARC and NESARC-III datasets were concatenated, adding a survey variable. To determine the change in SRP over time, logistic regression was used to model SRP as a function of survey (time) and sociodemographic control variables (age, race/ethnicity, gender, education, and urbanicity). A second model also controlled for past-year alcohol, tobacco, and stimulant use. Model-predicted standardized prevalence of SRP (i.e., back-transformed from the log scale with sociodemographic characteristics averaged between the surveys) was estimated for each of the two surveys, and the difference between the two prevalence estimates indicated the change over time.

Logistic regression was then used to evaluate the association of each cannabis-related predictor variable with SRP, modeling SRP as a function of the cannabis-related predictor, survey, cannabis-related predictor x survey interactions, and sociodemographic control variables. Model-predicted standardized prevalence of SRP was estimated in each survey by cannabis-related predictor status (yes vs. no). The difference in these prevalence estimates indicated association of the cannabis-related predictor with SRP within each survey. Whether the associations differed between the surveys (i.e., changed over time) was indicated by contrasts between these prevalence differences (difference-in-difference tests). Additive effects and interactions were evaluated because those are considered most appropriate from the public health perspective (51–53), since additive effects can indicate groups with the greatest population level risk (44). For readers more familiar with odds ratios, we also evaluated effects and interactions on the multiplicative scale. Using the logistic regression models described above, ratios of the odds of SRP in those with the cannabis-predictor divided by odds in those without was estimated within each survey. Multiplicative interaction was evaluated as the ratio of odds ratios, i.e., odds ratio for 2012-2013 divided by the odds ratio for 2001-2002, given by the estimated model coefficient for the multiplicative interaction term.

For all analysis, SUDAAN 11.0.1(54) was used, incorporating survey weights to adjust for the complex sampling design, to yield US adult population-representative estimates. Statistical tests were 2-tailed with significance based on p<0.05, as indicated by 95% confidence intervals (CI). Interpretation of the 95% CI differs for difference (additive) and relative (multiplicative) effects. For difference effects, a value of 0.0 indicates no difference, so an estimate with 95% CI not including 0.0 is statistically significant at p<0.05. For relative effects, a value of 1.0 indicates no difference, so an estimate whose 95% CI do not include 1.0 is statistically significant at p<0.05.

Two sensitivity analyses were conducted. First, to reflect the addition of cannabis withdrawal to the DSM-5, we added cannabis withdrawal to the dependence criteria and re-ran the models for CUD. Second, we added a covariate indicating state MCL status at the time of each survey and re-ran the models. Participants from the 42 states included in both surveys were included in this analysis (41,706 from NESARC; 36,309 from NESARC-III; total=78,015), as in previous studies(35).

RESULTS

Trend in self-reported psychotic disorders

The standardized prevalence of past-year SRP among US adults was 0.33% in 2001-2002 and 0.80% in 2012-2013, a significant increase of +0.47% (95% CI=0.33, 0.61); odds ratio of 2.46 (95% CI=1.89, 3.22) in 2012-2013 compared to 2001-2002. In an adjusted model, further controlling for past-year alcohol, tobacco, and stimulant use, results were essentially the same as the original model. Adjusted prevalence of SRP in 2001-2002 was 0.32% and in 2012-2013 was 0.79%, with a prevalence difference of +0.47 (95% CI= 0.32, 0.62), suggesting that the change observed was not primarily driven by alcohol, tobacco, or stimulant use.

Within-survey association of cannabis predictors with self-reported psychotic disorders

Any past-year non-medical cannabis use

Self-reported psychotic disorders were more prevalent among participants with non-medical cannabis use compared to non-users in 2001-2002 (1.65% vs 0.27%, prevalence difference [PD]=1.38, 95% CI 0.47,2.29) and in 2012-2013 (1.89% vs. 0.68%, PD=1.21, 95% CI 0.56,1.86; Tables 1, 2).

Table 1.

Standardized prevalences and standard errors of self-reported psychotic disorders (SRP) by survey and by past-year cannabis variables.

	2001-2002 (N=43,093)			2012-2013 (N=36,309)
	Prevalencea	n	SE a	Prevalencea	n	SE a
Total samples (N=178; 337)b	0.33	178	0.03	0.80	337	0.06
Past year non-medical cannabis use groups c
Without any cannabis use (N=41,490; 32,608)	0.27	151	0.03	0.68	271	0.06
With any cannabis use (N=1,603; 3,701)	1.65	27	0.47	1.89	66	0.33
With frequent cannabis use (N=465; 1,527)	1.00	6	0.46	2.79	39	0.62
With daily/near-daily cannabis use (N=348; 1,161)	0.88	5	0.43	2.52	28	0.66
With CUD (N=560; 1,086)	2.55	11	1.08	3.38	29	0.85
With proxy for DSM-5 CUDd (N=565; 1,104)	2.80	12	1.11	3.33	29	0.84

Note: non-medical cannabis use: without a prescription or other than prescribed, e.g., to get high.

^aStandardized prevalence are model-predicted prevalence of self-reported psychotic disorders, adjusted for sociodemographic covariates (age, gender, race/ethnicity, education level, and urbanicity), from logistic regression.

^bN’s indicate number of cases of self-reported psychotic disorders in each survey (2001-2002, 2012-2013).

^cN’s in parentheses indicate overall survey n’s in 2001-2002, 2012-2013.

^dProxy for cannabis withdrawal syndrome (3 or more of 5 symptoms: nervousness/anxiety; sleep disturbances; restlessness; depressed mood; and any physical symptoms: sweating/fast heartbeat; fever; shaking; nausea, vomiting, stomach pain; or headache) or using cannabis to avoid withdrawal symptoms.

Table 2.

Association of past-year cannabis variables with self-reported psychotic disorders (SRP) within surveys: prevalence differences and odds ratios

	2001-2002 (NESARC)				2012-2013 (NESARC-III)
	Prevalence Difference (PD)a	PD 95% CIa	Odds Ratio (OR)b	OR 95% CIb	Prevalence Difference (PD)a	PD 95% CIa	Odds Ratio (OR)b	OR 95% CIb
With any cannabis use	1.38	0.47, 2.29	6.16	3.41, 11.01	1.21	0.56, 1.86	2.83	1.92, 4.17
With frequent cannabis use	0.73	−0.15, 1.61	3.70	1.57, 8.69	2.11	0.89, 3.33	4.25	2.63, 6.87
With daily/near-daily cannabis use	0.61	−0.23, 1.45	3.26	1.25, 8.49	1.84	0.55, 3.13	3.82	2.21, 6.59
With CUD	2.28	0.18, 4.38	9.60	4.10, 22.58	2.70	1.03, 4.37	5.19	3.03, 8.89
With proxy DSM-5 CUD c	2.53	0.35, 4.71	10.64	4.77, 23.71	2.65	1.00, 4.30	5.12	2.98, 8.79

Note: non-medical cannabis use: without a prescription or other than prescribed, e.g., to get high.

^aEffects estimated on the additive scale: prevalence difference indicates the prevalence difference of psychotic disorders between those with the cannabis predictor and those without cannabis use in 2001-2002 and 2012-2103. These effects are significant when the 95% confidence interval (CI) does not include 0.

^bEffects estimated on the multiplicative scale: odds ratio indicates the ratio of the odds of psychotic disorders between those with and without the cannabis predictor in 2001-2002 and 2012-2103. These effects are significant when the 95% CI does not include 1.

^cProxy for cannabis withdrawal syndrome (3 or more of 5 symptoms: nervousness/anxiety; sleep disturbances; restlessness; depressed mood; and any physical symptoms: sweating/fast heartbeat; fever; shaking; nausea, vomiting, stomach pain; or headache) or using cannabis to avoid withdrawal symptoms.

Frequent non-medical cannabis use

Self-reported psychotic disorders were more prevalent among participants with frequent non-medical cannabis use compared to non-users in 2012-2013 (2.79% vs. 0.68%, PD=2.11, 95% CI 0.89, 3.33; Tables 1,2), but not in 2001-2002.

Daily/near-daily non-medical cannabis use

Self-reported psychotic disorders were more prevalent among participants with daily/near-daily non-medical cannabis use compared to non-users in 2012-2013 (2.52% vs. 0.68%, PD=1.84, 95% CI 0.55, 3.13; Tables 1,2), but not in 2001-2002.

DSM-IV cannabis use disorder (CUD)

Self-reported psychotic disorders were more prevalent among participants with DSM-IV CUD compared to nonusers in 2001-2002 (2.55% vs 0.27%, prevalence difference [PD]=2.28, 95% CI 0.18, 4.38) and in 2012-2013 (3.38% vs. 0.68%, PD=2.70, 95% CI 1.03, 4.37; Tables 1,2).

Relative scale

On the relative scale, any past year non-medical cannabis use, frequent use, daily/near-daily use, and DSM-IV CUD were all significantly associated with SRP in both 2001-2002 and 2012-2013 (Table 2). Except for any past year non-medical cannabis use, there were no changes in magnitude of association over time (Table 3). When withdrawal was added, CUD remained significantly associated with SRP in both time periods (Table 2), with no significant differences in the association strength (Table 3).

Table 3.

Association of past-year cannabis variables with self-reported psychotic disorders (SRP), by survey: differences and comparisons across survey years

Cannabis Outcome	Difference in Prevalence Differences (PD) 2012-2013 vs. 2001-2002a		Ratio of Odds Ratios (OR) 2012-2013 vs. 2001-2002b
	Difference in PDs	95% CI	Ratio of ORs	95% CI
With any cannabis use	−0.17	−1.24, 0.90	0.46	0.24, 0.90
With frequent cannabis use	1.38	−0.09, 2.85	1.15	0.44, 3.00
With daily/near-daily cannabis use	1.23	−0.28, 2.74	1.17	0.40, 3.47
With CUD	0.42	−2.15, 2.99	0.54	0.21, 1.41
With proxy DSM-5 CUD c	0.12	−2.47, 2.71	0.48	0.19, 1.21

Note: non-medical cannabis use: without a prescription or other than prescribed, e.g., to get high.

^aEffects estimated on the additive scale: prevalence difference indicates the prevalence difference of psychotic disorders between those with and without the cannabis predictor in 2001-2002 and 2012-2103, while difference in prevalence differences indicates the difference between those differences. These effects are significant when the 95% confidence interval (CI) does not include 0.

^bEffects estimated on the multiplicative scale: odds ratio indicates the ratio of the odds (likelihood) of psychotic disorders between those with and without the cannabis predictor in 2001-2002 and 2012-2103, while ratio of odds ratios indicates the ratio between those ratios. These effects are significant when the 95% CI does not include 1.

^cProxy for cannabis withdrawal syndrome (3 or more of 5 symptoms: nervousness/anxiety; sleep disturbances; restlessness; depressed mood; and any physical symptoms: sweating/fast heartbeat; fever; shaking; nausea, vomiting, stomach pain; or headache) or using cannabis to avoid withdrawal symptoms.

Between-survey change in strength of associations

Although frequent and daily-near daily use was not associated with SRP in 2001-2002 but was associated in 2012-2013, none of the difference-in-difference tests indicating between-survey change in the strength of the associations were significant (Table 3).

Sensitivity analyses

After adding cannabis withdrawal to the dependence criteria, CUD remained associated with SRP in both time periods (Table 2). When adding MCL as a covariate, results were similar to those from the original models (Supplemental Tables 1,2,3).

DISCUSSION

The current study examined associations between several cannabis use indicators and self-reported psychotic disorders (SRP), and changes over time in these associations, in the general adult US population. In recent decades the US cannabis landscape has shifted substantially, including increased public perception of cannabis as a safe substance and increasing state cannabis legalization. Although the nature of the cannabis-psychosis relationship has been debated, cannabis use is widely considered to play a partial role in the risk of psychosis(27, 28). Thus, investigating changes in associations between psychotic disorders and cannabis use indicators over time is warranted. The current study shows that the prevalence of SRP increased among US adults between 2001- 2002 and 2012-2013. Results demonstrate that all non-medical cannabis use indicators were associated with SRP in 2012-2013. Further, any non-medical cannabis use and CUD were associated with SRP in both 2001-2002 and 2012-2013. Nevertheless, none of these associations were shown to significantly change across survey years.

Our finding that the prevalence of past-year SRP increased significantly between 2001-2002 and 2012-2013 is the first reported change in prevalence of self-reported psychotic disorders based on large-scale, nationally representative samples of US adults. This finding contrasts with earlier studies based on hospitalization records, whose methods of recording may be imprecise and variable over time. The present study adds to the literature by providing evidence that psychotic disorders have been on the rise in the US in recent decades based on comparison of prevalence of SRP between two national surveys that used identical measures of psychosis.

Our finding that self-reported psychotic disorders were significantly more prevalent among any past-year cannabis users compared to non-users in both surveys is consistent with results from past studies(55–58) and adds to the literature by reporting standardized prevalences of psychotic disorders among past-year adult cannabis users. These findings can contribute to information for clinicians and policymakers of the increased likelihood of psychosis among these individuals. In addition, SRP was significantly associated with frequent and daily/near-daily cannabis use in the more recent survey, supporting previous findings on a dose-response relationship between cannabis use and psychotic disorders(59), which should be further investigated. While not possible with the available data, a study design that would allow assessment of a true dose-response relationship as a function of a more fine-grained measure of cannabis use frequency and quantity would shed further light on the matter. While none of these associations significantly changed across survey years on the absolute difference scale, on the relative scale, the odds of SRP among any past-year non-medical cannabis users was significantly weaker in 2012-2013 (OR=2.83) than 2001-2002 (OR=6.16). One possible explanation for the weaker OR in the more recent survey could be the higher proportion of non-frequent cannabis users among all users in 2012-2013 (5.84%) than in 2001-2002 (2.86%). Changing marijuana norms (e.g., decreased perception of marijuana use as risky) may have led to more experimental, one or two-time users in 2012-2013, who are less likely to be diagnosed with psychotic disorders compared to regular and frequent marijuana users, as indicated in numerous studies(22, 60–62).

Study findings indicate that participants with CUD are at increased risk of reporting being diagnosed with a psychotic disorder compared to non-cannabis users, a finding that is also reflected in previous non-US studies(25, 26). Notably, the highest absolute prevalence of self-reported psychotic disorders in this study (3.38%) was seen in past-year cannabis users reporting DSM-IV CUD in the 2012-2013 survey. Findings from sensitivity analyses show that CUD with withdrawal (a combination that is closer to the DSM-5 diagnostic criteria for CUD) was associated with self-reported psychotic disorders in both surveys. Although differences in associations across surveys were not significant, one plausible explanation for the high rates of self-reported psychotic disorders among those with CUD in 2012-2013 is the increase in availability of high-potency cannabis products, which have been associated with higher prevalence of psychosis(59, 63). In sensitivity analyses, including medical cannabis laws (MCL) in the model did not change the associations between cannabis use variables and psychosis over time. This could be because MCL may not be related to SRP. However, early evidence suggests that recreational cannabis laws (RCL) are expected to have a stronger effect in increasing cannabis use and associated problems than MCL(64). Future studies that examine RCL effects are warranted and may be highly valuable in informing policy makers, clinicians and researchers about increases in risk associated with RCL.

Study limitations are noted. First, self-reported psychotic disorders were indicated by a single item rather than physician assessment, as in a previous NESARC study(31). While future national studies of substance use should measure psychotic disorders more extensively, a growing number of studies have explored the validity and reliability of various self-reported measures of psychotic disorders, including the current study’s measure, and reported prevalences that are similar to studies using clinical diagnoses(41, 65, 66). Furthermore, compared to other large-scale national surveys, such as the National Survey on Drug Use and Health (NSDUH), which included self-reported psychosis in a broad measure of “severe mental illness”, NESARC is the only national epidemiologic study that utilized a distinct psychosis variable. Second, cannabis use variables could be subject to social desirability bias, since they were based on self-report(34). Further, the present study did not address self-reported psychotic disorders among individuals using marijuana exclusively for medical purposes. The NESARC did not include a question about medical use of marijuana, precluding examination of this question in NESARC data. While the NESARC-III did include such a question, very few NESARC-III participants (weighted percent, 0.22%, SE=0.04) used marijuana for medical purposes only and did not also use marijuana non-medically(67), and those who used exclusively for medical reasons were not asked about frequency of use or CUD criteria. Given the small numbers of medical-only users, their omission seems unlikely to have altered the relationships found. However, when relevant data become available, future studies should address changes in the association of psychotic disorders with cannabis variables over time among those using marijuana exclusively for medical purposes. Further, this study did not examine negative control psychiatric conditions (i.e., those unrelated to cannabis use such as autism or obsessive-compulsive disorder) because the data were unavailable, but future studies should do so. Third, directionality of the relationship cannot be determined in cross-sectional data. Additionally, since DSM-IV mental disorders were diagnosed in NESARC, while DSM-5 diagnoses were made in NESARC-III, we could not adjust for the presence of other psychiatric disorders. If national data with consistent DSM or ICD mental disorder diagnoses over time can be found, studies should explore such adjustments. This also meant that DSM-5 CUD could not be assessed in both surveys. However, an extensive literature(42) shows that the criteria for DSM-IV cannabis abuse and dependence are unidimensional, justifying their combination as is done in many other studies since then, and that cannabis disorder diagnoses in DSM-IV correspond closely with DSM-5 CUD(68). Fourth, the NESARC and NESARC-III survey items about psychosis did not differentiate between types of psychotic disorders. Therefore, we could not account for time trends in specific disorders or differentiate between primary and secondary psychotic disorders. Future studies should account for specific types of psychotic disorders. Additionally, considering increasing rates of cannabis use in women in recent years(69, 70) and, conversely, higher rates of psychosis among men compared to women(12) , associations reported in the current study may have differed by sex. Examining effect modification by sex was beyond the scope of the current study but should be addressed in future studies. Finally, the NESARC and NESARC-III were surveys of household residents that did not include medically institutionalized participants (perhaps less likely than the general population to use cannabis), or incarcerated participants (more likely to use cannabis and often mentally ill). Thus, study results are not generalizable to these populations.

Conclusions

The prevalence of self-reported psychotic disorders in the adult US population has significantly increased from 2001-2002 to 2012-2013. Non-medical cannabis use and cannabis use disorder were significantly associated with self-reported psychotic disorders, specifically in more recent times. The increasing perception of cannabis as a harmless substance may deter the general public and also healthcare providers from recognizing that non-medical cannabis use may play a role in exacerbating risk for psychotic disorders. Therefore, increasing public knowledge and educating providers about this risk may serve a useful function. In particular, identifying cannabis use disorders may help identify individuals at increased risk of psychotic disorders. This information can inform clinicians and addiction specialists about the need for evaluation and appropriate interventions and therapeutic modalities for individuals at risk. Further, although not directly examined in the current study, policy makers should be aware of the increase in cannabis use and CUD among US adults, and any possible subsequent rise in cannabis-related outcomes, such as psychotic disorders.