Marijuana use and use disorders in adults in the USA, : analysis of annual cross-sectional surveys

Transcription

1 Marijuana use and use disorders in adults in the USA, : analysis of annual cross-sectional surveys Wilson M Compton, Beth Han, Christopher M Jones, Carlos Blanco, Arthur Hughes Summary Background The study of marijuana use disorders is urgently needed because of increasing marijuana legalisation in multiple jurisdictions, the effect of marijuana use on future risk of psychiatric disorders, and deleterious effects of marijuana exposure. Thus, understanding trends of marijuana use and use disorders and examining factors that might drive these trends (eg, perceptions of harms from marijuana use) is essential. Methods We analysed data from US civilians aged 18 years or older who participated in annual, cross-sectional US National Surveys on Drug Use and Health from 2002 to The sample in each US state was designed to be approximately equally distributed between participants aged years, years, and 26 years or older. For each survey year, we estimated prevalence of marijuana use and use disorders, initiation of marijuana use, daily or near daily use, perception of great or no risk of harm from smoking marijuana, perception of state legalisation of medical marijuana use, and mean number of days of marijuana use in the previous year. Descriptive analyses, multivariable logistic regressions, and zerotruncated negative binomial regressions were applied. Findings adults participated in the surveys. Marijuana use increased from 10 4% (95% CI ) to 13 3% ( ) in adults in the USA from 2002 to 2014 (β=0 0252, ), and the prevalence of perceiving great risk of harm from smoking marijuana once or twice a week decreased from 50 4% ( ) to 33 3% ( ; β= , ). Changes in marijuana use and risk perception generally began in After adjusting for all covariates, changes in risk perceptions were associated with changes in prevalence of marijuana use, as seen in the lower prevalence of marijuana use each year during than in 2002 when perceiving risk of harm from smoking marijuana was included in models. However, marijuana use disorders in adults remained stable at about between 2002 and 2014 (β= , p=0 22). Interpretation Prevalence and frequency of marijuana use increased in adults in the USA starting in approximately 2007 and showing significantly higher results in multivariable models during (compared with 2002). The associations between increases in marijuana use and decreases in perceiving great risk of harm from smoking marijuana suggest the need for education regarding the risk of smoking marijuana and prevention messages. Lancet Psychiatry 2016 Published Online August 31, S (16) See Online/Comment S (16)30270-X National Institute on Drug Abuse, National Institutes of Health, Bethesda, MD, USA (W M Compton MD, C Blanco MD); Substance Abuse and Mental Health Services Administration, Rockville, MD, USA (B Han MD, A Hughes MS); and Office of the Assistant Secretary for Planning and Evaluation, US Department of Health and Human Services, Washington, DC, USA (C M Jones PharmD) Correspondence to: Dr Wilson M Compton, National Institute on Drug Abuse, National Institutes of Health, Bethesda, MD 20892, USA wcompton@nida.nih.gov Funding None. Introduction Laws and policies related to marijuana have been changing markedly in the USA over the past 20 years. Medical marijuana legalisation had been adopted by 24 states and the District of Columbia by In addition to medical marijuana legalisation, jurisdictions have adopted full legalisation for non-medical marijuana use in several states and the District of Columbia. 2 Many other countries (eg, Canada, India, Mexico, Spain, Germany, and the Netherlands) have decriminalised the possession of small quantities of marijuana. Understanding patterns of marijuana use and use disorders in adults in the USA and how they have changed over time is essential in this evolving legal landscape. Using two waves of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC), authors of a 2015 study found that marijuana use and use disorders doubled in the USA from to , 2 which could be explained partly by methodological differences in the NESARC waves. 3 Furthermore, none of the existing studies 2,3 examined when changes in marijuana use began and what factors (eg, risk perception of marijuana use) might drive the trends. To better inform ongoing marijuana policy discussions and help clinicians provide optimal care for patients using marijuana, this study aimed to examine trends in marijuana use and use disorders in the USA and assess factors associated with these trends. Methods Data sources We examined data from adults (aged 18 years or older) who participated in the National Survey on Drug Use and Health (NSDUH) undertaken by the Substance Abuse and Mental Health Services Administration (SAMHSA). NSDUH provides nationally and state representative data on marijuana use and use disorders in the US civilian population aged 12 years or older who are not living in institutions. The sample in each state was designed to be approximately equally distributed among three age groups, years, years, and 26 years or Published online August 31,

2 Research in context Evidence before this study We searched PubMed with the search terms marijuana trends in adults in the United States, cannabis trends in adults in the United States, marijuana use disorders in adults in the United States, cannabis use disorders in adults in the United States, trends in marijuana use disorders, and trends in cannabis use disorders, between April 1, 2011, and March 31, 2016, for studies of trends in marijuana use and disorders in the general US population and in attitudes towards marijuana. One study using the National Survey on Drug Use and Health (NSDUH) documented increasing overlap marijuana and tobacco use between 2003 and Other work with the NSDUH documented decreasing perceived harmfulness of use of marijuana among adults, and multiple studies have documented decreasing perceived harmfulness and disapproval of marijuana use among youth and young adults. A study of two waves of the National Epidemiologic Survey on Alcohol and Related Conditions found that marijuana use and use disorders doubled in the USA from to with an additional paper describing differences in the methods used in these two waves of data collection. None of the existing studies investigated when changes in marijuana use began. Added value of this study Only the NSDUH has annual information on drug use and drug use disorders for the overall US adult (aged 18 years and older) population, which is the source of data for this paper. These yearly data allowed us to examine when changes in the trends started. In addition to examining trends using yearly repeated cross-sectional surveys with large samples, we also examined how a reductions in risk perceptions were associated with changes in marijuana use, frequent marijuana use, and marijuana use disorders. Although changes in perceived risk of harm have previously been shown to be important predictors of adolescent marijuana trends, no previous research has examined this relationship in adults. Implications of all the available evidence Clinicians, researchers, and policy makers need to be aware that more adults in the USA are using marijuana in recent years than in 2002 and that the frequency of use has increased as well. Prevention efforts might need to target the reduction in perceived harm of using marijuana. Future research needs to examine the impact of the increased rates of marijuana use and to continue to examine how these trends might be related to the change in legal status of marijuana in the USA. older. The NSDUH data collection protocol was approved by the Institutional Review Board at RTI International. Details of the NSDUH methods are provided elsewhere. 4 NSDUH has had consistent survey methodology since 2002 and large sample sizes, 3,4 allowing for valid examination of marijuana use trends across every year in NSDUH had a state-based design with an independent, multistage area probability sample within each state and the District of Columbia. The eight states with the largest population were designated as large sample states (California, Florida, Illinois, Michigan, New York, Ohio, Pennsylvania, and Texas) and had a sample size of about 3600 each. For the remaining 42 states and the District of Columbia, the sample size was about 900 per state. Data were collected by interviewers during inperson visits to households and non-institutional group quarters. The interviews lasted about one hour each. Audio computer-assisted, self-administered interviewing was used, providing respondents with a private and confidential way to record answers. Measures NSDUH collected data on use of tobacco, alcohol, marijuana, cocaine, hallucinogens, heroin, and inhalants in the previous 12 months, non-medical use of prescription opioids, sedatives, and stimulants in the previous 12 months, and heavy alcohol use in the previous 30 days (drinking five drinks or more on the same occasion on each of 5 days or more). NSDUH also asked respondents: How much do people risk harming themselves physically and in other ways when they smoke marijuana once or twice a week? (hereafter called perceived risk of harm from smoking marijuana), perceived state legalisation of medical marijuana use (whether respondents think that medical marijuana use is legalised in their residing state), perceived ease of marijuana availability, and age at first marijuana use. Daily or near daily users were people who had used marijuana in the previous year reporting use on an average of 5 days or more per week, 20 days or more per month, or 240 days or more in the previous 12 months. 4 Using state and year information, we created a variable indexing state legalisation of non-medical marijuana use. The NSDUH estimated major depressive episode (in the previous 12 months) and each specific substance use disorder (dependence on or abuse of alcohol or illicit drugs [eg, marijuana, cocaine, and hallucinogens]) on the basis of assessments of individual DSM-IV diagnostic criteria. 5 Adults were defined as having a major depressive episode if they had a period of 2 weeks or longer in the previous 12 months when they experienced depressed mood or loss of interest or pleasure in daily activities, and they had at least some additional symptoms (eg, problems with sleep, eating, energy, concentration, or self-worth). 6 Nicotine dependence in cigarette smokers was assessed with the Nicotine Dependence Syndrome Scale (range 1 5). 7 Sociodemographic characteristics included age, gender, race or ethnicity, educational attainment, employment status, marital status, health insurance, metropolitan statistical area, and census region. 2 Published online August 31,

3 β coefficients and p values for trends Sample size All adults Eligible adults Initiates Users Prevalence, weighted percentage Marijuana use in adults Initiation of marijuana use in eligible adults Marijuana use disorders in adults Marijuana use disorders in users Marijuana use disorders in noninitiate users Daily or near daily use in adults Daily or near daily use in users great risk, adults great risk, users 10 4% ( ) 0 7% ( ) ( ) 14 8% ( ) 15 1% ( ) 1 9% ( ) 18 0% ( ) 50 4% ( ) 8 8% ( ) 10 1% ( ) 0 6% ( ) ( ) 15 1% ( ) 15 3% ( ) 2 0% ( ) 19 6% ( ) 51 7% ( ) 9 3% ( ) 10 1% ( ) 0 7% ( ) 1 6% ( ) 16 0% ( ) 16 4% ( ) 2 0% ( ) 19 3% ( ) 51 4% ( ) 10 0% ( ) 10 1% ( ) 0 8% ( ) ( ) 14 5% ( ) 14 9% ( ) 2 0% ( ) 20 2% ( ) 50 2% ( ) 9 2%** ( ) 10 0% ( ) 0 7% ( ) ( ) 15 0% ( ) 15 4% ( ) 2 0% ( ) 19 5% ( ) 50 0% ( ) 8 8% ( ) 9 9% ( ) 0 7% ( ) 1 4% ( ) 14 4% ( ) 14 7% ( ) 2 1% ( ) 20 8% ( )** 49 5% ( ) 8 6% ( ) 10 1% ( ) 0 8% ( ) ( ) 14 9% ( ) 15 4% ( ) 2 2%** ( ) 21 9% ( )** 47 4% ( )** 7 1% ( )** 11 1%** ( ) 0 8%** ( ) ( ) 13 8% ( ) 14 0% ( ) 2 4%** ( ) 21 7%** ( ) 44 5%** ( ) 6 6%** ( ) 11 3%** ( ) 0 9%** ( ) 1 6% ( ) 14 1% ( ) 14 4% ( ) 2 7%** ( ) 23 7%** ( ) 42 0%** ( ) 5 3%** ( ) 11 3%** ( ) 1 0%** ( ) 1 4% ( ) 12 6%** ( ) 12 9%** ( ) 2 7%** ( ) 24 0%** ( ) 41 3%** ( ) 5 1%** ( ) 12 0%** ( ) 0 9%** ( ) ( ) 12 4%** ( ) 12 7%** ( ) 3 0%** ( ) 24 9%** ( ) 39 4%** ( ) 4 2%** ( ) 12 5%** ( ) 1 0%** ( ) ( ) 11 8%** ( ) 12 2%** ( ) 3 1%** ( ) 25 0%** ( ) 36 1%** ( ) 3 5%** ( ) 13 3%** ( ) 1 1%** ( ) ( ) 11 0%** ( ) 11 4%** ( ) 3 5%** ( ) 26 3%** ( ) 33 3%** ( ) 2 8%** ( ) Overall: β=0 0252, ; : β= , p=0 12; : β=0 0451, Overall: β=0 0398, ; no joinpoint identified during this period Overall: β= , p=0 22; no joinpoint identified during this period Overall: β= , ; : β= , p=0 53; : β= , Overall: β= , ; no joinpoint identified during this period Overall: β=0 0553, ; : β=0 0136, p=0 18; : β=0 0741, Overall: β=0 0389, ; no joinpoint identified during this period Overall: β= , ; : β= , p=0 0014; : β= , Overall: β= , ; : β= 0 011, p=0 46; : β= , (Table 1 continues on next page) Published online August 31,

4 β coefficients and p values for trends (Table continued from previous page) no risk, adults no risk, users availability, adults availability, users state legalisation for medical use, adults state legalisation for medical use, users 5 6% ( ) 24 8% ( ) 58 3% ( ) 89 4% ( ) 17 9% ( ) 28 9% ( ) 5 2% ( ) 23 0% ( ) 59 4% ( ) 88 6% ( ) 17 7% ( ) 26 8% ( ) 5 1% ( ) 22 9% ( ) 58 2% ( ) 88 5% ( ) 17 5% ( ) 27 6% ( ) 5 4% ( ) 25 5% ( ) 58 0% (57,13 58,89) 87 4%** ( ) 17 2% ( ) 27 7% ( ) 5 7% ( ) 25 3% ( ) 57 9% ( ) 87 9% ( ) 18 2% ( ) 28 9% ( ) Weighted mean number of days of marijuana use in the previous 12 months Adults 10 0 ( ) Users 97 9 ( ) 10 1 ( ) ( ) 10 2 ( ) ( ) 10 5 ( ) 105 4** ( ) 10 2 ( ) ( ) 6 0% ( ) 27 7%** ( ) 58 4% ( ) 87 9% ( ) 19 6%** ( ) 31 8% ( ) 10 3 ( ) 106 4** ( ) 6 9%** ( ) 30 0%** ( ) 56 8%** ( ) 89 6% ( ) 21 3%** ( ) 33 7%** ( ) 10 9** ( ) 110 4** ( ) 8 3%** ( ) 34 2%** ( ) 56 9% ** ( ) 87 8% ( ) 23 7%** ( ) 34 8%** ( ) 11 9** ( ) 108 4** ( ) 9 6%** ( ) 39 9% ** ( ) 58 0% ( ) 89 4% ( ) 25 3%** ( ) 37 1%** ( ) 12 9** ( ) 115 8** ( ) 10 1%** ( ) 40 5%** ( ) 58 6% ( ) 89 4% ( ) 27 4%** ( ) 37 7%** ( ) 13 0** ( ) 117 6** ( ) 11 6%** ( ) 43 5%** ( ) 58 5% ( ) 89 5% ( ) 28 3%** ( ) 37 0%** ( ) 14 2** ( ) 121 0** ( ) 13 0%** ( ) 47 0%** ( ) 59 7%** ( ) 89 5% ( ) 30 9%** ( ) 4** ( ) 14 8** ( ) 120 3** ( ) 15 1%** ( ) 49 5%** ( ) 6** ( ) 90 1% ( ) 32 6%** ( ) 43 8%** ( ) 16 3** ( ) 124 9** ( ) Overall: β=0 1104, ; : β=0 0093, p=0 38; : β=0 1372, Overall: β=0 1105, ; : β=0 0181, p=0 22; : β=0 1690, ; : β=0 1045, Overall: β=0 0054, ; : β= , ; : β=0 0331, Overall: β=0 0128, p=0 0038; no joinpoint identified during this period Overall: β=0 0786, ; : β= , p=0 23; : β=0 0916, Overall: β=0 0627, ; no joinpoint identified during this period Overall: β=0 5106, ; : β=0 0660, p=0 38; : β=0 8093, Overall: β=2 1741, ; no joinpoint identified during this period Bivariable logistic regression models were applied for testing differences in percentage estimates in the table. Linear regression models were applied for testing differences in the mean numbers of days of marijuana use in the table. *Data source: Substance Abuse and Mental Health Services Administration s National Survey on Drug Use and Health data. Substance Abuse and Mental Health Services Administration requires that any description of overall sample sizes based on the restricted-use data files has to be rounded to the nearest 100 to minimise potential disclosure risk. Aged 18 years or older in the USA. Denominator used for calculating initiation of marijuana use; people who never used marijuana before the previous 12 months. People who used marijuana for the first time in their lives in the previous 12 months. Adults aged 18 years or older who had used marijuana in the previous year in the USA. Joinpoint identified in this year. **p<0 05 compared with the 2002 estimate (the reference year). Use of marijuana on 5 days or more each week, 20 days or more each month, or 240 days or more in the previous 12 months. a great risk associated with smoking marijuana once or twice a week. no risk associated with smoking marijuana once or twice a week. availability (fairly easy or easy to obtain marijuana). legalisation of marijuana for medical use in residing state. Table 1: Prevalence of marijuana use; marijuana use disorders; and perceptions of risk, availability, and state legalisation for medical use in the USA, * 4 Published online August 31,

5 Statistical analysis For each year of the survey, we estimated prevalence of marijuana use and use disorders, initiation of marijuana use in eligible adults (ie, people who had never used marijuana before the previous 12 months), daily or near daily use, perception of great or no risk of harm from smoking marijuana, perception of state legalisation of medical marijuana use, and mean number of days of marijuana use in the previous year. Also, applying the prevalence data to the US census-derived adult population, we estimated trends in the number of marijuana initiates, the number of marijuana users, and the number of users with daily or near daily marijuana use during this time period. We applied bivariable logistic regression models to estimate percentage prevalence, to test for differences between estimates for 2002 and each year in , and to test p values of β coefficients of the year variable. For mean numbers of days of marijuana use, we applied linear regression models to examine differences between estimates for 2002 and each year in and to test p values of β coefficients of the year variable. Importantly, we identified joinpoints indicating significant changes in non-linear trends using a Monte Carlo permutation method 8 and estimated β coefficients and p values for each segment separated by a joinpoint using segmented regression analyses. We applied bivariable and multivariable logistic regressions to assess unadjusted and adjusted relative risks for marijuana use in adults and for marijuana use disorders in marijuana users. Because data for major depressive episode were not collected in the NSDUH, 6 separate multivariable models were applied for with this additional variable included and for without the variable added. SUDAAN (version ) software was used for all these analyses to account for the complex sample design. We applied bivariable and multivariable zero-truncated negative binomial regression models to examine factors associated with number of days of marijuana use among users. Stata (version 13) was used to do the zerotruncated negative binomial regression to account for the complex sample design and sample weights. Multicollinearity and potential interaction effects between examined factors were assessed and were not identified in final multivariable models. During modelling, we used similar hierarchical multiple regression approaches specifying a fixed order of entry for variables (sociodemographic factors, then substance use or use disorders and related factors, major depressive episode, and perceptions of marijuana use) to test the effects of certain predictors independent of the influence of others and to identify factors that might be associated with changes in these trends. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BH had full access to all the data in the study. The corresponding author had final responsibility for the decision to submit for publication. Results The annual mean weighted response rate of the NSDUH was 66 0% (SD 3 72). Based on people aged 18 years or older sampled in the NSDUH, the prevalence of marijuana use in adults increased from 10 4% in 2002 to 13 3% in 2014; this upward trend started in 2007 (β=0 0451, ; table 1, figure). The proportion of eligible adults who initiated marijuana use Marijuana use in adults (%) Mean number of days of marijuana use in adults Perception of no risk of harm from marijuana use in adults (%) Daily or near daily marijuana use in adults (%) Marijuana use disorders in adults (%) Prevalence (%) Days Year 0 Figure: Trends in marijuana use patterns, marijuana use disorders, and perceived risk of harm Annual prevalence and trends in any marijuana use, daily or near daily marijuana use, marijuana use disorders, mean number of days of marijuana use, and perception of no risk of harm from marijuana use in adults in the USA. *Joinpoints indicate significant changes in non-linear trends. Published online August 31,

6 Bivariable model without controlling for any covariates, unadjusted RR Multivariable model without perceived risk of smoking marijuana, but with other covariates, adjusted RR Multivariable model with perceived risk of smoking marijuana and other covariates, adjusted RR Year ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 0 9 ( ) 0 9 ( ) ( ) 0 9 ( ) 0 9 ( ) ( ) 1 0 ( ) 0 9 ( ) ( ) 1 0 ( ) 0 9 ( ) ( ) 1 0 ( ) 0 9 ( ) ( ) 1 1 ( ) 0 9 ( ) ( ) 1 1 ( ) 0 9 ( ) ( ) 1 1 ( ) 0 9 ( ) ( ) 1 2 ( ) 0 9 ( ) risk of smoking marijuana once or twice per week No risk Slight risk ( ) Moderate ( ) risk Great risk ( ) Unspecified ( ) RR=relative risk. *Substance Abuse and Mental Health Services Administration requires that any description of overall sample sizes based on the restricted-use data files has to be rounded to the nearest 100 to minimise potential disclosure risk. Each multivariable logistic regression model adjusted for age, sex, race or ethnicity, education, employment, marital status, health insurance, metropolitan statistical area, region, tobacco use, heavy alcohol use, cocaine use, hallucinogen use, heroin use, inhalant use, non-medical use of prescription opioids, non-medical sedative use, non-medical stimulant use, age at first marijuana use, state legalisation of non-medical marijuana use, perceived state legalisation of medical marijuana use, and perceived marijuana availability. A major depressive episode in the past year was not associated with marijuana use and was excluded from the final models. We used similar hierarchical multiple regression approach specifying a fixed order of entry for variables (sociodemographic factors, then substance use and related factors, major depressive episode, finally perceptions of marijuana use) to test the effects of certain predictors independent of the influence of others and to identify factors that might be associated with changes in trends. Reference group. Prevalence significantly differs from the prevalence of the reference group. Table 2: 12-month unadjusted and adjusted RR of marijuana use in adults in the USA, (n= )* Bivariable model without controlling for any covariates, unadjusted RR Multivariable model without controlling for perceived risk of marijuana use, but adjusting for other covariates, adjusted RR Multivariable model controlling for perceived risk of marijuana use and other covariates, adjusted RR Year ( ) 1 0 ( ) 1 0 ( ) ( ) 1 1 ( ) 1 1 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 0 9 ( ) 0 9 ( ) ( ) 0 9 ( ) 0 9 ( ) ( ) 0 8 ( ) 0 9 ( ) ( ) 0 8 ( ) 0 8 ( ) ( ) 0 8 ( ) 0 8 ( ) ( ) 0 8 ( ) 0 8 ( ) risk of smoking marijuana once or twice per week No risk Slight risk ( ) Moderate ( ) risk Great risk ( ) Unspecified ( ) RR=relative risk. *Substance Abuse and Mental Health Services Administration requires that any description of overall sample sizes based on the restricted-use data files has to be rounded to the nearest 100 to minimise potential disclosure risk. Each multivariable logistic regression model adjusted for age, gender, race or ethnicity, education, employment status, marital status, health insurance, metropolitan statistical area, region, nicotine dependence, alcohol use disorders, cocaine use disorders, hallucinogen use disorders, heroin use disorders, inhalant use disorders, prescription opioid use disorders, sedative use disorders, stimulant use disorders, age at first marijuana use, state legalisation of non-medical marijuana use, perceived state legalisation of medical marijuana use, perceived marijuana availability, source of marijuana, and major depressive episode. We used similar hierarchical multiple regression approach specifying a fixed order of entry for variables (sociodemographic factors, then substance use disorders and related factors, major depressive episode, finally perceptions of marijuana use) to test the effects of certain predictors independent of the influence of others and to identify factors that might be associated with changes in trends. Reference group. Prevalence significantly differs from the prevalence of the reference group. Table 3: 12-month unadjusted and adjusted RR of marijuana use disorders in adult users in the USA, (n= )* for the first time increased from 0 7% in 2002 to 1 1% in 2014 (β=0 0398, ). The prevalence of perceiving great risk of harm from smoking marijuana in all adults decreased from 50 4% in 2002 to 33 3% in 2014 with a reduction starting in 2003 and accelerated reduction since 2007 ( : β= , p=0 0014; : β= , ). The prevalence of daily or near daily use of marijuana increased from 1 9% in 2002 to 3 5% in 2014, with an upward trend starting in 2007 (β=0 0741, ). The prevalence of perceiving state legalisation of medical marijuana use increased from 17 9% in 2002 to 32 6% in 2014, with an upward trend starting in 2004 (β=0 0916, ). However, the prevalence of marijuana use disorders in adults remained stable at about between 2002 and 2014 (β= , p=0 22). The number of adults who first used marijuana increased from in 2002 (95% CI ) to 1 4 million in 2014 (1 2 million 1 5 million). The 6 Published online August 31,

7 overall number of marijuana users increased from 21 9 million in 2002 (20 9 million 22 8 million) to 31 9 million in 2014 (30 8 million 32 9 million). The number of daily or near daily users increased from 3 9 million in 2002 (3 6 million 4 2 million) to 8 4 million in 2014 (7 8 million 8 9 million; appendix). The prevalence of marijuana use disorders in marijuana users decreased from 14 8% in 2002 to 11 0% in 2014, with a downward trend starting in 2008 (β= , ). In marijuana users who did not initiate use in the previous year, the prevalence of marijuana use disorders decreased from 15 1% in 2002 to 11 4% in 2014 (β= , ). The prevalence of perceiving great risk of harm from smoking marijuana in users decreased from 8 8% in 2002 to 2 8% in 2014, with a downward trend starting in 2007 (β= , ). The prevalence of daily or near daily use in users increased from 18 0% in 2002 to 26 3% in 2014 (β=0 0389, ; table 1). The mean number of days of marijuana use in adults increased from 10 0 days in 2002 to 16 3 days in 2014, with an upward trend starting in 2007 ( : β=0 8093, ). The mean number of days of marijuana use in marijuana users increased from 97 9 days in 2002 to days in 2014 (β=2 1741, ; table 1). After adjusting for covariates (except for perceived risk of harm from smoking marijuana), the prevalence of marijuana use in the adult population was lower in than in 2002, but was higher in each year during than in 2002 (table 2). These increases were associated with decreases in perceived great risk of harm from smoking marijuana, as seen in the lower prevalence of marijuana use each year during than in 2002 when perceiving risk of harm from smoking marijuana was included in models (table 2). The adjusted prevalence of marijuana use disorders in adult marijuana users was lower in each year during than in 2002 (adjusted relative risk ; table 3). Additional adjustments for the effect of a decreased perception of great risk of harm from smoking marijuana did not affect these results. After controlling for covariates (except for perceived risk of harm from smoking marijuana), marijuana was used more frequently in every year during than in 2002 in adult marijuana users (table 4). These increases in frequency of use were also associated with decreases in perceiving great risk of harm from smoking marijuana, as seen in the stable number of days of marijuana use during when perceived risk of harm from marijuana was included in models (table 4). Higher adjusted prevalence of marijuana use was found in adults aged years, men, non-hispanic black people, non-hispanic adults of more than one race, those without a high school diploma, people without health insurance, those employed part-time, unable to work because of disability, or unemployed, those residing in the west of the USA, those perceiving legalisation of medical marijuana use in their state, and those perceiving ease of marijuana availability than in each corresponding reference group (table 5). Marijuana use was higher in users of tobacco, cocaine, hallucinogens, and inhalants, heavy alcohol users, and non-medical users of psychotherapeutics than in those who did not use these substances (table 5). Bivariable model without controlling for any covariates, unadjusted incidence rate ratio Multivariable model without perceived risk, but with other covariates, adjusted incidence rate ratio Multivariable model with perceived risk and other covariates, adjusted incidence rate ratio Year ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 0 ( ) 1 0 ( ) ( ) 1 1 ( ) 1 0 ( ) ( ) 1 1 ( ) 1 0 ( ) ( ) 1 0 ( ) 0 9 ( ) ( ) 1 1 ( ) 1 0 ( ) ( ) 1 1 ( ) 1 0 ( ) ( ) 1 2 ( ) 1 0 ( ) ( ) 1 2 ( ) 1 0 ( ) ( ) 1 2 ( ) 1 0 ( ) risk of smoking marijuana once or twice per week No risk Slight risk ( ) Moderate ( ) risk Great risk ( ) Unspecified ( ) *Substance Abuse and Mental Health Services Administration requires that any description of overall sample sizes based on the restricted-use data files has to be rounded to the nearest 100 to minimise potential disclosure risk. Each multivariable zero-truncated negative binomial regression model adjusted for age, gender, race/ethnicity, education, employment status, marital status, health insurance, metropolitan statistical area, region, nicotine dependence, alcohol use disorders, cocaine use disorders, hallucinogen use disorders, heroin use disorders, inhalant use disorders, prescription opioid use disorders, sedative use disorders, stimulant use disorders, age at first marijuana use, state legalisation of non-medical marijuana use, perceived state legalisation of medical marijuana use, perceived marijuana availability, and source of marijuana. Major depressive episode in the past year was not associated with marijuana use frequency and was excluded from the final models. We used similar hierarchical multiple regression approach specifying a fixed order of entry for variables (sociodemographic factors, then substance use disorders and related factors, major depressive episode, finally perceptions of marijuana use) to test the effects of certain predictors independent of the influence of others and to identify factors that might be associated with changes in trends. Each day of use is considered a separate occurrence and is only counted once in calculating the adjusted incidence rate ratio. Reference group. Prevalence significantly differs from the prevalence of the model-predicted number of days of the reference group. Table 4: Modelled number of days of marijuana use in the past 12 months in adult users in the USA, (n= )* See Online for appendix Published online August 31,

8 Marijuana use in adults, adjusted RR Marijuana use disorders in adult marijuana users, adjusted RR Number of days of marijuana use in adult marijuana users, adjusted incidence rate Age, years ( ) 1 8 ( ) 1 0 ( ) ( ) 1 2 ( ) 0 9 ( ) Sex Men 1 1 ( ) 1 2 ( ) 1 1 ( ) Women Race or ethnicity Non-Hispanic white Non-Hispanic black 1 2 ( ) 1 4 ( ) 1 3 ( ) Non-Hispanic native American or 1 0 ( ) 1 2 ( ) 1 0 ( ) Alaska native Non-Hispanic Hawaiian or other 1 1 ( ) 1 0 ( ) 1 0 ( ) Pacific Islander Non-Hispanic Asian 1 0 ( ) 1 3 ( ) 0 8 ( ) Non-Hispanic more than one race 1 1 ( ) 1 2 ( ) 1 2 ( ) Hispanic 1 0 ( ) 1 3 ( ) 1 0 ( ) Education Less than high school High school 0 9 ( ) 0 8 ( ) 1 0 ( ) Some college 0 9 ( ) 0 8 ( ) 0 9 ( ) College graduate 0 9 ( ) 0 7 ( ) 0 7 ( ) Employment status Full-time employed Part-time employed 1 2 ( ) 1 2 ( ) 1 0 ( ) Unable to work because of disability 1 2 ( ) 1 2 ( ) 1 1 ( ) Unemployed 1 1 ( ) 1 3 ( ) 1 1 ( ) Marital status Married Widowed 1 1 ( ) 0 7 ( ) 1 1 ( ) Divorced or separated 1 2 ( ) 1 0 ( ) 1 0 ( ) Never married 1 4 ( ) 1 2 ( ) 1 1 ( ) Health insurance Private only No insurance coverage 1 1 ( ) 1 0 ( ) 1 2 ( ) Medicaid 1 0 ( ) 1 0 ( ) 1 2 ( ) Other 1 0 ( ) 1 0 ( ) 1 2 ( ) Metropolitan statistical area Large Small 1 0 ( ) 1 0 ( ) 1 0 ( ) Non-metropolitan 0 9 ( ) 1 0 ( ) 1 0 ( ) Region Northeast 1 0 ( ) 1 0 ( ) 0 9 ( ) Midwest 0 9 ( ) 1 0 ( ) 0 9 ( ) South 0 9 ( ) 1 0 ( ) 0 9 ( ) West Tobacco** Yes 1 4 ( ) 1 1 ( ) 1 3 ( ) (Table 5 continues on next page) Adjusted prevalence of marijuana use disorders in marijuana users was higher in people aged years, men, non-hispanic black people, Hispanics, non-hispanic Asians, non-hispanic users of more than one race, adults without a high school diploma, people who were employed part-time, unable to work because of disability or unemployed, never-married adults, people with specific substance use disorders (nicotine, alcohol, cocaine, hallucinogens, inhalants, and psychotherapeutics), and major depressive episode than in each corresponding reference group (table 5). Marijuana use disorders were more prevalent in those perceiving ease of marijuana availability than in those who did not perceive easy availability (table 5). Frequency of marijuana use in marijuana users was higher in men, non-hispanic black people and non- Hispanic people of more than one race, people without a high school diploma, people who were unable to work because of disability and unemployed, never-married adults, people who were uninsured or Medicaid beneficiaries, people residing in the west of the USA, people initiating marijuana use before the age of 18 years, and people who perceived marijuana as easy to obtain than in each corresponding reference group (table 5). Users with most other substance use disorders (nicotine, prescription opioids, cocaine, and hallucinogens) used marijuana more frequently than users without the corresponding disorders (table 5). Discussion Marijuana use and frequency of use increased in the USA during compared with 2002, even after adjusting for sociodemographic, geographical, and substance use factors. Initiation of marijuana use also increased during Furthermore, adults have perceived less risk of harm from marijuana use since , and these declining risk perceptions were associated with increases in marijuana use and frequency of use. By 2007, 12 US states had legalised medical marijuana use, and the cumulative effects of these policy changes might have led to changes in marijuana use and risk perceptions in US adults in Although shifts in perceived risk have historically been shown to be important predictors of adolescent marijuana trends, 9 no previous research has examined this relationship in adults. Nevertheless, we found neither an increase in marijuana use disorders nor an association between changes in perceived risk of harm and the prevalence of marijuana use disorders. Previous studies have shown that the quantity and frequency of marijuana use is highly correlated with symptoms of marijuana use disorders. 10 Similarly, increases in marijuana potency might have been related to increases in marijuana use disorders. 11 However, we did not observe increases in marijuana use disorders, despite finding increases in frequency of use during , when potency has 8 Published online August 31,

9 continued to increase. 12 We speculate that the many people who have recently (within the past year) started to use marijuana might be using the drug less intensely and have less psychopathology than people who have used marijuana for longer, which could decrease their risk of transition from use to use disorders. 13,14 Future research on trends in marijuana abuse and dependence and their relationships with perceived risk could help elucidate reasons for the discrepancy between marijuana use patterns and use disorders. Our results confirm increases in marijuana use suggested in the two-wave NESARC and NESARC-III study, 2 although our increases are lower in magnitude than NESARC (which showed that use increased by more than double). 2 Similar to NESARC, we found decreases in the proportion of users reporting marijuana use disorders, but the prevalence of marijuana use disorders in users in our study (14 8% in 2002 and 11 0% in 2014) was lower than in NESARC (35 6% in and 30 6% in ). 2 However, our results showed no increase in marijuana use disorders in the adult population, by contrast with the NESARC study finding of an increase of almost double from to ( to 2 9%). 2 Differences in study design and implementation between NESARC and NSDUH as well as between NESARC waves might explain differences. 4,15 First, NSDUHS s Audio Computer-Assisted Self Interview process might have resulted in a higher prevalence of marijuana use than with NESARC s Computer-Assisted Personal Interview because people using illegal substances might be more likely to self-report accurately when given more privacy. 4,16,17 Second, NSDUH gave each respondent who completed the interview a US$30 cash incentive during NESARC did not give any monetary incentive, whereas NESARC-III gave a $90 cheque incentive for each completed interview. 16,17 Third, NESARC-III was done in six languages, whereas NESARC was done only in English. 16,17 Fourth, a federal government agency (US Census Bureau) did NESARC, whereas a non-profit company (Westat) did NESARC- III. 16,17 This could have affected an individual s willingness to report illegal behaviours (ie, marijuana use) differently in each NESARC wave. Overall, these differences make it difficult to draw firm conclusions about trends based on NESARC waves when compared with trends using NSDUH data. Comparison with other studies is difficult because no studies share the same survey design and outcome assessment. The Monitoring the Future (MTF) study 18 consisted of annual surveys of US high-school graduates over time and found that annual prevalence of any marijuana use increased from 29 3% in 2002 to 31 6% in 2014 in people aged years. Daily marijuana use increased for the same age group from 4 5% to 6 9%. 18 However, MTF did not ascertain marijuana use disorders, and other nationally representative data are not available. Marijuana use in adults, adjusted RR Marijuana use disorders in adult marijuana users, adjusted RR Examination of trends in other outcomes related to marijuana use show increases in emergency department visits related to marijuana 19 and traffic fatalities in which marijuana was detected in the deceased. 20 Although none of these studies were done in the US household population, discrepancies between these detected increases and the absence of increases in marijuana use Number of days of marijuana use in adult marijuana users, adjusted incidence rate (Continued from previous page) Alcohol** Yes 1 2 ( ) 1 8 ( ) 1 0 ( ) Cocaine** Yes 1 8 ( ) 1 5 ( ) 1 1 ( ) Hallucinogen** Yes 2 0 ( ) 2 1 ( ) 1 3 ( ) Heroin** Yes 0 9 ( ) 0 9 ( ) 0 8 ( ) Inhalant** Yes 1 5 ( ) 1 7 ( ) 0 8 ( ) Non-medical prescription opioids** Yes 1 3 ( ) 1 7 ( ) 1 1 ( ) Non-medical sedative** Yes 1 4 ( ) 1 8 ( ) 1 1 ( ) Non-medical stimulant** Yes 1 4 ( ) 1 5 ( ) 1 1 ( ) Age at first marijuana use, years < ( ) 1 2 ( ) 1 4 ( ) ( ) 1 2 ( ) 0 7 ( ) State-legalised non-medical marijuana use Yes 1 1 ( ) 1 0 ( ) 1 1 ( ) state legalisation of medical marijuana use Yes 1 1 ( ) 1 0 ( ) 1 1 ( ) Not sure or unknown 0 9 ( ) 0 9 ( ) 0 8 ( ) marijuana availability, fairly or very easy Yes 1 4 ( ) 1 2 ( ) 1 2 ( ) (Table 5 continues on next page) Published online August 31,

10 Marijuana use in adults, adjusted RR Marijuana use disorders in adult marijuana users, adjusted RR Number of days of marijuana use in adult marijuana users, adjusted incidence rate (Continued from previous page) Source of marijuana (from survey definitions) Bought it ( ) 3 0 ( ) Traded for it ( ) 2 6 ( ) Got it for free or shared Grew it yourself ( ) 3 4 ( ) Method unspecified ( ) 2 1 ( ) Major depressive episode Yes ( ).. No RR=relative risk. *Substance Abuse and Mental Health Services Administration requires that any description of overall sample sizes based on the restricted-use data files has to be rounded to the nearest 100 to minimise potential disclosure risk. Results for each factor are from the multivariable logistic regression model that adjusts for all the factors listed as well as the survey year and perceived risk of marijuana use (see table 2). Source of marijuana and major depressive episode were not associated with marijuana use and so are not included in the model. Results for each factor are from the multivariable logistic regression models that adjust for all the factors listed as well as the survey year and perceived risk of marijuana use (see table 3). Results for each factor are from the multivariable zero-truncated negative binomial regression model that adjusts for all the factors listed as well as the survey year and perceived risk of marijuana use (see table 4). Each day of use is considered a separate occurrence and is only counted once in calculating the adjusted incidence rate ratio. Major depressive episode was not associated with the number of days of marijuana use and so is not included in the model. Model-adjusted prevalence significantly differs from the model-adjusted prevalence of the reference group. Reference group. **For marijuana use model, results are for use of each of the corresponding substances, except for heavy alcohol use rather than alcohol use. For the marijuana use disorders and number of days of marijuana use models, results are for substance use disorders (or nicotine dependence) related to each of the corresponding substances. The presented results for major depressive episode are based on a model using National Survey on Drug Use and Health data from 2005 to 2014 only after controlling for other covariates. Other results are from the National Survey on Drug Use and Health data. Table 5: Other correlates of marijuana use in adults (n= ) and other correlates of marijuana use disorders and of the number of days of marijuana use in adult users (n= ) in the USA, * disorders in NSDUH over a similar time period are intriguing. Further research on the overlaps and distinctions in these marijuana outcomes is needed. Associations of marijuana use and disorders with younger age, male sex, low education, and other than full-time employment examined in this study are consistent with previous research. 2,13 Suggestions that prevalence is higher in non-hispanic black people than in non-hispanic white people contrast with previous work 18 and suggest a shifting pattern of marijuana use in the USA. 2 The associations of marijuana use disorders with depression and tobacco and other substance use are generally consistent with previous research Such co-occurrences are a stark reminder that marijuana use disorders are often comorbid with psychiatric illness and co-occur with use of multiple substances. When one of these psychiatric and behavioural problems is identified, clinicians must carefully look for other related problems. 24 Findings that marijuana use is more prevalent in adults residing in states with legalisation of medical marijuana use than adults not residing in these states 1 and in adults who perceived that medical marijuana use was legal in their state suggest two reasons. These areas might be the locations where residents would favour such laws because of previous experience, or changes in the laws might have been related to increasing use. Evidence supporting both pathways has been found, suggesting an overall reciprocal relationship of social attitudes and use patterns. 25 Our study has several limitations. NSDUH excluded people who were homeless and not living in shelters or people residing in institutions (eg, people who were incarcerated), which could lead to underestimates in drug use and drug use disorders. Furthermore, associations between marijuana use and many specific psychiatric disorders could not be considered because these measures were not included in NSDUH. Also, because of the cross-sectional nature of NSDUH data, this study could not establish either temporal or causal relationships. Finally, NSDUH is a self-reported survey and is subject to recall bias. Key advantages of NSDUH include the consistent survey design, methodology, and questionnaire content since 2002 and large sample sizes, allowing detection of changes in marijuana use trends across every year during Good validity and reliability have been found for marijuana use disorders and other substance use measured by NSDUH. 26,27 We found that the number of adults who had used marijuana in the past year increased by 10 million and the number of daily or near daily users increased by 4 4 million in the USA during Heavy marijuana use is associated with unemployment, lower than average income, diminished life satisfaction, and criminal behaviour. 28 Understanding of these trends is relevant for policy makers who continue to consider whether and how to modify laws related to marijuana and for healthcare practitioners who care for patients using marijuana. risk of marijuana use is associated with patterns of high frequency and early onset use, 28 suggesting the increasing need for clinical intervention. Furthermore, our results indicate an increasing need to modify risk perceptions of marijuana use in adults through effective education and prevention messages. Marijuana use and use frequency increased in the USA during and when compared with Associations between increases in marijuana use and decreases in perceived risk of smoking marijuana suggest a potential benefit of education and prevention messages. Co-occurrence of marijuana use and marijuana use disorders with use of other substances and depression underscores the importance of screening across the full range of behavioural health issues. Contributors WMC, BH, CMJ, and CB contributed to the study concept and design. BH collected the data. WMC, BH, CMJ, CB, and AH interpreted the data. WMC and BH drafted the manuscript. WMC, BH, CMJ, CB, and AH revised the manuscript. BH and CMJ did statistical analysis. WMC and BH provided administrative, technical, or material support. BH had full access to all of the data in this study and takes responsibility for integrity of the data and the accuracy of the data analysis Published online August 31,