Recently, Farmer and colleagues have explored the natural outcome of CUDs in a 14‐year, multiple follow‐up investigation of 16‐year‐olds 28 . They reported that nearly 90% of the individuals diagnosed with a CUD remitted during this time‐period, with women recovering at a faster rate than men. Notably, this study did not include non‐abstinent remission (i.e. any use of cannabis without meeting criteria for CUD) as a possible natural outcome of CUDs, although research conducted in the last decade indicate that non‐abstinent recovery is a relatively common outcome in alcohol use disorders 29 , 30 . Furthermore, the study focused on a 16–30‐year‐old age group, although previous reports point to substantial rates of CUD initiation above 30 years of age 31 , 32 . In this study, we analysed data from waves 1 and 2 of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC), from which we selected those individuals diagnosed with a CUD at wave 1. Our goals were: (1) to explore the natural outcome of CUD in a 3‐year longitudinal follow‐up and (2) to identify baseline socio‐demographic and clinical factors associated with different outcomes of CUD.

As a result of the potential negative consequences of CUDs, research has been conducted to explore the progression as well as correlates of the natural outcome of CUD. Conceptually, most studies exploring the progression of substance use disorders (SUDs) have focused on recovery or remission, usually differentiating the two in terms of duration of which an individual maintains relief from pathological substance use 18 , 19 or the degree by which an individual has achieved global health 20 . However, no clear definition of these two clinical outcomes is currently available, and the terms are used indiscriminately. Furthermore, most studies exploring natural progression of CUD have focused predominately on abuse and dependence separately and have limited themselves to specific age groups. For example, research on cannabis dependence found that more than 90% of those who have met the criteria remit at some point in their life, while half of those do so within 6 years following the onset of dependence 21 . Research on cannabis abuse suggested that approximately 57% of those who meet the criteria for this disorder become either non‐users or users without disorders within a year 22 . However, relatively few large‐scale studies have investigated the correlates associated with CUD outcomes. Recovery from CUD was associated with female gender, current employment, shorter durations and older age at onset of dependence, having relatively more cannabis‐induced symptoms (e.g. depression, paranoia, trouble concentrating, etc.) and high self‐competence 23 , 24 . Research concerning changes in prevalence rates of CUDs due to the shift from DSM‐IV to DSM‐5 criteria presented conflicting results 25 - 27 , and almost no data are currently available on the natural outcome of CUDs according to DSM‐5 criteria.

Cannabis is the most widely used internationally regulated drug in the world, with an estimated 2.8–4.5% annual prevalence of cannabis use globally 1 , 2 . Life‐time and 12‐month prevalence of cannabis use in the United States has been reported to be between 21 and 43% and between 4 and 10%, respectively, in different large‐scale epidemiological studies 3 , 4 . Among cannabis users, up to one‐third may develop a cannabis use disorder (CUD), i.e. cannabis abuse or dependence 5 . This is particularly true of those who started using cannabis at an earlier age and those with concomitant use of other substances 6 . Research indicated increasing prevalence rates of CUD in the United States in the past decades 7 and the global impact of increasingly prevalent CUDs is illustrated by estimates of it having resulted in more than 2 million disability‐adjusted life years in 2012 8 . Reports indicate that cannabis users, and particularly those with CUDs, may be at increased risk for mortality 9 , 10 , yet other studies present conflicting findings 11 . In addition, CUDs may be associated with deficits in cognitive functions such as executive functions, working memory and both immediate and delayed memory 12 - 15 . The evidence regarding the extent to which these deficits improve during abstinence is conflicting 16 , 17 .

We conducted cross‐tabulations in order to determine the socio‐demographic characteristics of individuals with and without CUDs at wave 1. Additionally, their socio‐demographic and clinical characteristics were assessed using cross‐tabulations according to outcome at follow‐up; χ 2 analyses were performed in order to assess differences in socio‐demographic and clinical characteristics between outcome groups. We used two sets of multiple logistic regression analyses in order to compare baseline socio‐demographic and clinical characteristics among the three possible CUD outcome categories. In the first set of analyses, for each socio‐demographic or clinical measure we compared individuals with CUD remission (abstinent and non‐abstinent) to individuals with sustained CUD at follow‐up (used as the reference category). The second set of analyses compared individuals with non‐abstinent remission at follow‐up to those with abstinent remission (used as reference category) for each socio‐demographic or clinical measure. Independent‐sample t ‐tests were used to compare the average number of cannabis use days and average joints per day of use between individuals in the non‐abstinent remission group and individuals in the sustained disorder group. Analyses were conducted using Software for Survey Data Analysis (SUDAAN) version 10 39 , which uses Taylor series linearization to make adjustments for the NESARC's sample design characteristics.

Socio‐demographic and clinical characteristics of individuals with CUD at baseline were analysed separately according to outcome (i.e. abstinent remission, non‐abstinent remission and sustained CUD). Unless specified otherwise, these relate to the 12 months prior to wave 1. Socio‐demographic characteristics included sex, race, educational level, household income, marital status, age, urbanity and region. Job problems related to trouble with a boss or co‐worker and number of past‐year medical conditions related to the following chronic and acute conditions confirmed clinically by a health professional: hardening of arteries or arteriosclerosis, high blood pressure or hypertension, cirrhosis of liver, other form of liver disease, chest pain or angina pectoris, rapid heartbeat or tachycardia, heart attack or myocardial infraction, other form of heart disease, stomach ulcer, gastritis or arthritis. All socio‐demographic characteristics have been described in detail elsewhere 30 . Clinical characteristics included DSM‐IV related psychiatric disorders: mood disorders (major depressive disorder, dysthymia and bipolar disorders), anxiety disorders (social anxiety disorder, generalized anxiety disorder, specific phobias and primary panic disorder) and life‐time personality disorders (antisocial, schizoid, histrionic, dependent, obsessive‐compulsive, paranoid and avoidant), as diagnosed using the AUDADIS‐IV. Use of other drugs included any use of one of the following drugs: sedatives, tranquillizers, opioids, amphetamines, cocaine or crack, hallucinogens, inhalants, heroin or other drugs.

Remission from CUD indicates no active CUD meaning that abuse or dependence criteria have not been met fully during the 12‐month preceding wave 2 interview. Following previous work that differentiated different aspects of recovery from alcohol dependence 29 , 30 , individuals were categorized as being in abstinent remission if they reported no use of cannabis during this interval and did not meet criteria for diagnosis of a CUD. They were categorized as being in non‐abstinent remission if they reported any use of cannabis but did not meet criteria for diagnosis of a CUD. Individuals were categorized as having sustained CUD if they continued to meet criteria for either cannabis abuse or dependence in the 12 months prior to wave 2. Among individuals who did not terminate their use of cannabis at follow‐up, number of days of cannabis use, number of joints per day of use and prevalence of particular CUD symptoms in the 12 months prior to follow‐up were assessed.

Diagnoses of CUDs and other SUDs were made using the Alcohol Use Disorder and Associated Disabilities Interview Schedule, DSM‐IV version (AUDADIS‐IV). It has been reported to have high reliability for substance use disorders and other psychiatric disorders 29 , 36 . This analysis combines cannabis abuse and dependence into a single diagnostic category (CUD), including individuals who qualified for a diagnosis of at least one of these disorders at wave 1. This analytical strategy is consistent with evidence indicating that the abuse/dependence distinction is not diagnostically advantageous 37 , 38 .

We identified a subgroup of NESARC respondents ( n = 560), representing all those who qualified for a diagnosis of DSM‐IV cannabis abuse and/or dependence during the 12 months prior to their wave 1 interview, i.e. endorsement of at least three of the seven DSM‐IV cannabis dependence criteria or at least one of the four DSM‐IV cannabis abuse criteria 19 . This subgroup included all individuals suffering from CUD, regardless of treatment utilization, legal problems or mode of referral to treatment, from which 444 respondents (79.3%) were subsequently re‐interviewed at wave 2 and included in our analysis. This is within the range of the general re‐interview rate reported for the NESARC sample.

Our study used data from waves 1 and 2 of the NESARC, a survey designed by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) 33 . The NESARC is a longitudinal and nationally representative survey which targeted the non‐institutionalized adult population of the United States, including military personnel living off‐base and those in group housing (e.g. college dormitories, shelters). Wave 1 of the NESARC (2001–02) consisted of 43 093 civilian, non‐institutionalized adults aged 18 years and older in the United States, including all 50 states and the District of Columbia. Wave 2 (2004–05) consisted of 34 653 adults, 86.7% of those from wave 1 eligible for re‐interview. Survey data were collected during face‐to‐face interviews by 1800 experienced and extensively trained interviewers 34 . The sample was weighted to adjust for oversampling of African Americans, Hispanics and young adults (ages 18–24), probabilities of selection housing unit equivalents, the selection of one individual per household and non‐response at the household and person levels. The weighted data were then adjusted statistically to be representative of the demographic makeup of the United States based on the 2000 Decennial Census in terms of age, sex, region, race and ethnicity 35 . Additional details regarding data collection and statistical adjustments can be found elsewhere 7 , 35 . Informed consent and the research protocol received full ethical review and approval from the US Census Bureau and the US Office of Management and Budget, and this study was approved by the Sheba Medical Center's institutional review board committee.

Tables 2 and 3 present the OR for achieving remission compared to sustained CUD and for achieving non‐abstinent versus abstinent remission. Individuals of Hispanic or Latino origin were significantly more likely to achieve remission [odds ratio (OR) = 2.82; 95% confidence interval (CI) = 1.27–4.87; OR = 2.59; 95% CI = 1.09–4.53, respectively] compared to Caucasians. Individuals with two or more past‐year medical conditions were significantly more likely to achieve remission (OR = 8.40; 95% CI = 2.67–26.41) compared to individuals with no past‐year medical conditions. Furthermore, individuals with baseline daily or almost daily use of cannabis were significantly more likely to achieve any type of remission (OR = 1.91; 95% CI = 1.15–3.16) and particularly non‐abstinent remission (OR = 1.92; 95% CI = 1.05–3.51) compared to individuals with less than weekly use of cannabis. Odds for achieving any type of remission was also significantly higher among individuals with past‐year use of other drugs compared to non‐users (OR = 1.63; 95% CI = 1.04–2.56).

Among individuals with a CUD at wave 1, 42.4% (SE = 2.9) reached abstinent remission at wave 2, 24.7% (SE = 2.4) reached non‐abstinent remission and 32.9% (SE = 2.59) did not remit. In the 12 months prior to wave 2, individuals in the non‐abstinent remission group averaged 102.97 (SE = 14.44) days of cannabis use, compared to 195.31 (SE = 13.25) among individuals in the sustained CUD group ( t = –4.85, P < 0.0001), and 1.79 (SE = 0.22) joints per day of use, compared to 2.13 (SE = 0.16) among individuals in the sustained CUD group ( t = –3.54, P < 0.001). All individuals who reached abstinent remission at follow‐up did not meet any criterion of CUD in the 12 months prior to wave 2, compared to 65% (SE = 5.51) among individuals who reached non‐abstinent remission at follow‐up. Most prevalent CUD symptoms among non‐remitters at follow‐up were persistent desire or unsuccessful efforts to cut down or control substance use (18.51%), tolerance (6.57%) and continuous substance use, despite knowledge of having a persistent or recurrent physical or psychological problem that is likely to have been caused or exacerbated by the substance (4.54%). Exploring the differences in socio‐demographic and clinical characteristics between outcome groups, wave 1 variables that were associated significantly with CUD outcome were initiation of a first full‐time job ( P < 0.001), past‐year medical condition ( P < 0.05) (Table 2 ), use of substances other than cannabis and alcohol ( P < 0.05) (Table 3 ), frequency of cannabis use ( P < 0.05) and type of CUD (abuse or dependence) ( P < 0.05) (Table 4 ).

Life‐time and past‐year prevalence of CUDs in the NESARC wave 1 sample was 8.9% [standard error (SE) = 0.29] and 1.4% (SE = 0.09), respectively. Individuals with a past‐year CUD differed significantly from those without CUD in sex ( P < 0.0001), age ( P < 0.0001), household income ( P < 0.0001), marital status ( P < 0.0001) and region ( P < 0.05) (Table 1 ). The majority of individuals in the CUD group were men (70%) and women (53%) in the non‐CUD group; the most predominant age group was 18–29 (65%) among individual in the CUD group compared to 45–64 (33%) among individuals in the non‐CUD group. The majority of individuals in the CUD group were never married (57%), whereas the majority (64%) of individuals in the non‐CUD group were married at wave 1. In the 12 months prior to wave 1, individuals with a diagnosis of CUD averaged 4.72 (SE = 0.15) days of cannabis use per week and 2.65 (SE = 0.17) joints per day of use. The average age of CUD onset among these individuals was 20 (SE = 0.35) years, with an average of 4.08 (SE = 0.31) years from first use of cannabis to onset of CUD.

Discussion

Our results suggest that approximately 67% of individuals with a diagnosis of CUD remitted at a 3‐year follow‐up, compared to 33% who remained with a sustained diagnosis of CUD. Notably, approximately 37% of those who remitted at follow‐up did so without terminating their use of cannabis, averaging nearly half the cannabis‐use days in the 12 months prior to follow‐up compared to those with sustained CUD. Individuals who used drugs other than cannabis and individuals who used cannabis daily at baseline had greater odds to achieve any type of remission from CUD at follow‐up. Daily cannabis users were also significantly more prone to achieve non‐abstinent rather than abstinent remission compared to individuals who used cannabis weekly or less than weekly. Additionally, individuals who achieved non‐abstinent remission at follow‐up had lower rates of baseline cannabis dependence (rather than abuse) compared to individuals who achieved abstinent remission or no remission. Analysis of socio‐demographic correlates revealed that individuals with medical comorbidities compared to no medical comorbidities and individuals with Hispanic compared to Caucasian ethnicity had greater odds to achieve any type of remission from CUD at follow‐up. Additionally, individuals who achieved abstinent remission were more likely than non‐abstinent remitters and those with sustained CUD to start a first full‐time job in the 3 years prior to wave 1.

Farmer and colleagues previously reported high rates of recovery from CUDs during a 14‐year period 28. Our results indicate that high remission rates are already present after a 3‐year period, which suggests a somewhat mutable nature of CUD 40. Our findings demonstrate that a substantial portion of individuals who remit from CUD do so without completely abstaining from cannabis use. Similarly, previous research concerning remission from alcohol dependence found analogous results 30, which suggests that remission from substance use disorders may not necessarily involve full abstinence. As remission from CUD may be accompanied by continuous yet reduced use of cannabis, it is suggested that outcome measures evaluating the natural course and treatment efficacy of CUD should also include an evaluation of baseline and follow‐up frequency of use.

Examining the clinical correlates associated with CUD outcome, it is evident that baseline patterns of substance use are an important factor. First, individuals with CUD who used cannabis daily at baseline had significantly greater odds than those with weekly cannabis use to achieve any type of remission, particularly non‐abstinent remission. High rates of remission and particularly non‐abstinent remission among daily users may imply that heavy users are less prone to cease their use of cannabis but nevertheless are able to initiate substantial change in patterns of cannabis use. This is different from findings from alcohol‐related research associating increased ethanol intakes with decreased remission rates 29. Additionally, individuals who reported concurrent use of drugs other than cannabis at baseline had significantly greater odds to achieve any type of remission compared to those who used cannabis exclusively. These findings suggest that extensive drug use may be associated with increased rather than decreased remission rates, due possibly to greater exposure to physical 41 and mental 42 hazards associated with frequent and extensive substance use.

Finally, individuals who achieved non‐abstinent remission at follow‐up had lower prevalence rates of cannabis dependence (compared to cannabis abuse) compared to individuals who achieved abstinent remission or no remission. This may be attributed to differences between the two diagnoses. The impairment associated with cannabis abuse relates to environmental aspects of functioning (social, vocational), which are highly susceptible to change regardless of alteration in pattern of cannabis use, thus resulting in high rates of cannabis abuse among non‐abstinent remitters. On the contrary, the impairment associated with cannabis dependence is associated more closely to pathological patterns of cannabis use per se (tolerance, increased amount, etc.), therefore remission from cannabis dependence is less likely to occur in the midst of continuous cannabis use.

Two socio‐demographic correlates were also associated significantly with CUD outcome. First, individuals with baseline CUD who suffered from two or more baseline medical conditions had greater odds to achieve any type of remission at follow‐up. These findings may suggest that poor health, whether caused directly or worsened by excessive use of cannabis (e.g. respiratory pathology 41), may serve as a catalyst in remission from CUD. Individuals with baseline CUD who started a first full‐time job in the 3 years prior to wave 1 were equally prone to achieve abstinent remission or sustained disorder at follow‐up, yet none achieved non‐abstinent remission. This is in line with a previously reported association between work‐related stress and low rates of recovery from alcohol dependence 30, and may further imply that starting a new full time‐job can either perpetuate or eliminate pathological cannabis use, as two possible strategies of coping with work‐related stressors. Additional research is required to elucidate this relationship further.

This study has several limitations. First, the NESARC sample does not include individuals under the age of 18, those currently residing in institutional settings and individuals who change residency frequently 43. These populations may be at high risk for developing CUD 44, therefore the generalizability of our results is decreased. Notably, rates of cannabis use in the NESARC sample appear to be lower compared to parallel surveys conducted in North America, Europe and Australia 3; this may be attributed to differences in measurement tools and survey methodologies 45, as well as matters of privacy and anonymity in the NESARC sample which may be associated with an under‐report of socially undesirable behaviours 46. In addition, symptoms associated with CUD were assed via measures of self‐report, thus resulting in a potential bias in diagnosis of CUD at baseline and follow‐up due to subjective experience. The AUDADIS‐IV has shown good reliability in assessing CUDs globally 47, 48, yet it has been argued that research participants may provide an inaccurate estimation of substance‐related behaviour, intentionally or unintentionally 49, 50. Furthermore, no data are available on the subtype of cannabis used by participants despite research indicating that different types of cannabis may have unique effects on substance‐related psychopathology 51. Additionally, this study included legal problems as a criterion for diagnosis of CUD, even though it was excluded from DSM‐5 due to variation in cross‐jurisdiction assessment. In some countries cannabis use holds different legal status compared to the United States, therefore our results may not be generalized for countries implying more strict or liberal approaches towards cannabis users. It should also be considered that there may be additional downstream consequences of legal problems which may affect other aspects of life (such as occupational, interpersonal relations, etc.) and eventually manifest in additional CUD criteria, although this was not accounted for in the NESARC. Finally, research has pointed out the fluctuant nature of CUDs 28, which may imply for re‐occurring periods of cannabis use (with or without meeting the criteria for CUD) and abstinence. Currently, NESARC data include only two waves and thus time–series analysis, which could shed light on this supposed transitory nature, could not be conducted. The relatively long time‐period between the two measures of CUD increased the probability of additional factors, which may have influenced the naturalistic aspects of CUD outcome.

Nevertheless, our findings are important, as they shed additional light on the natural outcome of CUDs and suggest that CUDs may be characterized by high remission rates over a shorter time‐period than reported previously. Our findings may be important for defining desirable assessment and outcome measures in treatment of CUD Assessing patterns of cannabis use may add important information to the function‐based criteria for diagnosis and outcome of CUD.