DSM-IV is a multi-dimensional measure for diagnosing CUD and is well established in the literature

There is substantial need for improved knowledge in this domain, as the majority of active cannabis users should be encouraged to transition from smoking to alternative modes. Clear and consistent public health messaging from primary stakeholders is required to facilitate this. The literature unequivocally recognizes intensive or frequent cannabis use patterns as a primary predictor of acute/chronic adverse outcomes . Based on available data, between one and two in five cannabis users in Canada engage in frequent/intensive cannabis use, and form a distinct ‘high-risk’ group for potentially severe cannabis-related harm. While some of these ‘intensive’ users may be reached by simple prevention messaging emphasizing decreased frequency of use, a sizeable proportion are likely to feature criteria for CUD. These users are likely less receptive or able to follow simple behavior-change advice and instead may require professional help or treatment . Driving immediately following cannabis use involves  impairment, and is a behavior that about doubles risk for traffic crash involvement, and related injury and/or fatalities . Cannabis-impaired driving is also a main contributor to cannabis-related disease burden, as it provides a  cause of direct cannabis-related mortality, and thus represents a primary target for prevention . The Canadian surveys indicate that substantial minorities of users engage in cannabis-impaired driving, with a further subset of these  engaging in driving co-impaired by alcohol, which further amplifies risk for injury . Moreover, as the surveys relied on short and varying time periods for impairment risk , these rates likely represent under-estimates of the risk total.

Irrespectively, these risk behavior rates are high and disconcerting overall. They are likely facilitated by multiple factors, including common beliefs about nonexistent, or only very limited ‘impairment’ effects of cannabis grow tent, as well as a low likelihood of apprehension under current enforcement for cannabis-impaired driving . These circumstances urgently require intensified targeted education and enforcement efforts. These efforts should draw on crucial lessons from alcohol/drunk driving intervention strategies, which have achieved substantial decreases in alcohol-impaired driving and related crashes . While some evidence exists about cannabis use-related adverse reproductive/infant health outcomes during pregnancy and/or breastfeeding, rather small minorities of women reported ongoing use during these periods. While cannabis compounds may be passed on to the foetus via intrauterine transmission or through breast milk, some women use cannabis ‘therapeutically’ to combat pregnancy-related nausea . Overall, adverse outcomes for newborns are uncertain and likely limited . However, avoiding cannabis use during pregnancy and breastfeeding represents a relatively simple prevention effort of possible harm to others . Moreover, this recommendation aligns with other precautionary health behavior adjustments among pregnant women or new mothers . Overall, current indicator data from major surveys indicate that respective majorities of cannabis users in Canada – with the exception of ‘smoking’ as the primary mode of use – are generally mostly compliant with the main LRCUG’ recommendations for which such data exist. At the same time, the proportion of users non-compliant with other LRCUG recommendations represent sizeable sub-populations of the currently 4–5 million cannabis users, many of which likely engage in more than just one risk behavior, and thus face considerable risk for acute and/or chronic adverse health outcomes . While population-level harms for cannabis are more limited than those for alcohol or tobacco, the ensuing disease burden is substantial, also given that cannabis use disproportionately occurs among youth/young adults where key LRCUG-defined risk behaviors are commonly concentrated . Thus, in order to achieve legalization’s objective of improved public health outcomes, key cannabis-related risk behaviors need to be more effectively addressed. Active and widespread dissemination and promotion of the LRCUG recommendations may lead to increased awareness and adjustment of relevant risk-behaviors among users . The behavioral uptake potential of interventions such as the LRCUG is uncertain, for example among intensive, chronic users.

However, it should be emphasized that the LRCUG represent a targeted prevention measure, rather than a  treatment tool for individuals possibly characterized by CUD . Nevertheless, other complementary, targeted intervention measures combined with appropriate regulatory provisions focusing on specific risk behaviors  are required in order for a prevention tool like the LRCUG to be effective . In addition, the impacts of such targeted measures on cannabis related risk behaviors require consistent assessment and improved understanding . The data used in the present review feature some limitations. Specifically, the survey sources for the indicator data relied on different sampling frames, essential methods details and item design , limiting the surveys’ reference populations full comparability. Only some of the surveys are considered population representative; the CAMH Monitor is an Ontario-based survey, not generalizable to populations elsewhere in Canada. All the surveys rely on  self-report data, which also may be burdened by recall or other biases. In addition, survey items for certain indicators were based on differential operational definitions , or included subjective estimates with unknown reliability in select instances . These may entail limitations for possible intrinsic and extrinsic indicator data validity or comparability. Overall, while the scientific evidence behind the LRCUG is evolving, consistent population-level measurement of risk-behavioral indicators for cannabis use-related health outcomes is essential for effective monitoring of public health-related cannabis risks and harm outcomes, especially in the era of legalization as an ongoing ‘policy experiment’. Substance use disorders are currently a major public health crisis in the US . Cannabis is the most commonly used illicit substance in the world . With more than 200 million users of cannabis worldwide, its harmful health effects have become a serious global problem . During the past two decades, the laws and policies related to cannabis use have also changed drastically throughout the world. For example, countries such as Canada, Spain, and Germany have legalized cannabis for medical use while some have even legalized its non-medical use, e.g., Uruguay in 2015 and Canada in 2018 . Not surprisingly, the legalization trend continues in the US, with 33 states and the District of Columbia legalizing medical marijuana use, and 11 states and the District of Columbia legalizing adult non-medical marijuana use . Regardless of the developing accord about the usefulness of medical marijuana for several serious illnesses, there is a widespread concern that this may cause adverse effects . According to a study on the effects of medical marijuana laws, the likelihood of current as well as regular use of cannabis among people aged 21 or older has increased after the laws came into effect .

This also appears to have contributed to an increased prevalence of illicit cannabis use and cannabis use disorder . In particular, among adult males, arrests due to illegal marijuana possession in major cities have increased by 15–20% and the treatment provided in rehabilitation facilities for such arrests have increased by 10–20% . This article focuses on cannabis use disorder . Earlier, there was a consensus that CUD is rare, which is no longer true. It is estimated that about 34% of cannabis users develop CUD during their lifetime based on the 4th edition of the Diagnostic and Statistical Manual of Mental Disorders  . Furthermore, a recent study based on DSM-V criteria found that about 27% of cannabis users develop CUD during their lifetime . Another research shows that after legalizing marijuana for recreational use, the prevalence of CUD among past year cannabis users between the ages of 12 and 17 rose from 22.8% to 27.2% . Thus, given that the prevalence of CUD is expected to increase further, it is imperative to predict the risk of developing CUD for cannabis users, especially for adolescents and emerging adults, based on their personal risk factors. Identifying individuals at high risk of CUD will allow the possibility of applying early intervention, which may potentially help stem the increasing prevalence of the disorder. Several risk factors have been reported for substance use disorders in general and specifically for CUD. These include male sex, early exposure to traumatic events, early use initiation, family history of substance use, childhood depression, and conduct disorder symptoms . High impulsivity and certain personality traits are also associated with the disorders . In particular, work by coauthor Filbey’s lab showed that openness distinguishes cannabis-only users from nicotine-only users, co-morbid marijuana and nicotine users, and non-users . The results from this study also indicate that conscientiousness is lower among grow lights for cannabis users. Some brief screening tools such as BSTAD  and S2BI  have been developed for adolescents . For example, the cutoff for CUD based on BSTAD is at least two days of marijuana use in the past one year. A relatively lengthy tool, Transmissible Liability Index, assesses the inherited risk for disorders based on a 45-item questionnaire . Also, a recent study has developed a simple cumulative risk index for substance dependence in adulthood using risk factors in childhood and adolescence .

It can be used to screen adolescents who are likely to develop persistent disorder in adulthood. A similar study has developed a risk score by counting the number of early life risk factors present in an individual and associating it with cannabis use and CUD in early adulthood . However, a key limitation of the existing tools is that none of them provides a quantitative risk of developing the disorder based on personal risk factors, which restricts their practical utility. Models for predicting such risks have been developed for several diseases, including breast cancer , contralateral breast cancer , heart disease , depression , and psychiatric disorders , and they are in wide clinical use. However, currently there is no such quantitative risk prediction tool for CUD. In this study, we build upon the findings of Ketcherside et al. in a cannabis-using adult population and perform a secondary analysis of the data. More specifically, we build a preliminary quantitative risk prediction model to estimate the chance that a cannabis user will develop CUD based on various demographic, behavioral, psychiatric, and cognitive risk factors. The initial data set obtained after applying the inclusion criterion consisted of 118 cannabis users. We used CUD as the outcome variable, which was derived based on the DSM-IV criteria for dependence.The variable selection process to identify potential risk factors was the following. First, the variables with more than 50% missing values were discarded. Then, among the remaining variables, only those that remain relatively stable over time were chosen. Given the cross-sectional nature of the data, focusing attention on such type of variables protects against using risk factors that may actually be an effect of CUD. This resulted in 30 variables. These included measures of impulsivity and personality traits. The former were obtained using two questionnaires, namely, Impulsive SensationSeeking Scale , a 19-item self-reported questionnaire from the Zuckerman-Kuhlman Personality Questionnaire and Barratt Impulsivity Scale , a 30-item self-reported questionnaire where the items can be grouped into six first-order factors that measure different aspects of impulsivity . Both ImpSS and BIS were considered because there are some characteristics of impulsivity that are captured by ImpSS but not by BIS and vice versa, and the two have been used together in several studies . The personality traits were obtained using Neuroticism, Extraversion, and Openness inventory , a five-factor inventory for measuring five different dimensions of personality . The actual measures derived from these questionnaires were total score on the ImpSS questionnaire, scores on the six factors from the BIS questionnaire, and scores on the five factors from the NEO questionnaire. Only 46 of the 118 subjects had complete data on all 30 variables. To guard against loss of subjects due to missing data on potentially unimportant variables, univariate logistic regression models were fitted with each of these variables as a predictor. Thereafter, the predictors with univariate model p-value less than or equal to 0.3 were selected into the final set of potential predictors for a multivariate model . The resulting data set had 12 potential risk factors and 94 subjects with complete observations on them. This final data set was used for the rest of the model building exercise. The data analysis was performed using five common statistical and machine learning models for classification , namely, logistic regression with LASSO penalty, K-Nearest Neighbor , support vector machine  with radial kernel, random forest, and gradient boosting.