Cannabis is the most widely used illicit substance in the world.¹ Indeed, though U.S. estimates of past year cannabis use problems (i.e., abuse, dependence) have remained relatively stable over the past 10 years, it is estimated that 7.6% of the U.S. population aged 18 or older have used cannabis in the past month, a significant increase over prior years.² In Australia, estimates suggest that 35% of individuals aged 14 or older have used cannabis at some point in their life, with 5.3% and 3.5% of Australians reporting past-month and past-week use, respectively.³

With rates of cannabis use rising, particularly among certain vulnerable segments of the population (e.g., veterans⁴), there has been increased attention focused on the short- and long-term consequences associated with cannabis use. A recent review highlighted impairments in memory, motor coordination, and judgment as short-term consequences, and cognitive impairment and heightened risk of psychosis as long-term consequences of cannabis use.⁵ Beyond specific negative health effects, a considerable amount of research has now documented that heavy cannabis use, particularly when initiated during adolescence and used regularly, can lead to dependence and other associated negative consequences (e.g., poor life satisfaction or educational outcome.⁵ It is not, therefore, surprising that cannabis is second only to alcohol in terms of the primary substance for which individuals sought treatment in the past year.²

Unfortunately, although many treatments are available for individuals experiencing problems from cannabis use (e.g., contingency management⁶), a number of barriers preclude successful clinician diagnosis and treatment referral. In a study of Australian nurses and general practitioners, Norberg et al.⁷ highlighted significant barriers to cannabis screening and management, including lack of clinician time, interest, as well as knowledge and skills for treating cannabis problems.⁷ Consistent with these barriers, a study by Bonn-Miller et al.⁴ noted significant underdiagnosis of cannabis use disorders (CUDs) among a sample of U.S. military veterans seen within the Veterans Health Administration.⁸ More recent shifts in clinical practice models toward Screening Brief Intervention and Referral to Treatment (SBIRT) have been successful in addressing other health-risk behaviors, including problem drinking⁹; however one barrier to implementation for cannabis is the absence of a quick, reliable, and valid screening tool.

Equipping clinicians with a shortened screening tool can dramatically increase the feasibility of screening for substance use disorders in a more systematic fashion. For instance, the Alcohol Use Disorders Identification Test,¹⁰ a self-report measure developed by the World Health Organization to identify individuals with alcohol problems, was originally 10 items. Its length, therefore, rendered it less likely for providers to incorporate into routine patient interviews or include in general health history questionnaires.¹¹ In response to this concern, the three-item AUDIT-C¹¹ was developed and validated in the late 1990s and since 2006, it has been regularly administered to every patient seen in VA hospitals, the largest healthcare system in the United States.^12,13 Similarly, provision of a valid, practical screening tool for CUD may help to combat perceived systemic and clinician-level barriers to diagnosing and treating problematic cannabis use.¹⁴ Specifically, implementation of a more formal and systematic assessment approach can serve to increase clinician confidence to detect CUD, make appropriate referrals, and engage clients in the treatment process—ultimately improving the overall system of care.

In an attempt to address some of the existing barriers to clinician screening, a team of Australasian (i.e., Australian and New Zealand) researchers created and later revised the Cannabis Use Disorders Identification Test (CUDIT^15,16). The most recent version of the CUDIT (i.e., CUDIT-R¹⁶) is an 8-item screening tool that assesses individual cannabis consumption, problems associated with abuse, dependence, as well as psychological consequences (Appendix 1). The CUDIT-R shows high levels of specificity and sensitivity in terms of the identification of individuals with a current DSM-IV CUD.¹⁷ Adamson et al.¹⁶ were able to detect 91.3% of individuals with a CUD by using a cut point 13 or above on the CUDIT-R, whereas 90.0% of those without a CUD could be detected by using a cut point below 13.¹⁶

Although the sensitivity, specificity, and psychometric properties of the CUDIT-R are strong, the length of the CUDIT-R is not ideal for administration in busy clinical settings where providers face many competing priorities. Another limitation of the CUDIT-R is that it was developed and validated in accordance with DSM-IV.¹⁷ With the release of DSM-5,¹⁸ which contains significant changes to the classification of CUDs including the recognition and addition of cannabis withdrawal¹⁹ and craving, it is paramount that any new or revised screening tool is validated in accordance with the current diagnostic understanding of CUDs. A third limitation to existing work on the CUDIT-R relates to the population among which it has been validated. Though the CUDIT and CUDIT-R have been widely used in studies outside of Australasia (e.g., Ref.²⁰), aside from the validation of a modified 14-item version of the CUDIT among a Swiss sample,²¹ analyses of sensitivity and specificity have only been conducted among Australasian populations. Accordingly, the present study aimed at filling existing gaps in the literature by utilizing item response theory (IRT) to identify and test a shortened version of the CUDIT-R, which could be used in busy clinical settings (e.g., primary care) to quickly and accurately identify individuals meeting DSM-5¹⁸ criteria for CUD, among two populations: medical cannabis users in the United States, and a community sample of cannabis users in Australia.

Approaches to measurement employing IRT, and Rasch models in particular, demonstrate several advantages over methods based on classical test theory (e.g., Ref.²²). One such advantage, relevant to the current study, is that Rasch models yield a standard error for every parameter estimated (i.e., item and person-level parameters). This stands in contrast to classical test theory, whereby researchers usually rely on a single standard error of measurement applied equally to all ability levels (e.g., levels of problematic cannabis use²²). Such an approach does not allow for examination of the expected degree of measurement error on an item-by-item basis at different ability levels (i.e., levels of problematic cannabis use). Another advantage is the ease with which Rasch and IRT models are often utilized to facilitate a smaller selection of items to assess a latent trait (e.g., proficiency), thereby reducing test length and increasing efficiency (e.g., Refs.^23–25). Accordingly, the objective of the current study was to employ IRT, with an emphasis of Rasch models, to select a subset of CUDIT-R items that optimized the short form's capacity to reliably detect problematic cannabis use (i.e., the latent trait).