Revising the Cannabis Use Disorders Identification Test (CUDIT) by means of Item Response Theory

Abstract

Cannabis use among adolescents and young adults has become a major public health challenge. Several European countries are currently developing short screening instruments to identify 'problematic' forms of cannabis use in general population surveys. One such instrument is the Cannabis Use Disorders Identification Test (CUDIT), a 10-item questionnaire based on the Alcohol Use Disorders Identification Test. Previous research found that some CUDIT items did not perform well psychometrically. In the interests of improving the psychometric properties of the CUDIT, this study replaces the poorly performing items with new items that specifically address cannabis use. Analyses are based on a sub-sample of 558 recent cannabis users from a representative population sample of 5722 individuals (aged 13-32) who were surveyed in the 2007 Swiss Cannabis Monitoring Study. Four new items were added to the original CUDIT. Psychometric properties of all 14 items, as well as the dimensionality of the supplemented CUDIT were then examined using Item Response Theory. Results indicate the unidimensionality of CUDIT and an improvement in its psychometric performance when three original items (usual hours being stoned; injuries; guilt) are replaced by new ones (motives for using cannabis; missing out leisure time activities; difficulties at work/school). However, improvements were limited to cannabis users with a high problem score. For epidemiological purposes, any further revision of CUDIT should therefore include a greater number of 'easier' items.

Figure 1

CRC; CUDIT, supplemented (items 1–14: left to right, top–down). Note: n = 558; estimator = maximum likelihood with robust standard errors (MLR; Mplus). CRCs indicate the probability of endorsing the item at a given response category across locations on the latent trait. Well performing items have CRCs with sharper and more differentiated peaks across response categories.

Figure 2

IIC; CUDIT, supplemented. Note: n = 558; estimator = maximum likelihood with robust standard errors (MLR; Mplus). IIC show the information given by the single items along the latent trait. The higher an IIC is the more information is given.

Figure 3

TIC; comparison of CUDIT, original and CUDIT, revised. Note: n = 558; estimator = maximum likelihood with robust standard errors (MLR; Mplus); TIC show the information given by a whole test along the latent trait. The higher an TIC is, the more information the test delivers.