The Gambling Disorders Identification Test (GDIT): Psychometric Evaluation of a New Comprehensive Measure for Gambling Disorder and Problem Gambling

Prevalence estimates of GD are scarce. Prevalence investigations have traditionally built on the concepts of “problem gambling” or “at-risk gambling,” two public health-based terms, where the former has been defined as “excessive gambling behavior that creates negative consequences for the gambler, others in his/her social network, and for the community” (Blaszczynski & Nower, 2002, p. 488), and the latter has been defined as “[gamblers] heavily involved in gambling, . . . [who] may or may not have experienced adverse consequences from gambling” (Ferris & Wynne, 2001). A systematic analysis found large variations in past-year prevalence for problem gambling worldwide, ranging from 0.12% to 5.8%; prevalence estimates from different studies were difficult to compare, as a variety of gambling instruments, cut-offs and time frames had been used to assess problem gambling (Calado & Griffiths, 2016). These measurement issues have been noted in additional reviews (see, e.g., Caler et al., 2016; Williams et al., 2012). In the same vein, gambling treatment has been characterized by a wide range of diverse outcome measures and domains assessing problem gambling (Toneatto & Ladouceur, 2003), causing difficulties when comparing effectiveness of different treatment approaches (Pallesen et al., 2005).

To address these measurement issues, an expert panel of gambling researchers convened in 2004, at the Alberta Gambling Research Institute’s 3rd Annual Conference (Walker et al., 2006). A consensus-based framework known as the BCA, was formulated, specifying a set of minimal features of gambling outcome measures within three domains: (a) gambling behavior (net expenditures per month, frequency of gambling in days per month, and time spent thinking about gambling per month); (b) problems caused by gambling (mental health, relationships, financial, employment and productivity, and legal [a criterion later removed in the DSM-5; American Psychiatric Association, 2013]); and (c) treatment-specific measures of proposed mechanisms of change. Since then, the BCA has been influential as the basis for a proposed core set of reporting standards for gambling treatment studies, although it is unclear whether studies have actually adhered to these recommendations (Pickering et al., 2017), possibly due to construct underrepresentation (Spurgeon, 2017) in existing gambling measures.

While comprehensive reviews examining measurement content analysis in relation to the BCA seem to be lacking, both Pickering et al. (2017) and Molander et al. (2019) commented that most existing gambling measures appeared to fail to fulfill the recommendations of the BCA.

Using Domains 1 (gambling behavior) and 2 (problems caused by gambling) of the BCA as benchmarks, we analyzed the content of seven of the most frequently used gambling outcome measures identified in a systematic review (Pickering et al., 2017), as well as the frequently used public health measure problem gambling severity index (PGSI; Ferris & Wynne, 2001); see Table 1. Measures mostly included items for gambling-related financial, and relationship problems, but assessed health-related problems and gambling behavior more rarely. In addition, to fulfill the BCA recommendations on gambling behavior, it is necessary for item response alternatives to specify time units per month. Most measures, such as SOGS (Lesieur & Blume, 1987) and the NORC Diagnostic Screen for Gambling Problems (NODS; Wickwire et al., 2008), had dichotomous “Yes” or “No” responses or, such as in the PSGI (Ferris & Wynne, 2001), or verbal response alternatives, for example “not at all,” “rarely,” “sometimes,” or “often”; these are also aggravating factors for temporal ambiguity and poor item readability (i.e., difficulties in knowing how to interpret items; see Peter et al., 2018). In conclusion, none of the individual measures, nor any combinations of the measures analyzed seemed to sufficiently fulfill the features of BCA. Regarding Domain 3 of the BCA, this encompasses measures dependent on treatment-specific assumptions of therapeutic change, meaning that it was not feasible to capture all these possible theoretical constructs in a single instrument. This domain was therefore excluded from our analysis.

As a response to these measurement challenges, a process was initiated to develop the Gambling Disorder Identification Test (GDIT), as a gambling measure analogous to the Alcohol Use Disorders Identification Test (AUDIT; Saunders et al., 1993), and the Drug Use Disorders Identification Test (DUDIT; Berman et al., 2005). The GDIT has been developed in several interdependent steps, using the BCA (Walker et al., 2006) as an overall benchmark. In the first step (Molander et al., 2019), four gambling researchers selected 30 items for proposed inclusion in GDIT, based on content analysis and categorization of a pool of 583 unique items from 47 existing gambling measures. In the second step, preliminary construct and face validity were established. Sixty-one gambling experts from 10 countries rated the 30 items proposed for inclusion in the GDIT, in an online Delphi process (see Molander et al., 2020), which is a systematic iterative method to determine expert consensus for research questions that are not suitable for experimental or epidemiological research designs, such as determining collective values or defining foundational concepts (Jorm, 2015; Yücel et al., 2019). Gambling researchers and clinicians participated in subsequent consensus meetings, yielding a 14-item draft version of GDIT, within three proposed domains. Gambling behavior (GDIT_{Items 1-3}), corresponded to the features of gambling behavior in Domain 1 of the BCA. Gambling symptoms (GDIT_{Items 4-10}), included key GD symptoms, which were not recommended in the BCA. Negative consequences (GDIT_{Items 11-14.}), corresponded to the features of problems caused by gambling, in Domain 2 of the BCA. In addition, an appendix to the GDIT assessed gambling expenditures and gambling types separately. Domain 3 of the BCA, measures of the processes of change, was not included in the GDIT, as these features could include a range of items assessing a multitude of treatment-specific assumptions and theoretical constructs. In similarity with the AUDIT (Saunders et al., 1993) and the DUDIT (Berman et al., 2005), the GDIT used frequency and time-based multiple choice response alternatives, which corresponded to the temporal features of the BCA. Face validity and item readability of the GDIT draft version were evaluated by obtaining user feedback from eight individuals with experience of problem gambling, as well as eight treatment-seeking participants fulfilling the criteria of GD (see Molander et al., 2020, for more details). In the final step, evaluation of the GDIT regarding psychometric properties was conducted among 603 gamblers from treatment- and support-seeking contexts, self-help groups, and population samples. The aim of this article is to report the results of this final step.

This study evaluated the psychometric properties of a novel gambling measure, the GDIT. The GDIT was developed in a recent international Delphi and consensus process, aiming to establish a comprehensive measure which corresponded to a previous international research agreement regarding features of gambling outcome measures, known as the BCA. A further aim was to develop a measure that could identify and assess fulfilment of the revised diagnostic criteria for GD, for example, levels of symptom severity.

GDIT total scores increased continuously when comparing samples with increasing levels of presumed symptom severity, that is, recreational gamblers, support-seeking gamblers, gamblers in self-help groups, and treatment-seeking gamblers. The GDIT showed estimates of convergent and discriminant validity that corresponded to theoretical expectations, and excellent estimates on most reliability statistics. Factor loadings based on confirmatory factor analysis were excellent to very good with the exception of one item, and supported the three proposed theoretical domains: gambling behavior (GDIT_{Items 1-3}), gambling symptoms (GDIT_{Items 4-10}), and negative consequences (GDIT_{Items 11-14}). Finally, ROC and clinical significance cut-off estimates yielded approximate GDIT cut-off score ranges for assessment of gambling severity. Overall, we conclude that GDIT can be viewed as a valid and reliable measure to identify and predict severity of GD, as well as problem gambling.

Regarding item inclusion, some psychometric estimates indicated possible construct irrelevance (Spurgeon, 2017), suggesting that the GDIT could be shortened. The GDIT_{Item 14}, measuring school or work-related problems due to gambling, showed lower, but still acceptable, reliability estimates (i.e., corrected item-total correlations and factor loadings), in comparison with other GDIT items The GDIT_{Item 14} was also problematic in terms of measurement invariance for men and women, specifically scalar invariance. For young adults (18-30 years) and older, the analyses of measurement invariance showed differences for most of the items (GDIT_{Item 12}, GDIT_{Item 13}, and GDIT_{Item 14}) within the negative consequences domain. The finding of invariance for negative consequences was not surprising, as gambling-related negative consequences can be expected to affect younger and older individuals differently due to varying life circumstances. Also, several GDIT items had high corrected item-total correlations.

Although these findings all indicate that GDIT items could to some extent be eliminated, we are mindful that the BCA (Walker et al., 2006) recommended that the features covered by these items should be included in gambling measures; furthermore, the aforementioned items were also rated in the Delphi expert process as important to include in the GDIT (Molander et al., 2020). Therefore, we decided to retain the GDIT item structure that resulted from the consensus meetings, prioritizing content validity over performance. Still the lower performance of GDIT_{Item 14}, in relation to the other GDIT items is noteworthy. In the GDIT item selection (Molander et al., 2019), it became evident that many items from previous gambling instruments assessed gambling-related school, work, and relationships problems lumped together in single items, equivalent to current and previous DSM diagnostic criteria (American Psychiatric Association, 1994, 2013). This particular double- and triple-barreled phrasing issue was also observed in the Delphi process (Molander et al., 2020) and was later adjusted, during the consensus meetings, into two GDIT items, assessing relationship problems in one item (GDIT_{Item 13}), and gambling-related school or work problems in another (GDIT_{Item 14}). Although the BCA (Walker et al., 2006) recommends that gambling-related problems concerning employment and productivity should be measured as a single feature in gambling instruments, most psychometric studies have evaluated gambling-related problems regarding work and/or relationships combined (i.e., in single items using double-barreled phrasing). The lower performance of GDIT_{Item 14}, compared with GDIT_{Item 13}, might indicate that gambling-related work problems (quite apart from relationship problems), might be less relevant than previously assumed. Anecdotally, several participants expressed during the SCI-GD interviews that their gambling actually made them more efficient at work rather than causing problems, as working was their primary source of obtaining money to continue gambling.

While showing promising psychometric properties, the largest contribution of the GDIT to the gambling field will probably lie in content analysis from a theoretical point of view. The GDIT closes a gap with the inclusion of several constructs that have been recommended in previous consensus-based agreements among gambling researchers (Walker et al., 2006), but have been lacking in existing gambling instruments. Furthermore, the GDIT was developed analogously to the AUDIT (Saunders et al., 1993) and the DUDIT (Berman et al., 2005), in line with the DSM-5 decision to equate gambling with alcohol and substance use, all as addictive disorders (American Psychiatric Association, 2013). The GDIT includes frequency- and time-based assessments, which is an advantage compared with most existing gambling instruments. For example, the gambling behavior domain (GDIT_{Items 1-3}), assesses gambling behavior in frequency of events and duration of hours, using response scales similar to the AUDIT (Saunders et al., 1993) and the DUDIT (Berman et al., 2005). This enables future research to examine specific time-related cut-offs for gambling behavior in relation to gambling-related constructs, as well as to facilitate comparisons between gambling behavior and substance use. Furthermore, although outside the scope of this evaluation, GDIT encompasses additional assessment of gambling expenditures in relation to income, as well as differentiation between gambling types at online versus physical venues.

This study had several strengths. We included gamblers from several populations, enabling us to present sample-specific estimates within a broad range of gambling contexts. The psychometric evaluation was preceded by several documented and interdependent development steps, including results from an international Delphi process with a large proportion of currently active gambling researchers participating, and consensus-based agreements among gambling researchers (Molander et al., 2019, 2020). While this is the first study evaluating psychometric properties of GDIT, it is also one of the first psychometric studies estimating diagnostic accuracy of a gambling measure in relation to semistructured diagnostic interviews assessing GD, also considering levels of severity. This is important, as a recent systematic review of gambling screening instruments (Otto et al., 2020), concluded that there is a lack of evidence for GD diagnostic accuracy. This study showed different GDIT cut-off scores for each GD severity level, which is important from dual perspectives of clinical utility and public health. For example, it is now possible to concurrently conduct prevalence estimates of diagnosed GD in different populations, in addition to assessing the prevalence of problem or at-risk gambling.

The study also had some limitations. One limitation was that GDIT estimates of prevalence accuracy for at-risk and problem gambling were more uncertain than the estimates in relation to GD, as they did not rely on comparisons with semistructured interviews, but rather on other self-report instruments: that is, PGSI and PPGM. As a complementary analysis, we estimated clinical significance (Jacobson & Truax, 1992), which indicated that problem gambling, but not at-risk gambling, corresponded to the cut-off differentiating between recreational and help-seeking gambling populations. Overall, at-risk and problem gambling are more loosely defined terms than GD, and it is less clear how to validly establish cut-off scores for these constructs. A second possible limitation was that there were small differences in demographic characteristics between the total sample, and the subsamples that complemented the GDIT retest and diagnostic interviews. An exception was that the proportion of participants with gambling debts was higher in the subsample that completed the diagnostic interviews compared with the total sample (64% vs. 36%), but smaller compared with the treatment-seeking cohort (64% vs. 85%). The higher proportion of gambling debts among those who completed a diagnostic interview might indicate a higher gambling severity. This is not necessarily a limitation of the study, as the primary purpose of the diagnostic interview was to assess a sufficient subsample of participant within the full GD severity spectrum (i.e., no GD, mild GD, moderate GD, and severe GD), to be able to estimate diagnostic accuracy of the GDIT. A third limitation is the low overall internal consistency for the negative consequences scale, as well as the low test–retest estimates for the gambling behavior subscale among treatment-seeking gamblers. The lower internal consistency for the negative consequences domain, among the recreational and treatment-seeking gamblers, may have to do with Cronbach α being contingent on item number, where the negative consequences domain is only four items. This emphasized the need to use the GDIT as a full instrument, as opposed to using separate subscales for each domain. Fourth, test–retest estimates for gambling behavior were lower among treatment-seeking gamblers, compared with the total score and the other domains. This is not necessarily a limitation, but rather may reflect the inclusion in this domain of items assessing behaviors that capture short-term changes, which may be typical of treatment-seeking gamblers. The test–retest analysis thus includes a measure both of reliability, but also of actual behavior change, specific to the treatment-seeking group. A final limitation of the study was that assessment of gambling-related expenditures in the GDIT appendix was not validated; this assessment was included in the GDIT as a recommended feature from the BCA.

Future research should thus include validation of GDIT assessment of expenditures and corroboration of GDIT severity score ranges. In addition, future psychometric evaluation of GDIT could include examination of item structure using item response theory (Reise & Waller, 2009), or validation through comparisons of GDIT scores between different gambling types in physical or digital milieus. Of special interest are also international comparisons among different gambling samples.

The authors would like to acknowledge the following recruiting sites: the Centers for Dependency Disorders in Stockholm, Eskilstuna and Örebro, the Addiction Units in Mjölby and Visby, the Outpatient Care Center in Örebro municipality, the social services in Kristinehamns municipality, Game Over in Linköping, Junepol in Jönköping, the Gambling Group in Eskilstuna, the local Associations for Gambling Addiction in Stockholm and Borlänge, as well as the Swedish national gambling helpline, Stödlinjen. Furthermore, diagnostic interviews were conducted by Viktor Månsson, Martin Bjurek, Jenny Svensson, Pernilla Swartz; Amanda Bergvik, Carina Wigren conducted diagnostic interviews and a preliminary psychometric investigation of GDIT (Bergvik & Wigren, 2019) under Philip Lindner’s supervision. Statistical guidance regarding diagnostic assessment was contributed by Håkan Källmén. Per Binde, Kristina Sundqvist, Viktor Månsson and Rachel Volberg made significant contributions to the GDIT development process. Lastly, the authors would like to thank the anonymous reviewers and editor, who contributed with several valuable suggestions for improving the article.