Substance Use Disorders (SUD) and addictive disorders may be defined as a dysregulation of control over use of rewarding substances and behaviors that translates into repeated and excessive use despite negative consequences (Auriacombe et al., 2018). Among people who use substances, those with SUD concentrate a diversity of negative health consequences such as cardiovascular diseases, cancer and accidents. Prevention of such consequences is contingent on early more accurate diagnosis and early treatment access. Currently, the diagnostic criteria of SUD lump central criteria related to loss of control over use, and consequential criteria to excessive use (American Psychiatric Association, 2013, Auriacombe et al., 2018). There are conceptual and measurement problems with using consequences as diagnostic criteria (Martin et al., 2014). The current major diagnostic challenge for SUD and addictive disorders is to target its central or etiological criteria that should be exclusive for diagnostic purpose and are to be distinguished from its consequences that should have a value for severity assessment only.

Craving, defined as an unwanted persistent and intense desire to use despite negative consequences, was introduced in the DSM-5 revision of the SUD diagnostic criteria (American Psychiatric Association, 2013, Hasin et al., 2013). Daily fluctuations of craving among patients in treatment have been prospectively associated with substance use in following hours (Fatseas et al., 2015, Serre et al., 2015), and higher risk of return to problem use after use suspension in the following months or years (Cleveland et al., 2021), giving craving an etiological and prognostic value of use in addiction (Vafaie and Kober, 2022). Due to its predictive role, craving is already a treatment target for some authors (Auriacombe et al., 2018, Cavicchioli et al., 2020, Sayette, 2016, Serre et al., 2018), and is a potential central criterion for the diagnosis of SUD (Auriacombe et al., 2018, Drummond et al., 2000, Hasin et al., 2013, Hughes, 1992, Shmulewitz et al., 2021, Tiffany and Wray, 2012, Vafaie and Kober, 2022).

Latent variable methods such as item response theory (IRT), have been the preferred approach for assessing dimensional and structural validity of diagnostic criteria (Embretson and Reise, 2000), IRT analyses applied to the DSM-5 SUD criteria have shown that craving fit well with the other criteria on the underlying unidimensional latent SUD variable. Importantly, craving was the least difficult criterion and the most discriminant of all criteria when comparing across substances (Kervran et al., 2020) and samples of people with regular use (Castaldelli-Maia et al., 2018, Chung et al., 2012, Gilder et al., 2014, Shmulewitz et al., 2011, Shmulewitz et al., 2021). Craving appears to capture subjects with less severe SUD well, suggesting its potential utility as an early indicator of SUD (Chung et al., 2012, Kervran et al., 2020).

Recent research proposes that diagnostic criteria are dynamically directly related to each other, and that their individual analysis may provide valuable results compared to IRT method (Epskamp et al., 2017, Fried, 2015, Schlechter et al., 2021). Symptom network analysis explore the interrelationships between symptoms, considering that they are causally mutually dependent, and influence each other accordingly in these network conceptualizations (Schlechter et al., 2021). The symptom network model uses pairwise interactions among symptoms (i.e., diagnostic criteria) to represent a disorder as a web of mutually influencing symptoms (Borsboom and Cramer, 2013). This kind of model is now used in a large and diverse number of disorders (Fried, 2015, McNally et al., 2017), such as depression and anxiety (Beard et al., 2016) or post-traumatic stress disorder (McNally et al., 2014). Symptom networks based on diagnostic criteria provide an overview of the connections between symptoms. The inferences offered by such networks allow a better understanding of how disorders develop, are maintained and could be treated (Borsboom and Cramer, 2013). Especially, centrality in symptom networks could be compared across several substances to highlight the different positions occupied by each criterion in the different SUD. These positions are partially related to the distinct pharmacological and psychological properties unique to each substance and partially related to the core of loss of substance use control (disordered use) that is expected to be across the different addictions (Rhemtulla et al., 2016, Rutten et al., 2021).

Better understanding of the network structure of the SUD diagnostic criteria could be relevant to increase diagnostic accuracy. For the DSM-III-TR SUD diagnostic criteria, network analysis showed the centrality of the “Large amount” criterion (Fried et al., 2016, Rhemtulla et al., 2016) and more recently, network analysis comparing DSM-IV and DSM-5 SUD criteria showed a high centrality of the “Time spent” on substance use criteria (Rutten et al., 2021). However, in this study the DSM-5 network analyses did not include the recently added craving criterion, but only an estimate by dichotomizing the score of the abridged 5-items version of the self-administered Obsessive-Compulsive Drinking Scale (OCDS) (Rutten et al., 2021). It would be interesting to investigate the DSM-5 diagnostic criteria for SUD, with all 11 diagnostic criteria directly explored.

The main goal of this paper was to explore symptom centrality to identify which diagnostic criteria may be most important in SUD, by investigating symptom interconnectivity in cross-sectional network analyses of DSM-5 SUD diagnostic criteria for alcohol, opioids, cocaine, cannabis and tobacco. The original hypothesis was that the craving criterion, added in the DSM-5 revision, may play a significant role in these networks regardless of the substance. Its centrality is potentially fundamental in terms of occurrence, maintenance and treatment of addiction.