Binge drinking (BD) is a major public health problem that produces several harmful consequences (Rolland et al., 2017). As the most prevalent drinking habit among Western youth (Dormal et al., 2019; Maurage et al., 2020), binge drinking increases injuries and health risks, leading to brain and cognitive alterations (Squeglia et al., 2011, 2012; Petit et al., 2014). Beyond immediate health risks, BD produces long-lasting brain and psychological impacts, suggesting that this drinking pattern may be a first step toward drug disorders in the long term (Patrick et al., 2019; Tavolacci et al., 2019; Jaeger and Oshman, 2021). Despite this impact on public health, the research community is overwhelmed with a diversity of definitions and denominations of BD (Rolland et al., 2017). Irrespective of cultural influences on the patterns of consumption, such diversity in the definition of BD leads to high variability among the markers of prevalence (Varela-Mato et al., 2012; Mahmood et al., 2016). Differences in such definitions affect categorizations of BD both within and across studies, and these differences could have a profound impact on both the results and implications of the studies.

Globally, two main definitional streams have emerged among the proposed criteria for BD. The first concerns quantity or quantity-frequency based definitions (Wechsler and Nelson, 2001; Rolland et al., 2017). These definitions define BD with respect to quantity of alcohol consumed, blood alcohol concentration (BAC), and/or standard alcohol dose measures (Maurage et al., 2020) (i.e., the WHO definition of ≥60 g of ethanol per occasion and the NIAAA definition of ≥56 g for women and ≥ 70 g drinks for men in a 2-h interval and blood alcohol concentration ≥ 0.8 g/L). The strictly quantity-based definitions have two main drawbacks, despite their ease of use and prevalence in epidemiological studies. First, accurate estimates of alcohol consumption are difficult to achieve for several reasons: (i) the amount of alcohol in a “standard drink” varies between countries (e.g., 10 g in France, 20 g in Austria, 14 g in US, and 8 g in UK); (ii) young adults frequently consume non-standardized drinks (plastic cups, etc.) (Morrell et al., 2021) in private dwellings (e.g., at home) or in a public place (e.g., public park) (Labhart et al., 2013) and (iii) the number of standard doses consumed does not take into account the consumer’s physical characteristics (tolerance, sex and body mass index) that affect BAC. Second, the use of “cut-offs,” or delineating participant groups according to the number of drinks consumed (e.g., fewer than 5/4 drinks for men/women as “non-binge” drinking), generates erroneous dichotomization and inaccurate labeling (i.e., broad categorization into a group) (DeCoster et al., 2009; Pearson et al., 2016). The use of cut-offs underestimates real consumption and mitigates actual consequences. As a dichotomous variable, BD has little predictive value for public health impact whereas a clear dose effect links BD frequency, intensity, and many negative health consequences, including mortality (Donat et al., 2021). The addition of a frequency dimension to quantity (i.e., quantity-frequency) improves the drinking assessment, as frequency is highly related to severity of drinking consequences (Williams et al., 1994; Greenfield, 2000). For example, the Timeline Follow Back (TLFB) (Collins et al., 2008) has been used to track BD episodes during a period of 7 days to 12 months. The degree to which a given quantity of consumption occurs within that time period provides an important temporal dimension (Williams et al., 1994; Greenfield, 2000).

The second stream of BD definitions concerns behavioral patterns (Townshend and Duka, 2002), or more subjective behavioral drinking “phenotypes,” such as frequency of drunkenness and hangovers. Such definitions include the adaptations of known questionnaires to rank consumption (Tuunanen et al., 2007; Olthuis et al., 2011; Cortés-Tomás et al., 2016). For example, the AUDIT questionnaire is sometimes used to assess the severity of BD (Motos-Sellés et al., 2020), although it remains incomplete for several reasons (Shakeshaft et al., 1998; Tuunanen et al., 2007; Mota et al., 2010; Olthuis et al., 2011; Cortés-Tomás et al., 2016; Motos-Sellés et al., 2020). In particular, the AUDIT questionnaire does not assess the number of drunkenness episodes or the speed of drinking, which are essential characteristics of BD behavior (Maurage et al., 2020). In addition, cut-offs for AUDIT total scores require careful adjustment to remain valid in college populations (Olthuis et al., 2011). To deal with these limitations, other behavior-based models of BD have been developed to include physiological and subjective intoxication criteria such as drunkenness and frequency of drunkenness. For example, the BD score proposed by Townshend and Duka (2002) includes a sum of consumption speed, number of drunkenness episodes, and percentage of drunkenness episodes out of 10 drinking occasions. Because of their subjectivity, drunkenness and frequency of drunkenness are recognized as both highly predictive of social consequences and symptoms of alcohol dependence and alcohol-related harm, and therefore, may be more valid than the amount of alcohol by subject experience, biological factors and demographic criteria (Midanik, 1999; Paljärvi et al., 2012; Sznitman et al., 2017). Speed of drinking, also included in the “BD score,” is another index of severity (López-Caneda et al., 2013; Poulton et al., 2016), increasing more rapidly in high-risk drinking (Groefsema and Kuntsche, 2019). Despite these advantages of the BD score, its utility is hampered by the percentile method used to categorize groups of participants: The lowest third is said to constitute “social drinkers” and the upper third is considered “binge drinkers,” but the intermediate percentile remains unclear and designated “intermediate.” Thus, these cutoffs will necessarily vary depending on the distribution of consumption in a given sample (Townshend and Duka, 2002).

In addition to wide variability among BD definitions, there is also variability in the inclusion of gender in these definitions. Indeed, some studies use gender-specific measures (e.g., 5+ and 4+ US standard drinks, or 70 g+/56 g+, respectively for men and women) (Wechsler and Nelson, 2001), whereas others do not (e.g., the AUDIT uses 60 g of ethanol) (Saunders et al., 1993; Babor et al., 1994; Graham et al., 1998; Wilsnack et al., 2018). Thus, lowering a threshold to define BD among women may increase the prevalence of this behavior among women (Chavez et al., 2012). Even though physiological factors, such as drunkenness, depend on the metabolism of alcohol in accordance to sex, no study to date has been able to quantify binge drinking in a manner that is valid regardless of gender.

This lack of standardization and consensus in BD definitions illustrates the variability that makes it difficult to compare findings across studies (Cortés-Tomás et al., 2016). This methodological difficulty has become a barrier to advancing public health, given the pervasiveness of BD in the population and its spectrum of consequences (Rolland et al., 2017; Maurage et al., 2020). Taken together, the adverse consequences of BD, the lack of clarity regarding BD as a mode of consumption and resulting obfuscation of research findings justify the need for a novel instrument to define and detect BD (Cortés-Tomás et al., 2016).

Therefore, the current study had three research aims: (1) starting with recognized BD criteria, develop a better tool for characterizing BD behavior that takes into account behavioral, quantitative and physiological consumption features and that defines homogeneous groups keeping all individuals from the whole population (2) with this tool, provide a stable, specific, and reproducible way to investigate other BD-related questions, especially those regarding severity and gender variability; and (3) further characterize BD groups using alcohol-related consequences and behavioral economic indices (e.g., demand intensity and breakpoint).

The goal of the current study was to develop an objective and simple tool for identification and characterization of BD for both genders based on a progressive severity. To our knowledge, this is the first study to propose an integrative model of BD that captures comprehensive features of BD and has led to a global definition combining 5-item (Appendix, online access to two tools: a population sample classification tool,¹ and an individual characterisation tool²).

The analyses we performed on our student learning sample revealed 4 homogeneous groups, defined by 5 salient items that together, capture a set of key BD factors (Maurage et al., 2020) including quantitative aspects of consumption (quantity-frequency with “6-drink frequency”), behavior (speed of consumption), and physiology (frequency of drunkenness, frequency of hangover). As such, ours is the only existing model that combines all of these key factors in a single and consistent measure. Moreover, our use of the statistical PPOM validation has 2 key advantages: (i) it ensures an objective, reliable and reproducible model, and (ii) it avoids the duplication of highly correlated items and selects only salient and statistically significant factors (i.e., 6-drink frequency and not consumption frequency nor drinks per typical day). The internal validation and low classification error of the PPOM that we found confirm construct validity of the 4 groups (McAloney et al., 2013).

The magnitude of all 5 final items progressively increased from group 1 to group 4 (Table 3), with only minor exceptions between the two most severe groups (3 and 4). In addition, the regression equations effectively predicted a participant’s probability of belonging to each group. In contrast to the methods using cut-off delimitations in the identification of the population groups, the present approach does not constrain analysis of BD into a dichotomous variable (McAloney et al., 2013). Rather, our model describes a progressive severity to categorize the binge-drinking level of a given participant and accurately map the drinking pattern.

Following the PPOM validation, the external validation confirmed the reproducibility of our method and generalization to two additional samples (Debray et al., 2015; Ramspek et al., 2021). The external validation samples differed from the learning population in many ways, such as sample size, gender ratio, and alcohol consumption metrics (Table 2). Despite these differences, our findings reveal a strong reproducibility of our validation items and confirm that the model variables are robust and that the groups are consistent. Thus, unlike prior studies (Townshend and Duka, 2005), the current model provides a valid characterization of BD independent of the characteristics of the sample.

To further confirm the validity of the 4 groups, they were reassessed using known criteria related to alcohol use and consequences. Specifically, we found that percentage of participants in high consumption scores sub-groups (high BD and AUDIT scores) and high adverse consequences sub-groups (AUDIT 2, 4–10, 14; first drink before 12 years old; consumption of 4 or more drinks in one occasion) as well as means of consumption scores (Table 8), and behavioral economic indices (Table 9) gradually increased across groups. Our behavioral economic results confirm consumption criteria. Previous studies reported correlations between behavioral economic indices and risky drinking. APT indices are correlated with measure of alcohol use, alcohol-related consequences, and alcohol use disorder criteria (Gaume et al., 2022). In particular, the demand metrics of “intensity” (i.e., amount of alcohol consumed at zero or very low price, or “How much alcohol would you consume if alcohol were free?”) and Omax (i.e., maximum expenditure on alcohol in one occasion) predict clinical alcohol problems, beyond alcohol consumption alone (MacKillop and Murphy, 2007) and 6-month binge drinking and alcohol problems (Dennhardt et al., 2015). Furthermore, since we did not observe a difference in elasticity (alpha) between the groups, our results do not support addictive behavior in the binge drinking group.

These results highlight both the homogeneity within each group and the severity scale across groups. Therefore, we propose the following category labels based on the characteristics of each group: low-risk drinking (group 1), hazardous drinking (group 2), BD (group 3) and high-intensity BD (group 4). By differentiating binge drinking and high-intensity BD as we do for groups 3 and 4, respectively, our approach offers new opportunities to better identify and describe individuals with these drinking profiles (Gmel et al., 2007; Read et al., 2008; Patrick and Azar, 2018; Fairlie et al., 2019). Indeed, the present results reveal some distinctions. For example, compared to group 3, group 4 users are more likely to start drinking before the age of 12, consume more than 10 drinks (100 g) per occasion, fail to do what is expected, need a first drink in the morning, experience weekly black-outs, and drink 4 (40 g) or more drinks per hour. In addition, only group 4 users reported being drunk more than 30 times in the last 6 months meaning at least one episode of drunkenness per week (Table 8). Participants in group 4 also display greater behavioral economic demand, specifically with significantly higher demand intensity, consistently associated with clinical severity of alcohol use.

Results of consumptions items and behavioral economics items confirm the severity of the consumption-related consequences in the high-intensity BD. High-intensity BD is linked to alcohol-related injuries, alcohol poisoning, risky sexual behavior, vomiting, fainting, long-term damage (Patrick and Azar, 2018). The link between high-intensity BD and addictive behaviors will be further investigated. Because high-intensity BD is identified as a strong prospective marker of risk for AUD symptoms in adults (Patrick et al., 2021), our tool could be decisive in identifying this drinking profile early and thereby improving prevention (Enoch et al., 2006; Maurage et al., 2020). Considering the widespread use of BD in adolescents and young adults, the accurate identification of High-intensity BD is highly relevant for providing personalized feedback and awareness-raising to these populations.

Another strength of our model is its applicability across genders (Table 6). In fact, adding the gender factor to the regression equations had no effect on misclassification error, indicating stability of a participant’s classification in one of the 4 groups, regardless of gender. In our exploratory analyses of gender differences, we found differences in drinking behavior within groups 1 and 2, but not in the more severe groups (Table 11). Given that social factors impact drinking habits differently between genders (Graham et al., 1998; Wilsnack et al., 2018), these results may indicate that the impact of such factors is stronger in the less severe groups compared with the more severe ones. The reliability of our model for each gender opens up the ability to study gender-specific differences in consumption behavior with a high degree of confidence.

Our results reveal three additional aspects of interest. First, the frequency of hangovers differentiates low risk drinking from other profiles. Specifically, we found that groups 2, 3, and 4 reported significantly more frequent hangovers compared with group 1, suggesting that hangover is specific to risky drinking patterns; however, we also found that the frequency of hangover was significantly lower in group 4 compared with group 3. This finding indicates that hangover does not linearly increase with severity of consumption (Swift and Davidson, 1998) and that high-intensity BD group may not be motivated by experiencing hangover. Second, the frequency of drunkenness is weighted heavily and effectively differentiates hazardous drinking from BD and, even more powerfully, BD from high-intensity BD (Tables 3, 6). These findings confirm that drunkenness is a strong marker of BD severity and should be taken into account in such definitions, as noted by others (Lannoy et al., 2021). Third, and perhaps most importantly, our model confirms that BD should be considered over time and not as a single instance of behavior (i.e., not simply a single instance of 4+/5+ drinks). As noted by Gmel et al. (2011), the frequency of “risky single-occasion drinking” (RSOD) may differ widely between groups, with moderate drinkers displaying RSOD rarely (Gmel et al., 2011). A model that considers the chronicity of BD episodes is necessary to discriminate consequences of occasional BD from chronic BD, in terms of (a) severity of harm and tolerance and (b) experience in managing these effects at equivalent levels of blood alcohol concentrations (Gmel et al., 2011). Our model addresses both of these needs and fills an important gap noted by several others.

Despite its strengths, our model has some limitations. For example, the learning sample and both validation samples originated from only one country. Although the three samples differed in many ways (such as average level of consumption and gender distribution), replication is warranted in future studies, with external geographical validation, to confirm that the model remains valid despite cultural differences. Moreover, a larger sample would allow the recruitment of more participants displaying high-intensity BD to better characterize this type of drinking. Reliance on self-report is also a limitation (Creswell et al., 2020) since young adults may minimize or exaggerate their drinking levels. Another limitation of this study is that it is conducted in a student population. Future studies may look at other populations in terms of age and geographical region. Our model could indeed address the growing need to study BD in middle-aged adults, particularly among women (age 30–44), for whom the incidence of BD has nearly doubled in the past decade (Han et al., 2019; Patrick et al., 2019). Finally, the reference period for the AUDIT is 1 year, and thus the recall for past drinking experiences may have been less reliable on these items.

In conclusion, the current model provides a novel approach to characterizing BD based on a strong statistical model. This model combines salient items with strong clinical validity to offer an objective pattern of BD consumption, independent of consumption cut-offs and population type, that is applicable regardless of gender. We now have a useful instrument to assess each participant’s BD severity [Appendix and online access (see footnote 1 and 2)] or to identify 4 homogeneous groups of alcohol drinking in a whole sample and differentiate types of drinking, including BD and high-intensity BD with a gradual severity [Appendix and online access (see footnote 2)]. With a simple to use model, the BD severity of each subject, male or female, can be determined according to a global definition that includes consumption, behavioral, and physiological criteria.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Comité de Protection des Personnes Nord-Ouest II and Comité de Protection des Personnes Est I. The patients/participants provided their written informed consent to participate in this study.

JA and MN developed the study. JA, MN, MM, and MD contributed to the study design. JA and FG conducted the data collection and data analyses were performed in collaboration with JA, MN, MM, MD, and OO. JA drafted the manuscript in collaboration with MN, MM, and MD. OP provided critical revisions. FF provided the website. All authors approved the final versions of the manuscript.

The study has been supported by the INCa national institute of cancer and IReSP research institute on public health, grant N° INCa/IReSP 16134. The present study was supported by the INSERM and the University of Picardie Jules Verne.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.