Conduct problems and depressive symptoms in association with problem gambling and gaming: A systematic review.

BACKGROUND AND AIMS: Behavioral addictions such as gambling and gaming disorder are significant public health issues that are of increasing importance to policy makers and health care providers. Problem gambling and gaming behaviors have been identified as being associated with externalizing and internalizing problems, with theoretical models suggesting that both conduct problems and depressive symptoms may be significant risk factors in the development of problem gambling and gaming. As such, the purpose of this systematic review is to provide an overview of research identifying the relationship between conduct problems, depressive symptoms and problem gambling and gaming among adolescents and young adults.
METHODS: Systematic literature searches in accordance with PRISMA guidelines found 71 eligible studies that met the inclusion criteria, 47 for problem gambling, 23 for problem gaming and one for both problem behaviors.
RESULTS: Based on cross-sectional evidence, both problem gambling and gaming are consistently concurrently associated with conduct problems and depressive symptoms. Longitudinal evidence appears to be clearer for conduct problems as a risk factor for problem gambling, and depressive symptoms as a risk factor for problem gaming. However, both risk factors appear to increase the risk for these problem behaviors. DISCUSSION AND
CONCLUSIONS: Results from the literature review suggest that problem gambling and gaming are associated with the presence of conduct problems and depressive symptoms, with the potential of sharing common etiological factors. Additional research is necessary to confirm these longitudinal relationships with an emphasis on investigating the interaction of both early conduct problems and depressive symptoms.

Introduction

Gambling and gaming behaviors are generally described as being on a continuum ranging from occasional or recreational participation, through to problematic, excessive and disordered engagement (Griffiths, Kuss, Lopez-Fernandez, & Pontes, 2017; Shaffer & Korn, 2002). In the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association [APA], 2013), gambling disorder was reclassified as a non-substance-related disorder and Internet gaming disorder was included in Section III as a condition requiring further research (APA, 2013). Moreover, as of May 2019, the World Health Organization (WHO) officially included gaming disorder in the International Classification of Diseases – Eleventh Edition (ICD-11; WHO, 2019) alongside gambling disorder as the two formally recognized disorders due to addictive behaviors (WHO, 2019). According to the DSM-5 (APA, 2013), both disorders are defined as the persistent and recurrent engagement in the behavior (gambling or video gaming), leading to clinically significant impairment or distress. Recent research has identified that rates of problematic participation in both gambling and gaming are higher among adolescents and young adults, with 0.2–12.3% reporting problem gambling (PG; Calado, Alexandre, & Griffiths, 2017) and 5–10% reporting problem video gaming (PVG; Fam, 2018; Paulus, Ohmann, von Gontard, & Popow, 2018; Turner et al., 2012).

Gambling and gaming have been found to share a number of similarities with both activities holding the potential for convergence (Griffiths, 2008b; King, Delfabbro, & Griffiths, 2010). Indeed, multiple similarities have been noted at the structural (e.g., variable reinforcement schedules), esthetic (e.g., exciting and stimulating sounds and light effects) and motivational (e.g., escapism, socializing) levels (Hilgard, Engelhardt, & Bartholow, 2013; King, Gainsbury, Delfabbro, Hing, & Arabanel, 2015). The distinction between gambling and gaming has been muddled even further with the integration of microtransactions and “loot boxes” in many mobile and console games, allowing gamers to spend money for a chance at acquiring random virtual items of differing value or benefit (Griffiths, 2018; Li, Mills, & Nower, 2019; Richard, 2018; Zendle, Meyer, & Over, 2019). However, some differences between both behaviors are also present. For one, video games do not have any formally legislated age restrictions, whereas gambling is legally restricted to adults in most jurisdictions. Additionally, video games are most often played for points and/or status, with skill determining one's performance, whereas gambling can be engaged in for money with the outcome being predominantly random (Griffiths, 2005). Lastly, although variable reinforcement schedules in gambling and gaming can make these activities more addictive, some video games do not utilize immediate reinforcement, with more delayed rewards being presented to the player after long periods of strategic and goal-directed engagement (James & Tunney, 2017).

Various psychosocial risk factors for the development of problematic engagement in both gambling and gaming have been identified. However, a significantly greater amount of research has investigated PG as the growth of studies investigating PVG largely begun in the 2000s with the rise of online gaming (Griffiths, Kuss, & King, 2012). Recent literature reviews have indicated that males (Brezing, Derevensky, & Potenza, 2010; Dowling et al., 2017) and younger individuals (Derevensky, 2019; Johansson, Grant, Kim, Odlaug, & Gotestam, 2009) are at an increased risk of PG and PVG. Moreover, high sensation seeking (Dowling et al., 2017; Shead, Derevensky, & Gupta, 2010), emotion-focused coping styles (i.e., suppressive and reactive; Johansson et al., 2009; Kuss & Griffiths, 2012), depressive symptoms (Johansson et al., 2009; Kuss & Griffiths, 2012), attention problems (Shead et al., 2010; Sugaya, Shirasaka, Takahashi, & Kanda, 2019), delinquency (Johansson et al., 2009; Shead et al., 2010), and conduct problems (Dowling et al., 2017; Sugaya et al., 2019) have been identified as risk factors for both PG and PVG. As numerous risk factors have been identified, there is a serious need to understand which of these are of critical importance when investigating the development and maintenance of these problem behaviors.

Theoretical models for the development of substance use disorders and non-substance addictive behaviors can provide a rationale to specify which risk factors are predominant in the development of PG and PVG. The potential generalizability of these models is based on research demonstrating that there are etiological, neurobiological, clinical, and phenomenological similarities between substance and non-substance addictions (Brezing, Derevensky, & Potenza, 2010; Potenza, 2014; Yakovenko & Hodgins, 2018). Based on a developmental psychopathology framework, various pathways have been suggested to explain the development of substance use disorders (SUDs) including the externalizing pathway, internalizing pathway, and combined pathway. The externalizing pathway suggests that early externalizing problems (e.g., conduct problems) and behavioral disinhibition predict later SUDs (Zucker, 1994; Zucker, Heitzeg, & Nigg, 2011). The internalizing pathway proposes that early inhibited temperament, internalizing symptoms (e.g., depression) and difficulties with emotional coping predict later SUDs (Hussong, Jones, Stein, Baucom, & Boeding, 2011; Trucco, Villafuerte, Hussong, Burmeister, & Zucker, 2018). Lastly, the combined pathway proposes that a combination or interaction of externalizing and internalizing problems leads to the development of SUDs (Englund & Siebenbruner, 2012; Maslowsky, Schulenberg, & Zucker, 2014), which may be explained by the presence of shared mechanisms underlying psychopathology (Angold, Costello, & Erkanli, 1999; Caspi & Moffitt, 2018).

Recently, two of these pathways have been indirectly adapted to explain the development of gaming disorder. Based on case reports, Benarous et al. (2019) identified an “internalized pathway” (involving depressive symptoms during adolescence) and an “externalized pathway” (involving conduct problems during adolescence) towards the development of gaming disorder in young adulthood. Indeed, subtypes of problem gamers have been identified depicting greater aggressive and/or depressive symptoms (Lemmens, Valkenburg, & Gentile, 2015; Myrseth & Notelaers, 2018; Sanders & Williams, 2019; Yu & Cho, 2016), with longitudinal studies identifying the predictive role of greater internalizing and externalizing symptoms for problem gaming (Krossbakken et al., 2018; Liu et al., 2018).

Although these specific developmental pathways have not been empirically or theoretically investigated with regards to PG, Yakovenko and Hodgins (2018) discuss the potential for the SUD literature to act as a model for future research related to gambling disorder. Linking these pathways to what has been theoretically established for PG, these appear to parallel the two subgroups of problem gamblers identified within the Pathways Model (Blaszczynski & Nower, 2002), namely, the antisocial impulsivist (Pathway 3) and emotionally vulnerable (Pathway 2) problem gamblers. Importantly, the presence of these subgroups of problem gamblers has been empirically validated using cross-sectional community (Gupta et al., 2013; Nower, Martins, Lin, & Blanco, 2013), clinical (Moon, Lister, Milosevic, & Ledgerwood, 2017; Valleur et al., 2016), and longitudinal data (Allami et al., 2017; Dowd, Keough, Jakobson, Bolton, & Edgerton, 2018; Mader, Christensen, & Williams, 2019). Lastly, as a combined pathway has been identified within the SUD literature, it is possible that comorbid externalizing and internalizing problems may also be related to the development of behavioral addictions. This combined pathway would be consistent with the subgroup of adolescent problem gamblers identified by Gupta et al. (2013) reporting higher levels of both externalizing and internalizing symptoms. This combined pathway presently remains unexplored with regards to PVG.

Based on this theoretical rationale, conduct problems (CP) and depressive symptoms (DS) appear to be critical risk factors in the development of PG and PVG. CP typically include symptoms and problem behaviors associated with conduct disorder such as aggression, rule-breaking, antisocial behaviors and violent or delinquent (e.g., theft, vandalism) acts (APA, 2013; Fonagy & Luyten, 2018; Hawkins et al., 2002). As for DS, these typically include symptoms associated with major depressive disorder such as low mood, anhedonia, and feelings of worthlessness (APA, 2013; Dean & Keshavan, 2017; Rapee et al., 2019). Empirically, both CP and DS have been associated with PG (Giralt et al., 2018; Richard & Derevensky, 2017; Sagoe et al., 2017) and PVG (Müller et al., 2015; Myrseth & Notelaers, 2018), with the presence of both CP and DS rendering youth at an even greater risk of addiction later in life (Giralt et al., 2018; Jessor & Jessor, 1977; Khoddam, Jackson, & Leventhal, 2016; Strittmatter et al., 2015; Wiesner, Kim, & Capaldi, 2005).

Given the theoretical and empirical associations between CP, DS and behavioral addictions, a review of the literature investigating the role of these problems in the development of PG and PVG is deemed necessary. Previous reviews have broadly examined the role of various risk and protective factors for PG (Dowling et al., 2017; Johansson et al., 2009; Lorains, Cowlishaw, & Thomas, 2011; Yakovenko & Hodgins, 2018) and PVG (Burleigh, Griffiths, Sumich, Stavropoulos, & Kuss, 2019; Cheng, Cheung, & Wang, 2018; González-Bueso et al., 2018; Mihara & Higuchi, 2017). However, these reviews have not investigated both PG and PVG together, with an emphasis on DS and CP. With the inclusion of gaming disorder as the second formally recognized behavioral addiction (WHO, 2019), reviews investigating both PG and PVG are necessary to assess for similarities and differences in the risk factors and mechanisms underlying the development of each problem behavior (Marchica, Mills, Derevensky, & Montreuil, 2019). Furthermore, as adolescents and young adults present with a higher risk for the development of these problem behaviors, a review targeting this demographic is of primary interest. Several reviews have limited their sample to adolescents and young adults, yet these are limited in their broad examination of factors associated with PG (Calado, Alexandre, & Griffiths, 2017) or PVG (Paulus et al., 2018; Sugaya et al., 2019), with no focus on CP and DS. As such, the purpose of the present systematic review is to provide an overview of research identifying the relationship between CP, DS and both PG and PVG among adolescents and young adults.

Methods

Study selection

The methodology employed in this review is compliant with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA; Moher, Liberati, Tetzlaff, & Altman, 2009). Electronic databases including PsychINFO, Medline, Social Work Abstracts, Scopus and Web of Science were searched. Searches in all databases utilized a combination of Medical Subject Headings (MeSH) and multi-purpose terms where applicable, searching for the keywords in the title, abstract, keywords, headings and subject headings of articles. The search terms used a combination of keywords and Boolean terms relating to gambling, gaming, CP and DS. Keywords for the searches are presented in Table 1. The search was restricted to articles published from January 1994 to October 2019, to coincide with the publication of the DSM-IV (APA, 1994) which introduced a scientific and empirically-based understanding of pathological gambling (National Research Council, 1999, p. 18). The search strategy was peer-reviewed by a university librarian with experience conducting systematic reviews. The search parameters yielded the following number of results in each database: PsychINFO (296), MEDLINE (945), Social Work Abstracts (22), Scopus (1,623), and Web of Science (915).

Table 1.

Keywords for database searches

Grouping terms	Keywords
Gambling or gaming	(gambl* OR gambling OR “problem gambling” OR “disordered gambling” OR “gambling disorder” OR “pathological gambling” OR “video game” OR videogame OR games OR gamer OR “problem* use of video games” OR “Internet gaming disorder” OR “gaming disorder”) AND
Conduct problems or depressive symptoms	(conduct* OR “conduct problems” OR “conduct disorder” OR delinquency OR aggression OR externaliz* OR “externalizing problems” OR “externalizing symptoms” OR “oppositional defiant disorder” OR “antisocial personality disorder” OR “antisocial behavio*r” OR depression OR “major depression” OR “major depressive disorder” OR “major depressive episode” OR dysthymia OR “mood disorder” OR internaliz* OR “internalizing problems” OR “internalizing symptoms”)

Inclusion criteria were developed for the title/abstract screening stage and full text screening stage. Studies were considered eligible at the title/abstract stage if they were: (1) original peer-reviewed publications in English or French; (2) providing an estimate of the presence of CP (i.e., physical or verbal aggression, delinquency, antisocial behavior, externalizing problems) or DS (i.e., low mood, negative affect, anhedonia, internalizing symptoms) in association with either PG or PVG; (3) observational and descriptive studies (e.g., cross-sectional, case–control, longitudinal); (4) reviews if they included the calculation of new data (i.e., meta-analysis). Studies were not eligible at the title/abstract screening stage if they were: (1) validating the development of a psychometric measure/scale; or (2) investigating the impact of a treatment intervention.

Studies were included at the full-text screening stage if: (1) participants were between the ages of 12 and 25 years; (2) the study included a measure specific to problem/disordered gambling or problem/disordered gaming as the dependent variable; and (3) the study included a measure specific to CP or DS as the independent variable. Studies were excluded at the full-text screening stage if they: (1) only included a measure for impulsive behavior, sensation seeking or substance use (and not CP more specifically); (2) only included a measure of emotional problems based on a composite of both anxious and depressive symptoms without isolating the effect of DS; or (3) investigated Internet addiction/problems without isolating the effect of PVG.

Study assessment

A PRISMA flow diagram of the search results is displayed in Fig. 1. A total of 2,549 articles were identified after duplicate records were removed. Rayyan QCRI (Ouzzani, Hammady, Fedorowicz, & Elmagarmid, 2016) was utilized to perform the blind screening of identified articles. To ensure reliable title and abstract screening, approximately 20% (n = 527) of the titles and abstracts were randomly selected and reviewed by JR and EF. The inter-rater agreement across the reviewers was 95.64%. After the title/abstract screening, 571 full-text articles were deemed potentially eligible and were retrieved for review. To ensure reliable full-text screening, approximately 20% (n = 114) of the full-texts were randomly selected and reviewed by JR and EF. The inter-rater agreement across the reviewers was 94.74%. Discrepancies were resolved through group discussion with a third reviewer (CT) as arbiter and inclusion/exclusion criteria were specified or adjusted as necessary. After the full-text screening, 71 articles met the inclusion criteria for the study.

Fig. 1.

Flow diagram of paper selection process for the systematic review

Data extraction

JR and EF extracted data from the included studies. The following data were extracted from each of the included studies: (1) type of study (and study duration/number of time-points if longitudinal); (2) independent/predictor, dependent/outcome and control variables; (3) population source; (4) number of participants; (5) age and sex/gender of participants; (6) participant inclusion and exclusion criteria; (7) statistical analyses; and (8) main findings.

Results

Sample characteristics

Sample characteristics of selected studies are presented in Tables 2–4. Approximately 59% of the studies were published within the past five years (2014–2019), with the majority of the studies being carried out in Canada (n = 22), Europe (n = 22), and the United States (n = 16). Forty-seven studies examined PG (33 cross-sectional and 14 longitudinal), 23 studies examined PVG (18 cross-sectional and five longitudinal) and one cross-sectional study investigated both PG and PVG. A majority of the studies utilized convenience sampling methods (n = 49), with others applying random, stratified, or cluster sampling methods. As for the gender of individuals in study samples, 15 studies reported samples that were more than 60% male and seven studies reported samples that were more than 60% female. The most commonly sampled populations included high-school students (n = 23), adolescents/young adults (n = 14), and university students (n = 11).

Table 2.

Summary of research articles investigating the association between conduct problems and problem gambling and gaming

Authors	Country	Design and sampling method	Sample population	Sample characteristics	PG and PVG measure	CP measure	Findings
Problem gambling
Barnes, Welte, Hoffman, and Tidwell (2011)	United States	Cross-sectional survey	Representative household sample of adolescents and young adults	N = 2,274	SOGS-RA	DISC-C	CP were correlated with PG (r = 0.31, P < 0.001). Youth who endorsed 3 or more symptoms of conduct disorder (CD) were 4 times more likely to report PG compared to those not meeting the criteria for CD (31% vs. 8%).
		Randomly selected telephone sample from a sampling frame of all working telephone blocks in the United States		Gender NR			Logistic regression results controlling for gender, age, socioeconomic status, and race/ethnicity indicated that CD increased the odds of being a problem gambler by 4.4 times (P < 0.001). When alcohol, tobacco and marijuana problems were added to the model, CD remained significant, increasing the odds of being a problem gambler by 2.9 times (P < 0.001).
				Mage = NR, range 14–21
Brunelle et al. (2012a)	Canada (Quebec)	Cross-sectional survey	High-school students	N = 1,870	DSM-IV-MR-J (French version)	MASPAQ	Male and female problem gamblers had higher average scores in all domains of CP including severe delinquency, fraud and theft, and vandalism and interpersonal violence compared to non-gamblers and non-problem gamblers (all significant at P < 0.001).
		Convenience sampling		54.1% female
				Mage = 15.43 (SD = 0.97, range 14–18)
Brunelle et al. (2012b)	Canada (Quebec)	Cross-sectional survey	High-school students	N = 1,870	DSM-IV-MR-J (French version)	MASPAQ	In both Internet and non-Internet gamblers, CP were associated with a greater severity of PG (β = 0.29, P < 0.05; β = 0.15, P < 0.001, respectively).
		Convenience sampling		54.1% female
				Mage = 15.43 (SD = 0.97, range 14–18)
Cheung (2014)	China	Cross-sectional survey	High-school students	N = 4,734	DSM-IV-MR-J	Delinquency scale	Correlations between delinquency and gambling variables (problems, frequency, permissiveness) ranged from r = 0.22 to r = 0.28 (P < 0.001).
		Stratified random sampling		50.7% male			Logistic regression predicting PG indicated that delinquency predicted PG (AOR = 1.20 95% CI [1.17, 1.23]) while controlling for age, gender, SES, familial status.
				Mage = 16.39 (SD = 1.73, range 12–23)			Delinquency remained significant in the model that also included tobacco and alcohol use (AOR = 1.11 95% CI [1.08, 1.15]).
Cook et al. (2015)	Canada (Ontario)	Cross-sectional survey	High-school students	N = 4,851	SOGS-RA	Delinquency scale (violent and non-violent acts)	Correlation between delinquency and PG was significant (r = 0.24, P < 0.001). Violent and non-violent delinquent behaviors were more common in PG than non-PG, with ORs ranging from 3.4 to 19.6 depending on the delinquent act. Overall delinquency scores indicated 11.35 times (P < 0.001) higher likelihood of PG.
		Stratified (region and school type), two-stage (school, class) cluster sampling		53% female			A multivariate logistic regression indicated that higher overall delinquency resulted in youth being 5.9 times (P < 0.001) more likely to meet the criteria for PG compared to less delinquent youth when controlling for hazardous drinking, cannabis dependency, suicide attempt(s) and psychological distress.
				Mage = 14.6 (SD = NR, range NR)
Hayatbakhsh, Clavarino, Williams, Bor, and Najman (2013)	Australia	Cross-sectional study	Young adults	N = 3,512	PGSI	CBCL Young Adult Self-Report	Individuals in the top 10% of externalizing problems had a greater likelihood of being categorized as at-risk for problem gambling compared to being categorized as non-gamblers (OR = 5.4, 95% CI [3.1, 9.4]).
		Convenience sampling		47% male
				Mage = 20.6 (SD = 0.8), range 18–23)
Martins et al., (2013)	United States	14-year longitudinal study	Urban males from predominantly low SES neighborhoods	N = 310	Age 17, 19, & 20: SOGS-RA	Grade 1–3: Childhood aggressive behaviors (Teacher Observation of Classroom Adaptation-Revised)	General growth mixture modeling based on the longitudinal development of CP indicated that those who had chronically high CP throughout childhood were 2.6 times more likely (95% CI [1.06, 6.38]) to meet the criteria for at-risk or PG.
		Convenience sampling		100% male		Grade 6–10:
				Mage = NR (range 6–20)		Adolescent aggressive behaviors (Teacher Report of Classroom Behavior-Checklist Form)	Those with chronically high CP throughout adolescence were 3.19 times more likely (95% CI [1.18, 8.64]) to meet the criteria for at-risk or PG.
Martins et al. (2014)	United States	17-year longitudinal study	Urban youth from predominantly low SES neighborhood	N = 617	Age 17, 19, & 20: SOGS-RA	Age 13–17: DISC-C	A greater proportion of problem gamblers (65%) were arrested before the age of 23 compared to social (38%) and non-gamblers (24%). PG was significantly associated with the hazard of first arrest by age 23 in both the unadjusted (HR = 3.6, P < 0.001) and adjusted (covarying for gambling status, race, household structure, lunch status, intervention status, theft/property damage, illegal drug use; AHR = 1.6, P = 0.05) models
		Convenience sampling		53% male		Age 17–23: Arrest history
				Mage = NR (range 6–23)
Pace, Schimmenti, Zappulla, and Maggio (2013)	Italy	Cross-sectional study	High-school students	N = 268	SOGS	CBCL Youth Self-Report	The group of at-risk and pathological gamblers compared to non-gamblers did not endorse higher levels of CP (P > 0.05).
		Convenience sampling		100% male			In the discriminant function analysis, higher CP was one of the variables that best differentiated at-risk gamblers from pathological gamblers (pathological gamblers having slightly more CP) but did not differentiate between non-gamblers and at-risk gamblers.
				Mage = 16.23 (SD = 0.39, range 15–17)
Terrone et al. (2018)	Italy	Cross-sectional study	High-school students	N = 94	SOGS	CBCL Youth Self-Report	Utilizing attachment style as a moderator, there was a significant positive association between CP and PG only among the dismissing-detached group (P = 0.04), but not among the fearful-avoidant group.
		Convenience sampling		65.96% male
				Mage = 17.51 (SD = 0.82, range 16–20)
Vitaro, Brendgen, Ladouceur, and Tremblay (2001)	Canada (Quebec)	5-year longitudinal study (2 year period reporting CP and PG)	Adolescent boys from disadvantaged neighborhoods	N = 717	Age 16–17: SOGS-RA	Age 16–17: Self-Reported Delinquency Scale	Delinquency at age 16 was positively correlated with PG at age 16 (r = 0.29, P < 0.001) and this association remained significant at age 17 (r = 0.31, P < 0.001). The path model accounting for gambling frequency, PG and drug/alcohol use at age 16 and 17, indicated that delinquency at age 16 did not significantly predict PG one year later.
		Convenience sampling		100% male
				Mage = NR (range 13–17)
Wanner, Vitaro, Carbonneau, and Tremblay (2009)	Canada (Quebec)	7-year longitudinal study	Sample 1: Low SES youth	Sample 1:	SOGS-RA	Sample 1:	In both samples, there were significant correlations (P < 0.05) between CP and PG at age 16 (r = 0.22–0.25), age 23 (r = 0.21–0.31), and age 16 and 23 (r = 0.07(ns)-0.13).
		Convenience sampling	Sample 2: Community youth	N = 502		Self-Report Delinquency Questionnaire	Investigating the cross-lagged models, CP at age 16 were not prospectively linked to PG at age 23, when accounting for gambling participation, PG, and substance use at age 16.
				100% male		Sample 2:
				Time 1:		DISC-C (delinquency)
				Mage = 16.2 (SD = 0.6, range NR)
				Time 2:
				Mage = 22.8 (SD = 0.6, range NR)
				Sample 2:
				N = 663
				100% male
				Time 1:
				Mage = 16.2 (SD = 0.5, range NR)
				Time 2:
				Mage = 22.5 (SD = 0.5, range NR)
Welte, Barnes, Tidwell, and Hoffman (2009)	United States	Cross-sectional study	United States residents	N = 2,258	SOGS-RA	DISC-C	Those who had current CP had a 6.1% rate of current PG (vs. 1.7% in non-CP) and a 22.9% rate of current at-risk/PG (vs. 5.2% in non-CP).
		Stratified sample by county and telephone block within county across the United States		Gender NR			In the logistic regression, with each additional DISC-C symptom, odds of at-risk/PG increased (OR = 1.4 (95% CI [1.3, 1.6]). This effect was most striking for those aged 14–15, with an odds ratio of 1.8 (95% CI [1.3, 2.2]). By age 20–21, this relationship was no longer significant (P > 0.05).
				Mage = NR (range 14–21)			In the multinomial logistic regression predicting at-risk/PG age of onset, each additional DISC-C symptom increased the odds that one would have a gambling problem before age 14 (OR = 1.6, 95% CI [1.4, 1.8]), and age 15 and later (OR = 1.2, 95% CI [1.0, 1.4]).
Widinghoff et al. (2019)	Sweden	Cross-sectional study	Violent offenders in prison	N = 264	SCID DSM-IV	SCID Conduct Disorder	Rates of gambling disorder were not higher among those with a conduct disorder (P = 0.15).
		Convenience sampling		100% male
				Mage = 22.3 (SD = NR, range 18–25)
Willoughby, Chalmers, and Busseri (2004)	Canada (Ontario)	Cross-sectional study	High-school students	N = 7,290	SOGS-RA	Delinquency (minor and major) and aggression (direct and indirect)	Correlations between CP and PG were significant ranging from r = 0.16–0.18 (P < 0.001). Results from the confirmatory factor analysis indicated a three-factor model with a delinquency factor including major delinquency (β = 0.50, P < 0.001), minor delinquency (β = 0.57, P < 0.001), and gambling (β = 0.30, P < 0.001). This factor was significantly correlated with the aggression factor (r = 0.63). Lastly, there was a consistent co-occurrence of CP and PG across levels of severity (risk ratios ranging between 1.01 and 3.45, all P < 0.001).
		Convenience sampling		47.7% male
				Mage = 15.58 (SD = 1.33, range 13–18)
Problem video gaming
Kim et al. (2018)	South Korea	Cross-sectional study	First year middle-school students	N = 402	IGUESS	BPAQ	Correlation analyses indicated a positive correlation between CP and PVG (r = 0.32, P < 0.001). A mediation model was created with father-adolescent communication as a mediating variable between CP and PVG. CP were directly related to PVG (β = 0.29, P < 0.001), with a significant partial indirect effect through poorer father-adolescent communication (β = 0.19, P < 0.001). The total effect of the model was significant (β = 0.42, P < 0.001).
		Convenience sampling		55.5% male
				Mage = 13.0 (SD = 0.40, range NR)
Ong, Peh, and Guo (2016)	Singapore	Cross-sectional study	Adolescents presenting at an addiction treatment center (for substance or behavioral addictions)	N = 260	Pathological gaming based on DSM-IV-R-PG;	Delinquent behavior based on violent and non-violent crimes	Adolescents with a history of delinquency were less likely to report PVG compared to adolescents without a history of delinquency (P = 0.001).
		Convenience sampling		81.2% male	PIUQ;
				Mage = 15.48 (SD = 1.93, range NR-19)	GAS
Tejeiro, Gómez-Vallecillo, Pelegrina, Wallace, and Emberley (2012)	Spain	Cross-sectional study	High-school students	N = 737	PVP	Anti-Social Illegal Behaviors Questionnaire	The cluster analysis indicated three clusters in the data; 1) comorbid-PVG, 2) social-PVG and 3) non-PVG. The comorbid-PVG cluster, had significantly higher levels of CP compared to the non-PVG group (P < 0.001).
		Convenience sampling		52% male
				Mage = 14 (SD = 1.12, range 12–17)
Wartberg et al. (2017)	Germany	Cross-sectional study	Family dyads (adolescent and related caregiver). 98.8% of caregivers were biological parents (85% mothers)	N = 1,095	IGDS	RAASI subscale for antisocial behavior	Two regression models were conducted (linear and logistic) controlling for gender, anger control problems, self-esteem problems, hyperactivity/inattention, parental depression and anxiety.
		Convenience sampling		50.8% male			In the linear regression model, CP predicted PVG (β = 0.14, P < 0.001), and CP also predicted PVG in the logistic regression model (OR = 1.11, 95% CI [1.00, 1.22], P < 0.05).
				Mage = 12.99 (SD = 0.82, range 12–14)
Wartberg, Kriston, Zieglmeier, Lincoln, and Kammerl (2019)	Germany	1-year longitudinal study	Family dyads (adolescent and related caregiver). 98.8% of caregivers were biological parents (85% mothers)	N = 985	IGDS	RAASI subscale for antisocial behavior	CP and PVG were correlated at Time 1 (r = 0.44, P < 0.01), Time 2 (r = 0.46, P < 0.01) and between Time 1 and Time 2 (r = 0.28–0.30, P < 0.01).
		Convenience sampling		50.7% male			In the cross-lagged panel design (controlling for anger control problems, emotional distress, self-esteem, hyperactivity/inattention, parental depression and anxiety), CP at Time 1 did not predict PVG at Time 2 (P > 0.05).
				Time 1:
				Mage = 12.99 (SD = 0.82, range 12–14)
				Time 2:
				Mage = 13.89 (SD = 0.89, range NR)

Note. BPAQ = Buss-Perry Aggression Questionnaire, CBCL = Child Behavior Checklist, DISC-C = Diagnostic Interview Schedule for Children for Conduct Disorder, DSM-IV-R-PG = Diagnostic and Statistical Manual-IV-Revised-Pathological Gambling, CP = conduct problems, GAS = Game Addiction Scale, IGDS = Internet Gaming Disorder Scale, IGUESS = Internet Game Use-Elicited Symptom Screen, MASPAQ = Mesure de l'adaptation sociale et personnelle pour adolescents Quebecois, PIUQ = Problematic Internet Use Questionnaire, PG = problem gambling, PGSI = Problem Gambling Severity Index, PVG = problem video gaming, PVP = Problem Video Game Playing, RAASI = Reynolds Adolescent Adjustment Screening Inventory, SCID = Structured Clinical Interview for DSM-IV, SOGS = South Oaks Gambling Screen, SOGS-RA = South Oaks Gambling Screen – Revised Adolescent.

Table 3.

Summary of research articles investigating the association between depressive symptoms and problem gambling and gaming

Authors	Country	Design and sampling method	Sample population	Sample characteristics	PG and PVG measure	DS measure	Findings
Problem gambling
Afifi, Nicholson, Martins, and Sareen (2016)	Canada (Manitoba)	5-year longitudinal survey	Representative sample of young adults	Time 1:	PGSI	CIDI-SF	Cross-sectional analyses indicated that at-risk or PG was associated with an increased risk of major depressive disorder (AOR = 2.33, 95% CI [1.47, 3.68])
		Random sampling, snowball recruitment, and convenience sampling		N = 679 51.8% female			Longitudinal findings indicated that at-risk or PG at T1 was associated with increased odds of major depressive disorder at Time 2 through 4 (AOR = 1.98, 95% CI [1.14, 3.44]). Major depressive disorder at T1 was not significantly associated with increased odds of at-risk or PG at Time 2 through 4 (P = 0.56).
				Mage = 18.9 (SD = NR, range 18–20)
				Time 4:
				N = 517
Bilevicius et al. (2018)	Canada (Manitoba)	1-month longitudinal survey	University students	Time 1:	PGSI	DASS depression subscale	DS were correlated with PG symptoms at both T1 (r = 0.14, P < 0.05) and T2 (r = 0.21, P < 0.01).
		Convenience sampling through an online participant pool of psychology students		N = 497			Mediation analyses indicated that after controlling for baseline PG, there was a significant positive indirect relationship between DS and PG which was partially mediated by high levels of shame (β = 0.021, 95% CI [0.006, 0.046]).
				Gender NR
				Mage = NR
				Time 2: N = 210
				76% female
				Mage = 19.71 (SD = 3.83, range NR)
Chinneck, Mackinnon, and Stewart (2016)	Canada (Manitoba)	4-year longitudinal survey	Representative sample of young adults	Time 1:	PGSI	CIDI-SF	At T1, DS and PG were positively correlated (r = NR). However, DS at T1 were unrelated to changes in PG over time. Further, PG at T1 was unrelated to changes in DS over time.
		Random sampling, snowball recruitment, and convenience sampling		N = 679
				51.8% female
				Mage = 18.92 (SD = 0.79, range 18–20)
				Time 4:
				N = 530
				Mage = 22.23 (SD = NR, range 22–24)
Cosenza, Ciccarelli, and Nigro (2019)	Italy	Cross-sectional survey	High-school students	N = 425	SOGS-RA	DASS depression subscale	DS was correlated with PG (r = 0.23, P < 0.001).
		Convenience sampling		45.5% male			Those with PG had higher levels of DS compared to the non-PG group. In the regression model, DS predicted PG (β = 0.015, P < 0.001).
				Mage = 17.12 (SD = 1.42, range 14–19)
Delfabbro and Grabosky (2006)	Australia	Cross-sectional survey	High-school students	N = 926	DSM-IV-MR-J	Negative Mood Checklist	Those reporting PG had greater negative mood (d = 0.49, P < 0.01) compared to those who did not report PG.
		Convenience sampling		51% male			In the regression model, DS was non-significant in predicting PG when accounting for other psychosocial predictors including self-esteem, health, social alienation, and relative deprivation. Only social alienation was significant.
				Mage = 14.5 (SD = 1.64, range 11–19)
Dowd et al. (2018)	Canada (Manitoba)	Cross-sectional survey	Representative sample of young adults	N = 566	PGSI	CIDI-SF	Results from the latent class analysis indicated that 27.4% of sample were the emotionally vulnerable type of problem gambler, with higher levels of DS. This was compared to a larger class of non-problem gamblers (59.90%) and impulsive problem gamblers (12.72%).
		Random sampling, snowball recruitment, and convenience sampling		47.8% male
				Mage = 19.97 (SD = 0.82, range 18–22)
Dussault et al. (2016)	Canada (Quebec)	9-year longitudinal survey (PG and DS measured over a 6-year period)	Boys living in economically disadvantaged areas	N = 1,004	Age 17: SOGS-RA	Age 17: CDI	Correlations between DS and PG at age 17 and 23 were of r = 0.14 (P < 0.01). DS and PG at age 23 were correlated at r = 0.15 (P < 0.01).
		Convenience sampling		100% male	Age 23:	Age 23: DISC-D	Longitudinal associations indicated that PG at age 17 predicted increases in DS at age 23 (β = 0.151, P < 0.001). DS at age 17 predicted increases in PG at age 23 (β = 0.131, P < 0.001). PG and DS at age 17 (P < 0.38) and age 23 (P < 0.66) were not concurrently associated.
				Time 1:	SOGS
				Mage = NR, (range 14–17)
Edgerton, Melnky, and Roberts (2015)	Canada (Manitoba)	4-year longitudinal survey	Representative sample of young adults	Time 1:	PGSI	CIDI-SF	The increased probability of DS were associated with increased initial PG severity scores at Time 1 (β = 0.134, P < 0.05). However, this had no effect on the rate of change in PG severity.
		Random sampling, snowball recruitment, and convenience sampling		N = 679
				51.8% female
				Mage = NR (range 18–20)
Edgerton, Keough, and Roberts (2018)	Canada (Manitoba)	4-year longitudinal survey	Representative sample of young adults	Time 1:	PGSI	CIDI-SF	Correlations between DS and PG from Time 1 to 4 ranged from r = 0.001 (P > 0.05) to r = 0.09 (P < 0.05), the latter being between DS at T1 and PG at T4.
		Random sampling, snowball recruitment, and convenience sampling		N = 679			Five classes were identified, with only one class indicating PG: moderate and stable PG with no DS (2.06%). Overall, there was no evidence of reciprocal growth in PG and DS in any of the classes.
				51.8% female
				Mage = NR (range 18–20)
Ellenbogen, Derevensky, and Gupta (2007)	Canada (Ontario & Quebec)	Cross-sectional survey	High-school or junior college students	N = 5,313	DSM-IV-MR-J	RADS	Non-PG, females were more likely to report DS. However, among those with PG, both males and females reported higher rates of DS. Rates of DS among PG were approximately 2–4 times higher than for social gamblers.
		Convenience sampling		51.8% male
				Mage = 14.77 (SD = NR, range 12–18)
Martin, Usdan, Cremeens, and Vail-Smith (2014)	United States	Cross-sectional study	University students	N = 1,430	DSM-IV-MR-J	PHQ-9	Correlations for DS and PG were significant (r = 0.105, P < 0.01).
		Convenience sampling		64.4% female			When compared to non-PG, PG had higher rates of DS (40.0%, OR = 3.3, 95% CI [1.9, 5.6]).
				Mage = NR (first- or second-year university students)
Molde, Pallesen, Bartone, Hystad, and Johnsen (2009)	Western Norway	Cross-sectional study	High-school students (11th-13th grade)	N = 2,055	MAGS	HADS depression subscale	In the univariate logistic regression, DS predicted PG (OR = 14.4, P < 0.001). In the multivariate logistic regression, significant predictors of PG included gender, depression (OR = 9.23, P < 0.001), alcohol abuse and self-forgetting when gambling. DS had the largest odds ratio.
		Random sampling from total population of high-school students		52.9% male
				Mage = 17.3 (SD = 0.8, range 16–19)
Nigro, Cosenza, and Ciccarelli (2017)	Italy	Cross-sectional study	Middle- and high-school students	N = 1,010	SOGS-RA	DASS depression subscale	Significant correlations were reported between DS and PG (r = 0.279, P < 0.01). Problem gamblers also endorsed a greater mean score of DS compared to non-gamblers, non-problem gamblers, and at-risk gamblers (P = NR). In the linear regression, DS predicted severity of PG while controlling for gender, age, impulsivity, anxiety, and future/immediate implications of behavior (β = 0.129, P < 0.001).
		Convenience sampling		47.5% male
				Mage = 15.37 (SD = 2.05, range 12–19)
Nower, Gupta, Blaszczynski, and Derevensky (2004)	Canada (Quebec & Ontario)	Cross-sectional study	Three combined samples of high-school students (two in Quebec, one in Ontario)	N = 3,941	Sample 1 & 2 (S1 & 2): DSM-IV-MR-J	RADS	In all three samples, those with PG reported significantly higher DS compared to non-gamblers and social gamblers (P < 0.01; S1 = 23.1%; S2 = 24.5%; S3 = 20.4% clinically depressed).
		Convenience sampling		49.15% male	Sample 3 (S3): DSM-IV-MR-J &
				Mage = NR (range 12–18)
Pascual-Leone, Gomes, Orr, Kaploun, and Abeare (2011)	Canada (Ontario)	Cross-sectional study	University students	N = 200	SOGS	BDI-II	Descriptive statistics indicated that 7.5% (n = 15) of the sample were at-risk or problem gamblers. Correlations indicate that DS were not significantly correlated with PG scores (r = 0.117, P > 0.05). DS were not included in the regression analysis.
		Convenience sampling		88.5% female
				Mage = 21.41 (SD = 3.53, range NR)
Sanscartier, Edgerton, and Roberts (2018)	Canada (Manitoba)	Cross-sectional study	University students	N = 496	PGSI	CES-D	Latent class analysis results yielded a four-class solution: 1) casual gamblers; 2) skill-interactive gamblers; 3) chance-passive gamblers; and 4) extensive gamblers. Extensive gamblers and chance-passive classes had higher rates of PG compared to the casual and skill-based gamblers. Chance-passive gamblers had greater DS compared to casual gamblers (P < 0.05, η2 = 0.14). Chance-passive gamblers and extensive gamblers did not differ in DS.
		Convenience sampling		43.1% male
				Mage = 20.22 (SD = 1.77, range 18–25)
Stuhldreher, Stuhldreher, and Forrest (2007)	United States	Cross-sectional study	University students	N = 1,079	Four criteria for PG: gambling-related harms and help-seeking behaviors	BDI	Comparisons for DS were conducted for each of the four criteria for PG. Results indicate that for all four criteria there was a greater proportion of individuals with a positive score for DS.
		Convenience sampling		58% female
				Mage = 19.9 (SD = 1.6, range NR)
Wohl, Matheson, Young, and Anisman (2008)	Canada	Cross-sectional study	First-year university students reporting gambling in the past year	N = 125	DSM-IV-MR-J	BDI	DS scores varied between gamblers, where pathological gamblers endorsed significantly more DS (M = 14.87, SD = 5.70) than problem (M = 8.40, SD = 7.80) or recreational (M = 6.60, SD = 5.71) gamblers. This was the case for both males and females.
		Convenience sampling		58.4% male
				Mage = 20 (SD = 0.75, range NR)
Problem video gaming
Bonnaire and Baptista (2019)	France	Cross-sectional study	Online forums	N = 429	GAS Short Version	HADS depression subscale	Compared to non-PVG, PVG had higher DS (M = 15.4 vs M = 12.2, P < 0.001). This was significant for both males and females.
		Convenience sampling		71.3% male			In the logistic regression analysis, DS significantly predicted PVG (OR = 1.2, 95% CI [1.1–1.3], P < 0.001) while controlling for sex, anxiety and alexithymia.
				Mage = 20.7 (SD = 2.6, range 18–25)
Dang, Zhang, Leong, and Wu (2019)	China	1-year longitudinal study	University students with gaming experience	N = 282	DSM-5 criteria for IGD	DASS depression subscale	DS and PVG were significantly correlated at both time points (r = 0.26–0.42, P < 0.001). In the path analysis, DS were identified as a mediator in the relationship between trait emotional intelligence (TEI) and PVG. In the cross-sectional model, DS mediated the relationship between TEI and PVG, while also mediating the relationship between coping flexibility and PVG. Consistent findings are reported in the prospective model. The models indicated that DS were a significant concurrent mediator at both T1 and T2 (direct effects; T1: (β = 0.37, P < 0.001; T2: β = 0.29, P < 0.001).
		Convenience sampling		39.4% male
				Time 1:
				Mage = 20.47 (SD = 1.15, range NR)
Givron, Berrewaerts, Houbeau, and Desseilles (2018)	Belgium	Cross-sectional study	First-year university medical students	N = 210	PVP	MADRS-S	As the severity of DS increased (none, minor, moderate), the average PVG score increased from M = 1.9, M = 2.4, to M = 3.9, respectively. This change was significant (P = 0.001).
		Convenience sampling		29.5% male
				Mage = 18.5 (SD = 1.0, range 17–25)
Guillot et al. (2016)	United States	1-year longitudinal study (9–18-month range)	Emerging adults, former attendees of alternative high-schools, and prior participants in a school-based substance abuse prevention program	N = 503	Video Game Addiction (1 item)	Snaith-Hamilton Pleasure Scale	Anhedonia predicted greater levels of PVG one year later (OR = 1.33, 95% CI [1.11, 1.60], P = 0.003), while controlling for gender and high-school graduation.
		Convenience sampling		47.7% male
				Mage = NR (range 19–24)
Kircaburun, Griffiths, and Billieux (2019)	Turkey	Cross-sectional study	High-school students	N = 470	IGDT-10	SDHS	DS were positively correlated with PVG (r = 0.13, P < 0.001). A multiple mediation model was tested to examine the mediating role of DS (in addition to mindfulness and rumination) on the relationship between emotional intelligence (EI) and PVG. Results indicated that although the other mediators were significant, DS did not mediate the relationship between EI and PVG.
		Convenience sampling		40.4% male
				Mage = 16.29 (SD = 1.17, range 14–18)
Li, Liau, and Khoo (2011)	Singapore	Cross-sectional study	Adolescents with massively multiplayer online gaming experience	N = 161	Pathological gaming based on DSM-IV-R-PG	Asian adolescent depression scale	DS and PVG were positively correlated (r = 0.31, P < 0.01). Results of the path model, indicated that escapism mediated the relationship between DS and PVG (indirect effect [β = 0.09, P < 0.05]; direct effect from DS to escapism [β = 0.45, P < 0.01]), while DS mediated the relationship between actual-ideal self-discrepancies (AISD) and escapism (indirect effect [β = 0.09, P < 0.05]; direct effect from AISD to DS [β = 0.20, P < 0.01]), with escapism being directly related to PVG (β = 0.34, P < 0.01). The direct relationship between DS and PVG was non-significant (β = 0.13, P > 0.05).
		Convenience sampling		49.1% male
				Mage = 14.04 (SD = 0.73, range 13–15)
Liu et al. (2018)	China	Study 1: 4-year longitudinal study	Study 1: University students with experience playing online games, spending on average 20% of their daily time gaming	Study 1:	Chinese Internet Addiction Scale	SCL-90 depressive symptoms	Across the four time points, higher DS at T1 were associated with greater PVG severity from Time 2 to 4 (r = 0.25–0.30, P < 0.01). Higher PVG at T1 was associated with greater DS at Time 2 to 4 (r = 0.19–0.27, P < 0.01). Results from the cross-lagged path models indicated that although there is a temporal interrelationship between DS and PVG, the impact of DS on PVG (β = 0.118, 0.126, 0.127; P < 0.001) is greater than the impact of PVG on DS (β = 0.070, 0.066, 0.070; P < 0.05).
		Convenience sampling		N = 563
				78% male
				Time 1:
				Mage = 18.31 (SD = 0.89, range 16–21)
Männikkö, Billieux, and Kääriäinen (2015)	Finland	Cross-sectional study	Adolescents and young adults	N = 293	GAS	Depression (frequency of feeling depressed)	DS and PVG were positively correlated (r = 0.17, P < 0.01). Those with PVG endorsed greater DS than those with no-PVG (22.6% vs 6.5%, P < 0.001). In the linear regression model, DS were a significant predictor of PVG (β = 0.18, P < 0.01) when controlling for occupation, education level, age, gaming frequency, health, exercise, life satisfaction, and preference for online interaction.
		Random sampling stratified for age and gender		51% male
				Mage = 18.7 (SD = 3.4, range 13–24)
Vadlin Åslund, Hellström, and Nilsson (2016)	Sweden	Cross-sectional study	Sample 1:	Sample 1:	GAIT	DSRS-A	In the multivariable logistic regression analysis adjusting for sex, age, school bullying, and family maltreatment, attention problems, and anxiety, adolescents with DS were 2.47 times more likely to be PVG (95% CI [1.44, 4.25], P < 0.001).
		Sample 1: Total population sampling	Community sample of adolescents	N = 1,868
		Sample 2: Consecutive sampling at child and adolescent psychiatric clinics	Sample 2:	55.4% female
			Clinical sample of adolescents in psychiatric clinics	Mage = 13.9 (SD = NR, range 12–16)
				Sample 2:
				N = 242
				69.8% female
				Mage = 15.39 (SD = NR, range 12–18)
Van Rooij et al. (2014)	Netherlands	Cross-sectional study	High-school students	N = 8,478	VAT	Depressive Mood List (Dutch translation)	Analyses were conducted separately by gender. In males, DS were associated with an increase in PVG (d = 0.91, P < 0.001). In females, DS associated with an increase in PVG (d = 1.23, P < 0.001).
		Stratified sampling of schools based on region, urbanization and education level		49% male
				Mage = 14.2 (SD = 1.1, range NR)

Note. BDI = Beck Depression Inventory, CDI = Child Depression Inventory, CES-D = Center for Epidemiologic Studies Depression Scale, CIDI-SF = Composite International Diagnostic Interview-Short Form, DASS = Depression Anxiety Stress Scale, DISC-D = Diagnostic Interview Schedule for Children for Depressive Symptoms, DS = depressive symptoms, DSM-IV-R-PG = Diagnostic and Statistical Manual - IV - Revised - Pathological Gambling, DSRS-A = Depression Self-Rating Scale Adolescent Version, GAIT = Gaming Addiction Identification Test, GAS = Game Addiction Scale, HADS = Hospital Anxiety and Depression Scale, IGD = Internet Gaming Disorder, IGDT-10 = Internet Gaming Disorder Test, MADRS-S = Montgomery and Asberg Depression Rating Scale, MAGS = Massachusetts Gambling Screen DSM-IV subscale, PG = problem gambling, PGSI = Problem Gambling Severity Index, PHQ-9 = Patient Health Questionnaire-9, PVG = problem video gaming, PVP = Problem Video Game Playing, RADS = Reynolds Adolescent Depression Scale, SCL-90 = Symptom Checklist, SDHS = Short Depression Happiness Scale, SOGS = South Oaks Gambling Screen, SOGS-RA = South Oaks Gambling Screen – Revised Adolescent, VAT = Video Game Addiction Test.

Table 4.

Summary of research articles investigating the association between both conduct problems, depressive symptoms and problem gambling and gaming

Authors	Country	Design and sampling method	Sample population	Sample characteristics	PG and PVG measure	CP and DS measure	Findings
Problem gambling
Allami et al. (2017)	Canada (Quebec)	11-year longitudinal survey	Sample 1: Low SES francophone adolescents	Sample 1:	Age 16: SOGS-RA	CP: Age 12: Teacher report (antisocial & aggressive behavior)	Latent profile analysis results indicated 4 classes, with two classes having higher CP and DS respectively: a biologically vulnerable (BV) class and an emotionally vulnerable (EV) class
		Sample 1: Convenience sampling (high-risk)	Sample 2: Representative francophone school students	Time 1:	Age 23: SOGS	DS: Age 12: Teacher report (DS)	Longitudinal findings indicated that being in either the BV or EV subgroup at age 12 was a risk for PG at age 23, but not age 16. This relationship was strongest for the BV class, reporting significantly more PG than behaviorally conditioned gamblers and significantly more PG than EV gamblers. By age 23, the proportion of BV gamblers increased from what it was at age 16. For EV gamblers, class proportions were stable.
		Sample 2: Convenience sampling (partly probabilistic)		N = 1,033
				Time 4:
				N = 899
				100% male
				Mage = NR, range 12–23
				Sample 2:
				Time 1:
				N = 6,397
				Time 4:
				N = 3,017
				58% male
				Mage = NR, range 12–23
Allami et al. (2018)	Canada (Quebec)	11-year longitudinal survey	Sample 1: Low SES francophone adolescents	Sample 1:	Age 16: SOGS-RA	CP: Age 12: Teacher report (antisocial & aggressive behavior)	Correlation analyses indicated significant correlations between CP and PG (r = 0.08–0.16, P < 0.01) and DS and PG at 16 and 23 years (r = 0.07–0.09, P < 0.01).
		Sample 1: Convenience sampling (high-risk group)	Sample 2: Representative francophone school students	Time 1:	Age 23: SOGS	DS: Age 12: Teacher report (DS)	Latent profile analysis results identified an externalizing (high CP), an internalizing (high DS) and a comorbid class (high CP and DS).
		Sample 2: Convenience sampling (partly probabilistic)		N = 1,033			At age 16 and 23, both externalizing and comorbid classes had greater PG symptoms compared to the well-adjusted class. The internalizing class did not report greater PG compared to the well-adjusted class.
				Time 4:
				N = 939
				100% male
				Mage = NR, range 12–23
				Sample 2:
				Time 1:
				N = 6,397
				Time 4:
				N = 3,142
				58% male
				Mage = NR, range 12–23
Giralt et al. (2018)	Germany	Cross-sectional study	Adolescents	N = 9,309	DSM-IV-MR-J	CP & DS: Strengths and Difficulties Questionnaire	When covarying for age, DS (boys P < 0.001, η2 = 0.037; girls; P < 0.001, η2 = 0.012) and CP (boys P < 0.001, η2 = 0.111; girls P < 0.001, η2 = 0.043) were greater in those with at-risk/PG compared to those with no PG.
		Two samples based on random probability sampling with a stratification by school type and regional population density		49.4% male
				Mage = 15.3 (SD = 1.71, range 12–18)
Gupta et al. (2013)	Canada (Quebec & Ontario)	Cross-sectional study	High-school students reporting problem gambling	N = 109	DSM-IV-MR-J	CP & DS: Millon Adolescent Clinical Inventory	Results from the latent class analyses identified 5 classes. These classes support the groups identified within the Pathways Model (Blaszczynski & Nower, 2002), in addition to a DS only class (27%) and a combined (DS/CP) (10%) class.
		Convenience sampling		72% male
				Mage = 15.6 (SD = 1.10, range 14–18)
Hardoon, Gupta, and Derevensky (2004)	Canada (Ontario)	Cross-sectional study	High-school students	N = 2,336	DSM-IV-MR-J	CP & DS: Conners-Wells Adolescent Self-Report Scale	CP were the largest clinical problem for probable PG, with 55% of the latter reporting clinical levels of CP, which was greater than the CP reported by non-gamblers and social gamblers (P < 0.001). Similar findings were reported for DS, with 27.4% of probable PG reporting such problems and this being greater than the DS reported by non-gamblers and social gamblers (P < 0.001).
		Convenience sampling		42% male			In the regression model controlling for family problems, anger control problems, hyperactivity, and inattention, CP was significant in predicting at-risk/probable PG (β = 0.083, Exp(β) = 1.087, P < 0.001). However, DS was non-significant.
				Mage = 14.76 (SD = 1.91, range 12–19)
Kaminer, Burleson, and Jadamec (2002)	United States	Cross-sectional study	Adolescents diagnosed with psychoactive substance use disorders (DSM-III-R)	N = 97	MAGS	CP & DS: DISC-C & SCID	In the sample, 34% never gambled, 57% were social/non-problem gamblers, 8% (n = 8) were at-risk and 1% (n = 1) met the criteria for pathological gambling.
		Convenience sampling		66% male			Bivariate analyses indicated that there was no correlation between DS and PG (r = -0.09, P < 0.05) and between CP and PG (r = 0.09–0.18, P < 0.05).
				Mage = NR (range 13–18)
Kong et al. (2014)	United States	Cross-sectional study	High-school students	N = 3,901	MAGS	CP: Aggression (2 items; weapon carrying or serious fights)	Latent class analysis results indicated that compared to the low-risk gambling class, all three other classes (including the at-risk chasing gambling, at-risk negative consequences gambling, and problem gambling) were more likely to experience CP (P < 0.001).
		Non-random, yet all schools in Connecticut were invited to participate.		48.5% male		DS: Depression (1 item; feeling sad or hopeless almost every day for 2 or more weeks)	For DS, only the at-risk negative consequences and problem gambling classes were more likely to experience DS compared to the low-risk gambling group (P < 0.001).
				Mage = NR (range 14–21)
Langhinrichsen-Rohling, Rohde, Seeley, and Rohling (2004)	United States	Cross-sectional study	High-school students	N = 1,735	SOGS-RA	CP: 2 items (getting in trouble more than others your age; more trouble in school than others your age)	Five groups were identified from the discriminant function analyses: 1) non-gambler; 2) non-problem gambler; 3) at-risk gambler; 4) problem gambler and; 5) pathological gamblers. CP were a significant predictor in the linear progression from least to most PG, DS were not a significant predictor.
		Convenience sample		Gender NR		DS: CES-D	Pathological gamblers reported significantly more CP (M = 1.36) and more DS (M = 0.88) than did individuals in the other four groups. At-risk gamblers reported more CP (M = 0.67) but not more DS (M = 0.21) compared to non-gambles and non-problem gamblers.
				Mage = NR (range 14–19)
Petry and Tawfik (2001)	United States	Cross-sectional study	Adolescents entering treatment for marijuana abuse	N = 255	GAIN & DSM-IV-MR-J	CP & DS: GAIN	In the multivariate model examining differences between PG and non-PG (controlling for race, age and gender), PG were more likely (P < 0.05) to endorse CP. There was no difference in DS between the PG and non-PG groups (P > 0.05).
		Convenience sampling		83.92% male
				Mage = 15.8 (SD = 1.2, range 12–18)
Potenza et al. (2011)	United States	Cross-sectional study	High-school students	N = 2,006	MAGS	CP: Aggression (2 items; weapon carrying or serious fights)	Comparing at-risk PG and PG to low risk gamblers, at-risk PG were more likely to report DS among both Internet (29% vs 20%, P = 0.05) and non-Internet (28% vs 22%, P = 0.01) gamblers. Similar findings for Internet and non-Internet gamblers are noted for those reporting CP (P < 0.001) including serious fights (13.55–30% vs. 5.94–13.75%) and carrying a weapon (38.8–53% vs. 20.15–33%).
		Convenience sampling		59.92% male		DS: Depression (1 item; feeling sad or hopeless almost every day for 2 or more weeks)
				Mage = NR (range 14–18)
Sagoe et al. (2017)	Norway	2-year longitudinal study	Late adolescents/young adults	N = 1,277	PGSI	CP: BPAQ-SF	Latent class analyses indicated a three-class solution provided the best fit for patterns of gambling from age 17 to 19: 1) consistent non-gambling (71%); 2) consistent non-risk gambling (23.8%); 3) risk-and-problem gambling (5.1%).
		Random sampling based on national population registry.		67.1% female		DS: HADS depression subscale	Correlates of the risk-and-problem gambling class at age 17 include greater CP (physical aggression, OR = 1.58, P = 0.01; verbal aggression, OR = 1.54, P = 0.03) and greater DS (OR = 1.14, P = 0.03). At age 19, the risk-and-problem gambling class had the highest levels of DS (d = 0.53) and CP (physical, d = 0.91; verbal, d = 0.57).
				Mage = NR (range 17–19)
Tackett et al. (2017)	United States	Cross-sectional study	University students	N = 4,751	SOGS	CP: Zuckerberg-Kuhlman Personality Questionnaire: Aggression/Hostility & BSI	After accounting for the shared variance between PG and alcohol problems, personality correlates of PG among men included CP (aggression/hostility; r = 0.31, P = 0.001). This was non-significant among women.
		Convenience sampling		55.1% male		DS: BSI	For the mental health correlates, among men, DS were non-significant yet CP (hostility; r = 0.23, P < 0.001) were significant. For women, both DS (r = 0.27, P < 0.01) and CP (hostility; r = 0.46, P < 0.001) were significant.
				Mage = 20.34 (SD = 3.31, range NR)
Yip et al. (2011)	United States	Cross-sectional study	High-school students	N = 2,484	MAGS	CP: Aggression (2 items; weapon carrying or serious fights)	Individuals with PG were more likely to report past year DS in comparison to low-risk (OR = 4.15, P < 0.001) and at-risk gamblers (OR = 3.95, P < 0.001). Individuals with PG were also more likely to report serious fights (OR = 6.48, P < 0.001; OR = 6.45, P < 0.001) and carrying a weapon to school (OR = 3.40, P < 0.001; OR = 2.89, P < 0.001) compared to low- and at-risk gamblers.
		Convenience sampling		55.84% male		DS: Depression (1 item; feeling sad or hopeless almost every day for 2 or more weeks)
				Mage = NR (range 13–19)
Yücel et al. (2015)	Australia	6-year longitudinal study	Adolescents	N = 156	SOGS-RA	CP: CBCL Parent Report	There were no significant differences across gambling risk categories with regards to DS (P = 0.62–0.96) and CP (P = 0.41–0.59) across all time points. However, females with risky gambling reported greater CP than no-risk gamblers at T1 (P = 0.03). This difference was no longer significant at T2 (P =0.18).
		Convenience sampling		51.3% male		DS: CES-D	Longitudinal change scores in DS and CP did not predict at-risk gambling in the entire sample or in the sample separated by sex. In the model investigating interaction effects by sex, higher aggression predicted PG in females and lower aggression predicted PG in males (β = 0.36, P = 0.018, OR = 1.43, 95% CI [1.06, 1.92]).
				Time 1:
				Mage = 12.5 (SD = 0.4, range NR)
				Time 2:
				Mage = 16.7 (SD = 0.4, range NR)
				Time 3:
				Mage = 18.8 (SD = 0.5, range NR)
Problem video gaming
Desai, Krishnan-Sarin, Cavallo, and Potenza (2010)	United States	Cross-sectional study	High-school students	N = 4,028	Problem video gaming (3 items: unsuccessful attempts to cut back; irresistible urges to play; tension only relieved by gaming.	CP: Aggression (2 items; weapon carrying or getting into serious fights)	Among boys, PVG was associated with DS (11.37% vs. 4.3%, P < 0.001), and CP (serious fights, 13.95% vs. 4.69%, P < 0.001; carrying a weapon, 7.66% vs. 4.55%, P = 0.02). Among girls, PVG was also associated with DS (7.48% vs. 1.01%, P < 0.001) and CP (serious fights, 9.76% vs. 2.67%, P = 0.01), but not carrying a weapon (P = 0.37). In the logistic regression, DS (OR = 3.62, P < 0.001) and CP (OR = 2.97, P < 0.001) were associated with PVG when controlling for grades, extracurriculars, smoking, marijuana, alcohol, drug, and caffeine use.
		Convenience sampling		45.8% male		DS:
				Mage = NR (range 14–18)		Depression (1 item; feeling sad or hopeless almost every day for 2 or more weeks)
Krossbakken et al. (2018)	Norway	3-year longitudinal study	Adolescents and young adults	N = 1,277	GAS	CP: BPAQ-SF	In the unrestricted path model, DS were reciprocally related to PVG (as both an antecedent [β = 0.11–0.12, P < 0.01]; and a consequence [β = 0.12–0.13, P < 0.01]). However, physical aggression was only an antecedent to PVG (β = 0.08, P < 0.05). Based on a typology of gamers (engaged, problem, addicted), physical (OR = 1.10–1.19, P < 0.01) and verbal aggression (OR = 1.11, P < 0.01) and DS (OR = 1.11, P < 0.01) were antecedents of PVG over a 1-year, but not a 2-year interval.
		Random sampling		61.7% female		DS: HADS depression subscale
				Time 1:
				Mage = 17.5 (SD = NR, range NR)
				Time 2:
				Mage = 18.5 (SD = NR, range NR)
				Time 3:
				Mage = 19.5 (SD = NR, range NR)
Müller et al. (2015)	Europe (7 countries)	Cross-sectional study	High-school students	N = 12,938	Assessment of Internet and Computer Game Addiction-Gaming Module	CP: CBCL Youth Self-Report	Correlations indicate a positive relationship between PVG and CP (r = 0.30, P < 0.01). Adolescents in the PVG group reported significantly greater levels of CP (males, M = 24.5; females, M = 25.9) compared to non-problematic and non-gamers (M scores for males and females range between 10.9 and 11.5; η2 = 0.033, P < 0.001).
		Random probability clustered sample		52.9% male		DS: CBCL Youth Self-Report	Large effect sizes were identified within the specific domains of rule-breaking behavior (M = 10.47 vs. M = 4.43, η2 = 0.06, P < 0.001) and aggressive behavior (M = 14.29 vs. M = 6.71, η2 = 0.05, P < 0.001) when comparing the PVG group to the non-PVG group. For DS, a significant albeit smaller effect was noted (M = 5.95 vs. M = 2.90, η2 = 0.03, P < 0.001).
				Mage = NR (range 14–17)
Myrseth and Notelaers (2018)	Norway	Cross-sectional study	Adolescents and young adults	N = 2,055	GAS-A	CP: BPAQ-SF	Latent class analyses identified 5 classes: 1) no symptoms of PVG (46%); 2) rare symptoms of PVG (22%); 3) occasional symptoms of PVG (23%); 4) often symptoms of PVG (7%); 5) very often symptoms of PVG (1.2%). Differences in DS were reported between the classes (η2 = 0.04, P < 0.001), where class 5 endorsed a greater number of DS compared to all other classes. Similar findings are noted for physical (η2 = 0.05, P < 0.001) and verbal aggression (η2 = 0.024, P < 0.001).
		Random sampling		52.9% male		DS: HADS depression subscale
				Mage = 17.5 (SD = NR, range NR)
Stockdale and Coyne (2018)	United States	Cross-sectional study	University students (87 game addicts and 87 matched healthy controls)	N = 174	IGDS	CP: BPAQ-SF	Comparing the group with PVG to the non-PVG group, the PVG group reported greater DS (M = 17.58, SD = 7.0 vs. M = 12.76, SD = 5.26; η2 = 0.15, P < 0.001) and greater CP (M = 3.22, SD = 0.82 vs. M = 2.7, SD = 0.82; η2 = 0.08, P < 0.001).
		Convenience sampling		48.85% male		DS: PROMIS Emotional Distress-Depression-Short Form 8a.
				Mage = 20.23 (SD = 4.17, range NR)
Strittmatter et al. (2015)	European (5 countries)	Cross-sectional study	High-school students	N = 8,807	Young Internet Addiction Test	CP: DISC-C	The PVG group reported greater CP (M = 1.40, SD = 1.60 vs. M = 0.63, SD = 1.06; d = 0.71, P < 0.001) and DS (M = 11.27, SD = 9.96 vs. M = 6.90, SD = 7.37; d = 0.58, P < 0.001) than the non-PVG group.
		Random sampling stratified by school		45.5% male		DS: BDI-II	In a multinomial logistic regression, DS (RR = 1.25, 95% CI [1.10, 1.43], P = 0.001) and CP (RR = 1.24, 95% CI [1.14, 1.36], P < 0.001) were significantly associated with PVG when controlling for gender, emotional symptoms, hyperactivity, peer problems, well-being, suicidal behavior and self-injurious behavior.
				Mage = 15 (SD = 1.3, range NR)
Torres-Rodrígez, Griffiths, Carbonell, and Oberst (2018)	Spain	Cross-sectional study	Adolescents with a diagnosis of Internet Gaming Disorder seeking mental health treatment	N = 31	DSM-5 criteria for IGD & IGD-20	CP: CBCL Youth Self-Report	Within the sample of problem video gamers, 64.5% were in the clinical range for DS and 6.4% were in the borderline range (SCL-90).
		Convenience sampling		100% male		DS: SCL-90 depressive symptoms & CBCL Youth Self-Report	Based on the CBCL, there were correlations between DS and PVG (r = 0.522, P < 0.001), and CP and PVG (r = 0.665–0.692, P < 0.001). Those with PVG had an average score denoting a clinical status in externalizing problems (M = 61.77, SD = 8.32).
				Mage = 14.97 (SD = 1.74, range 12–18)
Yu and Cho (2016)	South Korea	Cross-sectional study	Middle-school students	N = 2,024	IGDS (Modified Korean adaptation)	CP: BPAQ	Analyses indicated that the PVG group reported a significantly higher mean score of DS (M = 14.21) compared to the non-gamer (NG; M = 12.0), at-risk gamers (M = 11.76), and regular gamer (RG; M = 11.06) groups. Based on the cut-off score for DS, 15.1% of PVG had DS, compared to 5.1% of NG, 2.5% of at-risk gamers and 2.8% of RG. Further, analyses indicated that the PVG group reported the highest mean score in CP (M = 16.18), followed by the at-risk gamers (M = 14.02), NG (M = 13.59), and RG (M = 12.53) groups.
		Multiple-stage cluster sampling with stratification by region, grade level and sex. Three or more schools from each of the 15 districts were randomly selected, then, one 8th and one 9th grade from each selected middle school was randomly selected.		50.6% male		DS: Short Version of Online Psychological Tests	Significant correlations are reported between DS and PVG (r = 0.21, P < 0.01) and CP and PVG (r = 0.26, P < 0.01).
				Mage = 14.5 (SD = 0.50, range 13–15)
Problem gambling and gaming
Walther, Morgenstern, and Hanewinkel (2012)	Germany	Cross-sectional survey	Secondary and vocational school students	N = 2,553	PG: SOGS-RA	CP: Rating Scale of Oppositional Defiant/Conduct Disorders	Univariate analyses indicate that both PG (P < 0.01) and PVG (P < 0.001) reported more CP, but not more DS compared to non-problem gamblers (P = 0.71) and non-problem gamers (P = 0.10). Multivariate analyses indicate that PVG reported more CP (OR = 1.65, 95% CI [1.01–2.69]) but not more DS than non-problem gamers. PG reported no significant differences in either CP or DS.
		Random sample of 15 public schools		50.7% male	PVG: Video Game Dependency Scale	DS: Depression scale adapted from Kandel & Davies (1982)
				Mage = 16.7 (SD = 3.04, range = 12–25)

Note. BPAQ-SF = Buss-Perry Aggression Questionnaire - Short Form, BSI = Brief Symptoms Inventory, CBCL = Child Behavior Checklist, CES-D = Center for Epidemiologic Studies Depression Scale, CP = conduct problems, DS = depressive symptoms, HADS = Hospital Anxiety and Depression Scale, IGDS = Internet Gaming Disorder Scale, IGD = Internet Gaming Disorder, IGD-20 = Internet Gaming Disorder Test, GAIN = Global Assessment of Individual Needs, GAS-A = Gaming Addiction Scale-Adolescents, MAGS = Massachusetts Gambling Screen DSM-IV subscale, PG = problem gambling, PGSI = Problem Gambling Severity Index, PVG = problem video gaming, SCID = Structured Clinical Interview for DSM-IV, SCL-90 = Symptom Checklist, SOGS = South Oaks Gambling Screen, SOGS-RA = South Oaks Gambling Screen – Revised Adolescent.

Measures

Problem gambling (PG) and problem video gaming (PVG)

Commonly used measures for PG include: adolescent and adult versions of the South Oaks Gambling Scale (SOGS; Lesieur & Blume, 1987; Winters, Stinchfield, & Fulkerson, 1993) (n = 20), Diagnostic and Statistical Manual of Mental Disorders-Fourth edition-Multiple Response-Juvenile for pathological gambling (DSM-IV-MR-J; Fisher, 2000) (n = 12), Problem Gambling Severity Index (PGSI; Ferris & Wynne, 2001) (n = 9), and Massachusetts Gambling Screen DSM-IV subscale (MAGS; Shaffer, LaBrie, Scanlan, & Cummings, 1994) (n = 5). For PVG, the most commonly used measures include: Game Addiction Scale (GAS; Lemmens, Valkenburg, & Peter, 2009) (n = 5), Internet Gaming Disorder Scale (IGDS; Lemmens et al., 2015) (n = 3), DSM-5 criteria for Internet Gaming Disorder (APA, 2013) (n = 2), and pathological gaming based on the DSM-IV-R criteria for pathological gambling (APA, 2000) (n = 2). Compared to PG, there appeared to be a greater variation in the instruments used to assess PVG, with multiple studies using scales developed specifically for the study in question.

Depressive symptoms and conduct problems

Of the included studies, 20 measured CP, 28 measured DS, and 23 measured both CP and DS. The most commonly used measures for CP include: youth, young adult, and parent versions of the Child Behavior Checklist (CBCL; Achenbach 1997; Achenbach & Edelbrock, 1987; Achenbach, Howell, Quay, Conners, & Bates, 1991) (n = 6), Buss-Perry Aggression Questionnaire (BPAQ; Buss & Perry, 1992; Diamond & Magaletta, 2006) (n = 6), Diagnostic Interview Schedule for Children for Conduct Disorder (DISC-C; Shaffer, Fisher, Lucas, Dulcan, & Schwab-Stone, 2000) (n = 6), and the Reynolds Adolescent Adjustment Screening Inventory (RAASI; Reynolds, 2001) (n = 2). As for DS, commonly used measures include: Composite International Diagnostic Interview-Short Form (CIDI-SF; Kessler, Andrews, Mroczek, Ustun, & Wittchen, 1998) (n = 5), Hospital Anxiety and Depression Scale (HADS, Zigmond & Snaith, 1983) (n = 5), Depression Anxiety Stress Scale (DASS-21; Lovibond & Lovibond, 1996) (n = 4), Beck Depression Inventory I & II (BDI; BDI-II; Beck & Beck, 1972; Beck, Steer, Ball, & Ranieri, 1996) (n = 4), Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, 1997) (n = 3), and Reynolds Adolescent Depression Scale (RADS; Reynolds, 1987) (n = 2).

Conduct problems

Problem gambling (PG)

A total of 15 studies (11 cross-sectional and four longitudinal) were included investigating the relationship between CP and PG (Table 2). Seven cross-sectional studies indicated that problem gamblers experienced greater CP compared to those with no gambling problems, with relative risk or odds ratios ranging from 1.01 to 19.6 (Barnes et al., 2011; Brunelle, Leclerc, Cousineau, Dufour, & Gendron, 2012a; Cheung, 2014; Cook et al., 2015; Hayatbakhsh et al., 2013; Welte et al., 2009; Willoughby et al., 2004). Additionally, Brunelle and colleagues (2012b) reported that CP were related to an increased severity of PG in both Internet and non-Internet gamblers. Terrone et al. (2018) investigated the relationship between CP and PG while measuring the moderating role of attachment style. The authors reported that although CP were associated with PG among those with a dismissing-detached attachment style, this was not the case among those with a fearful-avoidant attachment style. The remaining two studies reported non-significant associations between CP and PG (Pace et al., 2013; Widinghoff et al., 2019).

Evidence from cross-sectional studies indicate strong evidence for a positive association between CP and PG. The two cross-sectional studies that did not report significant associations between CP and PG were similar in that they included a relatively small sample size of males. In their study, Pace and colleagues (2013) report that at-risk (M = 1.09, SD = 0.54) or pathological gamblers (M = 1.19, SD = 0.68), did not differ from non-gamblers (M = 1.11, SD = 0.55) with regards to CP. However, CP did differentiate between at-risk and pathological gamblers. Given the study's small sample size, it is possible that the non-significant difference was related to the lack of statistical power within the analyses. As for Widinghoff et al. (2019), they investigated violent offenders in prison where rates of conduct disorder in the overall sample was of 79.2%. Due to the elevated rates of conduct disorder in the non-PG (77.4%) and PG (88.4%) groups, it is possible that there was a ceiling effect associated with abnormally elevated levels of CP.

Two out of the four longitudinal studies reported significant relationships between CP and PG. Martins et al. (2013) reported that boys with chronically high CP throughout adolescence were more likely to be at-risk or problem gamblers by early adulthood. Martins et al. (2014) expanded on these findings by noting that problem gamblers in early adulthood were more likely to have been arrested before the age of 23 compared to social or non-gamblers. In contrast to these results, two studies of a duration of 2- and 7-years, respectively, reported that CP at age 16 did not predict PG at age 17 (Vitaro et al., 2001) or 23 (Wanner et al., 2009). All four studies sampled predominantly male youth from disadvantaged neighborhoods, with some overlap in the included samples (i.e., Vitaro et al., 2001; Wanner et al., 2009). As there were differences in the times at which CP was measured between the studies and differences in the types of CP measured, it is possible that the presence of CP (primarily aggression) in early adolescence (Martins et al., 2013), as opposed to CP (primarily delinquency) in later adolescence (Vitaro et al., 2001; Wanner et al., 2009), is of greater importance when predicting PG in young adulthood.

Problem video gaming (PVG)

A total of five studies (four cross-sectional and one longitudinal) investigated the relationship between CP and PVG (Table 2). All four cross-sectional studies reported significant associations, with three studies finding a positive association between CP and PVG (Kim et al., 2018; Tejeiro et al., 2012; Wartberg et al., 2017), and one study finding that those with a history of CP were less likely to be problem gamers (Ong et al., 2016). Differing results may have been reported in the study by Ong and colleagues (2016) because they investigated a relatively small sample of adolescents presenting at an addiction treatment center in Singapore. Moreover, the authors measured CP dichotomously which does not account for the severity of the behaviors when compared to the other three studies that measured CP continuously.

Only one longitudinal study investigated the prospective relationship between CP and PVG. This study was an extension of Wartberg and colleagues (2017) and they reported that CP at baseline did not predict PVG one year later when accounting for other psychosocial predictors (Wartberg et al., 2019). Overall, there is a paucity of research investigating the association between CP and PVG. Although cross-sectional results are encouraging, longitudinal findings do not confirm the presence of a prospective relationship between CP and PVG.

Depressive symptoms

A total of 18 studies (12 cross-sectional and six longitudinal) investigated the relationship between DS and PG (Table 3). Ten cross-sectional studies reported significant positive associations between DS and PG. Notably, greater DS were reported among problem gamblers compared to non-problem gamblers (Dowd et al. 2018; Ellenbogen et al., 2007; Martin et al., 2014; Nower et al. 2004; Sanscartier et al., 2018; Stuhldreher et al., 2007; Wohl et al., 2008). Moreover, in linear and logistic regression analyses, DS were associated with greater PG severity after controlling for various psychosocial and demographic variables (Cosenza et al., 2019; Molde et al., 2009; Nigro et al., 2017).

The remaining two cross-sectional studies reported non-significant associations between DS and PG (Delfabbro and Grabosky, 2006; Pascual-Leone et al., 2011). Pascual-Leone and colleagues (2011) reported a non-significant correlation in their sample which may have been a result of the study's small sample size (N = 200) of university students, with low overall levels of PG (15 reported at-risk or probable pathological gamblers) and DS (23 reported moderate to severe depression). Further, although Delfabbro & Grabosky (2006) reported a moderate effect size of DS on the presence of PG (d = 0.49), this effect became non-significant when controlling for other psychosocial factors including self-esteem and social alienation (correlation coefficients not reported in the study). Overall, based on the cross-sectional studies, there appears to be good evidence for a positive association between DS and PG.

In the six longitudinal studies, two report significant relationships whereby DS predicted PG one-month (Bilevicius et al., 2018), and six-years later (Dussault et al., 2016). Although the remaining four studies indicated cross-sectional associations between DS and PG (Afifi et al., 2016; Chinneck et al., 2016; Edgerton et al., 2015, 2018), these relationships were non-significant in the prospective analyses. It is important to note that all four of these studies utilized the same cohort (Manitoba Longitudinal Study of Young Adults) over time periods ranging from one to five years. Furthermore, Bilevicius and colleagues (2018) only had a one-month period between both time points which brings into question whether these results would be maintained over longer periods of time. Lastly, as Dussault and colleagues (2016) focused solely on males living in economically disadvantaged areas, it is possible that these individuals present with different psychosocial developmental trajectories compared to young adults in the general population. Overall, longitudinal evidence for the relationship between DS and PG is mixed, with a trend towards DS being a poor predictor of PG.

A total of 10 studies (seven cross-sectional and three longitudinal) investigated the relationship between DS and PVG (Table 3). To begin, two studies reported that problem gamers endorsed greater DS compared to non-problem gamers (Givron et al., 2018; Van Rooij et al., 2014). Additionally, three studies indicated significant positive associations between DS and PVG when controlling for demographic characteristics and other psychosocial factors (Bonnaire & Baptista, 2019; Männikkö et al., 2015; Vadlin et al., 2016). As for structural equation models, Li and colleagues (2011) report that although the direct path between DS and PVG was non-significant, an indirect path existed via escapism and actual-ideal self-discrepancies. However, Kircaburun and colleagues (2019) identified that DS were a non-significant mediator in the relationship between emotional intelligence and PVG.

All three longitudinal studies reported significant relationships between DS and PVG. Investigating the cross-sectional associations between DS and PVG at each time point, Dang and colleagues (2019) report these associations were significant at baseline and one-year later. As for the prospective results, Guillot and colleagues (2016) report that anhedonia predicted greater levels of PVG one year later (OR = 1.33). Similarly, Liu et al. (2018) report that greater DS at baseline were associated with greater PVG one, two, three and four years later. Given these consistent positive associations in both the cross-sectional and longitudinal research, there appear to be good evidence for the concurrent and temporal associations between DS and PVG.

Conduct problems and depressive symptoms

A total of 14 studies (10 cross-sectional and four longitudinal) investigated the relationship between both CP, DS, and PG (Table 4). Three cross-sectional studies reported that at-risk and problem gamblers endorsed greater DS and CP compared to low and no risk gamblers (Giralt et al., 2018; Potenza et al., 2011; Yip et al., 2011). Furthermore, in the two studies utilizing latent class analyses, problem gamblers could be separated based on the presence of greater CP or greater DS (Gupta et al., 2013; Kong et al., 2014). Moreover, Gupta et al. (2013) also identified a class of problem gamblers with higher levels of both CP and DS.

Alternatively, three cross-sectional studies reported that although the association between CP and PG was significant, the association between DS and PG was not. Although Hardoon et al. (2004) reported that problem gamblers endorsed greater levels of CP and DS, only CP were significantly associated with at-risk/probable PG. In the other two studies, problem gamblers were more likely to endorse greater severity of CP but not DS compared to non-problem gamblers (Langhinrichsen-Rohling et al., 2004; Petry & Tawfik, 2001).

As for the remaining studies, Tackett et al. (2017) investigated these relationships by gender and reported that although both DS and CP were correlated with PG among women, only CP were significant for men. Finally, Kaminer et al. (2002) reported no significant association between either DS and PG, and CP and PG. It is important to note that the study included a small number (N = 97) of adolescents diagnosed with psychoactive substance use disorders, where only nine individuals were identified as being at-risk/pathological gamblers. Overall, results from the cross-sectional studies indicated that both CP and DS were generally positively associated with PG, with other studies indicating that only CP were associated with PG. None of the cross-sectional studies indicated that DS were solely associated with PG.

In the four longitudinal studies, two studies reported significant relationships between both DS, CP, and PG. Allami et al. (2017) identified two classes of problem gamblers with higher levels of CP and DS, respectively, and that membership within these classes at age 12 was a significant predictor of PG at age 23. Additionally, Sagoe et al. (2017) report that at-risk and PG classes had greater concurrent levels of CP and DS from age 17 to 19. In contrast to these findings, although Allami, Vitaro, Brendgen, Carbonneau, and Tremblay (2018) utilized the same samples reported in their 2017 study, they identified a third comorbid (high CP and DS) class of adolescents. However, only the high CP and comorbid classes had greater levels of PG at age 16 and 23. Finally, in a 6-year longitudinal study, no significant overall differences in DS and CP were noted across different levels of gambling risk across time. However, sex-dependent effects were identified whereby greater levels of parental-reported CP predicted PG among females, and lower levels of parental-reported CP predicted PG among males (Yücel et al., 2015). Overall, longitudinal findings appear to have established a greater consistency regarding the predictive role of both CP and DS for later PG, followed by CP.

A total of eight studies (seven cross-sectional and one longitudinal) investigated the relationship between both CP, DS, and PVG (Table 4). Generally, problem gamers reported greater CP (η2 = 0.03–0.08; d = 0.71) and DS (η2 = 0.03–0.15; d = 0.58) than non-problem gamers (Müller et al., 2015; Myrseth & Notelaers, 2018; Stockdale & Coyne, 2018; Torres-Rodrígez et al., 2018; Yu & Cho, 2016). Furthermore, two studies reported that after controlling for psychosocial variables, DS (OR = 3.62; RR = 1.25) and CP (OR = 2.97; RR = 1.24) were significantly associated with PVG (Desai et al., 2010; Strittmatter et al., 2015).

One longitudinal study investigated the predictive role of both CP and DS on later PVG. In this study, Krossbakken et al. (2018) reported that both DS (β = 0.11–0.12) and CP (β = 0.08) predicted PVG. Moreover, both CP (OR = 1.10–1.11) and DS (OR = 1.11) predicted membership to the group of problem gamers one year, but not two years later. Overall, all seven cross-sectional studies and the one longitudinal study reported significant associations between both DS, CP and PVG. Although there is strong cross-sectional evidence, with results indicating that the association may be stronger for DS compared to CP, the paucity of longitudinal research limits the extent to which conclusions can be drawn regarding the predictive role of CP and DS on PVG.

One cross-sectional study measured both PG and PVG, in addition to CP and DS (Table 4). In a sample of German students (Walther et al., 2012), multivariate analyses controlling for demographic characteristics and parental monitoring indicated that problem gamers reported greater CP, but not greater DS compared to non-problem gamers. However, problem gamblers did not report greater CP or DS in the multivariate models, even if those with PG reported greater CP in the univariate model. It is noteworthy that in this study, only a small proportion of individuals were identified as problem or probable pathological gamblers (1.3%).

Discussion

The aim of this systematic review was to provide an overview of research identifying the relationship between CP, DS and both PG and PVG among adolescents and young adults. Overall, the results from these studies indicate important distinctions in the role of CP and DS as risk factors for PG and PVG. Present findings are consistent with the wider literature identifying both CP and DS as significantly associated with PG (Dowling et al., 2017; Johansson et al., 2009; Shead et al., 2010) and PVG (Brezing et al., 2010; Kuss & Griffiths, 2012; Sugaya et al., 2019). What is novel in the present review is the identification of the extent to which these claims can be made among adolescents and young adults, with an emphasis on the quantity and methodological quality of studies and the differences between the concurrent and prospective associations. Based on theoretical models for the development of addictive behaviors, there appears to be greater evidence for a combined pathway (Englund & Siebenbruner, 2012; Maslowsky et al., 2014) for the development of both PG and PVG, an externalizing pathway (Zucker, 1994; Zucker et al., 2011) for the development of PG, and an internalizing pathway (Hussong et al., 2011; Trucco et al., 2018) for the development of PVG. These results suggest that although both CP and DS are generally associated with PG and PVG, they may exert different effects on the development of these problem behaviors.

The difference in the extent to which CP and DS are related to PG and PVG warrants further discussion. To begin, CP were identified as a greater risk factor for PG when compared to PVG. In the broader literature, CP have been frequently associated with PG under a general pattern of co-occurring problem behaviors as a result of shared etiological factors (Blaszczynski & Nower, 2002; Jessor & Jessor, 1977; Welte, Barnes, & Hoffman, 2004; Willoughby, Chalmers, and Busseri, 2004). For instance, risk factors for antisocial behaviors and conduct problems include impulsivity and behavioral disinhibition, which may explain why these individuals are at a greater risk for later PG. As impulsivity has been identified as increasing the risk for both adolescent PG (Secades-Villa, Martínez-Loredo, Grande-Gosende, & Fernández-Hermida, 2016) and conduct problems (Ahmad & Hinshaw, 2017; Olson, Schilling, & Bates, 1999), it may be an important factor in the etiology of both behaviors. Overall, these predispositions would be consistent with what has been identified for the externalizing pathway (Zucker, 1994; Zucker et al., 2011).

As for gaming, PVG has not typically been associated with CP, nor included under the classic umbrella of the problem behaviors (Welte, Barnes, & Hoffman, 2004; Willoughby, Chalmers, and Busseri 2004). Indeed, there are noteworthy differences between gambling and gaming which may differentiate the two including the legality of the activity for adolescents and the close connection of youth gambling with other delinquent acts (Burleigh et al., 2019; Welte, Barnes, & Hoffman, 2004; Willoughby, Chalmers, & Busseri, 2004). An important point of distinction here may be that video game playing in youth is not as strongly associated with delinquency when compared to gambling. Additionally, all five included studies investigating PVG and CP were comprised primarily of samples of youth in early adolescence, with aggressive behaviors being more prominent in childhood and early adolescence and delinquent behaviors typically peaking in late adolescence (Fonagy & Luyten, 2018; Hawkins et al., 2002). As such, the age of the samples could potentially explain why evidence for the association between delinquency and PVG is mixed. Nevertheless, it is possible that specific conduct problems such as aggression and rule-breaking pose a higher risk for PVG compared to others including violent or delinquent behaviors. A potential common factor underlying aggression, rule-breaking and problem gaming could be impulsivity (specifically, impulsive decision making and urgency), which has been found to be more prominent among problem gamers (Billieux et al., 2015, 2011; Tian et al., 2018). Given the paucity of research in this domain, further research is necessary to elucidate the differences between specific conduct problems and their predictive value for PVG.

DS appear to be more consistently associated with PVG in both cross-sectional and longitudinal studies. For one, it is possible that the difficulties with emotional coping and emotional dysregulation experienced by those with elevated DS is related to the development of PVG. As the included studies investigating DS and PVG were comprised of samples ranging from early adolescence to young adulthood, it would appear as though these associations are present across these developmental periods. These predispositions and risk factors would be consistent with the internalizing pathway (Hussong et al., 2011; Trucco et al., 2018). Moreover, symptoms of PVG as outlined in the DSM-5 (APA, 2013) and ICD-11 (WHO, 2019) include a loss of interest in previous hobbies/activities, negative emotions (e.g., irritability, sadness) when gaming it taken away, and the use of games to relieve a negative mood (e.g., guilt). These symptoms have important overlaps with those of depression, which include anhedonia, depressed mood, and feelings of worthlessness or inappropriate guilt (APA, 2013). As such, it is possible that these two disorders share symptomatology in a manner that creates a downward spiral of DS and PVG, leading to a significant deterioration of psychological wellbeing over time. However, evidence for these temporal relationships are limited to samples of university students and emerging adults, with additional research being necessary to establish these links in early adolescents. Although similar claims could be made for PG and DS, temporal interrelationships remain inconclusive with stronger evidence for their concurrent associations. Given that gambling is an age-restricted activity in most jurisdictions, it is possible that these prospective associations become clearer later in adulthood.

Based on the included studies, the presence of both CP and DS appear to increase the risk for the development of PG and PVG. Although CP and DS are etiologically heterogeneous (Angold et al., 1999), this would be consistent with the combined pathway (Englund & Siebenbruner, 2012; Maslowsky et al., 2014). A potential hypothesized mechanism underlying this comorbid association could be the role of poor impulse control and emotion dysregulation (Angold et al., 1999; Caspi & Moffitt, 2018). Of note, poor impulse control and emotion dysregulation could explain the shared variance of internalizing (e.g., depression) and externalizing (e.g., conduct) problems, their continuity over time, and the later emergence of addictive behaviors due to poor behavioral inhibition and continued engagement in the behavior as a maladaptive learned coping strategy. Moreover, it is also possible that the predisposition towards poor impulse control or emotion dysregulation is related to gender, with Tackett et al. (2017) reporting that PG among young adult males could be related to risk-taking and impulsivity (as represented by CP) and PG among young adult females being related to affective dysregulation (as represented by DS). However, heightened CP among females may also be indicative of maladjustment (Crick, 1997), which is consistent with the findings by Yücel et al. (2015) identifying that higher levels of aggressive behavior in early adolescent females predicted PG in late adolescence. Overall, further research investigating the effect of CP and DS at different time-points across adolescence and young adulthood, while isolating the effects for males and females will be essential to understand whether the presence of these symptoms at different times represents a greater or lesser risk for the development of PG and PVG.

A major limitation of the included research studies is the shortage of longitudinal research investigating PG and PVG. Although 14 studies were included investigating the predictive role of CP and/or DS on later PG, only five studies were included with regards to PVG. As such, additional studies are necessary to draw clear conclusions regarding the predictive role of both of these risk factors for later PVG. Furthermore, there is a pressing need for greater consistency in the instruments used to measure PVG in order to ensure that different researchers are measuring the same latent construct. Additionally, there is a need to investigate the interaction of these two variables in the way they predict PG or PVG. Understanding the interaction of CP and DS in predicting PG and PVG will lead to important insights regarding the complexity of these symptoms with the potential of identifying youth at the highest risk for these problem behaviors and more concrete evidence for the combined pathway. Moreover, only one study investigated both PG and PVG among adolescents and young adults (Walther et al., 2012). Further research is necessary to clarify the developmental pathways of behavioral addictions and whether youth who report problems with one behavior are at a greater risk of experiencing problems with another at the same or a later time. Although gambling and gaming disorder are the only two formally recognized behavioral addictions (APA, 2013; WHO, 2019), other potential behavioral addictions such as problematic use of the Internet, sex, shopping, tanning, and exercising (Grant, Potenza, Weinstein, & Gorelick, 2010; Petry, Zajac, & Ginley, 2018) could be considered in relation to the risk factors examined in this review in addition to their potential co-occurrence with PG and PVG.

Limitations

Although the present review identified and clarified the role of CP and DS on the development of PG and PVG, it is subject to a number of limitations. First, it is possible that some articles were missed in the systematic search due to relevant terms (i.e., keywords for CP and DS) not being included in the areas searched by the MeSH and multi-purpose terms. Although the present study sought to identify articles with CP and DS as independent variables, it is possible that some published articles did not mention these terms in the title, abstract, keywords, headings and/or subject headings, although they were included in the study. As such, certain articles may have been omitted although they could have reported on the associations between the variables of interest. Second, the methodology used in the present review was primarily descriptive with the aim of investigating a breadth of literature. As such, no statistical conclusions can be drawn from the results. Third, the review was limited to English and French peer-reviewed journal publications identified on the five selected databases. Given these criteria, the authors may have omitted studies from other countries that were unavailable through the selected databases. Fourth, although clear inclusion and exclusion criteria and a high inter-rater reliability was established, the review, screening, and selection process is subjective and prone to certain biases. Fifth, a diversity of instruments measuring the constructs of interest (i.e., CP, DS, PG and PVG) were utilized which may have resulted in the measurement of slightly different constructs between studies. Finally, this review focused on adolescents and young adults and did not include children or older adults. Future reviews should aim to investigate the relative role of CP and DS in predicting PG and PVG across the lifespan.

Conclusions

While research in the domain of behavioral addictions is still in its early stages, the results of this review provide preliminary evidence regarding the role of CP and DS as risk factors for PG and PVG. Additionally, this review presents a novel way of investigating the development of PG and PVG, by drawing from a theoretical framework informed by the SUD literature and applying it to non-substance addictive behaviors. Presently, it appears as though there is significant evidence suggesting the importance of a combined pathway, including the presence of both CP and DS, in the development of PG and PVG. However, there also appears to be some evidence for an externalizing pathway for PG, and evidence for an internalizing pathway for PVG. Given the potential for adolescents and young adults to experience early CP or DS, prevention and intervention efforts should focus on intervening in an integrative manner in order to address multiple, potentially interacting problems simultaneously.

Funding sources

Jérémie Richard received doctoral fellowships from the Social Sciences and Humanities Research Council of Canada (SSHRC) and from the Fonds de recherche du Québec – Société et Culture (FRQSC): Programme de bourse sur le jeu responsable. No funding agencies had input into the content of this manuscript.

Authors' contribution

JR, SB, JD and CT designed the study and wrote the protocol. JR conducted the literature search and JR and EF selected the included articles and extracted the data. JR wrote the first draft of the manuscript. All authors reviewed the literature search, data extraction and contributed to the manuscript. All authors contributed to and have approved the final version of the manuscript.

Conflict of interest

The authors declare no conflict of interest.