Effect of Cognitive-Behavioral Techniques for Problem Gambling and Gambling Disorder: A Systematic Review and Meta-Analysis

Abstract

Aims:

To measure the effect of cognitive-behavioral techniques (CBTs) on gambling disorder severity and gambling behavior at posttreatment and follow-up.

Method:

Seven databases and two clinical trial registries were searched to identify peer-reviewed studies and unpublished studies of randomized controlled trials. The Cochrane Risk of Bias tool assessed risk of bias in the included studies. A random effect meta-analysis with robust variance estimation was conducted to measure the effect of CBTs relative to minimally treated or no treatment control groups.

Results:

Twenty-nine studies representing 3,991 participants were identified. CBTs significantly reduced gambling disorder severity (g = −1.14, 95% confidence interval [CI] = −1.68, −0.60, 95% prediction interval [PI] = −2.97, 0.69), gambling frequency (g = −0.54, 95% CI = −0.80, −0.27, 95% PI = −1.48, 0.40), and gambling intensity (g = −0.32, 95% CI = −0.51, −0.13, 95% PI = −0.76, 0.12) at posttreatment relative to control. CBTs had no significant effect on follow-up outcomes. Analyses supported the presence of publication bias and high heterogeneity in effect size estimates.

Conclusions:

Cognitive-behavioral techniques (CBTs) are a promising treatment for reducing gambling disorder and gambling behavior; however, the effect of CBTs on gambling disorder severity and gambling frequency and intensity at posttreatment is overestimated, and CBTs may not be reliably efficacious for all individuals seeking treatment for problem gambling and gambling disorder.

INTRODUCTION

Cognitive-behavioral techniques (CBTs) have repeatedly been touted as the treatment elements with the most robust empirical support for treating problem gambling and gambling disorder (1–3). These techniques typically involve the identification of triggers and consequences of gambling, strategies for managing urges to gamble, the generation of alternative activities to gambling, and relapse prevention (3). CBTs are commonly supplemented with motivation enhancing interventions to target low treatment engagement (3–5).

Several meta-analyses, including those on face-to-face treatments (1,6–8) and self-guided, remote treatments (e.g., Internet modules, workbook chapters) (9–11), have supported the immediate effect of CBTs. Collectively, the meta-analyses suggest that the effect is greater than minimally treated (e.g., psychoeducation, referral to Gamblers’ Anonymous) and inactive control groups (e.g., assessment only, waitlists) at posttreatment, and that CBTs produce greater effects on gambling disorder symptom severity than on gambling behavior (e.g., frequency, intensity, duration) (1,6,7,10,11). Nascent evidence suggests that the effect of CBTs may endure at follow-up/in the months and years following treatment (7,12).

Despite the support of these meta-analyses, several methodological weaknesses limit our understanding of the effect of CBTs on gambling disorder severity and gambling behavior relative to minimally treated and inactive controls. First, prior meta-analyses have calculated weighted effect sizes at posttreatment and follow-up from both between-group (i.e., randomized controlled trials) and within-group (i.e., cohort) studies (7,10,11,13). The inclusion of within-group effect sizes introduces a problematic confound because approximately one-third of individuals may recover from gambling disorder without any formal treatment (14). Therefore, it is not possible to determine whether the observed treatment effects were attributable to CBTs or natural change (15).

Second, authors of these reviews have not thoroughly examined how publication bias might impact effect size estimates. Three previous meta-analyses have not included any analysis of publication bias (1,6,13), and other meta-analyses assessed publication bias with the fail-safe N, which determines the number of studies needed to nullify a treatment’s effect (7,8). This indicator is inadequate because there are no criteria to determine how many studies would constitute a small or large N, and the number of studies to nullify the significance of a treatment’s effect is not typically relevant in meta-analysis (16). Preferably, meta-analyses should integrate multiple indicators of publication bias and examine how publication bias influences the magnitude of a treatment effect. For example, it would be informative to examine asymmetry in funnel plots (17) and to determine the dispersion of studies finding significant versus nonsignificant effects. Such an examination would prove helpful for understanding how publication bias impacts estimates of true treatment effects.

Third, previous authors have not thoroughly examined the impact of heterogeneity in effect sizes on the estimate of treatment effects. Past meta-analyses have reported the I² and/or τ² values to reflect heterogeneity (1,6,7,10,11). However, these values have been criticized because they do not inform how heterogeneity impacts estimates of treatment effects in future studies (18,19). Instead, the inclusion of prediction intervals may be useful to supplement the overall summary effect size estimate and its confidence interval (19). Prediction intervals are an index of precision based on the standard deviation, whereas the confidence interval is an index of precision based on the standard error. These prediction intervals would provide highly probable values for the true effects of CBTs in future settings and whether these effects correspond to benefit, null effects, or even deterioration (19). Some authors have already speculated that the effect of CBTs is overestimated due to high between-study heterogeneity in effect sizes (1,7). Collectively, these methodological limitations raise questions about the field’s certainty regarding the magnitude of the effect of CBTs on gambling outcomes.

The purpose of the present systematic review and meta-analysis was to strengthen the understanding of the magnitude of the effect of CBTs on gambling outcomes. Specifically, the present meta-analysis estimated the effect of CBTs relative to inactive or minimally treated control groups at posttreatment and follow-up in randomized controlled trials. These control groups were used to control for potential history effects as well as recovery without treatment. Estimates of the effect of CBTs included thorough testing of the potential impact of heterogeneity and publication bias as detailed above.

METHOD

The methods of the current study were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (20) and A Measurement Tool to Assess Systematic Reviews-2 (21) guidelines. The study was also preregistered on International Prospective Register of Systematic Reviews (CRD42022288058). Analyses on non-gambling outcomes, which are detailed in the study pre-registration, are presented in a separate report (22). There were no deviations from the study preregistration.

Search Strategy

Searches were conducted in seven databases and two clinical trial registries during July 2022 (see Supplemental Table 1). The search strategy had no restrictions on start date or language. Reference lists of past meta-analyses were also searched (6,10,11,13,23,24).

Inclusion and Exclusion Criteria

Studies were included if they involved (a) participants 18 years of age and older; (b) participants with problem gambling or gambling disorder according to an assessment strategy with some validity evidence, or perception of problem gambling or gambling disorder; (c) a treatment identified as cognitive-behavioral (with or without motivational interventions) or a treatment that described key cognitive-behavioral elements/objectives like cognitive restructuring, imaginal desensitization, relapse prevention, or stimulus control (e.g., limiting gambling expenditures) and (d) trials with random assignment to two or more conditions with an inactive (i.e., no treatment, waitlist) or minimal treatment (i.e., assessment only, psychoeducation, referral to Gamblers’ Anonymous) condition. Studies were excluded if they involved: (a) planned study protocols; (b) participants taking pharmacotherapy concurrently with CBTs (either as part of the study protocol or as part of their medical care unrelated to the study) because pharmacotherapy could confound the effect of CBTs on gambling disorder symptoms and gambling behavior (25); and (c) a secondary analysis.

Study Identification

Identification of eligible studies occurred in two stages. In the first stage, two doctoral level clinical psychologists (RAP, SCP), screened studies at the title and abstract levels. In the second stage, one clinical psychology graduate student (DPF) and a doctoral level clinical psychologist (RAP), independently determined if articles were eligible based on inclusion and exclusion criteria at the full-text level. The interrater reliability in stage 1 was ᴋ = 0.93, and the interrater reliability at stage 2 was ᴋ = 0.94. Discrepancies were resolved through discussion.

Data Extraction

Recommendations from the Banff, Alberta Consensus guided the selection of several outcomes (26), and multiple outcomes were extracted to evaluate the efficacy at posttreatment and at the longest available follow-up. Gambling outcomes included gambling disorder severity, which was measured using a variety of measures with some reliability and validity support, including the gambling subscale of the Addiction Severity Index (27), Gambling Symptom Assessment Scale (28), the pathological gambling adaptation of the Yale-Brown Obsessive-Compulsive Scale (29), the Problem Gambling Severity Index (30), the South Oaks Gambling Screen (31), and the number of Diagnostic and Statistical Manual of Mental Disorders symptoms. Gambling outcomes also included several dimensions of gambling behavior, including frequency (e.g., number of times gambled), intensity (e.g., money spent on gambling), and duration (e.g., minutes or hours gambled). Two authors (RAP, DPF) extracted outcomes to verify their accuracy.

Risk of Bias Assessment

The Cochrane risk of bias tool was used to assess for possible bias in the included studies (32). Potential bias was assessed in the following domains: random sequence generation, allocation concealment, masking of outcome assessors, and completeness of outcome data. Each domain was assigned high risk of bias, low risk of bias, or unclear risk of bias. Two authors (RAP, DPF) independently rated risk of bias using a codebook. The interrater reliability was ᴋ = 0.92. Discrepancies were resolved through discussion.

Data Analysis Plan

Five corresponding authors were contacted via email with requests for results when information needed to conduct meta-analysis was not reported in their included study. A reminder email was sent two weeks later if there was no response. Two authors provided us with their results, two indicated that their data were destroyed due to age, and one did not respond.

Comprehensive Meta-Analysis version 3.3070 was used to compute between-group Hedges’s g effect sizes, where negative values indicated beneficial effects of CBTs, at posttreatment and the longest available follow-up. Hedges’s g was used to correct for sampling bias (33) because some studies had relatively small sample sizes (34,35).

All meta-analyses were conducted using the “robumeta” and “clubSandwich” packages in R version 4.2.1. Meta-regression models with robust variance estimates where ρ was assumed 0.8 were used to accommodate dependent effect sizes (36). Random-effects meta-regression models estimated the overall mean Hedges’s g effect sizes. Sensitivity analyses were conducted to examine differences in Hedges’s g effect sizes at posttreatment and follow-up where all outcomes were combined and then compared to each outcome (i.e., frequency, intensity, duration, and gambling disorder severity) separately. Results were only presented for models with adequate degrees of freedom after accounting for small sample size adjustments (37). To test for potential publication bias, contour-enhanced funnel plots were examined for asymmetry, and the Egger’s regression test tested asymmetry in the funnel plot (17).

RESULTS

Study Identification

Figure 1 displays a flowchart of the study identification process, including the primary reason a study was excluded. The initial search yielded 5,409 articles. After removing duplicates, 2,469 articles were screened at the abstract level. Seventy articles were identified for possible inclusion and reviewed at the full-text level. A total of 29 articles met the inclusion criteria.

Figure 1.

Study flowchart for identification of studies to be included in this review.

Study and Sample Characteristics

Table 1 displays the 29 studies that met inclusion criteria and their sample characteristics. The publication years of the studies ranged from 1997 to 2020. Of these 29 studies, between 0% and 41% were included in past meta-analyses (6,10,11,13,23,24) (see Supplemental Table 3). Most studies were conducted in Canada (k = 9), followed by the United States of America (k = 8), Australia (k = 5), Sweden (k = 2), China (k = 1), France (k = 1), Japan (k = 1), Nigeria (k = 1), and Norway (k = 1). Sixty-nine percent of studies indicated that they were supported by grant funding, and 31% did not. Thirty-one percent pre-registered their methods and data analyses, and 69% did not describe pre-registration. The mean age of study participants was 40.02 years, and 60% identified as male. Most participants (72%) identified as White.

Table 1.

Studies Included in the Systematic Review and Meta-Analysis

Study	Location	Funding source	Pre-registered?	Mean age (years)	% male	% Asian	% Black	% White	% Hispanic
Abbott et al. (2018)	AUS	Ministry of Health New Zealand	Y	39	47	NR	NR	NR	NR
Boudreault et al. (2018)	CAN	Ministry of Health and Social Services of Quebec	N	52	61	NR	NR	NR	NR
Carlbring & Smit (2008)	SWE	Swedish National Institute of Public Health	N	32	94	NR	NR	NR	NR
Carlbring et al. (2010)	SWE	Swedish National Institute of Public Health	N	40	84	NR	NR	NR	NR
Casey et al. (2017)	AUS	Australia Department of Employment, Economic Development and Innovation	N	44	42	2	NR	82	NR
Cunningham et al. (2019)	CAN	Canada Research Chairs	Y	36	45	NR	NR	78	NR
Dowling et al. (2007)	AUS	None	N	44	0	NR	NR	NR	NR
Ede et al. (2020)	Nigeria	None	N	23	70	0	100	0	0
Grant et al. (2009)	USA	National Institute of Mental Health	Y	49	62	NR	NR	91	NR
Harris & Mazmanian (2016)	CAN	None	N	34	53	NR	3	72	6
Hodgins et al. (2001)	CAN	Alberta Heritage Foundation for Medical Research	N	46	48	NR	NR	NR	NR
Hodgins et al. (2009)	CAN	Canadian Institutes of Health Research, Ontario Problem Gambling Research Centre	N	NR	45	NR	NR	NR	NR
LaBrie et al. (2012)	USA	Nevada Department of Health and Human Services, National Institute of Drug Abuse	N	46	58	NR	NR	70	9
Ladouceur et al. (2001)	CAN	National Center on Responsible Gaming	N	42	83	NR	NR	NR	NR
Ladouceur et al. (2003)	CAN	None	N	43	78	NR	NR	NR	NR
Larimer et al. (2012)	USA	National Institute on Mental Health	Y	21	65	28	NR	60	3
Luquiens et al. (2016)	France	Winamax	Y	35	92	NR	NR	NR	NR
Marceaux & Melville (2011)	USA	None	N	48	37	NR	11	89	NR
Myrseth et al. (2009)	Norway	None	N	37	86	NR	NR	NR	NR
Oei et al. (2010)	AUS	None	N	NR	NR	NR	NR	NR	NR
Oei et al. (2018)	AUS	Queensland Office of Liquor and Gaming Regulation	N	49	49	NR	NR	87	NR
Petry et al. (2006)	USA	National Institute on Mental Health	Y	45	55	NR	9	84	4
Petry et al. (2008)	USA	National Institute on Drug Abuse, National Institute on Mental Health	N	43	62	NR	23	59	14
Petry et al. (2009)	USA	National Institute on Drug Abuse, National Institute on Mental Health	Y	20	87	7	NR	91	NR
Petry et al. (2016)	USA	National Institute on Drug Abuse	Y	42	72	NR	50	28	20
So et al. (2020)	Japan	Japan Agency for Medical Research and Development	Y	36	79	NR	NR	NR	NR
Sylvain et al. (1997)	CAN	Conseil Québécois de la Recherche Sociale, Loto-Québec	N	40	96	0	0	100	0
Toneatto et al. (2014)	CAN	None	N	44	56	NR	NR	NR	NR
Wong et al. (2015)	China	None	N	NR	100	100	0	0	0

Notes. + = high quality; - = low quality; AUS = Australia; CAN = Canada; CBT = cognitive-behavioral techniques; N = no; NR = not reported; SWE = Sweden; Y = yes; USA = United States of America

Table 2 presents the relevant study conditions, the presence of posttreatment and follow-up assessments, and the outcomes reported across the 29 studies. The studies comprised a total of 3,991 participants who were allocated to 38 treatment-control comparisons. Of the 38 CB conditions, 9 (24%) were combined with motivational interventions, 11 (29%) were conducted in groups, and 18 (47%) were conducted remotely (i.e., Internet, self-help workbook, telephone).

Table 2.

Relevant Study Conditions, Assessment Time Points, and Outcomes

Study	Total N	Relevant study conditions (n)	Posttreatment outcomes (months since baseline)	Follow up outcomes (months since treatment termination)	Outcomes
Abbott et al. (2018)	350	MI + CB WB + booster calls (116) MI + CB WB (118) Gambling helpline (116)	Y (3)	Y (9)	Frequency Intensity
Boudreault et al. (2018)	62	CB WB (31) Waitlist (31)	Y (2.75)	N	Frequency Intensity Duration GD severity
Carlbring & Smit (2008)	66	Internet CBT (34) Waitlist (32)	Y (3)	N	GD severity
Carlbring et al. (2010)	96	Group CBT (50) Waitlist (46)	Y (3)	Y (9)	Frequency Intensity Duration GD severity
Casey et al. (2017)	115	Internet CBT (60) Waitlist (55)	Y (1.5)	N	Frequency Intensity GD severity
Cunningham et al. (2019)	321	Internet CBT (151) No treatment (170)	Y (1.5)	Y (4.5)	Frequency GD severity
Dowling et al. (2007)	56	CBT (14) Group CBT (17) Waitlist (25)	Y (3)	N	Frequency Intensity Duration
Ede et al. (2020)	40	Group CBT (20) Waitlist (20)	Y (2)	Y (1)	GD severity
Grant et al. (2009)	68	CBT (33) GA referral (35)	Y (2)	Y (1)	GD severity
Harris & Mazmanian (2016)	32	Group CBT (16) Waitlist (16)	Y (2)	N	GD severity
Hodgins et al. (2001)	102	MI + CB WB (32) CB WB (35) Waitlist (35)	Y (3)	N	Frequency GD severity
Hodgins et al. (2009)	314	MI + CB WB + booster calls (84) MI + CB WB (83) CB WB (82) Waitlist (65)	Y (1.5)	Y (10.5)	Frequency Intensity GD severity
LaBrie et al. (2012)	377	CB WB with therapist guidance (105) CB WB (108) Waitlist (164)	Y (1)	Y (2)	Frequency
Ladouceur et al. (2001)	64	Cognitive therapy (35) Waitlist (29)	Y (3)	Y (9)	Frequency Intensity GD severity
Ladouceur et al. (2003)	71	Group cognitive therapy (46) Waitlist (25)	Y (3)	N	Frequency Duration GD severity
Larimer et al. (2012)	95	Group CBT (44) Assessment only (51)	N	Y (2.5)	Intensity GD severity
Luquiens et al. (2016)	829	Emailed CBT (301) CB WB (264) Waitlist (264)	Y (1.5)	Y (1.5)	Frequency Intensity GD severity
Marceaux & Melville (2011)	27	Group CBT (18) Waitlist (9)	Y (2)	N	Frequency Intensity GD severity
Myrseth et al. (2009)	14	Group CBT (7) Waitlist (7)	Y (1.75)	Y (3)	Intensity GD severity
Oei et al. (2010)	102	MI + CBT (37) Group MI + CBT (37) Waitlist (28)	Y (1.5)	N	Frequency Intensity
Oei et al. (2018)	55	CB WB (23) Waitlist (32)	Y (1.75)	N	Frequency Intensity GD severity
Petry et al. (2006)	231	GA referral + CB (84) GA referral + CB WB (84) GA referral (63)	Y (2)	Y (10)	Frequency GD severity
Petry et al. (2008)	88	MET + CBT (40) Assessment only (48)	Y (1.5)	Y (7.5)	Intensity GD severity
Petry et al. (2009)	55	MET + CBT (21) Assessment only (34)	Y (1.5)	Y (7.5)	Frequency GD severity
Petry et al. (2016)	151	MET + CBT (82) Psychoeducation (69)	Y (2)	Y (22)	Frequency GD severity
So et al. (2020)	197	Text-message based CBT (96) Assessment only (101)	Y (1)	N	Frequency Intensity GD severity
Sylvain et al. (1997)	29	CBT (14) Waitlist (15)	Y (NR)	N	Frequency Intensity Duration GD severity
Toneatto et al. (2014)	18	Group mindfulness enhanced CBT (9) Waitlist (9)	Y (1.75)	N	GD severity
Wong et al. (2015)	38	Group CBT (18) Supportive counseling (20)	Y (2.5)	N	Frequency Intensity Duration GD severity

Notes. Studies highlighted in gray indicate studies that were included in the systematic review but could not be meta-analyzed due to insufficient reporting of results.

CB = cognitive-behavioral; CBT = cognitive-behavioral techniques; GA = Gamblers’ Anonymous; GD = gambling disorder; MET = motivational enhancement therapy; MI = motivational interviewing; N = no; Y = yes; WB = workbook

Of the 29 studies, 27 (93%) reported a posttreatment assessment. Posttreatment assessments occurred between one and three months since baseline (median = 2 months). Fourteen studies (48%) reported a long-term follow-up assessment. Follow-up assessments occurred between 1 and 22 months after CBTs were terminated/the posttreatment assessment (median = 7.5 months).

Risk of Bias Assessment

Table 3 presents the ratings for risk of bias on randomization sequence generation, allocation concealment, masking of assessors, and incomplete outcome data. Across the 29 studies, only three studies (10%) had low risk of bias ratings for each risk of bias domain (12,38,39). Most studies (86%) did not use an adequate method to mask outcome assessors to study conditions because they did not adequately describe how assessors were masked, relied entirely on self-report measures, or did not mask assessors. Approximately three-quarters of the studies (76%) did not describe an adequate method to conceal the allocation sequence to the study team or did not use an adequate method. More than half of the studies (66%) did not report complete outcome data or relied on completer analyses. About half of the studies (52%) did not adequately describe the method for generating the randomization sequence or did not use an adequate method.

Table 3.

Assessment of Study Quality for the Studies Included in the Meta-Analysis

Study	Randomization Sequence Generation	Allocation Concealment	Masking of Assessors	Outcome Data
Abbott et al. (2018)	+	+	+	+
Boudreault et al. (2018)	+	?	−	+
Carlbring & Smit (2008)	+	+	−	+
Carlbring et al. (2010)	+	+	−	+
Casey et al. (2017)	+	-	−	-
Cunningham et al. (2019)	+	+	−	+
Dowling et al. (2007)	?	?	−	+
Ede et al. (2020)	+	+	−	+
Grant et al. (2009)	?	?	−	+
Harris & Mazmanian (2016)	−	−	−	+
Hodgins et al. (2001)	?	?	−	−
Hodgins et al. (2009)	+	?	+	+
LaBrie et al. (2012)	+	?	−	−
Ladouceur et al. (2001)	?	?	?	−
Ladouceur et al. (2003)	?	?	?	−
Larimer et al. (2012)	?	?	−	+
Luquiens et al. (2016)	?	?	−	−
Marceaux & Melville (2011)	?	?	?	−
Myrseth et al. (2009)	?	?	?	−
Oei et al. (2010)	+	?	?	?
Oei et al. (2018)	?	?	-	−
Petry et al. (2006)	+	?	?	+
Petry et al. (2008)	+	+	+	+
Petry et al. (2009)	+	+	+	+
Petry et al. (2016)	+	?	−	+
So et al. (2020)	+	+	−	+
Sylvain et al. (1997)	?	?	?	−
Toneatto et al. (2014)	−	−	?	+
Wong et al. (2015)	?	?	−	+

Notes. + = low risk of bias; – = high risk of bias; ? = unclear risk of bias

Posttreatment Outcomes

Table 4 presents the effect of CBTs on all gambling outcomes at posttreatment. When all outcomes were combined (k = 25, 80 effect sizes), CBTs significantly reduced gambling disorder severity and gambling behavior compared to control at posttreatment (g = −0.79, 95% CI [−1.19, −0.47], p < .0001, τ² = 0.42). Approximately 88% of the variance was due to variance in true effects, suggesting that effects could vary considerably in future trials (95% PI = −2.18, 0.60) when combining all outcomes.

Table 4.

Effect of Cognitive-Behavioral Techniques at Posttreatment with Confidence Intervals and Prediction Intervals

Posttreatment
Outcome	k	# of effect sizes	Hedges’s g	95% CI	95% PI
All combined	25	80	−0.79	−1.12, −0.46	−2.18, 0.60
Severity	20	28	−1.14	−1.68, −0.60	−2.97, 0.69
Frequency	18	26	−0.54	−0.80, −0.27	−1.48, 0.40
Intensity	14	20	−0.32	−0.51, −0.13	−0.76, 0.12
Duration	5	6	‐	‐	‐
Follow-Up
Outcome	k	# of effect sizes	Hedges’s g	95% CI	95% PI
All combined	10	29	−0.44	−1.03, 0.15	−1.87, 0.98
Severity	8	12	−0.58	−1.52, 0.36	−2.58, 1.42
Frequency	7	12	−0.10	−0.30, 0.10	−0.44, 0.24
Intensity	4	5	‐	‐	‐
Duration	0	0	‐	‐	‐

Notes. Results were only presented for models with adequate degrees of freedom after accounting for small sample size adjustments to the robust variance estimates.

CI = confidence interval, k = number of studies, PI = prediction interval.

Sensitivity analyses were conducted to examine the effect of CBTs for each outcome at posttreatment. Analyses indicated that CBTs significantly reduced gambling disorder severity (k = 20, 28 effect sizes, g = −1.14, 95% CI [−1.68, −0.60], p < .001, τ² = 0.68) and gambling frequency (k = 18, 26 effect sizes, g = −0.54, 95% CI [−0.80, −0.27], p < .001, τ² = 0.18) relative to control. A large amount of the variance in the effects on gambling disorder severity (I² = 91%) and frequency (I² = 79%) was due to variance in true effects, suggesting that effects could vary considerably in future trials for both severity (95% PI = −2.97, 0.69) and frequency (95% PI = −1.48, 0.40). CBTs also significantly reduced gambling intensity (k = 14, 20 effect sizes, g = −0.32, 95% CI [−0.51, −0.13], p < .004, τ² = 0.03) relative to control. Approximately 34% of the variance was due to true effects, suggesting that effects could vary considerably in future trials (95% PI = −0.76, 0.12). An insufficient number of studies reported results on duration at posttreatment, and this outcome could not be examined.

Two tests suggested the presence of publication bias at posttreatment with all outcomes combined. An examination of a contour-enhanced funnel plot indicated asymmetry (Figure 2). The Egger’s regression test to funnel plot asymmetry was significant (p < .0001).

Figure 2.

Funnel Plot of All Hedges’s g Values at Posttreatment

Follow-Up Outcomes

Table 3 presents the effect of CBTs on all gambling outcomes at follow-up. CBTs had no significant effect on all outcomes combined (k = 10, 29 effect sizes, p = 0.13, τ² = 0.29, I² = 87%), gambling disorder severity (k = 8, 12 effect sizes, p = 0.19, τ² = 0.44, I² = 90%), or gambling frequency (k = 7, 12 effect sizes, p = 0.26, τ² = 0.01, I² = 17%). An insufficient number of studies reported results on intensity and duration at follow-up, and these outcomes could not be examined.

Two tests suggested the presence of publication bias at follow-up with all outcomes combined. An examination of a contour-enhanced funnel plot indicated asymmetry (Figure 3). The Egger’s regression test to funnel plot asymmetry was significant (p < .0001).

Figure 3.

Funnel Plot of All Hedges’s g Values at Follow-Up

DISCUSSION

The present study aimed to understand the effect of CBTs for problem gambling and gambling disorder at posttreatment and follow-up. This study consisted of 29 studies representing almost 4,000 participants and constituted a major methodological improvement upon past meta-analyses because it included randomized controlled trials only, thoroughly tested publication bias, and reported prediction intervals. The results of the robust variance estimation meta-analysis indicated that CBTs, relative to control, produced a significant, large reduction (g = −0.79) in a variety of outcomes compared to control at posttreatment, and these results are consistent with several past meta-analyses (6,10,11,13,23,24). CBTs significantly reduced gambling disorder severity (g = −1.14), gambling frequency (g = −0.54), gambling intensity (g = −0.32), and gambling duration (g = −0.70). There was no conclusive evidence for the effect of CBTs on any outcomes at follow-up.

Although these findings support that CBTs can be efficacious, the present meta-analysis challenged the prevailing assumption that there is a robust effect of CBTs on problem gambling and gambling disorder (6,10,11,13,23,24). Specifically, funnel plots indicated that studies with small sample sizes and large treatment effects dominate the literature. According to the risk of bias assessment, most studies on CBTs had high attrition paired with completer analyses. Thus, the effect of CBTs on posttreatment gambling outcomes in previous meta-analysis appears overestimated. One explanation for this publication bias may be the preference to publish studies with large treatment effects over studies with small or non-significant effects. The lack of conclusive evidence for the effect of CBTs on any outcome at follow-up is likely not simply due to publication bias but limits of study design, with a notable lack of studies reporting results beyond the posttreatment assessment.

The present meta-analysis also indicated that CBTs might not be reliably efficacious for everyone seeking treatment for problem gambling and gambling disorder. The present meta-analysis indicated there was a large amount of heterogeneity in posttreatment effect sizes, which was consistent with past meta-analyses (1,7). The prediction intervals collectively indicated that the magnitude of treatment effects in future trials could range from g = −2.58 to g = 1.42. Thus, participants in future trials may experience large reductions in gambling disorder severity and gambling behavior relative to minimal or no treatment. However, some participants may experience no change in outcomes or may even deteriorate. The span of these prediction intervals suggests that there is great promise for CBTs in reducing gambling disorder severity and gambling behavior, but the precise magnitude of effects at posttreatment is unclear.

The heterogeneity in the magnitude of effects of CBTs on gambling disorder severity and gambling behavior at posttreatment may stem from the lack of grant funding to support rigorous research on problem gambling and gambling disorder treatments (40). Moderator analyses indicated that studies with grant funding were significantly associated with smaller reductions in gambling frequency at posttreatment than studies without grant funding, suggesting that studies with grant funding produce more reliable estimates of treatment effects. As of the year 2022, no federal agency funds or guides programs to address gambling disorder in the United States of America. States funds for any research on gambling were limited to $14 million in the year 2016, and only $144,000 was available for research on gambling treatment (40). These total available state funds for any gambling research are 20 times smaller than the $554 million of federal funds available for alcohol research (41) and less than 1% of the $1.8 billion available for drug research in the year 2022 (42). The lack of funding for gambling treatment research probably translates to limited resources for generating proper randomization sequences, implementing sufficient allocation concealment, masking outcome assessors, and incentivizing participants to complete assessments beyond the termination of CBTs; only three studies in the present meta-analysis were rated low risk of bias across these four dimensions. More funding of treatment research is needed to provide precise estimates of effect of CBTs on gambling disorder symptom severity and dimensions of gambling behavior.

The heterogeneity in posttreatment effect sizes may also stem from other moderators like whether CBTs were therapist-assisted. Moderator analyses indicated that therapist-assisted treatments were significantly associated with larger reductions in gambling intensity than self-guided treatments, which is consistent with another meta-analysis (9). Furthermore, there were other possible explanations for effect size heterogeneity like the duration of treatment. Another meta-analysis has shown that the intended and received number of face-to-face sessions attended was significantly associated with more favorable treatment outcomes (6). Future meta-analyses should examine how different study design features and how heterogeneous combinations of CBTs moderate effect sizes.

Several limitations should be borne in mind when interpreting the results of the present meta-analysis. First, the procedures for testing publication bias in the present meta-analysis do not necessarily represent direct evidence of publication bias. The procedures only test whether the funnel plot is asymmetrical. However, the results related to publication bias suggest that it is highly likely that the magnitude of the effect of CBTs on gambling disorder severity and gambling behavior is overestimated. Future researchers are strongly encouraged to pre-register their randomized controlled trials in trial registries, as pre-registration will facilitate later investigations of publication bias.

Second, practically all outcome assessments were self-reported. Although there is generally strong concordance between self-reported gambling behavior and actual gambling behavior (43), emerging evidence suggests low concordance for online gambling behavior. Specifically, 4% of individuals can accurately recall online gambling losses and frequencies within a 10% margin (44). This research is concerning because opportunities to gamble online continue to proliferate. The increased use of objective gambling data (e.g., participant data pulled from gambling websites) represents an improved strategy for assessing the impact of interventions on actual behavior change compared with self-reported behavior (44). Furthermore, the complete reliance upon self-report questionnaires of gambling disorder symptoms is inherently limited by the degree to which individuals can understand the questions (45). Altogether, there are assessment-related limitations pervasive throughout the gambling treatment literature.

Third, the presence of comorbid conditions was not well documented in the included studies. Ten studies (34%) excluded participants experiencing comorbid conditions (e.g., depression, psychosis, substance use). However, this information was only useful for determining the absence rather than the presence of comorbid conditions. Thus, it was not possible to determine what impact the presence of comorbid conditions had on the results.

Fourth, the inclusion and exclusion criteria for the present meta-analysis had strengths and weaknesses that impact the understanding of the results. Studies were included if participants perceived that they had a gambling problem because many individuals experience gambling harms without meeting diagnostic criteria for gambling disorder (46). This inclusion criterion was selected because it would maximize the generalizability of individuals experiencing gambling harm, but it is possible that many participants were not actually experiencing gambling harm. Relatedly, studies were excluded if participants were taking pharmacotherapy concurrently with CBTs because there is evidence that use of pharmacotherapy reduces gambling disorder severity (25). This exclusion criterion was selected to isolate the effects of CBTs independent of pharmacotherapy, but it is possible we unintentionally reduced generalizability to samples of individuals with comorbid psychological disorders. Future meta-analyses should seek to include individuals with comorbid psychological disorders and understand the relative effect of CBTs and pharmacotherapies on outcomes.

Fifth, the present study was restricted to minimal and no treatment control conditions and did not compare the effect of CBTs to other active treatments. This restriction allowed us to understand the maximal effects of CBTs while controlling for history effects and spontaneous remission. However, this restriction prevented us from understanding whether CBTs reduce gambling behavior and gambling disorder severity relative to theoretically distinct treatments (e.g., twelve-step facilitation), different combinations of CBTs, or pharmacotherapy. Future meta-analyses should examine the effects of CBTs on these gambling outcomes relative to other active treatments to understand whether certain treatments are more or less efficacious than other treatments.

In conclusion, CBTs show promise for reducing gambling disorder severity and gambling behavior. However, the effect of CBTs on these outcomes is overestimated, and CBTs might not be reliably efficacious for everyone seeking treatment for problem gambling and gambling disorder. Future meta-analyses should further test potential moderators of heterogeneity. However, these meta-analyses require rigorous randomized clinical trials with large sample sizes and incentives that encourage the completion of study assessments at posttreatment and beyond the termination of CBTs. Such trials will not be possible without the United States federal government taking a more proactive role in funding gambling research (40).