Journal of Psychiatric Research 151 (2022) 86–94

Contents lists available at ScienceDirect

Journal of Psychiatric Research

journal homepage: www.elsevier.com/locate/jpsychires

Psychological online interventions for problem gambling and gambling

disorder – A meta-analytic approach
Christoph Augner a, b, *, Thomas Vlasak c, Wolfgang Aichhorn a, Alfred Barth c
a
Department of Psychiatry, Psychotherapy & Psychosomatics, Christian-Doppler Medical Centre, University Clinics of the Paracelsus Medical University, Salzburg,
Austria
b
Institute for Human Resources Research in Health Care, University Clinics of the Paracelsus Medical University, Salzburg, Austria
c
Institute for Psychology, Sigmund Freud University Linz, Austria

A R T I C L E I N F O A B S T R A C T

Keywords: Objective: The Covid-19 pandemic has reignited discussions about the prevalence of and treatment options for
Psychotherapy problem gambling and gambling disorder (PGGD). Since affected persons seldom seek professional help, online
Online intervention interventions can improve accessibility. Thus, this study aimed to investigate the effectiveness of psychological
Gambling addiction
online interventions on PGGD.
Internet
Methods: We conducted a systematic review and meta-analysis and searched in PubMed, PsycINFO, and Google
Scholar for peer-reviewed experimental and quasi-experimental research published between 2010 and 2021. We
calculated two meta-anlyses, one for treatment control comparisons (TCC), and one for pre-post-comparisons
(PCC).
Results: We included six studies (ten TCC and n = 2076) in meta-analysis 1 and five studies (six PCC and n = 781)
in meta-analysis 2. Online interventions turned out to be effective in both analyses with Hedges g = 0.41, 95%
confidence interval = [0.22 to 0.60], p < .001, for meta-anaylsis 1 and Hegdes g = 1.28, 95% confidence in­
terval = [0.85 to 1.71], p < .001, for meta-analysis 2.
Conclusions: We identified significant effects of online interventions on PGGD in both analyses, indicating the
potential of online applications. We discuss methodological aspects and further research directions.

1. Introduction Rumpf and Brandt, 2020). While gambling disorder has a clear defini­
tion, the term problem gambling is more diffuse and frequently defines
Problem gambling and gambling disorder (PGGD) are a major public conditions below the clinical threshold (Bucker et al., 2018; Cowlishaw
health issue. PGGD is associated with a preoccupation with gambling, et al., 2014).
the inability to control or reduce gambling, symptoms of restlessness and PGGD is highly comorbid with other mental health problems. First,
a dysphoric mood when trying to stop or cut down gambling, increasing the most established is the relationship to substance use disorders. A
financial losses and severe mental and interpersonal consequences recent study reported that 37% of substance use patients demonstrate
(Hodgins et al., 2011) (Livazovic and Bojcic, 2019). As recent research comorbid PGGD (Cowlishaw et al., 2014). Gambling disorder patients
indicates, these consequences include an approximately four times are reported to have an increased risk of 4.4 for substance abuse disorder
higher risk for suicide in PGGD than in the rest of the population (Wardle (Hodgins et al., 2011). Second, PGGD is strongly associated with
et al., 2020), and serious financial costs on a societal level that clearly depressive symptoms and depression. The risk of depression is three
exceed the tax revenues from gambling (Hofmarcher et al., 2020). times higher in patients with gambling disorder than in non-gamblers
Past year problem gambling rates ranges between 0.12 and 5.8% (Hodgins et al., 2011). Thirty percent of patients with gambling disor­
across different regions in the world (Calado and Griffiths, 2016). der have a comorbid major depression, thus psychological interventions
Recently, disordered gambling has been re-classified into addictive for depression may be effective in PGGD as well (Bucker et al., 2018).
disorders/addictive behaviors in DSM-5 and ICD-11 (SAMHSA, 2016; The SARS-CoV-2 (COVID-19) pandemic has intensified the

* Corresponding author. Department of Psychiatry, Psychotherapy & Psychosomatics, Christian-Doppler Medical Centre, University Clinics of the Paracelsus
Medical University, Salzburg, Austria.
E-mail address: [email protected] (C. Augner).

https://doi.org/10.1016/j.jpsychires.2022.04.006
Received 3 November 2021; Received in revised form 24 March 2022; Accepted 6 April 2022
Available online 9 April 2022
0022-3956/© 2022 Elsevier Ltd. All rights reserved.
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

discussion about PGGD and raised concerns about an intensification of (Moher et al., 2009) (Walther et al., 2011) we included a working title,
gambling problems during the health crisis. Closures of land-based review objective, search strategies, study selection criteria, in­
gambling in many countries during lockdown periods seem to have terventions and end points.
reduced overall gambling frequency. However, a vulnerable group have For the definition of psychological online interventions we used a
intensified their gambling via online gambling providers (Hodgins and recent conceptualization of van der Maas et al., “any prevention or
Stevens, 2021). This vulnerable group is characterized predominantly treatment program designed to reduce the harm of problem gambling that
by serious pre-existing gambling problems. In a Swedish sample, makes use of internet resources to deliver content or resources” (van der
increased gambling during the pandemic was associated with more se­ Maas et al., 2019), p.3.
vere gambling problems, increased alcohol consumption and psycho­ PGGD is the dependent variable in this meta-analysis and is
logical distress (Hakansson and Widinghoff, 2021). Psychological measured by standardized questionnaires representing a continuum of
interventions, especially those based on cognitive based psychotherapy problematic/pathological gambling.
(CBT), are effective in treating gambling disorder (Abbott, 2019; Cow­
lishaw et al., 2012; Pfund et al., 2020). However, serious concerns 2.2. Search strategy
remain. While participation and help-seeking rates are low (estimates
range from 6% to 30% of people with PGGD), reported dropout rates in This meta-analysis includes studies from a systematic literature
face-to-face interventions are around 40% (Casey et al., 2017; Gomes search in PubMed, PsycINFO, and Google Scholar. The search was
and Pascual-Leone, 2015; Rodda et al., 2017; van der Maas et al., 2019). limited to the time span between 2010 and 2021. We used the following
Internet-based interventions have revealed an effect size comparable search terms in PubMed and PsycINFO:
to face-to-face interventions (Barak et al., 2008). Several authors insist (gambl*) AND ((online therapy) OR (online intervention) OR
that online interventions have major advantages regarding access (internet-based therapy) OR (internet-based intervention) OR (com­
compared to face-to-face therapies. In particular, target populations that puter-assisted therapy) OR (computer-assisted intervention) OR
are not willing to engage in classical forms of psychotherapy could be (smartphone therapy) OR (smartphone intervention)). For Google
reached (Chebli et al., 2016; Giroux et al., 2017; van der Maas et al., Scholar we used *gambling online intervention* as a search term.
2019). Furthermore, we used a backward snowballing literature search from
Furthermore, recent research indicate the increasing relevance of the reference lists of existing reviews (Chebli et al., 2016; Giroux et al.,
online gambling for pathological gambling. Very involved players are at 2017; Humphreys et al., 2021; van der Maas et al., 2019) (Fig. 1).
high risk of Gambling disorder symptoms and associated addiction and
mental health problems. They often use all gambling channels available, 2.3. Inclusion criteria
offline and online (Marmet et al., 2021). Thus, internet based in­
terventions could pick up this group, where they gamble, i.e. online. We included studies in this meta-analysis that addressed the effec­
Internet-based interventions can be divided into guided and un­ tiveness of any prevention program and treatment/intervention/ther­
guided interventions. Guided therapy may consist of online textbooks, apy transmitted via the internet (including apps, e-mails, websites, chats
personalized e-mails and messages, video conferencing or integrated etc.) on PGGD. Thus, we included studies that:
telephone contact. A therapist guides the therapeutic process. Unguided
online interventions are those without therapist contact and guidance. (1) Evaluated the effect of a psychological prevention program/
Online therapy programs consist of computer-generated e-mails or intervention and/or treatment and/or psychotherapy on PGGD
messages (Karyotaki et al., 2021; McDonald et al., 2020). (2) Used standardized questionnaires to assess PGGD as the depen­
A recent meta-analysis indicated that online therapies can be effec­ dent variable
tive in the treatment of depression (Karyotaki et al., 2021). Unfortu­ (3) Used a (quasi-) experimental design that included an interven­
nately, evidence on psychological online interventions in relation to tion/control group and pre-post-designs without a control group
PGGD is scarce. A recent review focused on online and mobile in­ (4) Were published or online first in peer review journals
terventions for problem gambling, alcohol, and drugs. However, the (5) Reported relevant data in English.
authors reported that no study for problem gambling was retained for
the review due to methodological reasons (Giroux et al., 2017). Another We excluded studies that:
narrative review identified 27 articles evaluating internet-based in­
terventions for problem gambling, many of them using text-based in­ (1) Did not involve analysis of original research (reviews, study
teractions with a therapist, automated personalized and normative protocols)
feedback on gambling behavior, and interactive CBT. The authors (2) Tested feedback applications for gamblers or focused on face-to-
conclude that internet-based interventions were effective in reducing face interventions for online gamblers or internet therapies for
problem gambling scores and gambling behaviors. Unfortunately, no significant others of gamblers
meta-analysis was calculated to quantify these statements (van der Maas (3) Presented qualitative data only
et al., 2019). Further research included four studies on the effectiveness (4) Did not report sufficient statistical data to calculate effect sizes
of internet-based interventions on gambling and reported positive ef­ (5) Used assessment only as an intervention
fects of the interventions on gambling and associated negative side ef­
fects. Nevertheless, the authors did not calculate a meta-analysis to 2.4. Study selection
quantify these effects, probably due to methodological reasons and the
limited sample size (Chebli et al., 2016). Two authors conducted the study selection (CA, TV). We identified
Thus, the objective of this meta-analysis was to assess the effective­ relevant research via title and abstracts, and removed duplicates form
ness of psychological online interventions on PGGD. the search results. We imported relevant full texts into Endnote X9. Two
authors independently analyzed full texts for eligibility. Consensus was
2. Methods reached via discussion or – if not possible – by a senior author’s decision.

2.1. Definitions 2.5. Data extraction and quality assessment

For our review protocol in accordance with the Preferred Reporting The screening process and data extraction were carried out via a
Items for Systematic Reviews and Meta-Analysis method (PRISMA) coding scheme which we developed. Based on mean rating and a two-

87
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

Fig. 1. Study selection process (Adapted from Moher et al. (2009)).

way mixed-effects model (k = 3, absolute agreement) we achieved an recommended by Cujipers et al. (Cuijpers et al., 2017). The respective
inter-rater reliability of 0.85 while differences were solved by consensus. meta-analytic results are based on the post-intervention mean
The variables of data extraction consisted of the authors of the publi­ between-group differences between the exposure and control groups,
cation, year of publication, effect sizes as well as the characteristics of since they control for biased outcomes (Cuijpers et al., 2017). As sug­
the samples. If studies reported results for subgroups (e.g., gender) gested and shown in existing meta-analyses, the synthesis of pre-post
rather than for the total sample, groups were entered as independent measurement data adds a more in-depth interpretation of results when
samples. All three authors evaluated the quality of the treatment control comparing the effects of routine practice with randomized control trials
comparisons (TCC) using the Cochrane risk of bias tool version 2.00 for (Cuijpers et al., 2017; van der Lem et al., 2012). Therefore, we carried
randomized trials (Sterne et al., 2019). This assessment tool consists of out an additional meta-analysis focusing on the effect of routine practice
five domains of bias (e.g., randomization process, results, etc.) and by exclusively including studies with a pre-post design. These pooled
provides an overall score of either low, some, or a high risk of bias. An results are based on the mean difference between pre- and post mea­
inter-rater reliability of 0.88 was estimated for study quality. In contrast, surements of the participants. Hedge’s g was selected as the effect size
for the study quality for pre-post comparisons (PPC) we used the Na­ for the meta-analyses, since it leads to a more precise estimation of true
tional Institutes of Health (NIH) quality assessment tool for before-after effect sizes due to less bias compared to Cohen’s d (Hedges, 1981),
(pre-post) study with no control group. The assessment consists of 12 particularly in small sample sizes. Regarding the direction of effect,
major components (e.g. study question, follow-up baseline, etc.) and positive effect sizes indicate higher scores in the
provides an overall score of either good, fair or low quality. An exposure/post-intervention group, while negative effect sizes reflect
inter-rater reliability of .86 was achieved. higher values in the control/pre-intervention group for PGGD.
Effect sizes and appropriate 95% confidence intervals for PGGD were
pooled by using random effects models with a restricted likelihood
2.6. Statistical analysis
estimator and Hedge’s invariance weighting. The homogeneity of effect
sizes was addressed by Q statistic regarding significance levels and τ2 for
In general, we carried out two separate meta-analyses regarding the
overall quantification (Cochran, 1954). The influence of publication bias
different study designs examining the outcome variable. The main meta-
was examined by funnel plots and Egger’s Test (Egger et al., 1997).
analysis focuses on the estimation of the effects of the therapeutic in­
Sensitivity analysis also included addressing potential p-hacking via
terventions, including randomized control trials exclusively as

88
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

p-curve diagrams and analyses (Simonsohn et al., 2014). Additionally, one study with two conditions (19%) used cognitive bias modification
not only subgroup analyses but also weighted mixed-effects meta-re­ (Wittekind et al., 2019). Three conditions in our PPC analysis were
gressions were carried out to examine further moderating effects. Before guided by a therapist (43%) (Carlbring et al., 2012) (Erevik et al., 2020;
entering the analyses, we applied Cook’s distance to identify outliers in Myrseth et al., 2013). All other treatment conditions were unguided
the effect sizes (Viechtbauer and Cheung, 2010). All statistical analyses (57%).
were carried out with the “metafor” package via the software R In PPC studies, three different instruments to measure PGGD
(Viechtbauer, 2010). outcome were used. Two effect sizes (29%) used South Oaks Gambling
Screen (SOGS) (Erevik et al., 2020; Myrseth et al., 2013), 3 effects sizes
3. Results (43%) were measured by NORC DSM-IV Screen for Gambling Problems
(Carlbring et al., 2012; Hodgins et al., 2019), and 2 effect sizes (29%)
3.1. Study sample and characteristics were measured by Yale Brown Obsessive Compulsive Scale adapted for
Pathological Gambling PG-YBOCS (Wittekind et al., 2019) (see Table 2).
3.1.1. Treatment-control comparison (TCC)
In total, we included six TCC studies in meta-analysis 1, achieving 3.2. Meta-analytic results
ten effect sizes and a sample of n = 2051 participants consisting of n =
1034 in the intervention and n = 1017 in the control group (Table 1). 3.2.1. Treatment-control comparison (TCC) – Meta-analysis 1
The publication date ranged from 2016 to 2020 while a mean age of Test for homogeneity indicated heterogeneous effect sizes of the
32.74 (SD = 8.16) years was given. The median duration of therapeutic included TCC studies (Q(9) = 22.61, p = .007) with τ2 = 0.06. The
intervention was 350 min; however, this value is based on only three random-effect model meta-analysis across six studies including ten ef­
studies, since specific information was absent in the remaining majority. fect sizes showed significant differences between exposure and control
Overall, the risk of bias for the included studies was considered low to group with g = 0.41 (95% confidence interval from 0.22 to 0.60, p <
some levels of concern due to some missing information about the .001). Therefore, individuals participating in psychotherapeutic treat­
characteristics of the intervention process and the total sample. All ment show significantly lower scores in gambling than their controls.
intervention groups for TCC were CBT-based (100%) and included
behavioral and cognitive strategies aiming to reduce or eliminate 3.2.2. Pre-post comparison (PPC) – Meta-analysis 2
gambling. Most online programs in the TCC analysis did not include The heterogeneity of effect sizes indicated for the included PPC
guidance by a therapist (n = 7, 70%); only three intervention groups, studies by the test for homogeneity (Q(6) = 115.45, p < .001) with τ2 =
one in Luquiens et al. (2016) and both groups in Jonas et al. (2020) were 0.31. The random-effect model with five studies across seven effect sizes
interventions with therapeutic guidance (n = 3, 30%). showed significant differences between the pre- and post measurement
TCC Studies indicated high homogeneity regarding outcome in­ with g = 1.28 (95% confidence interval from 0.85 to 1.71, p < .001).
struments. In five effect sizes (50%) a version of the South Oaks Therefore, individuals showed significantly lower scores in gambling
Gambling Screen (SOGS) was used (Bucker et al., 2018; Canale et al., after the treatment than before.
2016; Casey et al., 2017), in five effect sizes (50%) the Problem
Gambling Severity Index (PGSI) (Jonas et al., 2020; Luquiens et al., 3.3. Sensitivity analysis
2016; So et al., 2020) (see Table 1).
3.3.1. Outliers
3.1.2. Pre-post comparison (PPC) One effect size from the treatment-control comparison (TCC) studies
In meta-analysis 2, five PPC studies were included accompanied by was detected as an outlier (Casey et al., 2017). Although removal of the
seven effect sizes, achieving a total sample of n = 781 participants outlier resulted in a reduced pooled effect size from g = 0.41 to 0.33
(Table 2). Date of publication ranged from 2012 to 2020 and the mean (95% confidence interval from 0.20 to 0.45), the random effects model
age of participants was 39.15 (SD = 6.10) years. The median duration of remained significant (p < .001) and a drop in the homogeneity of the
therapeutic interventions was 500 min; however, this estimation is effect sizes (Q(8) = 7.76, p = .46) τ2 from 0.06 to 0.00 was achieved. The
based only on three studies, while in most cases there was no informa­ reason for detection as an outlier might be due to the magnitude of the
tion. In general, the study quality was considered fair, while in most effect size, which is almost twice as large as the highest remaining one.
studies information about the intervention process, sample character­ Therefore, the outlier (Casey et al., 2017) may drive the heterogeneity of
istics and statistical analyses was lacking. effect sizes of TCC studies.
PPC analysis included seven intervention conditions. Four of them Regarding the pre-post comparison (PPC) studies, one effect size was
had an identifiable CBT basis (57%), Hodgins et al. (2019) used one also detected as an outlier (Carlbring et al., 2012). Exclusion of the
“light” condition that focused on normative feedback on gambling outlier from the analysis resulted in a slightly reduced pooled effect size
behavior and provided brief information on how to change (14%), and from g = 1.28 to 1.09 (95% confidence interval from 0.83 to 1.36, p <

Table 1
Study characteristics meta-analysis (1) Treatment control comparisons (TCC).
Author Intervention Group Outcome g SE n Mean Age SD Quality
Age

Bucker et al. (2018) Deprexis SOGS 0.30 0.17 140 35.71 10.21 1
Canale et al. (2016) NFG SOGS-RA 0.35 0.19 114 15.01 0.60 2
Canale et al. (2016) FG SOGS-RA 0.30 0.19 54 2
Casey et al. (2017) CBT SOGS 1.14 0.20 115 44.45 9.75 1
Casey et al. (2017) MFS SOGS 0.51 0.19 114 44.50 9.27 1
So et al. (2020) Gambot PGSI 0.14 0.14 197 36.35 9.80 2
Luquiens et al. (2016) CBT Self Help PGSI 0.11 0.20 528 34.70 10.10 2
Luquiens et al. (2016) CBT Weekly Email PGSI 0.10 0.29 565 2
Jonas et al. (2020) Check Out PGSI 0.61 0.19 111 33.50 10.60 2
Jonas et al. (2020) Email Counseling PGSI 0.53 0.19 113 2

Notes. NFG – non-frequent gamblers; FG – frequent gamblers; MFS – monitoring, feedback, support; PGSI – Problem Gambling Severity Index; SOGS(-RA) - South Oaks
Gambling Screen(-Revised for Adolescents); Quality rating: 1 = low risk of bias, 2 = some risk of bias.

89
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

Table 2
Study characteristics meta-analysis (2) Pre-post comparisons (PPC).
Author Intervention Group Outcome g SE n Mean Age SD Quality
Age

Carlbring et al. (2012) ICBT NODS 2.37 0.11 284 32.2 8.8 7
Erevik et al. (2020) ICBT SOGS-R 0.91 0.18 67 42.3 11.6 7
Hodgins et al. (2019) CYG NODS 1.07 0.16 93 46.8 11.8 6
Hodgins et al. (2019) SCT NODS 0.81 0.15 94 46.7 12.2 6
Myrseth et al. (2013) ICBT SOGS-R 1.71 0.16 112 35.7 10.24 7
Wittekind et al. (2019) ABM PG-YBOCS 1.00 0.18 66 36.62 10.32 6
Wittekind et al. (2019) ABM Sham PG-YBOCS 1.03 0.19 65 33.72 11.53 6

Notes. CYG - Check Your Gambling; SCT - Self Change Tools; ABM - Approach Bias Modification; ABM Sham - ABM Sham Intervention; NODS - NORC DSM-IV Screen for
gambling problems; SOGS-R – South Oaks Gambling Screen – Revised; PG-YBOCS - Yale Brown Obsessive Compulsive Scale adapted for Pathological Gambling;
Quality rating: 6.7 = fair.

.001), while the heterogeneity of effect sizes still remained (Q(5) = Fig. 2) nor by Egger’s test for funnel plot asymmetry (p = .54). The
20.11, p = .001) with τ2 = 0.08. Therefore, the outlier may only partially results of the p-curve analysis (see Fig. 3) showed a significant test for
drive the heterogeneity of effect sizes. right-skewness (p < .001) and a non-significant test for flatness (p =
.99). Our results regarding the meta-analytical effect size of TCC studies
3.3.2. Subgroup and moderator analysis are robust against publication bias and p-hacking, showing the true ef­
No significant subgroup differences in treatment-control comparison fect of therapeutic intervention.
(TCC) studies could be found regarding the risk of bias (Q(1) = 1.64, p = For pre-post comparisons (PPC) studies, publication bias was not
.20). Since therapeutic guidance was applied in only two conditions only visually indicated by the funnel plot (see Fig. 4) but also by Egger’s
(Jonas et al., 2020; Luquiens et al., 2016) of the TCC studies, no test (p = .009). When testing for p-hacking by conducting p-curve
meaningful subgroup analysis could be carried out with regard to dif­ analysis (see Fig. 5), effect sizes showed significant right-skewness (p <
ferences between guided and non-guided conditions. Neither the year of .001) and non-significant flatness (p = .99). Therefore, the results for
publication (Q(1) = 0.02, p = .90; r2 = 0.00, b = 0.00) nor the mean age PPC studies regarding the overall effect size of gambling do not seem to
(Q(1) = 1.18, p = .28; r2 = 0.01, b = 0.01) significantly moderated the be influenced by p-hacking; however, they might be affected by publi­
effect sizes. cation bias.
For pre-post comparison (PPC) studies no significant subgroup dif­
ferences were found between guided and non-guided conditions (Q(1) 4. Discussion
= 0.01, p = .92). Due to limited variation in study quality, no mean­
ingful analysis could be carried out regarding the included PPC studies. The objective of this meta-analysis was to assess the effectiveness of
Additionally, moderation analysis showed significant results for year of psychological online interventions on problem gambling and gambling
publication (Q(1) = 48.03, p < .001; r2 = 0.94, b = − 0.17) but not for disorder (PGGD). In total, we included eleven studies comprised of 17
mean age (Q(1) = 3.79, p = .06; r2 = 0.35, b = − 0.06). When both effect sizes in both meta-analyses, i.e. six studies including ten
variables entered the regression simultaneously, year of publication treatment-control comparisons (TCC) and five studies comprised of
remained significant in the model (X2(2) = 43.62, p < .001; r2 = 0.93, seven pre-post comparisons (PPC).
bpublicationyear = − 0.15, bmeanage = − 0.01). Therefore, earlier studies tend Our results indicate that psychological online interventions have
to show higher effect sizes than current studies, which may explain a moderate effects on PGGD in the short term. The random effects model
certain degree of heterogeneity of effect sizes. showed a significant positive pooled effect size of g = 0.41 for TCC, and
g = 1.28 for PPC.
3.3.3. Publication bias and p-hacking Compared to the results of a recent review of the efficacy of tradi­
With respect to the treatment-control comparison (TCC) studies, tional face-to-face therapies on gambling problems including 14 studies
publication bias was neither visually indicated by the funnel plot (see and revealing a Hedges g = 0.65, our study revealed lower effect sizes of

Fig. 2. Funnel plot TCC.

90
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

Fig. 3. P-curve TCC.

Fig. 4. Funnel plot PPC.

the online interventions included in our TCC meta-analysis, but higher fields of online intervention.
ones in our PPC meta-analysis (Pfund et al., 2020). In general, com­ We were not able to calculate a reasonable analysis of the effec­
parisons of psychological online interventions and face-to-face therapies tiveness of guided vs. unguided forms of online interventions, since only a
are scarce in all fields of psychiatry. However, Barak et al.’s minority of studies included in our review offered guided interventions.
meta-analysis indicated the non-inferiority of online interventions Nevertheless, there was no indication that guidance vs. no guidance
(0.39) vs. face-to-face-therapies across different mental illnesses (0.34, affects the outcome. There is a clear trend towards unguided and even
p > .05) (Barak et al., 2008). autonomous interventions which have the potential to complement or
Comparisons of our results with the effectiveness of psychological even substitute a therapist in person (McDonald et al., 2020). Naturally,
online interventions in other fields of psychiatry show heterogeneous re­ this has led to recent controversies about the role of the contact with
sults. A recent review revealed a weighted effect size of 0.53 for psy­ psychotherapists (Chebli et al., 2016). A recent review of the effective­
chological online interventions across different psychiatric disorders, ness of internet-based cognitive behavioral therapy on depression
the strongest effects for PTDS (0.88), panic and anxiety (0.80), and showed that in individuals guided vs. unguided interventions were
lower ones for stopping smoking (0.62), drinking (0.48) and for almost equally effective for low or moderate depression scores. For pa­
depression (0.32) (Barak et al., 2008). A more recent study on psycho­ tients with a higher depression score guided therapy forms were
logical online interventions reported typical effect sizes of 0.5–0.8 for considerably more effective (Karyotaki et al., 2021).
depression and anxiety in pre/post designs, with effect sizes decreasing Fourteen treatment conditions (82%) in our meta-analyses used CBT-
to 0.4 when a control group is included (McDonald et al., 2020). Thus, based interventions. Face-to-face CBT interventions for PGGD are widely
effect sizes in our meta-analyses are comparable with other application used in clinical practice (Smith et al., 2018). Recent reviews indicate the

91
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

Fig. 5. P-curve PPC.

superiority of online CBT formats compared to interventions with other treatment. They achieve this by means of practical and behavior-focused
theoretical backgrounds (Barak et al., 2008). For PGGD a recent strategies, e.g. avoidance of cues, and by taking part in activities
meta-analysis found higher but not significant effects sizes of incompatible with gambling (Hodgins et al., 2011). Effective online
face-to-face CBT in relation to other therapy modalities (Pfund et al., interventions can support and trigger these strategies in people with
2020). This has two implications: first, CBT works in the real world, so it PGGD and improve accessibility to interventions. Future online in­
works online too. Second, there a clear need for more studies addressing terventions will probably be autonomous and personalized. The patient
the efficacy of alternative therapeutic strategies. However, the theoret­ will be able to move in a virtual space which collects behavioral data
ical implications of neurophysiological evidence make it clearer why from them in order to use it in therapeutic interactions (McDonald et al.,
CBT-based interventions work in PGGD. The initiation and progress of 2020). Neuroimaging could provide direct und valid feedback about the
PGGD can be characterized by a monopolization of gambling-related process and further improve the effectiveness of the gambling-related
rewarding cues. The uncertainty of the outcome of gambling (not the intervention. Moreover, a hybrid of different promising applications,
win) is able to activate dopamine release and reward-associated brain e.g. non-invasive brain stimulation (Pettorruso et al., 2021), could
activity. Thus, PGGD is essentially a learning process (Clark et al., 2019). accompany future online psychotherapy. Recent studies show that vir­
CBT-based psychological interventions can directly tackle and change tual environments can help in the therapeutic process. Detecting
this process. high-risk situations, identifying cues that make relapses probable,
Recent research has dealt with the role of different sociodemographic simulating craving situation to develop personalized behavioral strate­
factors associated with gambling problems. Most prominently, young gies are important starting point for virtual reality applications (Bou­
sports gamblers tend to have high rates of addiction comorbidity, while chard et al., 2017; Mazza et al., 2021).
older women have higher rates of psychiatric comorbidities (Hodgins However, future research will additionally have to clarify fields of
et al., 2011). Thus, it would be plausible that age would be a significant application and the role of online interventions. This also includes the
factor in the efficacy of psychotherapy. Recent research indicates that role of the psychotherapist in these future models and the steps and
prevention and therapeutic interventions at an early age are funda­ situations in the therapeutic process that make direct interaction
mental, since gambling onset age is a predictor for future gambling necessary and that are not interchangeable with online formats.
problems. Furthermore, compared to middle-aged patients, young and
old gamblers have more damaging irrational beliefs that correlate with 4.1. Limitations
gambling severity (Granero et al., 2020). However, for TCC and PPC
moderator analysis, we found no significant effect of mean age on In general, our meta-analytic results showed significant effects of
treatment efficacy. Future studies should address age-related aspects of online therapy interventions in TCC (meta-analysis 1) and PPC (meta-
online interventions in order to clarify their relevance for the develop­ analysis 2) studies by reducing the problematic gambling behavior of
ment and application of psychotherapeutic approaches. their clients. However, some limitations need to be taken into account
TCC and PPC studies together only used four different standardized when interpreting our findings. First, the notable difference between the
questionnaires for PGGD. Thus, analysis consist of comparable and valid pooled effect sizes of the primary and secondary meta-analysis has to be
parameters. discussed. While TCC studies resulted in g = 0.41 and PPC in g = 1.28,
How will this issue proceed? Evidence suggests that the proportion of the discrepancy can be primarily explained by contrasting study designs.
people with PGGD who seek professional treatment is below 10% As stated in the literature, TCC studies are superior in estimating the true
(Hodgins et al., 2011). Online therapy has the potential to facilitate the effect of therapeutic interventions in comparison to PPC studies, due, for
accessibility and increase the adherence of psychological interventions example, to the randomization process, the control group, c for
at affordable costs, as outlined recently (McDonald et al., 2020). How­ cofounding variables, etc. (Cujipers et al., 2017). Therefore, when dis­
ever, although only a small proportion of individuals experiencing cussing our results regarding the true effect of internet-based in­
PGGD seek professional help, a third seem to recover even without terventions on problematic gambling, the pooled effect size of our

92
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

primary meta-analysis including TCC studies should be the point of Declaration of competing interest
focus.
Second, despite pooled estimations based on PPC studies tending to The authors declare no potential conflicts of interest.
be more biased, e.g. overestimating the true effect, vital information
about the therapeutic intervention is provided when it is synthesized References
(Cujipers et al., 2017; Kösters, 2017). In contrast to TCC studies, PPC
study designs are characterized by a high standard of external validity, Abbott, M.W., 2019. Professionally delivered interventions for gambling disorder. Curr.
Opin. Psychiatr. 32, 313–319.
reflecting effects in vivo rather than in vitro. Therefore, our meta-analytic Barak, A., Hen, L., Boniel-Nissim, M., et al., 2008. A comprehensive review and a meta-
results of the PPC studies may overestimate the experimental thera­ analysis of the effectiveness of internet-based psychotherapeutic interventions.
peutic effect size shown by the TCC studies in vitro, but confirm the J. Technol. Hum. Serv. 26, 109–160.
Bouchard, S., Robillard, G., Giroux, I., et al., 2017. Using virtual reality in the treatment
identical direction of effect in vivo. of gambling disorder: the development of a new tool for cognitive behavior therapy.
Third, the heterogeneity of effect sizes needs to be addressed for both Front. Psychiatr. 8, 27.
meta-analyses. While the outlier for TCC studies primarily drives inho­ Bucker, L., Bierbrodt, J., Hand, I., et al., 2018. Effects of a depression-focused internet
intervention in slot machine gamblers: a randomized controlled trial. PLoS One 13
mogenous results, sensitivity analysis revealed multiple sources for PPC (2018), e0198859, 13.
studies. The results of the secondary analysis seem to be affected not Calado, F., Griffiths, M.D., 2016. Problem gambling worldwide: an update and
only by outliers but also by publication bias that ultimately leads to systematic review of empirical research (2000-2015). J. Behav. Addictions. 5,
592–613.
possible overestimation of effects. Additionally, year of publication was
Canale, N., Vienoa, A., Griffiths, M.D., et al., 2016. The efficacy of a web-based gambling
found to be a significant moderator of effect size, showing that earlier intervention program for high school students: a preliminary randomized study.
studies found higher effect sizes. However, this relationship may be the Comput. Hum. Behav. 55, 946–954.
result of limited statistical data and needs to be addressed by further Carlbring, P., Degerman, N., Jonsson, J., et al., 2012. Internet-based treatment of
pathological gambling with a three-year follow-up. Cognit. Behav. Ther. 41,
research. 321–334.
Furthermore, detailed information regarding type of games and sex Casey, L.M., Oei, T.P.S., Raylus, N., et al., 2017. Internet-based delivery of cognitive
were restricted which may impact the meta-analytical results. There­ behaviour therapy compared to monitoring, feedback and support for problem
gambling: a randomised controlled trial. J. Gambl. Stud. 33, 993–1010.
fore, we recommend replication of analyses when a more detailed body Chebli, J.-L., Blaszczynski, A., Gainsbury, S.M., 2016. Internet-based interventions for
of literature is given to estimate specific pooled effect sizes. Conse­ addictive behaviours: a systematic review. J. Gambl. Stud. 32, 1279–1304.
quently, potential differences of online interventions concerning type of Clark, L., Boileau, I., Zack, M., 2019. Neuroimaging of reward mechanisms in Gambling
disorder: an integrative review. Mol. Psychiatr. 24, 674–693.
games and sex are better distinguished leading to a deeper under­ Cochran, W.G., 1954. The combination of estimates from different experiments.
standing for therapeutic application. Biometrics 101–129.
Finally, only a marginal proportion of the included studies reported Cowlishaw, S., Merkouris, S., Chapman, A., et al., 2014. Pathological and problem
gambling in substance use treatment: a systematic review and meta-analysis.
multiple repeated measurements. Therefore, our meta-analytic results J. Subst. Abuse Treat. 46, 98–105.
focus on significant short-term effects of online therapeutic in­ Cowlishaw, S., Merkouris, S., Dowling, N., et al., 2012. Psychological therapies for
terventions on problematic gambling. Due to the restricted periods, pathological and problem gambling. Cochrane Database Syst. Rev. 11, CD008937.
Cuijpers, P., Weitz, E., Cristea, I.A., et al., 2017. Pre-post effect sizes should be avoided in
future studies should focus on multiple follow-up measurements to
meta-analyses. Epidemiol. Psychiatr. Sci. 26, 364–368.
assess specific long-term effects of internet-based interventions for Egger, M., Smith, G.D., Schneider, M., et al., 1997. Bias in meta-analysis detected by a
PGGD. simple, graphical test. BMJ Br. Med. J. 315, 629–634.
In our study, we focused exclusively on PGGD as dependent variable Erevik, E.K., Pallesen, S., Mohn, M., et al., 2020. The Norwegian remote intervention
programme for problem gambling: short- and long-term outcomes. Nord. Stud.
measured via standardized and comparable instruments in order to (1) Alcohol Drugs 37, 365–383.
address the most important outcome of online intervention on gambling Giroux, I., Goulet, A., Mercier, J., et al., 2017. Online and mobile interventions for
and (2) to make sure heterogeneity of outcome measures was as little as problem gambling, alcohol, and drugs: a systematic review. Front. Psychol. 8, 954.
Gomes, K., Pascual-Leone, A., 2015. A resource model of change: client factors that
possible. However, this comes at the price that our analysis has a narrow influence problem gambling treatment outcomes. J. Gambl. Stud. 31, 1651–1669.
focus. We did not assess equally important outcome measures beyond Granero, R., Fernandez-Aranda, F., Valero-Solis, S., et al., 2020. The influence of
PGGD that e.g. address cognitive biases or social functioning. Thus, chronological age on cognitive biases and impulsivity levels in male patients with
gambling disorder. J. Behav. Addictions. 9, 383–400.
interpretation is therefore limited to gambling-related problems. Hakansson, A., Widinghoff, C., 2021. Changes of gambling patterns during COVID-19 in
Sweden, and potential for preventive policy changes. A second look nine months into
5. Conclusion the pandemic. Int. J. Environ. Res. Publ. Health 18.
Hedges, L., 1981. Distribution theory for Glass’s estimator of effect size and related
estimators. J. Educ. Stat. 6, 107–128.
Our study shows that online interventions in PGGD are effective in Hodgins, D.C., Cunningham, J.A., Murray, R., et al., 2019. Online self-directed
the short term. Future studies should address the role of demographics, interventions for gambling disorder: randomized controlled trial. J. Gambl. Stud. 35,
635–651.
access, duration, guided vs. non-guided interventions, and therapeutic
Hodgins, D.C., Stea, J.N., Grant, J.E., 2011. Gambling disorders. Lancet 378, 1874–1884.
approaches beyond CBT. Hodgins, D.C., Stevens, R.M.G., 2021. The impact of COVID-19 on gambling and
gambling disorder: emerging data. Curr. Opin. Psychiatr. 34, 332–343.
Declaration of conflicting interests Hofmarcher, T., Romild, U., Spangberg, J., et al., 2020. The societal costs of problem
gambling in Sweden. BMC Publ. Health 20.
Humphreys, G., Evans, R., Makin, H., et al., 2021. Identification of behavior change
The authors declare no potential conflicts of interest. techniques from successful web-based interventions targeting alcohol consumption,
binge eating, and gambling: systematic review. J. Med. Internet Res. 23, e22694.
Jonas, B., Leuschner, F., Eiling, A., et al., 2020. Web-based intervention and Email-
Funding counseling for problem gamblers: results of a randomized controlled trial. J. Gambl.
Stud. 36, 1341–1358.
The authors declare that they have not received any financial support Karyotaki, E., Efthimiou, O., Miguel, C., et al., 2021. Internet-based cognitive behavioral
therapy for depression A systematic review and individual patient data network
for the research, authorship and/or publication of this article. meta-analysis. JAMA Psychiatr. 78, 361–371.
Kösters, M., 2017. Every effect size has its place: a commentary on the avoidance of
Data availability pre–post effect sizes. Epidemiol. Psychiatr. Sci. 26 (4), 369–370. https://doi.org/
10.1017/S204579601700004X.
Livazovic, G., Bojcic, K., 2019. Problem gambling in adolescents: what are the
This study is a meta-analysis, and all data are available in the public psychological, social and financial consequences? BMC Psychiatr. 19.
domain. Luquiens, A., Tanguy, M.L., Lagadec, M., et al., 2016. The efficacy of three modalities of
internet-based psychotherapy for non-treatment-seeking online problem gamblers: a
randomized controlled trial. J. Med. Internet Res. 18.

93
C. Augner et al. Journal of Psychiatric Research 151 (2022) 86–94

Marmet, S., Studer, J., Wicki, M., et al., 2021. Online gambling’s associations with Smith, D.P., Fairweather-Schmidt, A.K., Harvey, P.W., et al., 2018. How does routinely
gambling disorder and related problems in a representative sample of young Swiss delivered cognitive-behavioural therapy for gambling disorder compare to "gold
men. Front. Psychiatr. 12, 703118. standard" clinical trial? Clin. Psychol. Psychother. 25, 302–310.
Mazza, M., Kammler-Sucker, K., Lemenager, T., et al., 2021. Virtual reality: a powerful So, R., Furukawa, T.A., Matsushita, S., et al., 2020. Unguided chatbot-delivered cognitive
technology to provide novel insight into treatment mechanisms of addiction. Transl. behavioural intervention for problem gamblers through messaging app: a
Psychiatry 11, 617. randomised controlled trial. J. Gambl. Stud. 36, 1391–1407.
McDonald, A., Eccles, J.A., Fallahkhair, S., et al., 2020. Online psychotherapy: Sterne, J.A.C., Savovic, J., Page, M.J., et al., 2019. RoB 2: a revised tool for assessing risk
trailblazing digital healthcare. BJPsych Bull. 44, 60–66. of bias in randomised trials. BMJ 366, l4898.
Moher, D., Liberati, A., Tetzlaff, J., et al., 2009. Preferred reporting items for systematic van der Lem, R., de Wever, W.W.H., van der Wee, N.J.A., et al., 2012. The
reviews and meta-analyses: the PRISMA statement. BMJ Br. Med. J. 339. generalizability of psychotherapy efficacy trials in major depressive disorder: an
Myrseth, H., Brunborg, G.S., Eidem, M., et al., 2013. Description and pre-post evaluation analysis of the influence of patient selection in efficacy trials on symptom outcome in
of a telephone and Internet based treatment programme for pathological gambling in daily practice. BMC Psychiatr. 12.
Norway: a pilot study. Int. Gambl. Stud. 13, 205–220. van der Maas, M., Shi, J., Elton-Marshall, T., et al., 2019. Internet-based interventions for
Pettorruso, M., Miuli, A., Di Natale, C., et al., 2021. Non-invasive brain stimulation problem gambling: scoping review. JMIR Ment Health 6, e65.
targets and approaches to modulate gambling-related decisions: a systematic review. Viechtbauer, W., 2010. Conducting meta-analyses in R with the metafor package. J. Stat.
Addict. Behav. 112. Software 36, 1–48.
Pfund, R.A., Peter, S.C., Whelan, J.P., et al., 2020. Is more better? A meta-analysis of dose Viechtbauer, W., Cheung, M.W., 2010. Outlier and influence diagnostics for meta-
and efficacy in face-to-face psychological treatments for problem and disordered analysis. Res. Synth. Methods 1, 112–125.
gambling. Psychol. Addict. Behav. 34, 557–568. Walther, S., Schuetz, G.M., Hamm, B., et al., 2011. Quality of reporting of systematic
Rodda, S.N., Hing, N., Hodgins, D.C., et al., 2017. Change strategies and associated reviews and meta-analyses: PRISMA (Preferred reporting items for systematic
implementation challenges: an analysis of online counselling sessions. J. Gambl. reviews and meta-analyses). Rofo-Fortschritte Auf Dem Gebiet Der Rontgenstrahlen
Stud. 33, 955–973. Und Der Bildgebenden Verfahren 183, 1106–1110.
Rumpf, H.-J., Brandt, D., 2020. Verhaltenssüchte in der ICD-11 [Behavioral Addictions in Wardle, H., John, A., Dymond, S., et al., 2020. Problem gambling and suicidality in
the ICD-11]. Suchttherapie 21, 140–143. England: secondary analysis of a representative cross-sectional survey. Publ. Health
SAMHSA, 2016. Impact of DSM-IV to DSM-5 Changes on the National Survey on Drug 184, 11–16.
Use and Health. Substance Abuse and Mental Health Services Administration (US), Wittekind, C.E., Bierbrodt, J., Ludecke, D., et al., 2019. Cognitive bias modification in
Rockville (MD). problem and pathological gambling using a web-based approach-avoidance task: a
Simonsohn, U., Nelson, L.D., Simmons, J.P., 2014. p-Curve and effect size: correcting for pilot trial. Psychiatr. Res. 272, 171–181.
publication bias using only significant results. Perspect. Psychol. Sci. 9, 666–681.

94