A systematic review of educational programs and consumer protection measures for gambling: an extension of previous reviews

Abstract

Introduction

Besides supply reduction, preventive interventions to reduce harm from gambling include interventions for the reduction of demand and to limit negative consequences. Several interventions are available for gamblers, e.g. limit-setting. Reviews have been published examining the evidence for specific measures as well as evaluating the effect of different measures at an overall level. Only a few of these have used a systematic approach for their literature review. The aim of this systematic review and meta-analysis is twofold. First, to assess the certainty of evidence of different preventive measures in the field of educational programs and consumer protection measures, including both land-based and online gambling. The second is to present shortcomings in eligible studies to highlight what type of information is needed in future studies.

Method

This systematic review included measures administered in both real-life settings and online. Twenty-eight studies fulfilled our inclusion criteria and had low or moderate risk of bias.

Results

The results showed that only two measures (long term educational programs and personalized feed-back) had an impact on gambling behavior. Follow-up period was short, and measures did not include gambling as a problem. The certainty in most outcomes, according to GRADE, was very low. Several shortcomings were found in the studies.

Discussion

We concluded that the support for preventive measures is low and that a consensus statement regarding execution and methods to collect and analyze data for preventive gambling research is needed. Our review can serve as a starting point for future responsible gambling reviews since it evaluated certainty of evidence.

Keywords:

• Prevention
• gambling
• systematic review
• intervention
• harm reduction

Introduction

Perspectives on harm and prevention in gambling

Gambling activities can be classified as nonproblematic, at-risk, or problematic with progressively increasing harm. Problem gambling is regarded as a public health issue (Abbott et al. 2018) and its harmful consequences have been researched extensively in clinical and epidemiological studies (Lorains et al. 2011; Dowling et al. 2015). Langham et al. (2015) conceptualized seven dimensions of harm emanating from gambling: financial harm, relationship disruptions, emotional or psychological distress, decrements to health, cultural harm, reduced performance at work or in academic study, and likelihood of criminal activity. Moreover, recent studies have found that even low-level gambling can result in some form of harm (Raisamo et al. 2015; Canale et al. 2016).

Preventive measures serve as one solution to decrease the likelihood of gambling-related harm. The fact that both high-risk and low-risk gambling can result in some form of harm substantiates the need for preventive measures. Therefore, there is a significant need to better understand methods which can prevent problematic gambling and how gambling-related harms can be limited. Understanding of preventive measures is becoming increasingly important due to increases in the incidence and scope of Internet gambling, which has resulted in gambling activities being permanently available in the home or on a smart phone (Abbott et al. 2018).

Currently, preventive interventions aimed at reducing the incidence of gambling, particularly problematic gambling, include demand reduction, harm reduction, and supply reduction. Demand reduction refers to interventions that reduce the demand for the availability of gambling activities through affecting changes in individuals’ knowledge concerning gambling, the harms related to gambling, personal motivation to gamble, or the social context in which gambling takes place. Examples of demand reduction include both universal and targeted programs directed at the youth and/or individuals with gambling addiction (Ladouceur et al. 2013; Keen et al. 2017).

Harm reduction refers to consumer protection measures, such as responsible gambling (RG). These measures are defined as policies and practices intended to reduce the potential harms resulting from gambling and are traditionally implemented by the gambling industry (Blaszczynski et al. 2004, 2008, 2011). Common features of RG measures include setting limits on the amount of time or money individuals can spend gambling and supplying personal feedback on gambling activities or self-exclusion (Ivanova et al. 2019). Williams et al. (2012, p. 6) stated the following regarding RG measures: ‘Unfortunately, the development, implementation, and evaluation of most of these initiatives have been a haphazard process. Most have been put in place because they ‘seemed like good ideas’ and/or were being used in other jurisdictions, rather than having demonstrated scientific efficacy or being derived from a good understanding of effective practices in prevention.’ This paradigm has been criticized for lacking consideration of the vulnerability of individuals who gamble and the need for gambling companies to take a more proactive stance to prevent problematic gambling (Hancock and Smith 2017a, 2017b).

Due to the criticisms of RG measures, researchers have called for a new consumer protection paradigm that prioritizes the duty of care of gambling operators, the integrity of operations, and consumer protection based on online and real-world preventive measures. Such preventive measures include setting limits on gambling activities or excluding individuals from gambling sites (Hancock and Smith 2017a, 2017b). Furthermore, some argue for preventive measures to take place in conjunction with a public health approach in terms of the regulation of gambling and the development of policies. In this review, we interpret RG from a consumer protection paradigm as a method of limiting harm resulting from gambling by assisting individuals engaging with gambling products to avoid potential gambling-related harms.

Another gambling-related harm preventive measure is supply reduction. This method has been evaluated in two recent systematic reviews (Meyer et al. 2018; McMahon et al. 2019). Both reviews state that there is a lack of research concerning supply reduction measures. Meyer et al. (2018) reported some inconsistences in the findings concerning supply reduction as a gambling-prevention measure. Nonetheless, the researchers stated that a reduction in the supply of gambling products/activities resulted in a decline in gambling participation, a reduced number of frequent gamblers, lower demand for therapy due to problematic gambling, and fewer problematic gambling cases. This claim is congruent with findings that suggest that supply reduction is the most effective preventative measure in reducing other problematic behaviors, such as alcohol consumption and tobacco use (Babor et al. 2010). Studies investigating the reduction of gambling supply are not included in our review since there is a need to limit the aim and scope of the study, and because existent research is very recent (Meyer et al. 2018; McMahon et al. 2019).

Reviews and meta-analysis focusing on preventive measures in gambling

An umbrella review focusing on preventive measures in gambling reported the publication of ten systematic reviews comprising 55 unique primary studies (McMahon et al. 2019). Most of the systematic reviews reported on studies of interventions aimed at harm reduction and focused specific preventive measures, such as limit-setting or interventions incorporated into the design of gambling activities (e.g. pop-messages). Support for most of the preventive measures considered was weak and the general conclusion drawn was that the studies concerning RG measures reported the absence of a significant effect in terms of decreasing gambling-related harm (McMahon et al. 2019). Furthermore, the reviews lacked scientific rigor. Only one out of the ten systematic reviews assessed the risk of bias in the primary studies examined and took bias into account when evaluating the effects of preventive measures on gambling. None of the studies used GRADE (Grading of Recommendations Assessment, Development and Evaluation) (Guyatt et al. 2011) or other methods of assessing the reliability and validity of the evidence. Another review indicated a need for further research in this area and increased methodological rigor in research through the use of specific guidelines when evaluating the effects of different preventive measures (Ladouceur et al. 2017). Two significant aspects emanating from the review are a lack of repeated measures methods and the use of different measures of gambling (Ladouceur et al. 2017). These two aspects may be the primary reasons for the lack of significant results.

Apart from the need for systematic reviews and meta-analyses that examine different types of preventive measures in gambling (e.g. educational programs and RG measures), there is also a need to address the shortcomings of existing research when discussing the different aspects of the study’s methods and how the results were reported. These shortcomings have resulted in the exclusion of studies in review papers and unclear results that are challenging to interpret.

Aim of this study

The aims of this systematic review and meta-analysis were two-fold. First, in the framework of a systematic review, we assessed the certainty of the evidence relating to different gambling preventive measures in the context of educational programs and consumer protection measures (e.g. responsible measures for both real-world and online gambling). Second, our goal was to present and discuss the shortcomings identified in eligible studies to better understand how preventive measures should be designed and tentatively identify the probable results of future studies.

Methods

This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Liberati et al. 2009).

Inclusion and exclusion criteria

The inclusion and exclusion criteria for the educational programs are presented in Table 1. The criteria for RG strategies are presented in Table 2.

Table 1. Inclusion and exclusion criteria for gambling prevention/educational programs.

Scope	To investigate the effectiveness of gambling prevention/educational programs
Participants	≥13 years of age. Studies with younger participants were included if ≥75% of the participants were 13 years or older, or if they reported results separately for relevant age. Studies where ≥75 % of the participants had severe gambling problems (e.g. Problem Gambling Severity Index, PGSI ≥8) were excluded.
Intervention	Education Excluded: information activities without a specific target group.
Control	None or any types of intervention
Outcome	Primary outcome: gambling frequency and gambling problem. Secondary outcomes: gambling knowledge and attitudes.
Study design	Randomized controlled trials (RCT) and non-randomized study (NRS) with at least 20 participants in each group.
Follow up	Primarily studies with at least 6 months follow up in accordance with Society for Preventive Research recommendations. However, studies with 1 month follow up were accepted.

Table 2. Inclusion and exclusion criteria for responsible gambling strategies.

Scope	The effectiveness of responsible gambling strategies
Participants	≥16 years of age. Studies with younger participants were only included if ≥75% of the participants were 16 years or older or if they reported results separately for relevant age. Studies where ≥75 % of the participants had severe gambling problems (e.g. Problem Gambling Severity Index, PGSI ≥ 8) were excluded, exception for studies evaluating self-exclusion programs
Intervention	Responsible gambling strategies such as pop-up messages, self-exclusion program, normative personalized (normative) feedback, limits
Control	A control was not required to be included in the review
Outcome	Primary outcome: gambling frequency measured with time or money and self-reported gambling problems
Study design	Randomized controlled trials (RCT) and non-randomized study (NRS) with at least 20 participants per group. For studies lacking a control group the number of participants had to be ≥100. Studies without control (cohorts) with ≥50% drop out were excluded. Studied with a drop out between 30 and 50% were discussed by the whole group. Case control studies were excluded.
Follow-up	No restriction

Literature search and databases

We conducted a literature search in the following databases: PsycINFO (EBSCO), PubMed (NLM), Scopus (Elsevier), and SocINDEX (EBSCO). Search strategies consisted of a combination of free text terms and controlled vocabulary. In addition, a simultaneous centralized search of multiple EBSCO databases was conducted. Databases included were: Academic Search Elite, CINAHL, ERIC, MEDLINE, PsycINFO, Psychology and Behavioral Sciences Collection, and SocINDEX. All databases were searched for studies that were published from January 2000 to October 2018. Peer reviewed articles in English, Danish, Norwegian, and Swedish were included. The reference lists presented in the reviews and identified studies were manually searched. Search strategies were developed by an information specialist in collaboration with the experts in the review team. Search strategies according to PRISMA are provided in Supplementary Material 1.

Selection of studies

Two experienced reviewers independently screened titles and abstracts. We retrieved the full text articles if either or both reviewers considered a study to be potentially eligible. For potentially eligible studies, both reviewers read the full texts and consensus was reached. In cases of disagreement, the project group discussed the article(s) and consensus was reached. Supplementary Material 2 lists the excluded articles and the reasons for exclusion.

Assessment of risk of bias

Two reviewers (experienced gambling researchers) independently assessed each eligible study for risk of bias, followed by discussions in the project group (by authors DF, JS, JO, and AP) until consensus regarding inclusion was reached. For randomized studies and nonrandomized controlled studies (NRSI), we used the Rob 2.0 checklist (Higgins et al. 2016). For NRSI, we listed the critical confounders: gender, age, gambling problem, and/or overall gambling behavior. These confounders were selected since it is established that gambling, gambling behaviors, and gambling problems (as outcome variables) differ with regard to age and gender (Merkouris et al. 2016). The confounds age and gender have also been used to assess bias in non-randomized studies of gambling interventions (Sterne et al. 2016). If the confounders were not reported in the selected study, or if it was not clearly stated how these were handled, the risk of bias was assessed as critical and no further domains were evaluated.

The first domain of the Rob 2.0 checklist, selection bias was replaced with the following: Were the groups, at baseline, similar in terms of important confounders? If not, were adjustment techniques used to attempt to correct the selection biases?

We developed a checklist for studies lacking control groups (see Supplementary Material 3). Only studies with low or moderate risk of bias were included in the analyses. Excluded studies with high risk of bias are listed in Supplementary Material 4. The characteristics of the studies included in this review are listed in Supplementary Material 5.

Meta-analysis

The meta-analysis was carried out using the Review Manager version 5.3. (2015), with the random effects model due to clinical heterogeneity in the studies. For dichotomous outcomes, the risk difference (RD) or risk ratio (RR) with 95% confidence intervals (CI) was calculated using the Mantel-Haenszel method (Kuritz et al. 1988). For continuous outcome variables, the mean difference (MD) or standard mean difference (SMD) with a 95% CI was calculated using the Inverse Variance method. In one case, we contacted the authors of the relevant studies and requested supplementary data from Canale et al. (2016). Examination of this data qualified the study for inclusion in the meta-analysis.

Educational programs were divided into two groups: short (programs taking place on one occasion) and long (programs taking place over several occasions, usually between four and six sessions) interventions. Subsequent analyses were based on these two groups. To merge data into the meta-analysis, the interventions had to be comparable in intensity and length. Therefore, occasional sessions were not comparable to programs lasting several weeks.

The certainty of evidence

The certainty of evidence was assessed with the GRADE (strong (⊕⊕⊕⊕), moderate (⊕⊕⊕Ο), low (⊕⊕ΟΟ), or very low (⊕ΟΟΟ)) (Guyatt et al. 2008). We applied the following rules in this systematic review:

• Education programs with a follow-up time of less than or equal to 6 months; indirectness –1.

• If an intervention was evaluated in only one smaller study (<350 participants) and the outcome was not statistically significant, the certainty of the evidence was assessed as very low (⊕ΟΟΟ) (Guyatt et al. 2011).

• When using descriptive or narrative analysis for several studies covering one intervention (since meta-analysis was not feasible due to lack of data), and the results were not statistically significant in the individual studies, the precision was −1.

• If an intervention was evaluated in only one study and published for ≥10 years: indirectness −1.

• The outcomes ‘knowledge’ and ‘attitudes’ are indirect measures: indirectness −1.

Results

Search results and flow charts

The literature search yielded a total of 6202 records. We excluded 5915 records based on citation and abstract screening. Thus, 294 full-text articles were further assessed for eligibility. Finally, a total of 37 articles were classified as relevant and eligible based on the inclusion and exclusion criteria.

Of the 37 studies, 28 had low or moderate risk of bias (Figure 1). The remaining nine studies presented with critical and high risk of bias and were not included in the analysis (Ladouceur et al. 2005, 2007; Turner et al. 2008; Jardin and Wulfert 2009; Nelson et al. 2010; Wohl et al. 2010; Stewart and Wohl 2013; Auer et al. 2014; Kotter et al. 2018). The shortcomings in the eligible studies are presented in the section below supplemented by Figure 2(a,b) outlining the risk of bias and subsequent exclusion.

Figure 1. Flow chart of the from the search results.

Figure 2. (a) RCT and NRS with critical or high risk for bias. (b) Cohort studies with critical or high risk for bias.

Educational programs

In total, 11 studies concerning educational programs were identified as eligible (Ladouceur et al. 2005; Williams and Connolly 2006; Doiron and Nicki 2007; Turner et al. 2008; 2008; Williams et al. 2010; Wohl et al. 2010; Lupu and Lupu 2013; Donati et al. 2014; Canale et al. 2016; St-Pierre et al. 2017). Of these, three were excluded from the analysis due to critical or high risk for bias (Ladouceur et al. 2005; Turner et al. 2008; Wohl et al. 2010). The determining reasons for the classification of critical or high risk of bias in these studies were:

1. no information reported concerning participants’ age and sex (one study contained students that were 10 years of age);

2. no information regarding gambling behaviors;

3. unclear randomization processes; and

4. lack of description of attrition rates.

Of the eight remaining studies, six were conducted in school environments (Turner et al. 2008; Williams et al. 2010; Lupu and Lupu 2013; Donati et al. 2014; Canale et al. 2016; St-Pierre et al. 2017). Long-term designs were used in five studies (Turner et al. 2008; Williams et al. 2010; Lupu and Lupu 2013; Canale et al. 2016; St-Pierre et al. 2017). One study was conducted using college students (Williams and Connolly 2006) and one targeted gamblers (Doiron and Nicki 2007). A summary of the studies is presented in Table 3 and the characteristics of the respective studies are shown in Supplementary Material 5.

Table 3. Characteristics of studies with low or moderate risk of bias regarding educational program strategies.

Study design, Country reference	Intervention	Population	Outcome	Follow up	Included in meta-analysis
Canale 2016 RCT, Italy Canale et al. (2016)	Education, school, longer program	14–18 years of age n = 168	Gambling problem, gambling frequency, loss/bets attitudes	2 months	Gambling frequency
Donati 2014 RCT, Italy Donati et al. (2014)	Education, school, longer program	14–18 years of age n = 91	Knowledge, attitudes	6 months
Lupu 2013 RCT, Romania Lupu and Lupu (2013)	Education, school, longer program	12–13 years of age n = 75	Knowledge	6 and 12 months
St-Pierre RCT, Canada St-Pierre et al. (2017)	Education, school, short program	13–18 years of age n = 280	Gambling frequency/attitudes	3 months
Turner 2008 RCT, Canada Turner et al. (2008)	Education, school, longer program	13–18 years of age n = 201	Knowledge	2 months
Williams 2010 RCT, Canada Williams et al. (2010)	Education, school, longer program	13–20 years n = 1686	Gambling Problems, gambling frequency, loss/bets, knowledge, attitudes	4 months	Gambling frequency
Williams 2006 NRS, Canada Williams and Connolly (2006)	Education, college, focused on probability	College students n = 300	No of gamblers, gambling frequency, bets, attitudes, knowledge	6 months
Doiron 2007 RCT, Canada Doiron and Nicki (2007).	Program stop and think!	Gamblers Mean age: 38 years Men = 60% n = 40	Money spent, hours played/no of sessions, gambling behavior	1 months

A meta-analysis of the identified literature was only possible for school programs with long-term interventions (spanning several weeks) and for the outcome frequency of gambling. In terms of outcome frequency, gambling had a positive decrease in frequency measured in days, SMD = 0.29 (95%CI −0.48, −0.11). The certainty of evidence was low in this case (Table 4). For the remaining outcomes, including amount of money spent on gambling, attitudes toward gambling, and knowledge (Turner et al. 2008; Williams et al. 2010; Lupu and Lupu 2013), the certainty of the evidence was very low (Table 3). This was also the case for educational programs aimed toward high school students (Williams and Connolly 2006) and interventions targeting gamblers (Doiron and Nicki 2007). The certainty of the evidence for the separate outcomes is shown in Tables 4 and 5.

Table 4. Summary of finding for educational program (long) compared to no intervention.

Outcome	Follow-up	Participants (no of studies)	Results	GRADE	Comments
Changes in gambling problems, SOGS-RA	2 months	168 (1 RCT) Canale et al. (2016)	MD: −0.19 (95%CI −0.37, −0.01)	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^b
Proportion of persons with gambling problems	4 months	659 (1 RCT) Williams et al. (2010)	RD: −0.02 (95%CI −0.05, 0.01)	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^c
	4 months booster	390 (1 RCT) Williams et al. (2010)	RD: 0.04 (95%CI 0.00, 0.07) RR: 0,31 (95%CI 0.09, 1.10)	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^c
Gambling frequency (number of days gambling)	2–4 months	827 (2 RCT) Canale et al. 2016; Williams et al. (2010)	SMD: −0.29 (95%CI −0.48, −0.11)	Low ⊕⊕ΟΟ	–One bias^a –Imprecision^d –Inconsistency^e
Gambling frequency (number of days gambling/90 days)	4 months booster	390 (1 RCT) Williams et al. (2010)	MD: −10,66 (95%CI −15.46, −5.86) reduction	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^c
Bets	2 months	168 (1 RCT) Canale et al. (2016)	No statistically significant difference	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^b
Losses	4 months	659 (1 RCT) Williams et al. (2010)	No statistically significant difference	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^c
	4 months booster	390 (1 RCT) Williams et al. (2010)	No statistically significant difference	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^c
Knowledge (different scales used). Not possible to merge in meta-analysis)	2 months	201 (1 RCT) Turner et al. (2008)	MD: 17.00 (95%CI 12.53, 21.47)	Very low ⊕ΟΟΟ	–Two indirectness^f –Two imprecision^b
	4 months	659 (1 RCT) Williams et al. (2010)	MD: 0.89 (95%CI 0.67, 1.11)	Very low ⊕ΟΟΟ	–One bias^a –One imprecision^h –Two indirectness^f
	6 months	147 (1 RCT) Donati et al. (2014)	No statistical difference	Very low ⊕ΟΟΟ	One small study without significant results
	6 months 12 months	75 (1 RCT) Lupu and Lupu (2013)	MD: 11.58 (95%CI 9.45,13.71)	Very low ⊕ΟΟΟ	–One bias^g –Two imprecision^b –One indirectness^h
			MD: 12.39 (95%CI 10.26, 14.52)
Attitude (different scales used)	2 months	168 (1 RCT) Canale et al. (2016)	MD: −1.97 (95%KI −4.06, 0.12)	Very low ⊕ΟΟΟ	–One bias^a –Two imprecision^b –Two indirectness^f
	4 months	659 (1 RCT) Williams et al. (2010)	MD: −0.54 (95%CI − 0.77, −0.31)	Very low ⊕ΟΟΟ	–One bias^a –One imprecision^h –Two indirectness^f
	6 months	147 (1 RCT) Donati et al. (2014)	No statistical difference	Very low ⊕ΟΟΟ	One small study without significant results

Note: CI: confidence interval; MD: mean difference; SMD: standard mean difference; RD: risk difference.

^aDrop out >20%.

^bVery few participants.

^cWide CI and only one study.

^dWide CI.

^eInconsistency among the programs.

^fUsed scales that did not measure gambling problems and short follow up (indirectness).

^gUnclear how the randomization was carried out.

^hOne study with limited number of participants.

Table 5. Summary of findings for educational program (short) compared to no intervention.

Outcome	Participants (no of studies) Reference	Results MD (95%CI)	GRADE	Comments
Attitudes (GAS, 12 items, score 12 − 60)	280 (1 RCT) St-Pierre et al. (2017)	MD: −0.96 (−2.54, 0.62)	Very low ⊕ΟΟΟ	One small study without significant results
Gambling Activities Questionnaire (max score 44)		MD: −0.24 (−2.43, 1.95)	Very low ⊕ΟΟΟ	One small study without significant results

Note: CI: confidence interval.

None of the selected studies evaluated the effects of educational programs for staff in gambling environments on individuals who gamble.

Responsible gambling measures

The review identified 26 publications which fulfilled the inclusion criteria (Table 6). Of these, 14 were randomized controlled trials (RCT) (Steenbergh et al. 2004; Floyd et al. 2006; Cunningham et al. 2012; Jardin and Wulfert 2012; Celio and Lisman 2014; Kim et al. 2014; Wohl et al. 2014; Martens et al. 2015; Neighbors et al. 2015; Rockloff et al. 2015; Auer and Griffiths 2016; Ginley et al. 2016; Broussard and Wulfert 2017; Caillon et al. 2019), two were non-randomized studies (NRS) with control groups (Sharpe et al. 2005; Wood and Wohl 2015) and four studies did not have a comparison group (cohort studies) (Nelson et al. 2008; Auer and Griffiths 2013, 2015a, 2015b; McCormick et al. 2018). A summary of these studies is shown in Table 6 and their characteristics are shown in Supplementary Material 5.

Table 6. Characteristics of studies with low or moderate risk of bias regarding consumer protection interventions.

Study design, country Reference	Intervention	Population setting	Outcome	Follow-up	Included in meta-analysis
Auer 2016 RCT, UK Auer and Griffiths (2016)	PNF		Theoretical loss	6 months
Celio 2014 RCT, USA Celio and Lisman (2014)	PNF	Laboratory setting	Gambling frequency, theoretical loss	Instantly after intervention
Cunningham 2012 RCT, Canada Cunningham et al. (2012)	PNF PF	Gamblers	Gambling frequency, theoretical loss	3, 6 and 12 months	Gambling frequency, theoretical loss
Martens 2015 RCT, USA Martens et al. (2015)	PNF	College	Gambling problem, gambling frequency, theoretical loss	3 months	Gambling problem, gambling frequency
Neighbors 2015 RCT, USA Neighbors et al. (2015)	PNF	College	Gambling problem, gambling frequency theoretical loss	3 and 6 months	Gambling problem, gambling frequency, theoretical loss
Wood 2015 NRS, Canada Wood and Wohl (2015)	PF	Problem gambling	Theoretical loss	6 months
Auer 2015 Cohort, UK Auer and Griffiths (2015)	PNF PF	Adults	Proportion of gamblers ending their online gambling session/slots	Instantly after intervention
Broussard, 2017 RCT, USA Broussard and Wulfert (2017)	Pop-up	Laboratory setting, college students, slots	Gambling frequency: no. of spins	Instantly after intervention	Gambling frequency: no. of spins
Floyd 2006 RCT, USA Floyd et al. (2006)	Pop-up	Laboratory setting, college students, roulette	Time spent gambling, bets	Instantly after intervention	Time spent gambling, bets
Ginley 2016 RCT, USA Ginley et al. (2016)	Pop-up	Laboratory setting, college students, slots	Gambling frequency: no. of spins	1 week	Gambling frequency: no. of spins
Jardin, 2012 RCT, USA Jardin and Wulfert (2012)	Pop-up	Laboratory setting, experienced gambler	Time spent gambling, bets	Instantly after intervention
Rockloff, 2015 RCT, Australia Rockloff et al. (2015)	Pop-up	Laboratory setting, gamblers without experience	Bets	Instantly after intervention
Steenbergh 2004 RCT, USA Steenbergh et al. (2004)	Pop-up	Laboratory setting, college students, roulette	Time spent gambling, bets	Instantly after intervention	Time spent gambling, bets
Kim 2014 RCT, Canada Kim et al. (2014)	Pop-up: Limitation in money	Laboratory setting, College students n = 43	Time spent gambling	Instantly after intervention
Wohl 2014 RCT, Canada Wohl et al. (2014)	Pop-up: Limitation in time	Laboratory setting, college students n = 56	Proportion of gamblers who didn’t exceed their limit regarding time	Instantly after intervention
Auer 2013 Cohort, Austria Auer and Griffiths (2013)	Limits in money and time	Online	Theoretical loss, gambling frequency	1 months
Nelson 2008 Cohort, USA Nelson et al. (2008)	Limits in money and time	Online	Loss	6 months
Sharpe 2005 NRS, Australia Sharpe et al. (2005)	Limits in time: Slots with different limits	Hotel and pubs,	Time spent gambling, bets, net loss	Instantly after intervention
Caillon, 2019 RCT, France Caillon et al. (2019)	Self exclusion 1 week	18% with gambling problems n = 60	Money, time spent gambling	14 days, 2 months
McCormick 2018 Cohort, Canada McCormick et al. (2018).	Self exclusion, 6-12 months	74% with gambling problems n = 269	Gambling problems (vilket instrument?)	6 and 12 months

Note: PFN: personalized normative feedback; PF: personalized feedback.

Meta-analyses were possible for two RG measures: personal feedback (PF) and pop-up messaging (Supplementary Material 6). The meta-analyses for PF demonstrated that the measure results in a decrease in gambling frequency at three months. However, the certainty of the evidence was low (Table 7) and the long-term effects are unknown. For the remaining RG measures and their outcomes, the certainty of evidence was very low (see Tables 7–11).

Table 7. Summary of finding for PFN compared to no intervention.

Outcome	Follow-up: participants (no of studies) Reference	Result (95%CI)	Absolute effect	GRADE	Comments
Changes in gambling problem (SOGS and PGSI), University students	3 months: 447 (2 RCT) Martens et al. (2015); Neighbors et al. (2015)	SMD −0.21 (−0.42, 0.01)		Very low ⊕ΟΟΟ	–Two imprecision^a –One inconsistency^b
	6 months: 226 (1 RCT) Neighbors et al. (2015)	SMD −0.06 (−0.32, 0.20)		Very low ⊕ΟΟΟ	One small study without significant results
Gambling frequency
No of days  spent gambling	3 months: 583 (3 RCT) Cunningham et al. (2012); Martens et al. (2015); Neighbors et al. (2015)	MD 1.07 (−1.25, 3.38)		Very low ⊕ΟΟΟ	–Two inconsistencies^c –Two imprecisions^d
	6 months: 365 (2 RCT) Cunningham et al. (2012) Neighbors et al. (2015)	MD: 0.90 (−1.50, 3.30)		Very low ⊕ΟΟΟ	–Two imprecisions^d –One inconsistency^f
No of days spent  gambling (days spent gambling within 30 days, general population)	12 months: 139 (1 RCT) Cunningham et al. (2012)	Full PNF: no statistically significant difference Partial PNF: MD: −2.5 (−4.98, −0.02)		Very low ⊕ΟΟΟ Low ⊕⊕ΟΟ	One small study without significant results –Two imprecisions^a
Proportion of  gamblers ending their gambling session, Online gamblers/slots	Within one gambling session. Approximately 11,232 (1 cohort) Auer and Griffiths (2015a)	RD: 0.01 (0.00, 0.01) RR: 2.13 (1.63, 2.79)	10 more that stopped gambling compared to the control group.	Very low ⊕ΟΟΟ	–Two bias^e –One imprecision^g
Monetary losses
Losses	3 months: 366 (2 RCT) Cunningham et al. (2012); Neighbors et al. (2015)	MD: 14.36 (−74.44, 103.16)		Very low ⊕ΟΟΟ	–Two imprecisions^a –One inconsistency^c
	6 months: 366 (2 RCT) Cunningham et al. (2012); Neighbors et al. (2015)	MD: −17.36 (−36.76, 2.04)		Very low ⊕ΟΟΟ	–Two imprecisions^a –One inconsistency^c
	12 months: 139 (1 RCT) Cunningham et al. (2012)	No statistically significant difference		Very low ⊕ΟΟΟ	One small study without significant results
Theoretical loss Online gamblers	6 months: 5528 (1 RCT) Auer and Griffiths (2016)	PNF (χ^2= 32,08, df 5, p = 0.0001)		Very low ⊕ΟΟΟ	–Two inconsistencies^h –One bias^i
Deposit money,  Online gamblers	6 months: Low risk gamblers n = 623 At risk gamblers: n = 101 Problem gamblers n = 55 (1 NRS) Wood and Wohl (2015)	Statistically significant difference PNF for at risk gamblers No effect for other groups		At risk gamblers Very low ⊕ΟΟΟ Low risk gamblers/Problem gamblers very low ⊕ΟΟΟ	–Two bias^j –Two imprecision^k
					One small study without significant results
Bets ($),  University students	333 (1 RCT) Martens et al. (2015)	MD: −86.54 (−197.74, 24.66)		Very low ⊕ΟΟΟ	One small study without significant results

^aFew participants and wide CI. ^bDifferent scales and clinical heterogeneity. ^cThe results were inconsistent and evaluated/measured in different ways and the populations used did not have the same sample characteristics. ^dWide CI. ^eA cohort without control group, evaluation only once before and after. ^fThe results were inconsistent and evaluated/measured in different ways and the populations used did not have the same sample characteristics. ^gFew events. ^hMerging of results from five different intervention groups that have received little different actions. ⁱShortcomings in the result report. ^jRisk for confounding. ^kFew participants.

Table 8. Summary of finding: pop-up message.

Author (Participants, no of studies) Reference	Gambling mode/population	Results	GRADE	Comments
Outcome: time spent gambling
Broussard (2017) and Ginley (2016)  (116, 2 RCT) Broussard et al. (2017), Ginley et al. (2016)	Slots Students at university	MD: −28.13 (−82.77, 26.51)	Very low ⊕ΟΟΟ	−Two imprecision^a −One inconsistency^b −One indirectness^c
Floyd (2006)  (122, RCT) Floyd et al. (2006) & Stenbergh (2004 (101, RCT) Steenbergh et al. (2004)	Roulette	Heterogeneous results	Very low ⊕ΟΟΟ	−One inconsistency^d −Two imprecision^e −One indirectness^c
Jardin (2012)  (80, RCT) Jardin and Wulfert (2012)	Wheel of luck	Decreased, p < 0.05 in favor for intervention	Very low ⊕ΟΟΟ	−Two imprecision^f −One indirectness^c
Outcome: bets
Floyd (2006) (122, RCT) Floyd et al.  (2006) & Steenbergh (2004) (101, RCT) Steenbergh et al. (2004)	Roulette	Heterogeneous results	Very low ⊕ΟΟΟ	−One inconsistency^d −Two precision^e −One indirectness^c
Jardin (2012) (80, RCT) Jardin and Wulfert (2012)	Wheel of luck	Decreased, p < 0.05 in favor for intervention	Very low ⊕ΟΟΟ	−Two imprecision^f −One indirectness^c
Rockloff (2015) (130, RCT) Rockloff et al. (2015)	Slot	No difference regarding spins	Very low ⊕ΟΟΟ	One small study without significant results

Notes: CI: confidence intervals; MD: mean difference.

^aFew participants and wide CI.

^bDifference between the interventions.

^cLaboratory setting.

^dThe results were inconsistent, difference between the interventions).

^eNarrative analysis and few participants.

^fVery few participants and only one study.

Table 9. Summary of finding of pop up message: limitation in time or money as compared to control group/standard message.

Outcome	Participants (no of studies) Reference	Result MD/RD (95%CI)	Absolut effect per 1000	GRADE	Comments
Intervention
Length of gambling  session (minutes)	43 (1 RCT) Kim et al. (2014)	MD: −4.48 (−8.49, −0.47) SMD: −0.63 (−1.24, −0.01)		Very low ⊕ΟΟΟ	–Two imprecision^a –One indirectness^b
Intervention
Number of gamblers that  adhered to limits	56 (1 RCT) Wohl et al. (2014)	RD: 0.27 (0.04, 0.50)	270 (40 till 500 more)	Very low ⊕ΟΟΟ	–Two bias^c –Two imprecision^a –One indirectness^b

Notes: CI: confidence interval; MD: mean difference; RD: risk difference; SMD: standard mean difference.

^aVery few participants, wide confidence interval and only a single study.

^bPerformed in a laboratory setting.

^cRandomization unclear.

Table 10. Summary of findings of Limitation in time or money, as compared to no intervention.

Outcome	Participants (no of studies) Reference	Result MD/RD (95%CI)	GRADE	Comments
Online—cash in limit
Gambling frequency	441 (1 cohort) Nelson et al. (2008)	MD: −4.72 (−8.27, −1.17)	Very low ⊕ΟΟΟ	–Two bias^a –One imprecision^b –One indirectness^c
Bets/day	441 (1 cohort) Nelson et al. (2008)	MD: −1.51 (−2.66 till −0.36)	Very low ⊕ΟΟΟ	–Two bias^a –One imprecision^b –One indirectness^c
Bets (Euro)	441 (1 cohort) Nelson et al. (2008)	MD: 0.28 (−2.87, 2.31)	Very low ⊕ΟΟΟ	–Two bias^a –Two imprecision^d –One indirectness^c
% Losses	452 (1 cohort) Nelson et al. (2008)	MD: −0.01 (−0.07,0.05)	Very low ⊕ΟΟΟ	–Two bias^a –Two imprecision^c –One indirectness^c
Loss (theoretical loss)  Changes in deposit per months or week	500 (1 cohort) Auer and Griffiths (2013)	t value: −4.47 p < 0.0001	Very low ⊕ΟΟΟ	–Two bias^a –One imprecision^b
Online—limitation in time
Loss (theoretical loss)  Changes in deposit per day	500 (1 cohort) Auer and Griffiths (2013)	t value: −3.19 p < 0.0001	Very low ⊕ΟΟΟ	–Two bias^a –One imprecision^b
Hotels and other gambling venues
Time spent gambling	634 (1 NRS) Sharpe et al. (2005)	4.6 min difference, mean 3 min p < 0.001	Very low ⊕ΟΟΟ	–Two bias^e –One imprecision^f
Bets	634 (1 NRS) Sharpe et al. (2005)	11.7 more bets control, mean 211, p < 0.01	Very low ⊕ΟΟΟ	–Two bias^e –One imprecision^f
Losses (net)	634 (1 NRS) Sharpe et al. (2005)	5.2 Australian dollars less in mean loss compared to the control group 19.85, p < 0.001	Very low ⊕ΟΟΟ	–Two bias^e –One imprecision^f

Notes: CI: confidence interval; MD: mean difference; RD: risk difference. ^aA cohort without control group, evaluation of effect only once before and after. ^bLimited number of participation and only one study. ^cThe intervention took place 2005 and the ability/possibility to monitor gambling behavior has improved. ^dOne study with few participants and wide CI. ^eNo information about baseline data from both groups. ^fOnly one study with limited number of participants.

Table 11. Summary of findings for self-exclusion (short or long).

Outcome	Follow-up	Participants (no. of studies) Reference	Result MD/RD (95%CI)	GRADE	Comments
Self-exclusion, 6–12 months
Gambling problem (PGSI score)	6 months	269 (1 cohort) McCormick et al. (2018)	Mean value 1.3	Very low ⊕ΟΟΟ	–Two bias^a –One imprecision^b
	1 year	235 (1 cohort) McCormick et al. (2018)
Self-exclusion (1 week)
Money	2 months	60 (1 RCT) Caillon et al. (2019)	No statistically significant difference	Very low ⊕ΟΟΟ	One small study without significant results
Time spent gambling	2 months	60 (1 RCT) Caillon et al. (2019)	No statistically significant difference	Very low ⊕ΟΟΟ	One small study without significant results

Notes: CI: confidence intervals.

^aDrop-out.

^bFew participants.

Shortcomings in the eligible studies

Studies with high or critical risk of bias

The studies with high or critical risk of bias lacked required information in several different domains (Figure 2(a,b)) and were, therefore, excluded from the systematic review. The primary reasons for high or critical risk of bias were: (1) a lack of information concerning randomization of groups resulting in risk of confounding factors in NRS and cohort studies; (2) a lack of outcome data and information concerning the specific measurement of the effect; (3) the applied intervention deviated from the intended interventions resulting in a high risk for bias (Sharpe et al. 2005); and (4) flaws in terms of data analysis and presentation.

Studies with low or moderate risk of bias

Very few studies had a low risk for bias, primarily due to a lack of information regarding randomization and methodology (how and on what basis was the study done), including the absence of descriptive information on how the randomization process was carried out. Missing data was a prevalent factor in several cases (Figure 3(a)). For NRS, risk for bias was related to a lack of information regarding confounding issues (Figure 3(b)). Additional shortcomings were the use of several different outcome measurements, evaluation or presentation of the results, and different lengths of time between follow-ups. This resulted in an inability to perform meta-analysis and thus, an inability to assess progress in the program. The most pervasive shortcoming across studies was missing outcome data. In several cases, means, standard deviations, and attrition rates were not reported. These factors resulted in an inability to perform a meta-analysis on these studies and, thus, an overall analysis of the effects of different RG measures was not possible. Studies with low and moderate risk for bias were included in the systematic review.

Figure 3. (a) RCT and NRS with low or moderate risk for bias. (b) Cohort—low or moderate risk of bias.

Discussion

Despite the availability of a relatively large number of review studies in the field of prevention of gambling problems (e.g. Williams et al. 2012; Ladouceur et al. 2013; Motka et al. 2018), to our knowledge, this is the first systematic review that assesses the certainty of evidence in a structured manner. We present the shortcomings evident in the reviewed studies on preventive education programs and RG measures. Our findings suggest that there is a low level of certainty of evidence for studies making use of long-term school interventions and PF, indicating a potential effect of these interventions. It is important to note that, in these cases, the follow-up period was short. For the remaining interventions and their outcomes, the certainty of evidence was very low. In such cases, it is not possible to draw firm conclusions concerning the effects of the intervention (i.e. whether it reduces or increases gambling or harm from gambling).

The absence of evidence across measures and studies does not imply that the intervention is not effective (Alderson 2004). Our inability to draw conclusions regarding the effects of the RG measures and educational programs does not necessarily suggest that the interventions had no effects. Rather, there were insufficient eligible studies to positively or negatively evaluate the effects. Thus, the absence of evidence is not evidence of absence (Alderson 2004). This distinction is important for future systematic reviews of consumer protection measures, including RG tools and educational programs designed to diminish gambling-related harm.

One explanation for the absence of effects of interventions is that prevention measures curtail a progressive gambling trajectory. Continuation of the trajectory would cause the gambler lose increasing amounts of money over time and result in increased risk of harm. The effect of some measures might be the halting of a maladaptive gambling pattern without decreasing the amount of money or time spent gambling. Rather, the monetary and time metrics remain static, but the preventive measures diminish the progression of the problem.

A comparison to previous systematic reviews

Despite differences in the inclusion and exclusion criteria, assessment of risk of bias, and certainty in evidence, our conclusion regarding the need for increased and higher-quality research is similar to that of McMahon et al. (2019). Almost all previous reviews on this topic have implemented more liberal inclusion criteria than ours. For example, in our review, studies that did not report results in the form of an estimate with confidence intervals were excluded if it was not possible to calculate the result based on data presented.

Our systematic review revealed two primary results. Firstly, there appears to be a reduction in gambling consequent to an educational program and PF. These two interventions had an impact on gambling behavior, but not in terms of loss of money due to gambling. In these cases, the levels of certainty were low. Another systematic review suggested that PF is effective (Marchica and Derevensky 2016). However, an important point to note is the question of whether our results present with practical implications since they were not accompanied by a reduction in monetary losses. Moreover, we are currently unaware of the effects of these interventions on problem gambling or harm due to gambling. Another problem prevalent in this study was that some educational program studies did not measure gambling problems. This lack of data makes drawing inferences regarding reduction of harm in gambling prevention challenging.

The follow-up period for the studies on school-based interventions was short (up to 6 months). Therefore, the long-term effect of these interventions is difficult to determine or estimate. Our finding that educational programs may reduce gambling behaviors among students is congruent with findings in a review by Keen et al. (2017) which stated that five out of nine studies reviewed reported a significant effect of educational programs on gambling behaviors. All of Keen et al.’s (2017) reviewed studies (n = 9) reported an effect of the intervention on cognition or knowledge. However, this review, as in other systematic reviews on school-based interventions, did not include meta-analyses (Keen et al. 2017; Oh et al. 2017).

All the reviews considered stressed the methodological flaws present in their included studies. Ladouceur et al. (2017) used an a priori approach in their review to evaluate the effects present within the factors of self-exclusion, limit-setting, training of venue employees, RG, specific game features, and behavioral characteristics of RG. They included all studies that conducted in a real-world gambling environment (excluding online gambling) and fulfilled at least one of the following criteria: (1) existence of control groups; (2) repeated measures designs; or (3) validated measurement scales. Of the 29 eligible studies, only six fulfilled all three criteria, indicating a lack of scientific rigor.

Shortcomings in the reviewed studies

Despite meeting the inclusion criteria, several flaws were present in the reviewed studies: weakness in, or lack of, reliable data, including the measurement issues; small numbers of participants; and brief follow-up periods. Furthermore, there was a lack of studies in general.

In several previous reviews, it was not possible to perform meta-analysis, which makes drawing conclusions regarding the suitable types of prevention for excessive gambling difficult. Moreover, most reviews did not attempt to conduct a meta-analysis, although some examined bias according to the Quality Assessment Tool for Quantitative Studies (National Collaborating Centre for Methods and Tools, 2008). In these cases, the studies concluded that the global rating would be considered weak and, therefore, study quality could not be systematically evaluated. In McMahon et al’s (2019) study, only 1 out of 10 systematic reviews assessed the risk of bias of the included primary studies and took this into account when evaluating the effects of interventions. In our review, we excluded studies with high risk of bias from the analysis because the results may be unreliable.

Overall, it is difficult to determine the effect measures or programs in the field of RG. Apart from the shortcomings mentioned earlier in the review, there are a multitude of factors that need to be assessed when considering these effects. These include risk level for the gambler, whether online gambling is prevalent or land-based venues are used (or both), and what type of gambling activity the individual engages in (specific or several activity types). Use of the interventions is also an important factor; e.g. there is low use among gamblers who have engaged with the RG tool ‘Playscan’ (Forsström et al. 2016, 2017, 2020). There are also several demographic factors to incorporate into analyses as they are known to influence outcomes of interventions, including age and gender. Several studies have demonstrated the importance of gender in the study of gambling and problematic gambling (Merkouris et al. 2016). Across the studies included in the systematic reviews, very few sub-group analyses were carried out, or the results presented the analyzed effects for men and women separately. This should be addressed in future research (Romild et al. 2016). It is also necessary to investigate if RG measures operate in a different manner for vulnerable groups, such as people at economic disadvantage (low socio-economic status) or persons with neuropsychiatric disorders.

Another important consideration is the setting/context where the study takes place. Ladouceur et al. (2017) argued that only studies in real-world gambling settings provide useful information and, therefore, used this as an a priori inclusion criteria. However, although there is a definitive need for more studies in real-world settings and there are shortcomings inherent in laboratory-based research, there are also potential benefits associated with laboratory research, such as the possibility of providing the researchers with a greater level of control. This would make it possible to examine specific factors and replicate research.

Toward a consensus in gambling prevention research

There is a framework for reporting outcomes in gambling treatment research (Walker et al. 2006). The framework signifies a starting point for treatment outcome studies, providing a rationale for research and making it possible to compare the effects of the treatment of gambling disorders. Preventive measures for gambling are diverse and this could partly explain the lack of definitive results and conclusions emanating from the systematic reviews. It is possible that the lack of consensus on basic measurement issues could be avoided by coming to an agreement regarding guidelines which would support the process of building evidence. Ladouceur et al. (2017) recently attempted to establish guidelines for gambling-prevention research by using an a priori approach. However, including studies that fulfill at least one of the three stated criteria is problematic when conducting meta-analyses.

The process of developing research on gambling prevention requires collaboration and census concerning standards of studies. Factors to consider include what outcome measure to use and an established follow-up period. Additionally, very few sub-group analyses are carried out, at least based on the studies included in our systematic review. Most of the studies used risk as a grouping variable, but there is a need for further research in the field based on gender and type of gambling activity. One plausible hypothesis is that RG measures will be used differently, and have different effects, for sub-samples of individuals that gamble. If this is the case, the discourse established in reviews stating that there is an absence of effectiveness potentially maintains the status quo regarding preventive measures and suggests that new studies will not provide additional, valuable information since they are likely to use the same short follow-up periods or different outcome measures. Williams et al.’s (2012) statement clearly depicts the state of the prevalent discourse in gambling prevention. Our systematic review drew similar conclusions, despite using the PRISMA guidelines: the empirical support for gambling preventive measures is weak and there is a need for an increased number of studies, higher-quality research, and expanding on the discourses surrounding these measures.

Limitations of this study

There are several limitations to this review study. The bias ratings were performed and matched. However, interrater reliability was not checked. Therefore, the raters may have investigated bias in the studies using different standards, thereby compromising the accuracy of the review. In such a case, studies might have been excluded or included on the wrong premises. However, the risk of bias for all articles was independently assessed by the same two researchers, followed by discussions within the project group until consensus was reached. There were few instances of disagreement between the two reviewers and consensus was reached in all the cases without significant debate or deviation.

Future research

Researchers have suggested that RG measures need to be more proactive (Hancock and Smith 2017a, 2017b). Jonsson et al. (2019) investigated the use of behavioral feedback via telephone aimed at excessive gamblers. Gamblers that received feedback via telephone had a 29% decrease in theoretical losses compared to 3% for the control group. This result was obtained at a 12-week follow-up (this study was published after the literature search for our review was performed). These findings indicate that further research involving proactive measures is required.

Future efforts in the study of gambling should focus on developing a consensus statement similar to that concerning the treatment of pathological gambling/gambling disorders (Walker et al. 2006). Ladouceur et al. (2017) also state that gambling prevention research must be theory driven. Hence, future studies should attempt to build on the existent theory and literature to find optimal methods of facilitating gambling-related behavioral change.

Using the existing frameworks on criteria for efficacy and effectiveness in prevention and public health interventions, such as the framework from the Society for Prevention Research, could be a useful first step in enhancing the quality of research. The objective of these standards is to articulate a set of principles to identify prevention programs and policies that are empirically validated and merit being labeled as ‘tested and efficacious.’ However, an important distinction to consider is the difference between efficacy and effectiveness. Effectiveness refers to the program effects when delivered under real-world conditions. To claim effectiveness, studies must meet all the conditions of efficacy trials as well as fulfilling the following criteria: program definitions, such as manuals and technical support, should be present; and theory and real-world conditions should be accounted for.

Conclusions

This systematic review showed that there is little support for RG measures and educational programs for gambling in the studies which met the inclusion criteria. Most interventions had a very low certainty of evidence. This review could serve as a model for developing educational programs and RG reviews as it evaluated the certainty of evidence and provided a systematic evaluation of the included studies. Nonetheless, more research is required to understand whether existing interventions are effective. Furthermore, consensus on the best method to conduct preventive gambling studies is required to effectively compare studies and perform meta-analyses. This is essential to further develop guidelines for the prevention of gambling-related harms and to inform gamblers and gambling companies of the presence of effective preventive measures.

Disclosure statement

No potential conflict of interest was reported by the author(s).