Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Pharmacological interventions for the treatment of disordered and problem gambling

Authors' declarations of interest

Version published: 22 September 2022 Version history

https://doi.org/10.1002/14651858.CD008936.pub2
Collapse all Expand all

Background

Pharmacological interventions for disordered and problem gambling have been employed in clinical practice. Despite the availability of several reviews of the efficacy of pharmacological interventions for disordered or problem gambling, few have employed systematic search strategies or compared different categories of pharmacological interventions. Systematic reviews of high‐quality evidence are therefore essential to provide guidance regarding the efficacy of different pharmacological interventions for disordered or problem gambling.

Objectives

The primary aims of the review were to: (1) examine the efficacy of major categories of pharmacological‐only interventions (antidepressants, opioid antagonists, mood stabilisers, atypical antipsychotics) for disordered or problem gambling, relative to placebo control conditions; and (2) examine the efficacy of these major categories relative to each other. 

Search methods

We searched the Cochrane Common Mental Disorders Specialised Register, the Cochrane Central Register of Controlled Trials (CENTRAL), Ovid MEDLINE, Embase, and PsycINFO (all years to 11 January 2022).

Selection criteria

We included randomised trials evaluating a pharmacological intervention for the treatment of disordered or problem gambling. Eligible control conditions included placebo control groups or comparisons with another category of pharmacological intervention.

Data collection and analysis

We used standard methodological procedures, including systematic extraction of included study characteristics and results and risk of bias assessment. Our primary outcome was reduction in gambling symptom severity. Our secondary outcomes were reduction in gambling expenditure, gambling frequency, time spent gambling, depressive symptoms, anxiety symptoms, and functional impairment; and responder status. We evaluated treatment effects for continuous and dichotomous outcomes using standardised mean difference (SMD) and risk ratios (RR), respectively, employing random‐effects meta‐analyses. A minimum of two independent treatment effects were required for a meta‐analysis to be conducted (with only meta‐analytic findings reported in this abstract).

Main results

We included 17 studies in the review (n = 1193 randomised) that reported outcome data scheduled for end of treatment. Length of treatment ranged from 7 to 96 weeks. 

Antidepressants: Six studies (n = 268) evaluated antidepressants, with very low to low certainty evidence suggesting that antidepressants were no more effective than placebo at post‐treatment: gambling symptom severity (SMD −0.32, 95% CI −0.74 to 0.09, n = 225), gambling expenditure (SMD −0.27, 95% CI −0.60 to 0.06, n = 144), depressive symptoms (SMD −0.19, 95% CI −0.60 to 0.23, n = 90), functional impairment (SMD −0.15, 95% CI −0.53 to 0.22, n = 110), and responder status (RR 1.24, 95% CI 0.93 to 1.66, n = 268).

Opioid antagonists: Four studies (n = 562) evaluated opioid antagonists, with very low to low certainty evidence showing a medium beneficial effect of treatment on gambling symptom severity relative to placebo at post‐treatment (SMD −0.46, 95% CI −0.74 to −0.19, n = 259), but no difference between groups in responder status (RR 1.65, 95% CI 0.86 to 3.14, n = 562).

Mood stabilisers: Two studies (n = 71) evaluated mood stabilisers (including anticonvulsants), with very low certainty evidence suggesting that mood stabilisers were no more effective than placebo at post‐treatment: gambling symptom severity (SMD −0.92, 95% CI −2.24 to 0.39, n = 71), depressive symptoms (SMD −0.15, 95% CI −1.14 to 0.83, n = 71), and anxiety symptoms (SMD −0.17, 95% CI −0.64 to 0.30, n = 71).

Atypical antipsychotics: Two studies (n = 63) evaluated the atypical antipsychotic olanzapine, with very low certainty evidence showing a medium beneficial effect of treatment on gambling symptom severity relative to placebo at post‐treatment (SMD −0.59, 95% CI −1.10 to −0.08, n = 63).

Comparative effectiveness: Two studies (n = 62) compared antidepressants with opioid antagonists, with very low certainty evidence indicating that antidepressants were no more effective than opioid antagonists on depressive symptoms (SMD 0.22, 95% CI −0.29 to 0.72, n = 62) or anxiety symptoms (SMD 0.21, 95% CI −0.29 to 0.72, n = 62) at post‐treatment. Two studies (n = 58) compared antidepressants with mood stabilisers (including anticonvulsants), with very low certainty evidence indicating that antidepressants were no more effective than mood stabilisers on depressive symptoms (SMD 0.02, 95% CI −0.53 to 0.56, n = 58) or anxiety symptoms (SMD 0.16, 95% CI −0.39 to 0.70, n = 58) at post‐treatment.

Tolerability and adverse events:  Several common adverse effects were reported by participants receiving antidepressants (e.g. headaches, nausea, diarrhoea/gastrointestinal issues) and opioid antagonists (e.g. nausea, dry mouth, constipation). There was little consistency in the types of adverse effects experienced by participants receiving mood stabilisers (e.g. tiredness, headaches, concentration difficulties) or atypical antipsychotics (e.g. pneumonia, sedation, increased hypomania). Discontinuation of treatment due to these adverse events was highest for opioid antagonists (10% to 32%), followed by antidepressants (4% to 31%), atypical antipsychotics (14%), and mood stabilisers (13%).

Authors' conclusions

This review provides preliminary support for the use of opioid antagonists (naltrexone, nalmefene) and atypical antipsychotics (olanzapine) to produce short‐term improvements in gambling symptom severity, although a lack of available evidence precludes a conclusion regarding the degree to which these pharmacological agents can improve other gambling or psychological functioning indices. In contrast, the findings are inconclusive with regard to the effects of mood stabilisers (including anticonvulsants) in the treatment of disordered or problem gambling, and there is limited evidence to support the efficacy of antidepressants. However, these conclusions are based on very low to low certainty evidence characterised by a small number of included studies, high risk of bias, modest pooled sample sizes, imprecise estimates, moderate between‐study heterogeneity, and exclusion of participants with psychiatric comorbidities. Moreover, there were insufficient studies to conduct meta‐analyses on many outcome measures; to compare efficacy across and within major categories of interventions; to explore dosage effects; or to examine effects beyond post‐treatment. These limitations suggest that, despite recommendations related to the administration of opioid antagonists in the treatment of disordered or problem gambling, pharmacological interventions should be administered with caution and with careful consideration of patient needs. A larger and more methodologically rigorous evidence base with longer‐term evaluation periods is required before definitive conclusions can be drawn about the effectiveness and durability of pharmacological treatments for disordered or problem gambling.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Pharmacological treatments for disordered and problem gambling

Background

Gambling problems can lead to severe consequences for gamblers, their family members and friends, and the community. A range of medications are used to treat people with gambling problems, but there are few high‐quality reviews of the research evidence to guide which ones should be used in practice.

Review question

We investigated whether different types of medications are effective in reducing gambling symptoms compared to no treatment (i.e. placebo, or dummy treatment) or other types of medications.

Study characteristics

We included randomised trials (gold‐standard trials for evaluating the effectiveness of treatments where participants are randomly assigned to one of two or more treatments) published before January 2022. We included 17 studies in the review with a total of 1193 participants. These studies compared antidepressants (medications used to prevent or treat depression), opioid antagonists (medications that reverse and block the effects of opioids like pain relievers and heroin; e.g., naltrexone, nalmefene), mood stabilisers (medications that treat and prevent highs (mania) and lows (depression)), and atypical antipsychotics (medications used to treat schizophrenia or to help treat other psychiatric disorders) with either a no‐treatment control group (i.e. placebo) or with each other. We explored the effectiveness of these treatments on a range of outcomes: gambling symptom severity, gambling expenditure, gambling frequency, depressive symptoms, anxiety symptoms, functional impairment (limitations due to the illness), and responder status (i.e. positive response to treatment based on gambling symptoms or behaviour, or both).

Key results

Antidepressants: We combined the results of six studies (268 participants) investigating the effectiveness of antidepressants. At the end of treatment, there were no clear differences between antidepressants and no treatment on any measure where there was more than one available study: gambling symptom severity, gambling expenditure, depressive symptoms, functional impairment, or responder status.

Opioid antagonists: We combined the results of four studies (562 participants) investigating the effectiveness of opioid antagonists. More than one study evaluated gambling symptom severity and responder status. At the end of treatment, these medications were more helpful than no treatment in improving gambling symptom severity, but there were no clear differences in terms of responder status.

Mood stabilisers: We combined the results of two studies (71 participants) investigating the effectiveness of medications with mood‐stabilising properties. At the end of treatment, there were no clear differences between mood stabilisers and no treatment on any measure where there was more than one available study: gambling symptom severity, depressive symptoms, or anxiety symptoms.

Atypical antipsychotics: We combined the results of two studies (63 participants) investigating the effectiveness of the atypical antipsychotic olanzapine. At the end of treatment, this medication was more helpful than no treatment in reducing gambling symptom severity.

Comparisons between medications: We identified very few studies that compared the effectiveness of different types of medications. Two studies compared antidepressants with opioid antagonists; two studies compared antidepressants with mood stabilisers; and one study compared opioid antagonists and mood stabilisers. At the end of treatment, there were no clear differences between any of these medications on any measure.

Adverse effects: Several common adverse effects (side effects) were reported by people receiving antidepressants (e.g. headaches, nausea, diarrhoea/gastrointestinal issues) and opioid antagonists (e.g. nausea, dry mouth, constipation). Adverse effects reported by people receiving mood stabilisers (e.g. tiredness, headaches, concentration difficulties) or atypical antipsychotics (e.g. pneumonia, sedation, elevated mood) were not consistent. Those most likely to stop treatment due to adverse effects were receiving opioid antagonists, followed by antidepressants, atypical antipsychotics, and mood stabilisers.

Quality of evidence

Only a limited number of studies with a small number of participants investigated the effectiveness of each type of medication. We considered the quality of the evidence across most of the outcomes in this review as very low or low, meaning that we are uncertain about the results. 

Conclusion

Based on a small amount of low‐quality evidence, we conclude that opioid antagonists and atypical antipsychotics (but seemingly not antidepressants) may be effective in reducing gambling symptom severity. There was insufficient information to determine whether these medications can improve other gambling and psychological symptoms. The results relating to mood stabilisers are uncertain. We do not know how effective these medications are in the long term. More research is needed before we can draw any firm conclusions about the effectiveness of medications for gambling problems.

Authors' conclusions

Implications for practice

The findings of this review support the assertion that opioid antagonists demonstrate promising results in the treatment of disordered or problem gambling (Goslar 2019). Consistent with gambling treatment guidelines in Australia (Thomas 2011), the current review provides preliminary support for the use of opioid antagonists, such as naltrexone and nalmefene, in reducing gambling symptom severity, but not response to treatment. However, this conclusion was based on modest and uncertain evidence that illustrated only short‐term effects of treatment. To date, there is insufficient evidence to ascertain whether effects observed immediately after treatment are maintained. Moreover, because of a lack of evidence, the degree to which these pharmacological agents can improve other indices of gambling or psychological impairment remains unclear. These limitations suggest that, despite recommendations related to the administration of opioid antagonists in the treatment of disordered or problem gambling, these pharmacological interventions should be administered with caution and with careful consideration of patient needs (Thomas 2011). 

The available evidence also suggests potential benefits from olanzapine, an atypical antipsychotic, on gambling symptom severity. However, this conclusion was based on two trials characterised by small samples and evidence of very low certainty. Moreover, given the lack of available evidence, it is unknown whether this beneficial effect extends to other indices of gambling behaviour or psychological functioning. There was also an absence of longer‐term follow‐up evaluation periods. Consequently, additional research is needed before definitive statements regarding the efficacy of atypical antipsychotics such as olanzapine can be made.

The review findings regarding the effects of mood stabilisers (including anticonvulsants) were inconclusive. Although the effect size estimate was indicative of a beneficial effect on gambling symptom severity, it is possible that these interventions were no more effective than placebo; the evidence base is also very small and of very low certainty. Moreover, mood stabilisers do not appear to have a beneficial effect on psychological outcomes, such as depressive and anxiety symptoms. Further research is therefore required to investigate the efficacy of mood‐stabilising agents in the treatment of problem or disordered gambling before any conclusions can be drawn.

There is little evidence for the efficacy of antidepressants in reducing gambling symptom severity, as well as other indices such as gambling expenditure, depressive symptoms, functional impairment, and responder status. As previously suggested, the effect size for antidepressants may fail to cross the threshold for statistical significance due to the smaller samples employed in these trials (Bartley 2013). Moreover, anomalous results from small and less methodologically sound studies seemed to have had an undue influence on the study findings. Trials with larger samples and lower risk of bias are required before any definitive statements regarding the efficacy of antidepressants can be made.

Consistent with the meta‐analysis conducted by Bartley 2013, the findings from this review provide limited support for the efficacy of pharmacological treatments for problem or disordered gambling. The most promising findings are for the efficacy of opioid antagonists, such as naltrexone and nalmefene, and, to a somewhat lesser extent, atypical antipsychotics such as olanzapine, on gambling symptom severity immediately following treatment. A lack of available data precluded the conduct of many meta‐analyses for secondary outcomes, such as other gambling indices and psychological functioning. As such, psychological interventions, namely cognitive‐behavioural therapy and motivational interviewing, remain 'best practice' and the first‐line response for the treatment of gambling problems (Cowlishaw 2012).

Implications for research

Although opioid antagonists and atypical antipsychotics show promise for the reduction of gambling symptom severity, the conclusions of this review were derived from a limited evidence base, suggesting the need for further research evaluating opioid antagonists (naltrexone, nalmefene) and atypical antipsychotics (olanzapine). Larger and methodologically rigorous studies investigating antidepressants, mood stabilisers (lithium), and anticonvulsants (topiramate) may also be warranted. Moreover, the trials of opioid antagonists have been conducted by one research group in the United States. Independent replication studies are therefore needed to evaluate opioid antagonists before they can be viewed as an empirically supported gambling treatment, at which time they can serve as a benchmark for other pharmacotherapies. Given the seemingly inconsistent findings identified in this review, further research comparing pharmacological agents across different categories is also required. Moreover, further research exploring the tolerability and adverse effects of pharmacological agents is required, particularly for young people. Finally, the available evidence does not enable conclusions regarding the long‐term efficacy of any pharmacological intervention, as only one primary study reported long‐term efficacy (Rosenberg 2013). This study evaluated the long‐term efficacy of antidepressants (bupropion and escitalopram), an opioid antagonist (naltrexone), and an anticonvulsant (topiramate), at 24 months after treatment completion on key outcomes including depression and anxiety. Whilst no inferential testing was conducted, the descriptive statistics suggest that further reductions were identified at 24 months after treatment completion for depression and anxiety. Given evidence that the treatment of gambling problems is associated with high rates of relapse, even after the delivery of 'best‐practice' psychological interventions (Cowlishaw 2012), further research employing follow‐up evaluation periods is required. 

Future research should aim to recruit sample sizes large enough to provide sufficient statistical power to identify small‐to‐medium treatment effects. Studies should also use appropriate methods of randomisation and allocation concealment, and employ procedures that allow for the blinding of participants, personnel, and outcome assessors, all of which are clearly described in the associated publications. Studies that employ analytic approaches that align with intention‐to‐treat principles (e.g. multilevel modelling) are required. They should report means and standard deviations for a range of outcomes that are consistent with trial registrations or protocol papers (Walker 2006). Consistency in the selection of outcome measures across studies assessing pharmacological interventions and other gambling treatments, such as psychological and self‐help interventions, would allow for meaningful comparisons across this growing evidence base. Given the limited available evidence base, future research would benefit from adopting measures that extend beyond gambling symptom severity, such as other indices of gambling behaviour, psychological functioning, and functional impairment. The insufficient number of studies available to evaluate reporting bias may also have increased the risk of deriving incorrect conclusions regarding the efficacy of these interventions, which is an avenue for further research.

This literature is at an early stage where it is appropriate to conduct trials in tightly controlled homogenous samples. It remains unclear, however, how these interventions would operate in more heterogeneous samples, suggesting the need for future pragmatic trials that relax the exclusion criteria (e.g. on the basis of psychiatric comorbidity). Further research is also required to identify for whom these treatments work best and the mechanisms underpinning such effects. For example, it remains unclear whether antidepressants work best for gamblers with comorbid depression, or whether mood stabilisers such as lithium work best for gamblers with comorbid bipolar disorder (Dowling 2016b). Moreover, given that psychological interventions are currently 'best practice' for the treatment of gambling problems, research would benefit from exploring the efficacy of pharmacotherapies in combination with these psychological interventions, to determine whether combined treatments produce better effects than single treatment options (Myserth 2011). Finally, it is also important for research to transition from efficacy to effectiveness studies, to determine the degree to which these pharmacological agents are effective in real‐world settings, with real‐world clients (e.g. Ward 2018). In this regard, researchers would benefit from using the PRagmatic Explanatory Continuum Indicator Summary (PRECIS‐2) tool to ensure that trials are designed for the purpose they are intended (Loudon 2015).

Summary of findings

Open in table viewer
Summary of findings 1. Antidepressants compared to placebo for the treatment of disordered and problem gambling

Antidepressants compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with antidepressants

Gambling symptom severity
assessed with: various clinician‐administered or self‐report measures
follow‐up: range 8 weeks to 24 weeks

The mean gambling symptom severity was 18.2 for the placebo group at post‐treatment.

SMD 0.32 lower
(0.74 lower to 0.09 higher)

‐

225
(5 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure
assessed with: various self‐report or author‐derived measures
follow‐up: range 8 weeks to 24 weeks

The mean gambling expenditure was 47.3 for the placebo group at post‐treatment.

SMD 0.27 lower
(0.6 lower to 0.06 higher)

‐

144
(3 RCTs)

⊕⊕⊝⊝
Low 1 3

Gambling frequency
assessed with: visual analogue scale
follow‐up: 24 weeks

The mean gambling frequency was 15.2 for the placebo group at post‐treatment.

SMD 0.08 lower
(0.59 lower to 0.42 higher)

‐

60
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Time spent gambling
assessed with: Timeline Follow Back
follow‐up: 10 weeks

The mean time spent gambling was 42.8 for the placebo group at post‐treatment.

SMD 0.17 lower
(0.8 lower to 0.46 higher)

‐

39
(1 RCT)

⊕⊕⊝⊝
Low 4

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 8 weeks to 16 weeks

The mean depressive symptoms was 3.6 for the placebo group at post‐treatment.

SMD 0.19 lower
(0.6 lower to 0.23 higher)

‐

90
(3 RCTs)

⊕⊕⊝⊝
Low 1 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 9 weeks

The mean anxiety symptoms was 3.8 for the placebo group at post‐treatment.

SMD 0.23 higher
(0.38 lower to 0.85 higher)

‐

41
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: range 10 weeks to 16 weeks

The mean functional impairment was 7.2 for the placebo group at post‐treatment.

SMD 0.15 lower
(0.53 lower to 0.22 higher)

‐

110
(2 RCTs)

⊕⊕⊝⊝
Low 1 3

Responder status
assessed with: improvement based on the Clinical Global Impression ‐ Improvement scale or abstinence from gambling
follow‐up: range 8 weeks to 24 weeks

Study population

RR 1.24
(0.93 to 1.66)

268
(6 RCTs)

⊕⊝⊝⊝
Very low 1 4 5

396 per 1000

491 per 1000
(368 to 657)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). 

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence 

High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. 

Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. 

Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.

Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded two levels for inconsistency: high heterogeneity as indicated by I2 and point estimates including both beneficial and harmful effects.
3Downgraded one level for imprecision: small sample size.
4Downgraded two levels for imprecision: small sample size and very wide CIs.
5Downgraded one level for inconsistency: minor heterogeneity, but point estimates include both beneficial and harmful effects.

Open in table viewer
Summary of findings 2. Opioid antagonists compared to placebo for the treatment of disordered and problem gambling

Opioid antagonists compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: opioid antagonists
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with opioid antagonists

Gambling symptom severity
assessed with: various clinician‐administered or self‐report measures
follow‐up: range 10 weeks to 16 weeks

The mean gambling symptom severity was 21.3 for the placebo group at post‐treatment.

SMD 0.46 lower
(0.74 lower to 0.19 lower)

‐

259
(3 RCTs)

⊕⊕⊝⊝
Low 1 2

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 18 weeks

The mean depressive symptoms was 9.1 for the placebo group at post‐treatment.

SMD 0.76 lower
(1.29 lower to 0.23 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 18 weeks

The mean anxiety symptoms was 9.6 for the placebo group at post‐treatment.

SMD 1.39 lower
(1.96 lower to 0.83 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: 18 weeks

The mean functional impairment was 8.4 for the placebo group at post‐treatment.

SMD 0.53 lower
(1.06 lower to 0.01 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Responder status
assessed with: improvement on various measures or gambling abstinence
follow‐up: range 10 weeks to 16 weeks

Study population

RR 1.65
(0.86 to 3.14)

562
(4 RCTs)

⊕⊝⊝⊝
Very low 1 2 4

402 per 1000

664 per 1000
(346 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for imprecision: small sample size.
3Downgraded two levels for imprecision: small sample size and very wide CIs.
4Downgraded two levels for inconsistency: substantial to considerable heterogeneity, and study‐specific estimates include both beneficial and harmful effects.

Open in table viewer
Summary of findings 3. Mood stabilisers compared to placebo for the treatment of disordered and problem gambling

Mood stabilisers compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: mood stabilisers
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with mood stabilisers

Gambling symptom severity
assessed with: Pathological Gambling ‐ Yale Brown Obsessive Compulsive Scale
follow‐up: range 10 weeks to 14 weeks

The mean gambling symptom severity was 20 for the placebo group at post‐treatment.

SMD 0.92 lower
(2.24 lower to 0.39 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3 4

Gambling expenditure
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean gambling expenditure was 317.9 for the placebo group at post‐treatment.

SMD 0.33 lower
(1.07 lower to 0.41 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Gambling frequency
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean gambling frequency was 3.4 for the placebo group at post‐treatment.

SMD 0.49 higher
(0.26 lower to 1.24 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Time spent gambling
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean time spent gambling was 149.3 for the placebo group at post‐treatment.

SMD 0.33 lower
(1.07 lower to 0.41 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Depressive symptoms
assessed with: various self‐report measures
follow‐up: range 10 weeks to 14 weeks

The mean depressive symptoms was 5.6 for the placebo group at post‐treatment.

SMD 0.15 lower
(1.14 lower to 0.83 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 3 4 5

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 10 weeks to 14 weeks

The mean anxiety symptoms was 6.1 for the placebo group at post‐treatment.

SMD 0.17 lower
(0.64 lower to 0.3 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 3 6

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: 14 weeks

The mean functional impairment was 10.5 for the placebo group at post‐treatment.

SMD 0.34 lower
(0.95 lower to 0.27 higher)

‐

42
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Responder status
assessed with: improvement based on the Clinical Global Impression ‐ Improvement scale
follow‐up: range 10 weeks to 14 weeks

Study population

RR 2.69
(1.14 to 6.32)

40
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

227 per 1000

611 per 1000
(259 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: mostly insufficient reporting with some trials at high risk of bias.
2Downgraded one level for inconsistency: moderate to substantial heterogeneity.
3Downgraded one level for indirectness: the population within this comparison had strict inclusion criteria relating to comorbid bipolar disorder.
4Downgraded two levels for imprecision: small sample size and very wide CIs.
5Downgraded two levels for inconsistency: moderate to substantial heterogeneity, and point estimates include both beneficial and harmful effects.
6Downgraded one level for imprecision: small sample size.

Open in table viewer
Summary of findings 4. Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling

Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: atypical antipsychotics
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with atypical antipsychotics

Gambling symptom severity
assessed with: various clinician‐administered measures
follow‐up: range 7 weeks to 13 weeks

The mean gambling symptom severity was 6.6 for the placebo group at post‐treatment.

SMD 0.59 lower
(1.1 lower to 0.08 lower)

‐

63
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean gambling expenditure was 42.0 for the placebo group at post‐treatment.

SMD 0.16 lower
(1.03 lower to 0.7 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling frequency
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean gambling frequency was 2.5 for the placebo group at post‐treatment.

SMD 0.04 lower
(0.9 lower to 0.83 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Time spent gambling
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean time spent gambling was 1.5 for the placebo group at post‐treatment.

SMD 0.26 lower
(1.13 lower to 0.6 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 7 weeks

The mean depressive symptoms was 3.5 for the placebo group at post‐treatment.

SMD 0.12 higher
(0.74 lower to 0.99 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 7 weeks

The mean anxiety symptoms was 2.8 for the placebo group at post‐treatment.

SMD 0.84 higher
(0.07 lower to 1.75 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: improvement based on the Clinical Global Impression scale
follow‐up: 13 weeks

Study population

RR 0.93
(0.62 to 1.40)

42
(1 RCT)

⊕⊕⊝⊝
Low 1 4

714 per 1000

664 per 1000
(443 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: mostly insufficient reporting with some trials at high risk of bias. 
2Downgraded one level for indirectness: the population within this comparison had strict inclusion criteria relating to gambling modality.
3Downgraded two levels for imprecision: small sample size and very wide CIs. 
4Downgraded one level for imprecision: small sample size.

Open in table viewer
Summary of findings 5. Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling

Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: opioid antagonists

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with opioid antagonists

Risk with antidepressants

Gambling symptom severity
assessed with: Clinical Global Impression ‐ Severity Scale
follow‐up: 12 weeks

The mean gambling symptom severity was 12.2 for the opioid antagonist group at post‐treatment.

SMD 0.08 lower
(0.86 lower to 0.71 higher)

‐

25
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 12 weeks to 2 years

The mean depressive symptoms was 7.8 for the opioid antagonist group at post‐treatment.

SMD 0.22 higher
(0.29 lower to 0.72 higher)

‐

62
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 12 weeks to 2 years

The mean anxiety symptoms was 8.2 for the opioid antagonist group at post‐treatment.

SMD 0.21 higher
(0.29 lower to 0.72 higher)

‐

62
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: gambling abstinence
follow‐up: 12 weeks

Study population

RR 1.01
(0.54 to 1.87)

36
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 4

526 per 1000

532 per 1000
(284 to 984)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.
4Downgraded one level for imprecision: small sample size.

Open in table viewer
Summary of findings 6. Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling

Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: mood stabilisers

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with mood stabilisers

Risk with antidepressants

Gambling symptom severity
assessed with: Pathological Gambling ‐ Yale Brown Obsessive Compulsive Scale
follow‐up: 12 weeks

The mean gambling symptom severity was 12.5 for the mood stabiliser group at post‐treatment.

SMD 0.07 higher
(0.64 lower to 0.77 higher)

‐

31
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 12 weeks to 2 years

The mean depressive symptoms was 8.6 for the mood stabiliser group at post‐treatment.

SMD 0.02 higher
(0.53 lower to 0.56 higher)

‐

58
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 12 weeks to 2 years

The mean anxiety symptoms was 10.4 for the mood stabiliser group at post‐treatment.

SMD 0.16 higher
(0.39 lower to 0.7 higher)

‐

58
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: gambling abstinence
follow‐up: 12 weeks

Study population

RR 0.63
(0.29 to 1.33)

31
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 4

600 per 1000

378 per 1000
(174 to 798)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.
4Downgraded one level for imprecision: small sample size.

Open in table viewer
Summary of findings 7. Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling

Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: opioid antagonists
Comparison: mood stabilisers

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with mood stabilisers

Risk with opioid antagonists

Gambling symptom severity ‐ not measured

‐

‐

‐

‐

‐

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 2 years

The mean depressive symptoms was 10.3 for the mood stabiliser group at post‐treatment.

SMD 0.71 lower
(1.61 lower to 0.2 higher)

‐

24
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 2 years

The mean anxiety symptoms was 4.8 for the mood stabiliser group at post‐treatment.

SMD 0.26 lower
(1.15 lower to 0.62 higher)

‐

24
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status ‐ not measured

‐

‐

‐

‐

‐

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.

Background

Description of the condition

Gambling is a highly accessible and common recreational activity in some countries, with up to 80% of adults gambling at least once per year (Calado 2016). Gambling statistics suggest that Australia has the highest global expenditure per capita, which is approximately twice as much as the average in other Westernised countries (The Economist 2018). However, the largest gambling market in terms of absolute losses is the United States, followed by China and Japan. Although the highest global expenditure is for casino gambling, online gaming is the industry’s fastest‐growing sector, accounting for over 10% of global gambling losses. High‐intensity electronic gambling machines (EGMs) contribute to large proportions of gambling losses in many countries (Queensland Government Statistician's Office 2019).

Gambling problems are characterised by impaired control and negative sequelae including financial harm, relationship dysfunction, emotional difficulties, health concerns, cultural harm, decreased work or academic performance, and criminality (Langham 2015). Financial problems, which are the most commonly reported gambling‐related harm by gamblers and their family members (Dowling 2014a; Langham 2015), include reduced financial capacity (e.g. reduced discretionary spending, erosion of savings), activities to manage short term cash‐flow issues (e.g. additional employment, additional credit, pawning, payday loans), and reduced expenditure on items with non‐immediate (e.g. insurance, repairs) and immediate (e.g. education, medical care, clothing, food) consequences (Langham 2015). Financial harms, such as inability to meet essential needs, loss of accommodation, loss of major assets, and bankruptcy, are often the 'tipping point' to seek assistance (Langham 2015). 

Gambling problems commonly co‐occur with other mental health conditions, including personality disorders, mood and anxiety disorders, and alcohol and substance use disorders (Dowling 2015a; Dowling 2015b; Lorains 2011). Gambling problems have also been consistently associated with both the perpetration and victimisation of intimate partner and family violence (Dowling 2014b; Dowling 2016a; Dowling 2018; Roberts 2018; Roberts 2020), as well as coercive control and economic intimate partner violence (Hing 2021; Langham 2015). Several risk factors for the development of gambling problems have been established, such as depression, substance use (e.g. cannabis use, tobacco use, frequency of alcohol use), number of gambling activities, peer antisocial behaviours and violence, personality traits (e.g. impulsivity, sensation seeking, poorly controlled temperament), and low academic performance (Dowling 2017). Male sex is also a consistent risk factor for the development of gambling problems (Dowling 2017), although the ratio of males‐to‐females of people with gambling problems is only approximately two‐to‐one in many jurisdictions (Merkouris 2016). Moreover, prevalence studies have indicated an increasing level of gambling participation amongst women, along with gradually increasing rates of gambling‐related harm (McCarthy 2019). Although speculative, these changes in women’s gambling and risk of harm have been attributed to various potential factors that include the ‘feminisation’ of some forms of gambling, as well as the increased targeting of women by the gambling industry in marketing and promotional campaigns (McCarthy 2019).

Definitions of the condition

The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM‐5) reclassified gambling disorder (formerly pathological gambling) from an impulse control disorder to an addiction and related disorder, alongside alcohol and substance use disorders (APA 2013). The DSM‐5 defines gambling disorder as persistent and recurrent problematic gambling behaviour in a 12‐month period which leads to substantial impairment or distress. The DSM‐5 also includes an exclusion criterion of a manic episode, as well as specifiers for course (episodic, persistent), remission (early, sustained) and severity (mild, moderate, severe). The International Statistical Classification of Diseases and Related Health Problems, 11th Revision (ICD‐11) recently classified gambling as a "disorder due to addictive behaviours" that is defined by impaired control over gambling, increasing priority given to gambling over other life interests and activities, and the continuation of escalation of gambling despite negative consequences (WHO 2018).

In psychiatric diagnostic settings, the term 'problem gambling' is often used to describe subclinical presentations of gambling disorder. In contrast, many jurisdictions have adopted a public health framework in which gambling is positioned within a whole of population approach that can inform policy for prevention and intervention practices (Korn 1999). Using this framework, gambling problems are conceptualised across a continuum of risk, ranging from no risk (i.e. gambling behaviour has no associated health or social problems) to extreme risk (i.e. gambling behaviour is associated with serious problems) (Korn 2003). In this context, the term 'problem gambling' has been used to describe any problems in restricting money and/or time spent on gambling that negatively affects the gambler, significant others, or the broader community (Neal 2005). For current purposes, the term 'problem gambling' will be used to refer to the severe and clinically significant end of the continuum of gambling‐related harm, inclusive of the diagnostic category of gambling disorder/pathological gambling, but excluding subclinical problems (which are not the indicated targets of pharmacological interventions).

Measuring or diagnosing the condition

Many self‐report measures have been developed to screen for problem gambling, with varying levels of classification accuracy. The first instrument developed was the South Oaks Gambling Screen (SOGS; Lesieur 1987), which was designed to screen for problem gambling in clinical settings. More recently, the Problem Gambling Severity Index (PGSI; Ferris 2001) has been the measurement tool of choice in population‐level research (Dellis 2014; Neal 2005; Williams 2014). Several diagnostic interviews are also available to confirm diagnostic status, such as the World Mental Health Composite International Diagnostic Interview (WMH‐CIDI; Kessler 2004), the Diagnostic Interview for Gambling Schedule (DIGS; Winters 2002), and the Structured Clinical Interview for Pathological Gambling (SCI‐PG; Grant 2004).

A framework for the minimum features of reporting gambling treatment efficacy, known as the Banff consensus, proposes that treatment outcome studies measure gambling behaviour (frequency, expenditure, and time), gambling harms (which can be complemented by additional measures of quality of life), and the processes of change (the hypothesised mechanisms of therapeutic change) (Walker 2006). More recently, measures intended to evaluate the efficacy of treatment on gambling symptom severity have been developed, including the Gambling Symptom Assessment Scale (G‐SAS; Kim 2009), the Gambling Treatment Outcome Monitoring System (GAMTOMS; Stinchfield 2007), the Gambling Follow‐up Scale (Galetti 2017), the Clinical Global Impression Scale (CGI; Hollander 1998; Kim 2001b), and the Pathological Gambling Modification of the Yale‐Brown Obsessive‐Compulsive Scale (PG‐YBOCS; Hollander 1998).

Prevalence of the condition

The estimated prevalence of problem gambling varies across jurisdictions due to many variants of survey methodology, as well as true differences in underlying rates that may reflect factors including the availability of different gambling products. In the last decade, the average 12‐month prevalence of problem gambling internationally has ranged from 0.1 to 5.8 per cent (Calado 2016). Generally, the highest standardised prevalence rates of problem gambling are observed in Asia; intermediate prevalence estimates are identified in North America and Australia; and the lowest rates tend to occur in Europe (Williams 2012).

Causal conceptions of the condition

No single conceptual model sufficiently accounts for the biological, dispositional, developmental, psychological, ecological, and commercial determinants of disordered or problem gambling. There is, however, a growing body of research indicating that both genetic and environmental factors are influential (Gyollai 2014; Xuan 2017). Meta‐analytic evidence of twin data has found moderate additive genetic (50%) and non‐shared environmental influences (50%) on the development of gambling problems, with the magnitude of genetic influences higher for adults than adolescents, and for male gambling than female gambling (Xuan 2017). These findings suggest that risk factors related to social environment, such as family and peer influences, make larger contributions to the development of gambling problems in adolescents and women. Broader contextual influences, which have been the subject of increasing attention, also include features of national regulatory environments (which largely determine the nature and availability of gambling activities) and the commercial practices of the gambling industry (e.g. marketing and advertising strategies targeting gambling attitudes and consumption intentions of adults and children) (Adams 2009; Deans 2017).

Multiple neurochemical systems and brain regions are implicated in the development of problem gambling. Gene association data have primarily reported the involvement of the dopaminergic system, hypothesised to contribute to reward and reinforcement, and the serotonergic system, hypothesised to contribute to the regulation of behavioural inhibition (Gyollai 2014). Involvement of other neurotransmitter systems, such as the norepinephrine, opioid, cortisol, and glutamate systems, have also been observed (Potenza 2013). Moreover, neuroimaging studies suggest that brain circuits, particularly reward‐related brain regions such as the ventral prefrontal, ventral striatal, and limbic regions, may play a role in the development of gambling problems (Potenza 2013).

Description of the intervention

The current review will assess evidence regarding pharmacological‐only interventions designed to reduce symptoms of disordered or problem gambling as described in Types of interventions. Pharmacological treatment options for gambling problems are generally classified into four main categories: (1) antidepressants, (2) opioid antagonists, (3) mood stabilisers (including anticonvulsants), and (4) atypical antipsychotics. Antidepressant interventions employed in the treatment of disordered or problem gambling include selective serotonin reuptake inhibitors (SSRIs), such as fluvoxamine, paroxetine, sertraline, and escitalopram. Studies have examined the efficacy of other antidepressant interventions, including norepinephrine‐dopamine reuptake inhibitors (NDRIs), such as bupropion; tricyclic antidepressants, such as clomipramine; and atypical antidepressants or serotonin‐norepinephrine‐dopamine reuptake inhibitors (SNDRIs), such as nefazodone. Opioid antagonists, such as naltrexone and nalmefene, are also commonly employed in the treatment of disordered or problem gambling. Other studies have examined the use of mood stabilisers, such as lithium, and anticonvulsants, such as topiramate, valproate, and carbamazepine; and atypical antipsychotics, such as olanzapine. Less commonly evaluated pharmacological agents include glutamatergic agents, such as N‐acetyl cysteine and memantine; atypical stimulants, such as modafinil; anti‐Parkinsonian medications, such as amantadine and tolcapone; and anticraving medications, such as baclofen and acamprosate.

Why it is important to do this review

Pharmacological interventions for disordered or problem gambling have been employed in clinical practice given limited resources for psychological interventions (Ward 2018). Whilst many reviews of the efficacy of pharmacological interventions for disordered/problem gambling or across the addictions (published in English) are available (Achab 2011; Asevedo 2014; Bartley 2013; Bullock 2013; Christensen 2018; Deepmala 2015; Goslar 2019; Grant 2014b; Hollander 2005; Kraus 2020; Labuzek 2014; Leung 2009; Lupi 2014; Minarini 2017; Mouaffak 2017; Pallesen 2007; Pettorruso 2014; Van den Brink 2012; Vasiliu 2017; Victorri‐Vigneau 2018), only a subset have employed systematic search strategies (Achab 2011; Asevedo 2014; Bartley 2013; Christensen 2018; Deepmala 2015; Goslar 2019; Grant 2014b; Kraus 2020; Lupi 2014; Minarini 2017; Mouaffak 2017; Pallesen 2007; Pettorruso 2014; Vasiliu 2017). Systematic reviews of high‐quality evidence are essential to provide guidance about the quality and effectiveness of different types of treatment. Only a small number of these systematic reviews have conducted quantitative syntheses of evidence using meta‐analyses (Bartley 2013; Goslar 2019; Mouaffak 2017; Pallesen 2007), with somewhat different conclusions.

Pallesen 2007,  the earliest of these studies, conducted a meta‐analysis on the post‐treatment effects of pharmacological interventions on combined gambling outcomes (including gambling symptom severity, cravings, expenditure, frequency, and time spent gambling) in both placebo‐controlled and uncontrolled studies. They found that pharmacological interventions were more effective than no treatment/placebo, yielding an overall effect size of d = 0.78, and that there were no differences in outcomes between antidepressants, opioid antagonists, and mood stabilisers/anticonvulsants. The authors concluded that pharmacological interventions may be an adequate treatment for gambling problems.

Bartley 2013 investigated the efficacy of pharmacological interventions using only randomised, double‐blind, placebo‐controlled trials. They found that opioid antagonists were associated with a small but significant decrease in symptom severity (in which effect sizes were pooled across gambling symptoms, expenditure, and frequency) compared to placebo at post‐treatment, with an effect size (standardised mean difference (SMD)) of 0.22. However, the authors reported that the efficacy of opiate antagonists was significantly associated with non‐adherence to intention‐to‐treat (ITT) analyses, the proportion of randomised participants, and earlier year of publication. They argue that the treatment effect for this class of pharmacological agents may be driven by several early trials that employed non‐ITT analyses. In a systematic review exploring the efficacy of naltrexone specifically in double‐blind randomised controlled trials, Mouaffak 2017 identified a similar, but non‐significant, post‐treatment effect size compared to control conditions on combined gambling outcomes (gambling symptom severity, gambling severity) (SMD −0.22), presumably because of the smaller‐sized sample included in the analysis. In their review, Bartley 2013 also note that other pharmacological agents, such as antidepressant medications (SSRIs, NDRIs) (SMD 0.18) and antipsychotic agents (olanzapine) (SMD 0.23), have similar effect sizes to opioid antagonists but fail to cross the threshold for statistical significance due to the smaller samples employed in these trials. They concluded that trial data included in their review provide limited support for the efficacy of any pharmacological agent in the treatment of gambling problems.

In a meta‐analysis of data from placebo‐controlled trials, Goslar 2019 found that pharmacological interventions were associated with medium to large and significant pre‐post improvements in gambling symptom severity (Hedge’s g = 0.41) and expenditure (Hedge’s g = 0.22), but not frequency (Hedge’s g = 0.11). Opioid antagonists produced a significant and medium effect size for the improvement of gambling symptom severity (Hedge’s g = 0.46), but not expenditure (Hedge’s g = 0.15) or frequency (Hedge’s g = −0.001). In contrast to the Bartley 2013 review, the analyses relating to opioid antagonists suggested between‐study homogeneity, but with no identifiable moderators influencing effect sizes. The difference in findings across reviews may be because this recent review differentiated between distinct aspects of gambling behaviour, rather than employing pooled effect sizes across multiple gambling indices (Goslar 2019). For medications with mood‐stabilising properties (e.g. lithium, topiramate, valproate, carbamazepine, olanzapine), significant and medium effect sizes were observed for gambling symptom severity (Hedge’s g = 0.53) and expenditure (Hedge’s g = 0.53), but not frequency (Hedge’s g = 0.31). Antidepressants (SSRIs, NDRIs, SNDRIs, and other antidepressants) did not produce significant effect sizes for any of the gambling indices: gambling symptom severity (Hedge’s g = 0.37), expenditure (Hedge’s g = 0.09), or frequency (Hedge’s g = 0.05). Despite these differences, effect sizes were not clearly moderated by type of pharmacological treatment. Moreover, effect sizes were not moderated by the type of data analysis or the methodological quality of studies. However, significantly larger effect sizes were found in gambling symptom severity for studies published most recently. The authors concluded that opioid antagonists and medications with mood‐stabilising properties demonstrated promising results in the treatment of gambling problems.

Whilst these meta‐analytic reviews provide information regarding the potential efficacy of pharmacological interventions, Pallesen 2007 pooled effect sizes across within‐group and controlled study designs, which is problematic given evidence that the magnitude of effect sizes is lower in controlled study designs, suggesting high rates of placebo response (Bartley 2013; Goslar 2019; Pallesen 2007). The influence of methodological characteristics implicated in risk of bias (as an overall quality score) on the magnitude of treatment effects and the comparative superiority of pharmacological intervention has only been explored using moderator analyses in one meta‐analytic review (Goslar 2019). Each available meta‐analytic review combined mood stabilisers, such as lithium; anticonvulsants, such as topiramate; and atypical antipsychotics, such as olanzapine, into different major categories of pharmacological interventions, making comparisons between reviews difficult. Specifically, Pallesen 2007 combined mood stabilisers and anticonvulsants; Bartley 2013 separated atypical antipsychotics from anticonvulsants; and Goslar 2019 combined mood stabilisers, anticonvulsants, and atypical antipsychotics into a category of medications with mood‐stabilising properties. Most of the available meta‐analytic reviews included studies that evaluated combined pharmacological and psychological treatments (Bartley 2013; Goslar 2019; Mouaffak 2017), which precludes conclusions about the efficacy of pharmacology‐only conditions. Finally, all of the available studies have evaluated the efficacy of pharmacological interventions on gambling‐related variables, such as symptom severity and gambling behaviours (expenditure, frequency, and time spent gambling), Goslar 2019, or on composite measures of these variables (Bartley 2013; Mouaffak 2017; Pallesen 2007). This is a limitation given that the Banff consensus, which outlines the minimum features of reporting efficacy of gambling treatments, highlights the importance of evaluating outcomes that extend beyond gambling behaviour (Walker 2006).

The current systematic review and meta‐analysis will provide an updated and expanded evidence base regarding the efficacy of any type of pharmacological‐only intervention for disordered or problem gambling. This review employed comprehensive systematic search strategies and quantitative syntheses of randomised controlled trials using meta‐analyses. Moreover, this synthesis systematically compared the efficacy of major categories of pharmacological interventions; conducted analyses to consider the effect of different methodological characteristics of studies associated with risk of bias on outcomes; and considered a broader range of outcome measures that extend beyond gambling symptom severity and behaviour. 

Objectives

The primary aims of the study were to:

  1. Examine the efficacy of major categories of pharmacological‐only interventions for disordered or problem gambling relative to placebo control conditions; and

  2. Examine the efficacy of these major categories relative to each other.

Secondary exploratory aims were to:

  1. Evaluate the clinical characteristics of participants that can explain variability in treatment effects; 

  2. Explore the methodological characteristics of studies associated with risk of bias and their influence on treatment effects; and

  3. Describe the tolerability and adverse events associated with each major category of intervention.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised trials comparing the efficacy of any category of pharmacological‐only treatment with a placebo or another category of pharmacological treatment. Cross‐over trials were eligible if random allocation to treatment sequence occurred; however, only results for the between‐group comparisons from the first treatment stage were utilised. Quasi‐randomised trials, in which an alternative form of treatment allocation was used (e.g. sequential allocation), were ineligible. We excluded studies if participant allocation into the randomised trial occurred after an open‐label phase, whereby all potential participants received the pharmacological intervention and only ‘responders’ were eligible for the randomisation phase. We did not use sample size and language of the report to determine eligibility.

Types of participants

Participant characteristics

Participants were males and females of any age and ethnicity.

Diagnosis

Participants were required to meet criteria for gambling disorder/pathological gambling or problem gambling using standardised diagnostic or assessment instruments. These included general clinical interviews based on DSM criteria (APA 2013) or structured clinical interviews, such as the Diagnostic Interview for Gambling Schedule (DIGS; Winters 2002) and the Structured Clinical Interview for Pathological Gambling (SCI‐PG: Grant 2004). We also included studies that employed validated self‐report measures, such as the Problem Gambling Severity Index (PGSI; Ferris 2001), the South Oaks Gambling Screen (SOGS; Lesieur 1987), and the National Opinion Research Centre DSM Screen for Gambling Problems (NODS; Gernstein 1999).

Comorbidities

There were no restrictions on the inclusion of studies utilising samples with comorbid psychiatric disorders. However, studies comprising samples of participants diagnosed with Parkinson’s disease were excluded from this review, as gambling problems in this population can be symptomatic of anti‐Parkinsonian medication (Heiden 2017).

Setting

There were no restrictions on the inclusion of studies based on study setting.

Types of interventions

Pharmacological interventions

We considered any type of pharmacological intervention specifically targeting the reduction of gambling symptom severity or behaviour. Several major categories of pharmacological interventions were identified a priori and were used to organise the review. Consistent with all of the available meta‐analyses, we decided to combine mood stabilisers and anticonvulsants into one major category, as both classes of drugs are commonly used to treat mood instability. Indeed, a number of anticonvulsants, including sodium valproate and carbamazepine, have long been recognised in national practice guidelines as mood stabilisers for the treatment or prevention of mood episodes in bipolar disorder (Ceron‐Litvoc 2009; Melvin 2008). Newer anticonvulsants, such as lamotrigine, have also been found to be effective in the treatment of bipolar depression (Calabrese 1999; Goodnick 2007). Topiramate has also been shown to be an effective add‐on treatment for bipolar depression as well as treatment‐resistant depression (McIntyre 2002; Mowla 2011); however, a 2016 Cochrane Review was inconclusive due to the inclusion of few high‐quality studies (Pigott 2016). In contrast, olanzapine was analysed separately, which is consistent with the majority of available meta‐analyses (Bartley 2013; Pallesen 2007). The final major categories of pharmacological interventions therefore included the following.

  • Antidepressants, including selective serotonin reuptake inhibitors (SSRIs; e.g. fluvoxamine, paroxetine, sertraline, escitalopram, citalopram, fluoxetine), serotonin and noradrenaline reuptake inhibitors (SNRIs; e.g. venlafaxine, desvenlafaxine, duloxetine), reversible inhibitors of monoamine oxidase (RIMAs; e.g. moclobemide), norepinephrine–dopamine reuptake inhibitors (NDRIs; e.g. bupropion), tricyclic antidepressants (TCAs; e.g. clomipramine, nortriptyline, dothiepin, imipramine, amitriptyline, desipramine, doxepin, protriptyline, trimipramine), noradrenaline and specific serotoninergic antidepressants (NASSAs) (also called tetracyclic antidepressants) (e.g. mirtazapine, mianserin), noradrenaline reuptake inhibitors (NARIs; e.g. reboxetine), monoamine oxidase inhibitors (MAOIs; e.g. tranylcypromine, phenelzine, isocarboxazid, selegiline), melatonergic antidepressants (e.g. agomelatine), serotonin modulators (e.g. vortioxetine), and atypical antidepressants or serotonin‐norepinephrine‐dopamine reuptake inhibitors (SNDRIs) (e.g. nefazodone).

  • Opioid antagonists, such as naltrexone, nalmefene, and naloxone.

  • Mood stabilisers, including mood stabilisers (e.g. lithium) and anticonvulsants (e.g. topiramate, valproate, carbamazepine, lamotrigine).

  • Atypical antipsychotics (e.g. olanzapine, quetiapine, risperidone, aripiprazole, ziprasidone, paliperidone, asenapine).

Interventions that were not considered

Studies in which pharmacological interventions were delivered in combination with a psychological therapy intended to reduce gambling symptom severity or behaviour were beyond the scope of this review.

Comparator interventions

We considered the following control or comparison conditions for this review.

  • Placebo: any placebo condition, including active and inert placebo conditions.

  • Comparator interventions: one of the aims of this review was to consider the comparative superiority of pharmacological interventions. We thus included comparisons between interventions comprising the major drug categories defined in this review.

Control and comparator interventions that were not considered

Comparisons within the same drug category defined in this review (e.g. comparison of an SSRI and a TCA) were out of the scope of this review. Future updates of this review will endeavour to consider interventions within drug categories as studies become available. We also did not consider comparisons between pharmacological interventions and psychological interventions.

Types of outcome measures

We considered one primary outcome measure and seven secondary outcome measures in the review. These included measures of gambling symptom severity, gambling behaviours (expenditure, frequency, time spent gambling), psychological functioning (depressive and anxiety symptoms), and functional impairment, as well as participants who were classified as having 'responded' to the intervention (responder status). These measures are generally consistent with the Banff consensus, which outlines the minimum features of reporting efficacy of gambling treatments (Walker 2006).

Primary outcomes

  • Reduction in severity of gambling symptoms: assessed using standardised patient‐ or clinician‐rated measurement tools. These included the Gambling Symptom Assessment Scale (G‐SAS; Kim 2009), the Clinical Global Impression – Severity Scale (CGI‐S; Hollander 1998; Kim 2001b), the South Oaks Gambling Screen (SOGS; Lesieur 1987), and the Yale Brown Obsessive Compulsive Scale adapted for Pathological Gambling (PG‐YBOCS; Pallanti 2005). When a study provided data on multiple measures, we only used one measure. We gave preference to multi‐item patient‐rated measures (e.g. G‐SAS), followed by multi‐item clinician‐rated measures (e.g. PG‐YBOCS) and single‐item clinician‐rated measures (e.g. CGI‐S).

Secondary outcomes

  • Reduction in gambling expenditure: assessed using instruments such as gambling diary records, Timeline Follow‐Back interviews (Sobell 1992), visual analogue scales (Tiffany 1993), or single‐item self‐report scales (e.g. amount spent or lost gambling in the past week or month).

  • Reduction in gambling frequency: assessed using instruments such as gambling diary records, Timeline Follow‐Back interviews (Sobell 1992), visual analogue scales (Tiffany 1993), or single‐item self‐report scales (e.g. number of sessions or days gambled in the past month).

  • Reduction in time spent gambling: assessed using instruments such as gambling diary records, Timeline Follow‐Back interviews (Sobell 1992), visual analogue scales (Tiffany 1993), or single‐item self‐report scales (e.g. number of minutes or hours gambled in the past month).

  • Reduction in depressive symptoms: assessed using standardised and validated measures (e.g. Hamilton Depression Rating Scale) (Hamilton 1960).

  • Reduction in anxiety symptoms: assessed using standardised and validated measures (e.g. Hamilton Anxiety Rating Scale) (Hamilton 1960).

  • Reduction in functional impairment: assessed using standardised and validated measures (e.g. Sheehan Disability Scale) (SDS; Sheehan 1983).

  • Responder status: assessed using any categorisation of gambling symptom severity or gambling behaviour (e.g. a score of improved on the CGI‐Improvement Scale (CGI‐I), gambling abstinence (based on gambling frequency), or positive endorsement on items of the Criteria for Control of Pathological Gambling Questionnaire (CCPGQ). When a study provided data on multiple measures, we used only one measure, with preference given to categorical data based on CGI‐I, which was the most commonly used measure.

Search methods for identification of studies

Cochrane Specialised Register (CCMDCTR)

The Cochrane Common Mental Disorders Group (CCMD) maintain two clinical trials registers, namely a references‐ and a studies‐based register, known as the CCMDCTR. The CCMDCTR‐References Register contains over 34,000 reports of trials in mood and anxiety disorders, eating disorders, self‐harm, and other mental health disorders. Over half (60%) of these references have been tagged to individual coded trials (based on the EU‐Psi coding manual), which are homed in the CCMDCTR‐Studies Register. Herein, records are linked between the two registers via unique Study ID tags. 

Reports of trials to be included in the CCMD’s registers are sourced from routine (weekly), generic searches of MEDLINE (1950 onwards), Embase (1974 onwards), and PsycINFO (1967 onwards); quarterly searches of the Cochrane Central Register of Controlled Trials (CENTRAL, all years); and review‐specific searches of additional databases. Reports of trials were also sourced from international trials registers, courtesy of the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (apps.who.int/trialsearch/), drug companies, and the handsearching of key journals, conference proceedings, and other (non‐Cochrane) systematic reviews and meta‐analyses. Details of CCMD’s generic search strategies can be found in the ‘Specialised Register’ section of the CCMD website.

Electronic searches

A CCMD Information Specialist searched the group specialised register (Appendix 1) together with the main biomedical databases (all years to 11 January 2022), using relevant subject headings (controlled vocabularies) and search syntax as appropriate for each resource (Appendix 2).

  1. Cochrane Common Mental Disorders Controlled Trials Register (CCMDCTR) (archived) (all years to 14 June 2016).

  2. Cochrane Central Register of Controlled Trials (CENTRAL; 2022, Issue 1), in the Cochrane Library (searched 12 January 2022). 

  3. Ovid MEDLINE (1946 to 11 January 2022).

  4. Ovid Embase (1974 to 11 January 2022).

  5. Ovid PsycINFO (all years to January Week 1 2022).

We also searched the international trials registries (to 12 January 2022) via the WHO ICTRP (www.who.int/ictrp/) and ClinicalTrials.gov (clinicaltrials.gov/) to identify unpublished or ongoing studies.

We applied no restrictions on date, language, or publication status.

Searching other resources

We manually searched the following journals to identify other relevant studies, as these journals were not indexed in any of the included databases (or specific years were not indexed).

  • Gambling Research (2003 to 2014).

  • Journal of Gambling Issues (2000 to 2006).

We manually searched the reference lists of all primary studies and narrative or systematic reviews to identify other potentially relevant studies.

Data collection and analysis

Selection of studies

To determine study eligibility, several members of our review team (ND, SM, SC) independently screened the title and abstract of every record retrieved, where available, with two review authors reviewing every record. We retrieved the full‐text articles for assessment if information was insufficient to determine eligibility, or suggested that the study:

  • included males and females of any age and ethnicity, who had a diagnosis of gambling disorder/pathological gambling or problem gambling as described above in Types of participants;

  • examined pharmacological interventions for gambling problems as described in Types of interventions; and

  • used a systematic process for random allocation of participants to groups.

We resolved study selection discordance through discussion (Higgins 2008a), with agreement reached in all cases.

Dealing with duplicate publications

We identified duplicate publications, listed them with the primary publication, and examined them for any relevant information not mentioned in the primary report. Where multiple publications reported on the same outcome (e.g. gambling behaviour), we used the publication that reported the most complete information (such as means, standard deviations (SDs), and n’s) for each meta‐analysis.

Data extraction and management

We employed double‐data extraction, whereby two researchers independently extracted information from all primary studies (SM, EO). Using a standard data extraction template, we extracted the following information from each primary study.

  • Study descriptive characteristics, including title, authors, year of publication, and country.

  • Study design features, including setting, intervention, control or comparison, eligibility criteria, and assessment of diagnostic criteria.

  • Characteristics of participants, including sample size, age, sex, and comorbidities.

  • Characteristics of the intervention(s) and control/comparison, including drug category, dose, frequency, duration, tolerance, and adverse effects.

  • Missing data strategies, including loss to follow‐up and ITT or completers‐only analysis.

  • Outcome measures, including all primary and secondary outcomes.

  • Results, including means, SDs, and frequency counts.

To evaluate the efficacy of each drug category, we planned several comparisons a priori, including the following.

  • Antidepressants versus placebo

  • Opioid antagonists versus placebo

  • Mood stabilisers versus placebo

  • Atypical antipsychotics versus placebo

  • Antidepressants versus opioid antagonists

  • Antidepressants versus mood stabilisers

  • Antidepressants versus atypical antipsychotics

  • Opioid antagonists versus mood stabilisers

  • Opioid antagonists versus atypical antipsychotics

  • Mood stabilisers versus atypical antipsychotics

Assessment of risk of bias in included studies

During data extraction, two researchers (SM, EO) independently assessed the risk of bias of each primary study, classifying each study as at low, high, or unclear risk of bias across the following key sources of bias in clinical trials (Higgins 2011; Westphal 2007).

  • Random sequence generation: quasi‐randomised trials (i.e. trials with a high risk of bias for random sequence generation given systematic methods of sequence generation (e.g. date of presentation)) were not eligible for inclusion in the review. As such, we classified studies either as having a low (e.g. random number table or computer‐generated randomisation) or unclear (i.e. insufficient detail provided) risk of bias.

  • Allocation concealment: appropriate allocation concealment is dependent on the use of an appropriate random sequence generation procedure (Schulz 2002). We classified studies as having a high risk of bias if allocation to intervention before random assignment was not adequately concealed. An example of adequate allocation concealment is the use of sequentially numbered, sealed, opaque envelopes.

  • Blinding of participants and personnel: we classified studies that failed to blind participants and personnel as having a high risk of bias. We classified studies that were stated to be ‘double‐blind’ with no specific information relating to who was blinded from treatment allocation as having an unclear risk of bias.

  • Blinding of outcome assessment: we classified studies that failed to blind outcome assessors as having a high risk of bias.

  • Incomplete outcome data: studies reporting missing data due to attrition generally reported either completers‐only analyses (i.e. only including data from participants who provided complete information) or ITT analyses (i.e. data from all participants were included via missing data strategies, such as those based on principles of maximum likelihood or multiple imputation, as well as older (and potentially biased) imputation strategies, such as mean imputation and last observation carried forward (LOCF)). We classified studies as having a high risk of bias if they violated any of the principles of ITT analyses: (1) keeping participants in the intervention groups to which they were randomised, regardless of the intervention they received; (2) measuring outcome data on all participants; and (3) including all randomised participants in the intervention (Higgins 2017).

  • Selective reporting: we classified studies as having a high risk of bias if: (1) a study protocol or trial registration was available, and listed outcomes or measures that were not reported in the results; or (2) outcomes were reported with insufficient detail to permit inclusion in the meta‐analyses. We necessarily classified studies where study protocols or trial registrations were not published as having an unclear risk of bias.

  • Other bias: pharmaceutical industry involvement, either through funding of the study or provision of medication, may suggest a conflict of interest which can impact on research findings. We thus classified studies as having a high risk of bias if: (1) the study was funded by a pharmaceutical company; and/or (2) a pharmaceutical company provided the medication for the study. We classified studies as having an unclear risk of bias if they did not report the source of the study funding.

Any disagreements arising from the assessment of risk of bias of studies were resolved through discussion (Higgins 2008a), with agreement reached in all cases. We considered each source of bias independently (Juni 1999). We also collated summary assessments for risk of bias for each outcome across studies using the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008a).

Measures of treatment effect

To evaluate continuous outcomes (e.g. reduction in gambling expenditure) of treatment effects, we used the standardised mean difference (SMD) and associated 95% confidence interval (CI). Because there are no recognised guidelines for interpreting the magnitude of SMD point estimates in studies of pharmacological gambling interventions, we used the Cohen 1992 benchmarks as a preliminary point of reference for effects that are small (SMD 0.20), medium (SMD 0.50), and large (SMD 0.80). To evaluate dichotomous outcomes (e.g. responder status) of treatment effect, we used the risk ratio (RR) and associated 95% CI. Oliver 2017 suggests analogous benchmarks for interpreting small (RR 1.22), medium (RR 1.86), and large (RR 3.00) effects for estimates of the RR.

Managing skewed data

We expected skewed distributions for the gambling behaviour outcomes (gambling expenditure, gambling frequency, and time spent gambling), which are frequently characterised by long tails reflecting small numbers of extreme high values (e.g. indicating high gambling losses). In primary studies, numerous strategies may be employed to address skewed data, such as a commonly used logarithmic transformation of raw data. Where transformations occur, the descriptive statistics reported in the primary study may be in original or geometric forms (based on log‐transformed data). Meta‐analyses involving skewed data can be challenging, given that: (a) results from untransformed and log‐transformed data cannot be synthesised (Deeks 2008); and (b) skewed distributions with long tails can also inflate estimates of variability and thus homogenise the SMD.

We managed skewed data by extracting available information on skew and strategies to address skew from each primary study. Although log transformed data were reported in some instances for measures of gambling behaviour, the untransformed statistics were most consistently available and were used initially in meta‐analyses. Where log transformed data was the only information available, we converted these to approximate raw statistics using methods outlined by Higgins (Higgins 2008b).

Unit of analysis issues

Studies with multiple treatment groups

Where studies reported comparisons involving multiple treatments across different drug categories defined in this review, we evaluated only one intervention within each meta‐analysis. We considered comparisons involving other treatment types (e.g. alternative categories of pharmacotherapy) from the same study in separate analyses. Where different treatment arms reflected the same drug across multiple doses or different drugs within the same category, but data were reported separately, we pooled data from these groups to provide a single overall mean and SD (Higgins 2008c).

Studies using cluster‐randomised designs

Where cluster‐randomised trials were identified (Higgins 2008c), we extracted the methods used to analyse the data. We used the inflated standard error approach to adjust standard errors for non‐independence of observations (Higgins 2008c). We extracted the degree of non‐independence, as reflected in the intra‐class correlation (ICC), to facilitate adjustments. We assumed the ICC to be 0.05 when it was not reported.

Studies using cross‐over designs

Where cross‐over trials were identified, in which participants were randomly allocated to treatment sequence, we only used the between‐group comparison from the first treatment stage in the analyses.

Dealing with missing data

Missing information on study design and results/statistics

Where information about results or statistics (e.g. means, SDs, effective sample size at each time point) was missing or inadequately reported in the primary study, we examined secondary or duplicate publications for the relevant information before making contact with study authors. If these strategies were unsuccessful, we attempted to obtain the missing statistics from other results (e.g. standard errors, t‐values, F‐statistics). We also obtained approximate estimates of certain statistics (e.g. means) from figures presented in publications. As a last resort, we excluded the study from the specific analysis.

Missing observations from primary studies due to attrition

We determined the consideration of 'completers only' versus ITT data by the information reported, for example if required statistics from the ITT analyses were not reported. Where both types of data were available, we gave preference to the ITT sample. The influence of using completers‐only data informed sensitivity analyses (see Sensitivity analysis).

Assessment of heterogeneity

Clinical homogeneity

The decision to conduct a meta‐analysis and pool results across studies was based, primarily, on the clinical homogeneity of the studies. For studies that were clinically heterogeneous or that presented insufficient information for pooling, we presented a narrative summary of the results.

Statistical homogeneity

We quantified statistical heterogeneity in effect size estimates using the I2 statistic, which represents the percentage of total variability across studies that is attributable to between‐study differences (Huedo‐Medina 2006). We also reported the Chi2 statistic and associated significance test (P value), but as this lacks statistical power (Deeks 2008), we placed greater emphasis on the I2. Despite arbitrary thresholds for I2, there are overlapping bands suggesting minor (0% to 40%), moderate (30% to 60%), substantial (50% to 90%), and considerable (75% to 100%) heterogeneity (Deeks 2008). We also qualified interpretation by evaluating the pattern of variability, and whether individual studies indicated beneficial, null, or harmful effects. Where there was strong evidence of true heterogeneity, we considered the pooled effect to a limited best estimate which we qualified through discussion of diversity.

Assessment of reporting biases

We searched multiple databases and trial registers to minimise reporting biases. We evaluated reporting bias via production of funnel plots when there were sufficient studies (k ≥ 10).  

Data synthesis

We conducted random‐effects meta‐analyses in Review Manager 5 to provide a weighted estimate of the effects of each intervention compared to placebo or another category of intervention (Review Manager 2020). We narratively summarised results of studies that were not included in the meta‐analysis due to heterogeneity or insufficient data (i.e. fewer than two studies).

Subgroup analysis and investigation of heterogeneity

In the context of statistical heterogeneity and sufficient numbers of studies, we planned to conduct subgroup analyses to examine factors that explain between‐study variability on the primary outcome measure (gambling symptom severity). These analyses considered potential differences in treatment effects according to participant and methodological characteristics, including:

  • medication type within category;

  • study eligibility based on psychiatric comorbidity (e.g. participants with major comorbidities excluded during participant selection compared to participants with comorbidities included);

  • year of publication (prior to 2008 compared to 2008 onwards);

  • diagnosis of disordered or problem gambling (self‐report compared to clinically administered);

  • outcome (self‐report compared to clinically administered);

  • multicentre trials compared to single‐centre trials;

  • participant sex (male compared to female).

We planned to calculate estimates of treatment effect and precision for each subgroup when at least 10 studies were available, with between‐group differences assumed when 95% CIs did not overlap (Hunter 2004). 

Sensitivity analysis

We conducted sensitivity analyses on the primary outcome (gambling symptom severity) for each comparison to examine methodological characteristics associated with risk of bias. The main characteristics considered are specified in Assessment of risk of bias in included studies. We conducted sensitivity analyses for every comparison across every risk of bias domain only with studies with low risk of bias. We narratively summarised results of sensitivity analyses with fewer than two studies at low risk of bias.

Tolerability and adverse events

The tolerability and adverse events associated with pharmacological interventions are described in Characteristics of included studies. We subjected these data to narrative synthesis.

Summary of findings and assessment of the certainty of the evidence

We created summary of findings tables for each comparison, including antidepressants versus placebo (summary of findings Table 1), opioid antagonists versus placebo (summary of findings Table 2), mood stabilisers versus placebo (summary of findings Table 3), atypical antipsychotics versus placebo (summary of findings Table 4), antidepressants versus opioid antagonists (summary of findings Table 5), antidepressants versus mood stabilisers (summary of findings Table 6), and opioid antagonists versus mood stabilisers (summary of findings Table 7). These tables include outcome‐specific information about the magnitude of the effect estimates and evaluations of the certainty of the evidence, which reflect the confidence in the estimated effect sizes as accurate and unbiased (Guyatt 2008). We evaluated the certainty of evidence using the GRADE approach (Guyatt 2008; Schunemann 2006), which classifies the overall certainty of evidence as high, moderate, low, or very low based on the following five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Results

Description of studies

Results of the search

A number of searches were conducted for this review, the latest of which was in January 2022. Over the years 5871 records were retrieved with 3508 duplicates removed. We screened 2363 records based on title and abstract, of which 2256 records were excluded and 107 records were carried forward to full‐text screen. A total of 17 primary studies, reported across 17 articles, met our inclusion criteria and were included in the review. A PRISMA flow diagram is shown in Figure 1.

Figure 1
Study flow diagram.

Study flow diagram.

We contacted seven authors for additional information regarding results or statistics during the review process. Two authors responded; one provided data for inclusion in the meta‐analysis (Black 2007), and one indicated that the data had been destroyed in line with university policy (Kim 2001a; Kim 2002).

Included studies

We included a total of 17 studies, published from 2000 onwards, in the review (Characteristics of included studies).

Design

All 17 studies were randomised trials, with one study utilising a cross‐over randomised trial design (Hollander 2005). Thirteen studies had a placebo control group, with three studies comparing multiple doses of the same medication with placebo control (Grant 2006a; Grant 2008; Grant 2010). The remaining three studies compared two or more different drug categories (Dannon 2005a; Dannon 2005b; Rosenberg 2013). Treatment duration ranged from 7 to 96 weeks (average of around 18 weeks). All studies reported collection of outcome data that was scheduled for the end of treatment, with only one study conducting a longer‐term post‐treatment follow‐up (Rosenberg 2013). We did not include these 24‐month follow‐up data in the current review.

Sample sizes

A total of 1193 participants were randomly allocated to groups. The average sample size was 68, ranging from 13, in Hollander 2000, to 233, in Grant 2010.

Setting

The majority of  studies were conducted in the United States (k = 12), followed by Israel (k = 3) and Spain (k = 2). Four trials involved multiple sites or centres (Berlin 2013; Grant 2003; Grant 2006a, Grant 2010). Eleven studies clearly reported that trials were conducted in outpatient settings. The remaining studies did not describe the trial setting.

Participants

All studies utilised adult samples, with a mean age of 42.71 years. The majority of studies included males and females (k = 13), with four trials including male‐only samples (Dannon 2005a; Dannon 2005b; Hollander 2000; Rosenberg 2013). Fong 2008 recruited pathological gamblers who reported a primary problem concerning video poker, and Hollander 2005 restricted their trial to pathological gamblers with a comorbid bipolar spectrum disorder. With the exception of Rosenberg 2013, for which exclusion criteria were not provided, all trials excluded participants with common comorbid mental health disorders. Several trials excluded individuals on the presence of any Axis I diagnosis (Berlin 2013; Blanco 2002; Dannon 2005a; Dannon 2005b; Fong 2008; Grant 2003; Grant 2006a; Grant 2010; Kim 2002), whereas other trials specified exclusions based on the presence of psychotic disorders (Black 2007; Grant 2008; Hollander 2000; Hollander 2005; McElroy 2008; Saiz‐Ruiz 2005), bipolar disorders (Black 2007; Grant 2008; Hollander 2000; Hollander 2005 (bipolar I only); McElroy 2008; Saiz‐Ruiz 2005), substance abuse or dependence (Black 2007; Grant 2008; Hollander 2000; Hollander 2005; Kim 2001a; McElroy 2008; Saiz‐Ruiz 2005), alcohol abuse or dependence (Kim 2001a; Saiz‐Ruiz 2005), personality disorders (Berlin 2013; Kim 2001a; McElroy 2008), or eating disorders (Black 2007). Several studies excluded individuals based on elevated baseline depressive (Berlin 2013; Black 2007; Grant 2003; Grant 2006a; Grant 2008; Kim 2002) and anxiety (Grant 2003; Grant 2006a; Grant 2008; Kim 2002) symptomatology.

All studies required that participants met criteria for gambling disorder/pathological gambling or problem gambling using the DSM criteria or validated assessment measures based on the DSM criteria, with some studies employing multiple measures. The most commonly used measure was the South Oaks Gambling Screen (SOGS) (score of 5 or more; k = 8) (Black 2007; Dannon 2005a; Dannon 2005b; Hollander 2000; Hollander 2005; Kim 2001a; Kim 2002; McElroy 2008), followed by a range of structured clinical interviews for pathological gambling (k = 7) (Berlin 2013; Fong 2008; Grant 2003; Grant 2008; Hollander 2005; McElroy 2008; Saiz‐Ruiz 2005), the National Opinion Research Centre DSM Screen for Gambling Problems (NODS) (score of 5 or more; Black 2007), and the Yale Brown Obsessive Compulsive Scale adapted for Pathological Gambling (PG‐YBOCS) (score of 21 or more; Grant 2010). Seven studies indicated that participants were required to meet DSM‐IV or DSM‐III‐R criteria for pathological gambling, but did not report the specific means of ascertaining diagnoses (Blanco 2002; Dannon 2005a; Dannon 2005b; Hollander 2000; Kim 2001a; Kim 2002; Rosenberg 2013). We identified no studies that comprised only a subset of participants who were eligible for inclusion in the review.

Interventions

In all studies except one (Hollander 2005), pharmacological treatment was delivered to target gambling specifically, rather than associated psychiatric comorbidity. Hollander 2005 delivered pharmacological treatment to assess reduction in both pathological gambling and bipolar spectrum disorder symptoms, although the primary outcomes related to gambling outcomes, whilst the secondary outcomes related to depressive symptoms, affective instability, and impulsivity severity.

Nine studies examined the efficacy of antidepressant pharmacotherapy. Seven evaluated SSRIs, specifically fluvoxamine (Blanco 2002; Dannon 2005b; Hollander 2000), paroxetine (Grant 2003; Kim 2002), sertraline (Saiz‐Ruiz 2005), and escitalopram (Rosenberg 2013). The remainder of the studies examined the efficacy of the NDRI bupropion (Black 2007; Dannon 2005a; Rosenberg 2013). Six studies investigated the efficacy of opioid antagonists, including naltrexone, Dannon 2005a; Grant 2008; Kim 2001a; Rosenberg 2013, and nalmefene (Grant 2006a; Grant 2010). Four studies examined the efficacy of drugs classified within the mood stabiliser category of this review. Specifically, one study examined sustained‐release lithium, a mood stabiliser (Hollander 2005), whilst the remaining studies examined topiramate, an anticonvulsant (Berlin 2013; Dannon 2005b; Rosenberg 2013). Finally, two studies examined an atypical antipsychotic, which was olanzapine (Fong 2008; McElroy 2008).

Outcomes

Sixteen of 17 studies reported measures of gambling symptom severity, with some studies using more than one measure. Fourteen studies used clinician‐administered measures, including the PG‐YBOCS, Berlin 2013; Black 2007; Dannon 2005a; Dannon 2005b; Grant 2003; Grant 2006a; Grant 2008; Grant 2010; Hollander 2000; Hollander 2005; McElroy 2008, and the Clinical Global Impression – Severity Scale (CGI‐S) (Fong 2008; Grant 2003; Grant 2008; McElroy 2008; Saiz‐Ruiz 2005). Nine studies used self‐report measures of gambling symptom severity, including the Gambling Symptom Assessment Scale (G‐SAS), Berlin 2013; Black 2007; Grant 2003; Grant 2006a; Grant 2008; Grant 2010; Kim 2001a; Kim 2002, and the SOGS (Saiz‐Ruiz 2005).

Six of 17 studies assessed gambling expenditure, using various self‐report measures and timeframes. One study used the Timeline Follow Back to measure money wagered per week (Black 2007), and one study used a 100‐millimetre visual analogue scale to measure amount gambled (Saiz‐Ruiz 2005). Blanco 2002 assessed money spent on gambling per week; Fong 2008 assessed money lost per day; Hollander 2005 assessed money lost per week; and Kim 2002 assessed reduction in money spent from baseline to post‐treatment.

Five studies assessed gambling frequency, also using various self‐report measures and timeframes. One study used a 100‐millimetre visual analogue scale (Saiz‐Ruiz 2005), and one study used take‐home diaries to assess weekly gambling frequency (McElroy 2008). Fong 2008 specified that gambling frequency was assessed as the number of days gambled per week, whereas Dannon 2005b only reported number of times gambled without specifying a timeframe. Hollander 2005 assessed gambling episodes per week.

Six studies assessed time spent gambling, again using various self‐report measures and timeframes. One study used the Timeline Follow Back to assess time spent gambling per week in minutes (Black 2007), and one study used take‐home dairies to assess hours spent gambling per week (McElroy 2008). The remaining three studies assessed number of hours spent gambling per week in the previous month (Blanco 2002); average time spent per day (Fong 2008); time spent per gambling in minutes (Dannon 2005a); and time spent per gambling episode in minutes (Hollander 2005).

Fourteen studies reported measures of depressive symptoms, with one study using more than one measure (Fong 2008). The most commonly used measure was the Hamilton Depression Rating Scale (k = 13; Black 2007; Dannon 2005a; Dannon 2005b; Fong 2008; Grant 2003; Grant 2006a; Grant 2008; Hollander 2000; Hollander 2005; Kim 2001a; Kim 2002; McElroy 2008; Rosenberg 2013), followed by the Beck Depression Inventory (Fong 2008), and the Montgomery‐Asberg Depression Rating Scale (Berlin 2013).

Eleven studies reported measures of anxiety symptoms, with all studies employing the Hamilton Anxiety Rating Scale (Berlin 2013; Dannon 2005a; Dannon 2005b; Fong 2008; Grant 2003; Grant 2006a; Grant 2008; Hollander 2005; Kim 2001a; Kim 2002; Rosenberg 2013).

Six studies reported measures of functional impairment, with all studies using the Sheehan Disability Scale (SDS) (Berlin 2013; Black 2007; Grant 2003; Grant 2006a; Grant 2008; Grant 2010).

Fifteen studies reported an assessment of responder status, using various definitions. Responder status was most commonly defined as a score of 1 (very much improved) or 2 (much improved) on the clinician‐rated CGI‐Improvement Scale (CGI‐I) (k = 8; Berlin 2013; Black 2007; Grant 2003; Grant 2006a; Hollander 2000; Hollander 2005; Kim 2001a; Kim 2002). Four studies classified participants as responders if they reported abstinence for a prespecified period of time (Blanco 2002; Dannon 2005a; Dannon 2005b; Grant 2008). One study classified participants as responders when prespecified items on the Criteria for Control of Pathological Gambling Questionnaire (CCPGQ) (i.e. ability to control gambling urges and substantial decrease in gambling problems) were positively endorsed (Saiz‐Ruiz 2005); one study required a reduction (≥ 35%) in scores on the PG‐YBOCS (Grant 2010); and one study classified participants as responders if there was a reduction in PG‐YBOCS scores and a score of 1 or 2 on the CGI‐I (McElroy 2008).

Excluded studies

See Characteristics of excluded studies.

Twenty‐three studies were identified as potentially eligible for this review but were excluded after closer examination. Ten of these were excluded because they evaluated a pharmacological intervention combined with a psychological intervention (Alho 2022; Echeburua 2011; Grant 2014a; Kovanen 2016a; Myrseth 2011; NCT00249431; NCT00326807; NCT01528007; NCT01843699; Toneatto 2009a), which was beyond the scope of this review. We excluded five studies because not all participants met the criteria for disordered/pathological or problem gambling (Grant 2017; McGrath 2013; Pallanti 2010; Porchet 2013; Potenza 2018). We excluded four studies as they compared the efficacy of two pharmacological interventions within the same drug category (Alho 2018; Dannon 2011; NCT03223896; Pallanti 2002). We excluded three studies because randomisation occurred after an initial open‐label phase, whereby only participants who responded positively to the initial pharmacological intervention were eligible to continue on the randomised controlled trial phase (Grant 2006b; Grant 2007; NCT00967005). Finally, we excluded one study that examined the efficacy of a pharmacological intervention for individuals with Parkinson’s disease and gambling problems (Thomas 2010).

Ongoing studies

See Characteristics of ongoing studies.

We identified one study as ongoing. This study is comparing milk thistle, also known as silymarin, with a placebo control group (NCT02337634). This eight‐week randomised controlled trial is utilising the PG‐YBOCS as the primary outcome, with secondary outcomes for potential inclusion in future versions of the review including the CGI‐I and CGI‐S scales, the G‐SAS, the Hamilton Depression Rating Scale, the Hamilton Anxiety Rating Scale, and the SDS.

Studies awaiting classification

See Characteristics of studies awaiting classification.

We identified two studies as awaiting classification. Both studies are comparing naltrexone with a placebo control group (NCT01057862; NCT04738773). One study is a 24‐week randomised controlled trial utilising the PG‐YBOCS as the primary outcome, with the G‐SAS as a secondary outcome (NCT01057862). The other study is a 12‐week controlled trial assessing gambling symptom severity, major comorbid psychiatric disorders and related psychopathology (impulsivity, craving, and locus of control), and eye‐tracking patterns prior to and one hour after administering the first dose, one week after, and at treatment completion (NCT04738773). Participants in both groups will receive psychoeducational sessions on weeks 2, 4, 6, and 8, during which they will have access to audio‐visual material and receive self‐help material.

Risk of bias in included studies

For details of the risk of bias judgements for each study, see Characteristics of included studies. A graphical representation of the overall risk of bias in included studies is presented in Figure 2 and Figure 3.

Figure 2
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Randomisation

We classified four studies as having a low risk of bias as they described an appropriate method of randomisation. Three of these conducted block randomisation in sizes of eight, using computer‐generated randomisation with no clinical information provided (Grant 2006a; Grant 2008; Grant 2010), and one study used a computer‐generated table of random numbers (Grant 2003). We classified the remaining studies as having an unclear risk of bias as they did not provide sufficient information about the randomisation procedure.

Allocation concealment

We classified only three studies as having a low risk of bias as  they described an appropriate method of allocation concealment. Black 2007 had a pharmacy department prepare the drug and look‐alike placebo and develop the randomisation scheme. Randomisation for all sites in Grant 2003 was completed by an operator with no clinical contact who kept the code during the trial. Finally, allocation concealment in McElroy 2008 involved the use of a research pharmacy performing the randomisation, packaging the study medication, and maintaining integrity of the blinded information. We classified the remaining 14 studies as having an unclear risk of bias as they did not indicate processes used to conceal allocation.

Blinding

Blinding of participants and personnel

We classified six studies as having a low risk of bias as they clearly and explicitly described the blinding of participants and personnel (Black 2007; Grant 2006a; Grant 2010; Hollander 2005; Kim 2001a; McElroy 2008). We assessed two studies as having a high risk of bias as they indicated that only the rater was blinded to treatment allocation  (Dannon 2005a; Dannon 2005b). Although the remaining nine studies were described as double‐blind trials, we classified them as having an unclear risk of bias as they did not explicitly describe the blinding of participants and personnel.

Blinding of outcome assessors

We classified six studies as having a low risk of bias as they clearly and explicitly described the blinding of the outcome assessors (Black 2007; Dannon 2005a; Dannon 2005b; Grant 2006a; Grant 2008; McElroy 2008). The remaining 11 studies were described as double‐blind trials, but were classified as having an unclear risk of bias as they did not explicitly describe that outcome assessors were blinded.

Incomplete outcome data

We classified eight studies as a having a low risk of bias as they utilised a missing data strategy that facilitated ITT analysis. Three of these studies used an LOCF approach (Hollander 2005; Kim 2002; Saiz‐Ruiz 2005). Three studies stated that an ITT approach was utilised and that all randomised participants with any available data were included in the analysis (Berlin 2013; Grant 2008; McElroy 2008). Finally, two studies used a linear mixed‐modelling approach, which also uses all the available data on each participant (Black 2007; Blanco 2002).

We classified three studies as having an unclear risk of bias as they did not provide sufficient detail on how missing data were handled in the analysis  (Dannon 2005a; Dannon 2005b; Rosenberg 2013). We classified the remaining six studies as having a high risk of bias as they only conducted a completers‐only analysis (Fong 2008; Hollander 2000), or did not include all randomised participants in the analyses (Grant 2003; Grant 2006a; Grant 2010; Kim 2001a). For example, Grant 2006a included participants with at least two postrandomisation observations in the analysis, and Kim 2001a included participants who had received the prespecified drug dosage for a minimum of two weeks.

Selective reporting

We classified 11 studies as having an unclear risk of bias as we could identify no protocol papers or trial registrations (Black 2007; Dannon 2005a; Dannon 2005b; Fong 2008; Grant 2006a; Grant 2008; Hollander 2000; Hollander 2005; Kim 2002; Rosenberg 2013; Saiz‐Ruiz 2005). We classified one study as having a low risk of bias as the reported outcomes coincided with the trial registration (Berlin 2013). We classified five studies as having a high risk of bias as the secondary outcomes listed in the methods section of the published paper were not employed in the analyses (Grant 2003; Grant 2010; Kim 2001a), or the main outcomes were not reported in sufficient detail to be included in the meta‐analysis of the current review (Blanco 2002; McElroy 2008). McElroy 2008 reported findings on secondary outcomes that were not specified in the trial registry.

Other potential sources of bias

Industry involvement

We classified 12 studies as having a high risk of bias due to the receipt of financial support from a pharmaceutical company for the conduct of the study (Berlin 2013; Blanco 2002; Dannon 2005b; Fong 2008; Grant 2003; Grant 2006a; Grant 2010; Hollander 2000; Hollander 2005; Kim 2002; McElroy 2008; Saiz‐Ruiz 2005). We classified three studies as having a low risk of bias as they reported no industry involvement (Black 2007; Grant 2008; Kim 2001a). Finally, we classified two studies as having an unclear risk of bias as they did not report where financial support for the study came from (Dannon 2005a; Rosenberg 2013).

Effects of interventions

See: Summary of findings 1 Antidepressants compared to placebo for the treatment of disordered and problem gambling; Summary of findings 2 Opioid antagonists compared to placebo for the treatment of disordered and problem gambling; Summary of findings 3 Mood stabilisers compared to placebo for the treatment of disordered and problem gambling; Summary of findings 4 Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling; Summary of findings 5 Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling; Summary of findings 6 Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling; Summary of findings 7 Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling

Comparison 1: Antidepressants versus placebo

This comparison comprised data from six studies including 268 participants evaluating the SSRIs fluvoxamine (Blanco 2002; Hollander 2000), paroxetine (Grant 2003; Kim 2002), and sertraline (Saiz‐Ruiz 2005), and the NDRI bupropion (Black 2007). See summary of findings Table 1.

Primary outcome
1.1 Gambling symptom severity

There was very low certainty evidence from five studies and 225 participants indicating that antidepressants were no more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: standardised mean difference (SMD) −0.32 (95% confidence interval (CI) −0.74 to 0.09) (Analysis 1.1). There was moderate statistical heterogeneity across studies (I2 = 54%), which included one very small trial that produced an effect favouring the placebo condition (Hollander 2000).

Secondary outcomes
1.2 Gambling expenditure

There was low certainty evidence from three studies and 144 participants indicating that antidepressants were no more effective than a placebo condition when conditions were compared on gambling expenditure at post‐treatment: SMD −0.27 (95% CI −0.60 to 0.06) (Analysis 1.2). Study results were generally consistent (I2 = 0%). 

We could not use data from one pilot double‐blind, placebo‐controlled study of 32 participants (Blanco 2002), which reported findings from a linear mixed‐effects analysis that could not be quantitatively synthesised through meta‐analyses. This study found that antidepressants (SSRIs) were no more effective than a placebo condition on gambling expenditure at the end of the six‐month treatment.

1.3 Gambling frequency

There was very low certainty evidence from one double‐blind, flexible‐dose, placebo‐controlled study of 60 participants (Saiz‐Ruiz 2005), which reported that antidepressants (SSRIs) were no more effective than placebo when conditions were compared on gambling frequency at the end of the six‐month treatment using LOCF analysis or mixed random regression models: SMD −0.08 (95% CI −0.59 to 0.42) (Analysis 1.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available. 

1.4 Time spent gambling

There was low certainty evidence from one randomised, double‐blind, placebo‐controlled, flexible‐dose study of 39 participants (Black 2007), which found that both the placebo and antidepressant (NDRI) conditions showed an approximately 9% weekly decrease in time spent gambling at the end of the 12‐week treatment, indicating that antidepressants were not superior to placebo: SMD −0.17 (95% CI −0.80 to 0.46) (Analysis 1.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

We could not use data from one pilot double‐blind, placebo‐controlled study of 32 participants (Blanco 2002), which reported findings from a linear mixed‐effects analysis that could not be quantitatively synthesised through meta‐analyses. This study found that antidepressants (SSRIs) were no more effective than a placebo condition on time spent gambling at the end of the six‐month treatment.

1.5 Depressive symptoms

There was low certainty evidence from three studies and 90 participants indicating that antidepressants were no more effective than placebo when conditions were compared on depressive symptoms at post‐treatment: SMD −0.19 (95% CI −0.60 to 0.23) (Analysis 1.5). Study results were generally consistent (I2 = 0%). 

1.6 Anxiety symptoms

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 41 participants (Kim 2002), which found that, despite low baseline levels of anxiety across all participants, there was an approximately 50% decrease in anxiety symptoms in both the placebo and antidepressant (SSRI) conditions at the end of the eight‐week treatment, indicating that antidepressants were not superior to placebo: SMD 0.23 (95% CI −0.38 to 0.85) (Analysis 1.6). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

1.7 Functional impairment

There was low certainty evidence from two studies and 110 participants indicating that antidepressants were no more effective than placebo when conditions were compared on functional impairment at post‐treatment: SMD −0.15 (95% CI −0.53 to 0.22) (Analysis 1.7). Study results were generally consistent (I2 = 0%). 

1.8 Responder status

There was very low certainty evidence from six studies and 268 participants indicating that antidepressants were no more effective than placebo when conditions were compared on responder status at post‐treatment: risk ratio (RR) 1.24 (95% CI 0.93 to 1.66) (Analysis 1.8). There were small amounts of statistical heterogeneity across studies (I2 = 15%). Notwithstanding that much of the between‐study variability could thus be accounted for by sampling error, one study, Black 2007, produced a point estimate favouring placebo (although the CI included values favouring the intervention).

Comparison 2: Opioid antagonists versus placebo

Primary outcome

This comparison comprised data from four studies including 562 participants evaluating nalmefene, Grant 2006a; Grant 2010, and naltrexone, Grant 2008; Kim 2001a. See summary of findings Table 2.

2.1 Gambling symptom severity

There was low certainty evidence from three studies and 259 participants indicating that opioid antagonists were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.46 (95% CI −0.74 to −0.19) (Analysis 2.1). The point estimate for the SMD indicated a medium effect and potential benefit of opioid antagonists. Study results were generally consistent (I2 = 0%), and all suggested beneficial effects of opioid antagonists. 

We could not use data from one multicentre, double‐blind, placebo‐controlled study of 233 participants (Grant 2010), which reported findings from a linear mixed model but did not provide sufficient information from which means and SDs could be calculated. This study found that neither dose of opioid antagonist (nalmefene; 20 mg/day, 40 mg/day) demonstrated superiority to a placebo condition on gambling symptom severity at post‐treatment.

Secondary outcomes
2.2 Gambling expenditure

None of the primary studies evaluated gambling expenditure, and no analyses could be conducted for this outcome.

2.3 Gambling frequency

None of the primary studies evaluated gambling frequency, and no analyses could be conducted for this outcome.

2.4 Time spent gambling

None of the primary studies evaluated time spent gambling, and no analyses could be conducted for this outcome.

2.5 Depressive symptoms

There was low certainty evidence from one double‐blind, placebo‐controlled study of 77 participants (Grant 2008), which found that, relative to the placebo condition, the opioid antagonist (naltrexone) condition produced a beneficial effect in mean depressive symptoms at the end of the 18‐week treatment, despite low symptom levels throughout the study in both conditions: SMD −0.76 (95% CI −1.29 to −0.23) (Analysis 2.2). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

2.6 Anxiety symptoms

There was low certainty evidence from one double‐blind, placebo‐controlled study of 77 participants (Grant 2008), which found that, relative to the placebo condition, the opioid antagonist (naltrexone) condition produced a beneficial effect in mean anxiety symptoms at the end of the 18‐week treatment, despite low anxiety levels throughout the study in both conditions: SMD −1.39 (95% CI −1.96 to −0.83) (Analysis 2.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

2.7 Functional impairment

There was low certainty evidence from one double‐blind, placebo‐controlled study of 77 participants (Grant 2008), which found that, relative to the placebo condition, the opioid antagonist (naltrexone) condition produced a beneficial effect in overall psychosocial functioning, as well as the three functional domains (work/school, social life/leisure activities, and family life/home responsibilities), at the end of the 18‐week treatment: SMD −0.53 (95% CI −1.06 to −0.01) (Analysis 2.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

We could not use data from one multicentre, randomised, dose‐ranging, double‐blind, placebo‐controlled study of 207 participants (Grant 2006a), as SDs could not be calculated from the information provided. However, this study identified a beneficial effect of the opioid antagonist (nalmefene) interventions (combined 25 mg, 50 mg, and 100 mg dose interventions) and a placebo condition on functional impairment at the end of the 16‐week treatment.

2.8 Responder status

There was very low certainty evidence from four studies and 562 participants indicating that opioid antagonists were no more effective than placebo when conditions were compared on responder status at post‐treatment: RR 1.65 (95% CI 0.86 to 3.14) (Analysis 2.5). There was evidence of high levels of statistical heterogeneity across studies (I2 = 82%), with point estimates from three studies favouring the intervention, and a fourth large study favouring placebo (although the CI included zero). 

Comparison 3: Mood stabilisers versus placebo

This comparison comprised data from two studies including 71 participants evaluating a mood stabiliser (sustained‐release lithium), Hollander 2005, and an anticonvulsant (topiramate), Berlin 2013. See summary of findings Table 3.

Primary outcome
3.1 Gambling symptom severity

There was very low certainty evidence from two studies and 71 participants indicating that mood stabilisers were no more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.92 (95% CI −2.24 to 0.39) (Analysis 3.1). Although there was high statistical heterogeneity across studies (I2 = 84%), point estimates for both trials were consistent with beneficial effects of mood stabilisers (although the CI for one study included values which could favour placebo).

Secondary outcomes
3.2 Gambling expenditure

There was very low certainty evidence from one randomised, double‐blind, placebo‐controlled study of 29 participants with gambling disorder and comorbid bipolar spectrum disorders (Hollander 2005), which found the mood stabiliser (sustained‐release lithium) was no more effective than placebo when conditions were compared on gambling expenditure at the end of the 10‐week treatment: SMD −0.33 (95% CI −1.07 to 0.41) (Analysis 3.2). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

3.3 Gambling frequency

There was very low certainty evidence frequency from one randomised, double‐blind, placebo‐controlled study of 29 participants with gambling disorder and comorbid bipolar spectrum disorders (Hollander 2005), which found the mood stabiliser (sustained‐release lithium) condition was no more effective than placebo when conditions were compared on gambling frequency at the end of the 10‐week treatment: SMD 0.49 (95% CI −0.26 to 1.24) (Analysis 3.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

3.4 Time spent gambling

There was very low certainty evidence from one randomised, double‐blind, placebo‐controlled study of 29 participants with gambling disorder and comorbid bipolar spectrum disorders (Hollander 2005), which found the mood stabiliser (sustained‐release lithium) was no more effective than placebo when conditions were compared on time spent gambling at the end of the 10‐week treatment: SMD −0.33, (95% CI −1.07 to 0.41) (Analysis 3.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

3.5 Depressive symptoms

There was very low certainty evidence from two studies and 71 participants indicting that mood stabilisers were no more effective than placebo when conditions were compared on depressive symptoms at post‐treatment: SMD −0.15 (95% CI −1.14 to 0.83) (Analysis 3.5). There was high statistical heterogeneity across studies (I2 = 75%), and the point estimate from one study of an anticonvulsant (topiramate), Berlin 2013, produced an effect size point estimate that favoured placebo.

3.6 Anxiety symptoms

There was very low certainty evidence from two studies and 71 participants indicating that mood stabilisers were no more effective than placebo when conditions were compared on anxiety symptoms at post‐treatment: SMD −0.17 (95% CI −0.64 to 0.30) (Analysis 3.6). Study results were generally consistent (I2 = 0%).

3.7 Functional impairment

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 42 participants (Berlin 2013), which found the anticonvulsant (topiramate) was no more effective than placebo when conditions were compared on functional impairment at the end of the 14‐week treatment: SMD −0.34 (95% CI −0.95 to 0.27) (Analysis 3.7). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

3.8 Responder status

There was very low certainty evidence from one randomised, double‐blind, placebo‐controlled study of 40 participants with gambling disorder and comorbid bipolar spectrum disorders (Hollander 2005), which found that the mood stabiliser (sustained‐release lithium) (92%) was superior to placebo (35%) when conditions were compared on responder status at the end of 10‐week treatment: RR 2.69 (95% CI 1.14 to 6.32) (Analysis 3.8). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Comparison 4: Atypical antipsychotics versus placebo

This comparison comprised data from two studies including 63 participants evaluating the atypical antipsychotic olanzapine (Fong 2008; McElroy 2008). See summary of findings Table 4.

Primary outcome
4.1 Gambling symptom severity

There was very low certainty evidence from two studies and 63 participants indicating that atypical antipsychotics (olanzapine) were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.59 (95% CI −1.10 to −0.08) (Analysis 4.1). The point estimate for the SMD indicated a medium effect and potential benefit of atypical antipsychotics. Study results were generally consistent (I2 = 0%).

Secondary outcomes
4.2 Gambling expenditure

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 21 participants with gambling problems on electronic gaming machines (Fong 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on gambling expenditure at the end of the seven‐week treatment: SMD −0.16 (95% CI −1.03 to 0.70) (Analysis 4.2), although there were reductions across both groups over time. It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

4.3 Gambling frequency

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 21 participants with gambling problems on electronic gaming machines (Fong 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on gambling frequency at the end of the seven‐week treatment: SMD −0.04 (95% CI −0.90 to 0.83) (Analysis 4.3), although there were reductions across both groups over time. It was not possible to conduct meta‐analyses or appraise statistical heterogeneity given that multiple studies were not available.

4.4 Time spent gambling

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 21 participants with gambling problems on electronic gaming machines (Fong 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on time spent gambling at the end of the seven‐week treatment: SMD −0.26 (95% CI −1.13 to 0.60) (Analysis 4.4), although there were reductions across both groups over time. It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

4.5 Depressive symptoms

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 21 participants with gambling problems on electronic gaming machines (Fong 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on depressive symptoms at the end of the seven‐week treatment: SMD 0.12 (95% CI −0.74 to 0.99) (Analysis 4.5). The point estimate for the effect size favoured the placebo condition. It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

4.6 Anxiety symptoms

There was very low certainty evidence from one double‐blind, placebo‐controlled study of 21 participants with gambling problems on electronic gaming machines (Fong 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on anxiety symptoms at the end of the seven‐week treatment: SMD 0.84 (95% CI −0.07 to 1.75) (Analysis 4.6). The point estimate for the effect size favoured the placebo condition. It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

4.7 Functional impairment

None of the primary studies evaluated functional impairment, and no analyses could be conducted for this outcome.

4.8 Responder status

There was low certainty evidence from one outpatient, randomised, double‐blind, parallel‐group, flexible‐dose study of 42 participants (McElroy 2008), which found that atypical antipsychotics (olanzapine) were no more effective than placebo when conditions were compared on responder status at the end of the 12‐week treatment when both ITT and completer analyses were employed: RR 0.93 (95% CI 0.62 to 1.40) (Analysis 4.7). When defined as the mean final CGI‐I score at endpoint using the ITT sample, 66.7% of the olanzapine group and 71.4% of the placebo group were categorised as treatment responders. It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Comparison 5: Antidepressants versus opioid antagonists

This comparison comprised data from two studies including 62 participants, one comparing the NDRI bupropion with naltrexone (Dannon 2005a), and the other comparing the combined effects of the SSRI escitalopram and the NDRI bupropion with naltrexone (Rosenberg 2013). See summary of findings Table 5.

Primary outcome
5.1 Gambling symptom severity

There was very low certainty evidence from one study of 25 male treatment completers (Dannon 2005a), which found that antidepressants (sustained‐release NDRI) were no more effective than opioid antagonists (naltrexone) when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.08 (95% CI −0.86 to 0.71) (Analysis 5.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Secondary outcomes
5.2 Gambling expenditure

None of the primary studies evaluated gambling expenditure, and no analyses could be conducted for this outcome.

5.3 Gambling frequency

None of the primary studies evaluated gambling frequency, and no analyses could be conducted for this outcome.

5.4 Time spent gambling

None of the primary studies evaluated time spent gambling, and no analyses could be conducted for this outcome.

5.5 Depressive symptoms

There was very low certainty evidence from two studies and 62 participants indicating that antidepressants were no more effective than opioid antagonists when conditions were compared on depressive symptoms at post‐treatment: SMD 0.22 (95% CI −0.29 to 0.72) (Analysis 5.2). Study results were generally consistent (I2 = 0%). 

5.6 Anxiety symptoms

There was very low certainty evidence from two studies and 62 participants indicating that antidepressants were no more effective than opioid antagonists when conditions were compared on anxiety symptoms at post‐treatment: SMD 0.21 (95% CI −0.29 to 0.72) (Analysis 5.3). Study results were generally consistent (I2 = 0%). 

5.7 Functional impairment

None of the primary studies evaluated functional impairment, and no analyses could be conducted for this outcome.

5.8 Responder status

There was very low certainty evidence from one study of 36 male participants (Dannon 2005a), which demonstrated that approximately three‐quarters of the 25 treatment completers were considered full responders (i.e. improved gambling symptom severity and no gambling for two weeks) in the antidepressant (NDRI) condition (9 of 12 completers; 9 of 17 ITT sample) and the opioid antagonist (naltrexone) condition (10 of 13 completers; 10 of 19 ITT sample), with the remaining completers considered partial responders, at the end of the 12‐week treatment: RR 1.01 (95% CI 0.54 to 1.87) (Analysis 5.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Comparison 6: Antidepressants versus mood stabilisers

This comparison comprised data from two studies including 58 participants, one comparing the SSRI fluvoxamine with the anticonvulsant topiramate (Dannon 2005b), and one comparing the combined effects of the SSRI escitalopram and the NDRI bupropion with the anticonvulsant topiramate (Rosenberg 2013). See summary of findings Table 6.

Primary outcome
6.1 Gambling symptom severity

There was very low certainty evidence from one study of 31 male participants  (Dannon 2005b), which found that antidepressants (SSRIs) were no more effective than mood stabilisers (topiramate) when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD 0.07 (95% CI −0.64 to 0.77) (Analysis 6.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

6.2 Gambling expenditure

None of the primary studies evaluated gambling expenditure, and no analyses could be conducted for this outcome.

6.3 Gambling frequency

None of the primary studies evaluated gambling frequency, and no analyses could be conducted for this outcome.

6.4 Time spent gambling

None of the primary studies evaluated time spent gambling, and no analyses could be conducted for this outcome.

6.5 Depressive symptoms

There was very low certainty evidence from two studies and 58 participants indicating that antidepressants were no more effective than mood stabilisers when conditions were compared on depressive symptoms at post‐treatment: SMD 0.02 (95% CI −0.53 to 0.56) (Analysis 6.2). Study results were generally consistent (I2 = 0%), although the point estimates from individual studies suggested contrasting small effects which favoured antidepressants, Rosenberg 2013, and mood stabilisers, Dannon 2005b, respectively. 

6.6 Anxiety symptoms

There was very low certainty evidence from two studies and 58 participants indicating that antidepressants were no more effective than mood stabilisers when conditions were compared on anxiety symptoms at post‐treatment: SMD 0.16 (95% CI −0.39 to 0.70) (Analysis 6.3). Study results were generally consistent (I2 = 0%). 

6.7 Functional impairment

None of the primary studies evaluated functional impairment, and no analyses could be conducted for this outcome.

6.8 Responder status

There was very low certainty evidence from one study of 31 male participants (Dannon 2005b), which found that, of the 20 treatment completers, 75% were considered full responders (i.e. improved symptom severity and no gambling for two weeks) in the mood stabiliser (topiramate) condition (9 of 12 completers; 9 of 15 ITT sample) and the antidepressant (SSRI) condition (6 of 8 completers; 6 of 16 ITT sample) at the end of the 12‐week treatment, with the remaining completers considered partially remitted: RR 0.63 (95% CI 0.29 to 1.33) (Analysis 6.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Comparison 7: Antidepressants versus atypical antipsychotics

None of the primary studies compared antidepressants with atypical antipsychotics, and no analyses could be conducted for this comparison. 

Comparison 8: Opioid antagonists versus mood stabilisers

This comparison comprised data from one study including 24 participants (Rosenberg 2013), which compared naltrexone with the anticonvulsant topiramate. See summary of findings Table 7.

Primary outcome
8.1 Gambling symptom severity

None of the primary studies evaluated gambling symptom severity, and no analyses could be conducted for this outcome.

Secondary outcomes
8.2 Gambling expenditure

None of the primary studies evaluated gambling expenditure, and no analyses could be conducted for this outcome.

8.3 Gambling frequency

None of the primary studies evaluated gambling frequency, and no analyses could be conducted for this outcome.

8.4 Time spent gambling

None of the primary studies evaluated time spent gambling, and no analyses could be conducted for this outcome.

8.5 Depressive symptoms

There was very low certainty evidence from one long‐term study of 24 participants who received treatment for two years with an additional two‐year follow‐up without medication (Rosenberg 2013). There was a demonstrated improvement in depressive symptoms across the opioid antagonist (naltrexone) condition at the end of the study period, but not the mood stabiliser (topiramate) condition: SMD −0.71 (95% CI −1.61 to 0.20) (Analysis 8.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

8.6 Anxiety symptoms

There was very low certainty evidence from one long‐term study of 24 participants who received treatment for two years with an additional two‐year follow‐up without medication (Rosenberg 2013). There was a demonstrated improvement across both the opioid antagonist (naltrexone) and mood stabiliser (topiramate) conditions at the end of the study period. At the end of the four‐year study period, lower mean anxiety symptoms were observed in the opioid antagonist condition relative to the mood stabiliser condition: SMD −0.26 (95% CI −1.15 to 0.62) (Analysis 8.2). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

8.7 Functional impairment

None of the primary studies evaluated functional impairment, and no analyses could be conducted for this outcome.

8.8 Responder status

None of the primary studies evaluated responder status, and no analyses could be conducted for this outcome.

Comparison 9: Opioid antagonists versus atypical antipsychotics

None of the primary studies compared opioid antagonists with atypical antipsychotics, and no analyses could be conducted for this comparison. 

Comparison 10: Mood stabilisers versus atypical antipsychotics

None of the primary studies compared mood stabilisers with atypical antipsychotics, and no analyses could be conducted for this comparison. 

Subgroup analyses

There were insufficient studies available to conduct subgroup analyses.

Sensitivity analyses

We conducted sensitivity analyses for the primary outcome of gambling symptom severity across the four intervention categories (antidepressants, opioid antagonists, mood stabilisers, and atypical antipsychotics) when compared with placebo. These were conducted separately for each risk of bias indicator, whereby studies with a high or unclear risk of bias were excluded from the analyses to examine inferences supported by studies characterised by low risk of bias only. There were very few studies addressing other comparisons that were assessed as at low risk of bias.

Comparison 11: Antidepressants versus placebo

Random sequence generation:  There was evidence from only one study assessed as at low risk of bias for random sequence generation, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a multicentre, randomised, double‐blind, flexible‐dose, placebo‐controlled trial of 71 participants (Grant 2003), antidepressants (SSRIs) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 16‐week treatment: SMD −0.10 (95% CI −0.56 to 0.37) (Analysis 11.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Allocation concealment:  There was evidence from only two studies (110 participants) assessed as at low risk of bias for allocation concealment. The evidence indicated that antidepressants were no more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.07 (95% CI −0.45 to 0.30) (Analysis 11.2). Study results were generally consistent (I2 = 0%). 

Blinding of participants/personnel:  There was evidence from only study assessed as at low risk of bias for blinding of participants/personnel, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a randomised, double‐blind, placebo‐controlled, flexible‐dose study of 39 participants (Black 2007), antidepressants (NDRIs) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.02 (95% CI −0.65 to 0.61) (Analysis 11.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Blinding of outcome assessors:  There was evidence from only one study assessed as at low risk of bias for blinding of outcome assessors, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a randomised, double‐blind, placebo‐controlled, flexible‐dose study of 39 participants (Black 2007), antidepressants (NDRIs) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.02 (95% CI −0.65 to 0.61) (Analysis 11.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Incomplete outcome data:  There was evidence from three studies (144 participants) assessed as at low risk of bias for incomplete outcome data. The evidence indicated that antidepressants were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.52 (95% CI −1.04 to −0.01) (Analysis 11.5). The point estimate for the SMD suggested a medium effect and potential benefit of antidepressants. There was evidence of moderate statistical heterogeneity across studies (I2 = 57%), with point estimates from two studies suggesting beneficial effects of antidepressants (Kim 2002; Saiz‐Ruiz 2005), and a third study suggesting an effect approaching zero (Black 2007).

Selective reporting:  No studies comparing antidepressants with placebo on gambling symptom severity were assessed as at low risk of bias for selective reporting, therefore no analyses could be conducted for risk of bias domain.

Other bias – industry funding:  There was evidence from only study assessed as at low risk of bias for other bias (industry funding), and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a randomised, double‐blind, placebo‐controlled, flexible‐dose study of 39 participants (Black 2007), antidepressants (NDRIs) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.02 (95% CI −0.65 to 0.61) (Analysis 11.6). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Comparison 12: Opioid antagonists versus placebo

Random sequence generation:  There was evidence from only two studies (223 participants) assessed as at low risk of bias for random sequence generation. The evidence indicated that opioid antagonists were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.49 (95% CI −0.79 to −0.20) (Analysis 12.1). The point estimate for the SMD suggested a medium effect and potential benefit of opioid antagonists. Study results were generally consistent (I2 = 0%). 

Allocation concealment:  No studies comparing opioid antagonists with placebo on gambling symptom severity were assessed as at low risk of bias for allocation concealment, therefore no analyses could be conducted for risk of bias domain.

Blinding of participants/personnel: There was evidence from only two studies (182 participants) assessed as at low risk of bias for blinding of participants/personnel. The evidence indicated that opioid antagonists were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.43 (95% CI −0.74 to −0.11) (Analysis 12.2). The point estimate for the SMD suggested a medium effect and potential benefit of opioid antagonists. Study results were generally consistent (I2 = 0%). 

Blinding of outcome assessors:  There was evidence from only two studies (223 participants) assessed as at low risk of bias for blinding of outcome assessors. The evidence indicated that opioid antagonists were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.49 (95% CI −0.79 to −0.20) (Analysis 12.3). The point estimate for the SMD suggested a medium effect and potential benefit of opioid antagonists. Study results were generally consistent (I2 = 0%). 

Incomplete outcome data: There was only one study assessed as at low risk of bias for incomplete outcome data, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a double‐blind, placebo‐controlled study of 77 participants (Grant 2008), opioid antagonists (naltrexone) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 18‐week treatment: SMD −0.57 (95% CI −1.10 to −0.04) (Analysis 12.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Selective reporting:  No studies comparing opioid antagonists with placebo on gambling symptom severity were assessed as at low risk of bias for selective reporting, therefore no analyses could be conducted for risk of bias domain.

Other bias – industry funding:  There was evidence from only two studies (113 participants) assessed as at low risk of bias for other bias (industry funding). The evidence indicated that opioid antagonists were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.47 (95% CI −0.89 to −0.06) (Analysis 12.5). The point estimate for the SMD suggested a medium effect and potential benefit of opioid antagonists. Study results were generally consistent (I2 = 0%). 

Comparison 13: Mood stabilisers versus placebo

Random sequence generation:  No studies comparing mood stabilisers with placebo on gambling symptom severity were assessed as at low risk of bias for random sequence generation, therefore no analyses could be conducted for risk of bias domain.

Allocation concealment: No studies comparing mood stabilisers with placebo on gambling symptom severity were assessed as at low risk of bias for allocation concealment, therefore no analyses could be conducted for risk of bias domain.

Blinding of participants/personnel:  There was only one study assessed as at low risk of bias for blinding of participants/personnel, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a randomised, double‐blind, placebo‐controlled study of 29 participants with gambling disorder and comorbid bipolar spectrum disorders (Hollander 2005), mood stabilisers (sustained‐release lithium) were significantly more effective than placebo when conditions were compared on gambling symptom severity at the end of the 10‐week treatment: SMD −1.63 (95% CI −2.50 to −0.77) (Analysis 13.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Blinding of outcome assessors: No studies comparing mood stabilisers with placebo on gambling symptom severity were assessed as at low risk of bias for blinding of outcome assessors, therefore no analyses could be conducted for risk of bias domain.

Incomplete outcome data: There was evidence from two studies (71 participants) assessed as at low risk of bias for incomplete outcome data. The evidence indicated that mood stabilisers were more effective than placebo when conditions were compared on gambling symptom severity at post‐treatment: SMD −0.92 (95% CI −2.24 to 0.39) (Analysis 13.2). The point estimate for the SMD exceeded a large effect and potential benefit of mood stabilisers. There was evidence of large amounts of statistical heterogeneity across studies (I2 = 84%), although point estimates from all studies suggested beneficial effects of mood stabilisers. 

Selective reporting: There was only one study assessed as at low risk of bias for selective reporting, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In a double‐blind, placebo‐controlled study of 42 participants (Berlin 2013), mood stabilisers (topiramate) were no more effective than placebo when conditions were compared on gambling symptom severity at the end of the 14‐week treatment: SMD −0.29 (95% CI −0.90 to 0.32) (Analysis 13.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Other bias – industry funding:  No studies comparing mood stabilisers with placebo on gambling symptom severity were assessed as at low risk of bias for other bias (industry funding), therefore no analyses could be conducted for risk of bias domain.

Comparison 14: Atypical antipsychotics versus placebo

Random sequence generation: No studies comparing atypical antipsychotics with placebo on gambling symptom severity were assessed as at low risk of bias for random sequence generation, therefore no analyses could be conducted for risk of bias domain.

Allocation concealment: There was only one study assessed as at low risk of bias for allocation concealment, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In an outpatient, randomised, double‐blind, parallel‐group, flexible‐dose study of 42 participants (McElroy 2008), atypical antipsychotics (olanzapine) were more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.64 (95% CI −1.26 to −0.02) (Analysis 14.1). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Blinding of participants/personnel: There was only one study assessed as at low risk of bias for blinding of participants/personnel, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In an outpatient, randomised, double‐blind, parallel‐group, flexible‐dose study of 42 participants (McElroy 2008), atypical antipsychotics (olanzapine) were more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.64 (95% CI −1.26 to −0.02) (Analysis 14.2). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Blinding of outcome assessors: There was only one study assessed as at low risk of bias for blinding of outcome assessors, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In an outpatient, randomised, double‐blind, parallel‐group, flexible‐dose study of 42 participants(McElroy 2008), atypical antipsychotics (olanzapine) were more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.64 (95% CI −1.26 to −0.02) (Analysis 14.3). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Incomplete outcome data: There was only one study assessed as at low risk of bias for incomplete outcome data, and findings should be interpreted considering the scarcity of trials at low risk of bias for this domain. In an outpatient, randomised, double‐blind, parallel‐group, flexible‐dose study of 42 participants (McElroy 2008), atypical antipsychotics (olanzapine) were more effective than placebo when conditions were compared on gambling symptom severity at the end of the 12‐week treatment: SMD −0.64 (95% CI −1.26 to −0.02) (Analysis 14.4). It was not possible to conduct a meta‐analysis or appraise statistical heterogeneity given that multiple studies were not available.

Selective reporting: No studies comparing atypical antipsychotics with placebo on gambling symptom severity were assessed as at low risk of bias for selective reporting, therefore no analyses could be conducted for risk of bias domain.

Other bias – industry funding: No studies comparing atypical antipsychotics with placebo on gambling symptom severity were assessed as at low risk of bias for other bias (industry funding), therefore no analyses could be conducted for risk of bias domain.

Reporting bias

There were insufficient studies (k < 10) available to evaluate reporting bias.

Tolerability and adverse events

Antidepressants

The nine studies that evaluated the efficacy of an antidepressant provided varying levels of detail regarding the type and number of adverse effects experienced (Black 2007; Blanco 2002; Dannon 2005a; Dannon 2005b; Grant 2003; Hollander 2000; Kim 2002; Rosenberg 2013; Saiz‐Ruiz 2005). Five studies reported the most common adverse effects experienced by participants, with headaches (Blanco 2002; Grant 2003; Hollander 2000; Kim 2002; Saiz‐Ruiz 2005), nausea (Blanco 2002; Grant 2003; Hollander 2000; Kim 2002), diarrhoea/gastrointestinal issues (Blanco 2002; Hollander 2000; Saiz‐Ruiz 2005), insomnia (Blanco 2002; Hollander 2000; Saiz‐Ruiz 2005), dizziness/lightheadedness (Blanco 2002; Hollander 2000; Saiz‐Ruiz 2005), and dry mouth (Grant 2003; Hollander 2000) reported consistently in more than one study. Three studies reported adverse effects for dropouts only (Dannon 2005b; Saiz‐Ruiz 2005), or for dropouts and completers separately (Dannon 2005a). Participants that dropped out more commonly reported diarrhoea/gastrointestinal tract disturbances (Dannon 2005a; Dannon 2005b; Saiz‐Ruiz 2005), dizziness (Dannon 2005a; Dannon 2005b), vertigo (Dannon 2005a), headaches (Dannon 2005b), loss of appetite (Dannon 2005a) and hypertension (Saiz‐Ruiz 2005), whereas completers more commonly reported dry mouth, nervousness, and dizziness (Dannon 2005a). One study did not report common side effects but reported that participants in the bupropion group experienced more headaches, nervousness, stomach discomfort, and dry mouth then participants in the placebo group (Black 2007). Finally, one study did not report the specific adverse effects experienced by participants, instead reporting that these events were minor in nature and that the medications were very well tolerated (Rosenberg 2013). 

There was also variability in the level of detail provided relating to participant dropout due to these adverse effects. Five studies indicated that participants dropped out due to adverse effects, with dropout ranging from 4.3% to 31.3% of participants in the antidepressant conditions (Dannon 2005a; Dannon 2005b; Grant 2003; Kim 2002; Saiz‐Ruiz 2005); one study indicated that there were no dropouts due to adverse effects (Rosenberg 2013); and the remaining three studies did not refer to dropout due to adverse effects (Black 2007; Blanco 2002; Hollander 2000). Finally, two studies referred to differences in adverse effects between the antidepressant and placebo groups, with one study indicating that participants receiving paroxetine experienced more adverse effects (Grant 2003), and another study indicating that there were no differences between the sertraline and placebo groups in adverse effects experienced (Saiz‐Ruiz 2005).

Opioid antagonists

The six studies evaluating the efficacy of opioid antagonists provided varying levels of detail regarding the type and number of adverse effects experienced (Dannon 2005a; Grant 2006a; Grant 2008; Grant 2010; Kim 2001a; Rosenberg 2013). Three studies reported the most common adverse effects experienced by participants, with nausea (Grant 2006a; Grant 2008; Kim 2001a), dry mouth (Grant 2006a; Grant 2008; Kim 2001a), constipation (Grant 2006a; Grant 2008; Kim 2001a), insomnia (Grant 2006a; Kim 2001a), dizziness (Grant 2006a; Grant 2008), headache (Grant 2008; Kim 2001a), and diarrhoea (Grant 2008; Kim 2001a) reported consistently in more than one study. Dannon 2005a reported adverse effects for dropouts and completers separately, with dropouts more commonly reporting vomiting, nausea, dizziness, agitation, insomnia, and orthostatic hypotension, and completers more commonly reporting sleep disturbance, gastrointestinal tract irritation, dry mouth, and subjective sense of mental dullness. Finally, two studies did not report the specific adverse effects experienced by participants, instead reporting that these adverse effects were not unusual, Grant 2010, or were minor in nature (Rosenberg 2013).

There was also variability in the level of detail provided relating to participant dropout due to these adverse effects. Two studies indicated that participants dropped out due to adverse effects, with dropout ranging from 10.4% to 31.6% of participants in the opioid antagonist conditions (Dannon 2005a; Grant 2010); one study indicated that there were no dropouts due to adverse effects (Rosenberg 2013); and the remaining three studies did not refer to dropout due to adverse effects (Grant 2006a; Grant 2008; Kim 2001a). Finally, no studies referred to differences in adverse effects between the opioid antagonist and placebo groups.

Mood stabilisers

The four studies evaluating the efficacy of mood stabilisers provided varying levels of detail regarding the type and number of adverse effects experienced (Berlin 2013; Dannon 2005b; Hollander 2005; Rosenberg 2013). Berlin 2013 did not differentiate between the types of adverse effects reported by participants in the topiramate and placebo groups, but reported that 48.5% of reported adverse effects were experienced by participants in the topiramate group. The most common adverse effects were tiredness and headaches, with most adverse effects classified as mild to moderate in severity. One study reported only the adverse effects for participants who dropped out of the study, citing concentration, dizziness, and vertigo (Dannon 2005b). Hollander 2005 reported detailed adverse effects for the lithium and placebo groups separately, with dry mouth and sedation the most common effects in the lithium group. Finally, Rosenberg 2013 did not report the specific adverse effects experienced by participants, stating only that they were minor adverse effects and that topiramate was very well tolerated. 

There was also variability in the level of detail provided relating to participant dropout due to these adverse effects. Specifically, Dannon 2005b reported that 13.3% of participants dropped out due to adverse effects, and Berlin 2013 reported a reduction in the drug dose delivered due to adverse effects. Of the remaining two studies, one study reported no participant dropout due to adverse effects (Rosenberg 2013), and the other study did not report if there was any dropout due to adverse effects (Hollander 2005). Finally, only one study referred to differences in adverse effects between the mood stabiliser and placebo groups. Specifically, Hollander 2005 indicated that there were no differences in adverse effects between the lithium and placebo groups.

Atypical antipsychotics

The two studies evaluating the efficacy of the atypical antipsychotic olanzapine provided contrasting findings (Fong 2008; McElroy 2008). McElroy 2008 reported only the adverse effects for participants who dropped out of the study, citing pneumonia, sedation, and increased hypomania. In contrast, Fong 2008 reported that there were no serious adverse effects. 

McElroy 2008 reported that 14.3% of participants dropped out due to adverse effects, whilst Fong 2008 reported that there were no adverse effects experienced by participants that could have affected dropout. Finally, McElroy 2008 indicated that there were no differences in the number of adverse effects between the olanzapine and placebo groups, but that more participants dropped out from the olanzapine group due to these adverse effects compared to the placebo group.

Discussion

Summary of main results

Antidepressants

The weighted mean of effects from six studies indicated that antidepressants were no more effective than placebo, with a small‐to‐medium beneficial effect on the primary outcome (gambling symptom severity) immediately following treatment relative to placebo. However, because this effect was based on very low certainty evidence, it remains uncertain whether antidepressants can improve gambling symptom severity. Moreover, sensitivity analyses also suggested that effects were reduced and approached zero when restricted to the small numbers of studies assessed as having a low risk of bias for allocation concealment. Similarly, although antidepressants were also no more effective than placebo on some secondary outcomes (gambling expenditure, depressive symptoms, functional impairment, and responder status), with small beneficial effects, the very low to low certainty of this evidence precludes definitive conclusions about the degree to which antidepressants can improve these outcomes. 

These findings relating to antidepressants versus placebo should be interpreted cautiously given other important features and limitations of the evidence. First, whilst this comparison was associated with the largest number of studies in this review (k = 6), the individual trials were small and supported only a modest pooled sample of participants (n = 268). As such, this comparison may have lacked the power to detect modest effects of the intervention. Second, only a single study provided data on secondary outcome measures including gambling frequency, time spent gambling, and anxiety symptoms. Third, there was moderate between‐study heterogeneity and one small trial favouring placebo (Hollander 2000). In contrast, a sensitivity analysis restricted to studies assessed as having a low risk of bias for incomplete outcome data (i.e. using ITT analyses) revealed a medium beneficial effect for antidepressants on gambling symptom severity. This suggests that unusual results from small and methodologically limited studies that violated the principles of ITT analyses may have had undue influence on the findings. Fourth, three of the six studies comparing antidepressants with placebo systematically excluded participants with Axis I disorders (including mood disorders) (Grant 2003; Kim 2002) or elevated depressive symptomatology (Black 2007; Grant 2003; Kim 2002). As such, the findings do not generalise to the large numbers of people with gambling problems who exhibit high rates of comorbid depression (Dowling 2015a; Lorains 2011). Finally, there were insufficient studies to compare the efficacy of different classes of antidepressants; to explore the effects of different dosages; or to examine the longer‐term effects of antidepressants. 

Opioid antagonists

Based on low certainty evidence, comparisons suggested that opioid antagonists were associated with a medium beneficial effect, relative to placebo, for gambling symptom severity immediately after treatment, suggesting that opioid antagonists may improve gambling symptom severity. However, because this effect was based on low certainty evidence, the degree to which opioid antagonists can improve gambling symptom severity remains uncertain. Although there were few studies assessed as having a low risk of bias for random sequence generation, blinding of participants/personnel, blinding of outcome assessors, and industry funding, sensitivity analyses based on these domains supported similar conclusions. However, opioid antagonists were no more effective than placebo on responder status, with a small‐to‐medium beneficial effect, although the very low certainty of the evidence precludes definitive conclusions about the degree to which opioid antagonists can improve this secondary outcome.

Findings relating to comparisons between opioid antagonist and placebo conditions should also be interpreted cautiously. First, the comparison was associated with only four studies. Of note, however, was that whilst the number of studies informing the comparison of opioid antagonist and placebo conditions was smaller than the comparison involving antidepressants, the individual trials produced a larger pooled sample size (for example, three studies provided data on the primary outcome, involving n = 259). Second, only a single study provided data for some secondary outcome measures (depressive symptoms, anxiety symptoms, and functional impairment), and no studies reported on other secondary outcome measures (gambling expenditure, gambling frequency, and time spent gambling). Third, the point estimate in the small‐to‐medium range for the responder status outcome was potentially due to the large amount of statistical heterogeneity across studies. Finally, there were insufficient studies to compare the efficacy of nalmefene and naltrexone; to explore the effects of different dosages; or to examine the longer‐term outcomes of these pharmacological interventions.

Mood stabilisers

The review findings were inconclusive with regard to the effects of mood stabilisers (including anticonvulsants). Whilst the weighted mean effect from two studies suggested a medium‐to‐large effect on gambling symptom severity relative to placebo, the benefits of these interventions remain unclear given the very low certainty of the evidence from a small number of studies and a comparison characterised by high statistical heterogeneity and wide CIs that include zero, indicating the possibility of a null effect. Because both studies were assessed as having a low risk of bias for incomplete outcome data, sensitivity analyses based on this domain supported a similar conclusion. Moreover, mood stabilisers were no more effective than placebo conditions on secondary outcomes, with small weighted mean effects (depressive symptoms, anxiety symptoms). Although the very low certainty of the evidence precludes definitive conclusions about the degree to which mood stabilisers can improve these secondary outcomes, such findings may suggest that any beneficial effects of mood stabilisers do not necessarily extend to indices of psychological functioning.

There were several notable features and limitations of the evidence regarding mood stabilisers. First, the studies comprised small trials and samples. Only a single study provided data on most secondary outcome measures (gambling expenditure, gambling frequency, time spent gambling, and functional impairment). The small number of trials precluded an exploration of the influence of risk of bias; for example, no studies were assessed as having a low risk of bias for random sequence generation, allocation concealment, or industry funding, and only a single study was classified as having a low risk of bias for blinding of participants/personnel and selective outcome reporting. Such features highlight the preliminary nature of the current evidence and thus the need for additional research. Studies informing this comparison also had various exclusion criteria relating to bipolar disorders or symptoms, whereby participants were excluded on the basis of bipolar disorder, bipolar I (in the context of a study on bipolar II) (Hollander 2005), and scores of > 15 on the Young Mania Rating Scale (Berlin 2013). As such, the findings do not generalise to individuals with comorbid gambling problems and bipolar or psychotic disorders (Dowling 2015a; Haydock 2015; Jones 2015). Finally, there were insufficient studies to compare the efficacy of different pharmacological agents in this category; to explore dosage effects; or to examine the longer‐term outcomes of these pharmacological interventions. 

Atypical antipsychotics

Based on two studies, the weighted mean effect suggests that the atypical antipsychotic olanzapine was associated with a medium beneficial effect on gambling symptom severity following treatment relative to placebo. Moreover, study results were generally consistent. Although these findings suggest that olanzapine may be effective in improving gambling symptom severity, the results must be interpreted cautiously due to the very low certainty of the evidence. There were only two trials, both of which had small sample sizes. Also, only one of these studies provided data on almost all secondary outcome measures (gambling expenditure, gambling frequency, time spent gambling, depressive symptoms, anxiety symptoms, and responder status), and neither study evaluated functional impairment. Moreover, an exploration of the influence of risk of bias was precluded by only one of the studies having been assessed as having a low risk of bias on almost all indicators (allocation concealment, blinding of participants/personnel, blinding of outcome assessors, incomplete outcome data), and neither study having been assessed as having a low risk of bias on the remaining domains (random sequence generation, selective outcome reporting, other bias – industry funding). As in the mood stabiliser category, these findings highlight the need for future high‐quality research investigating the efficacy of olanzapine in the treatment of disordered or problem gambling. The studies included in this major drug category also applied exclusion criteria relating to psychiatric comorbidity, whereby participants were excluded on the basis of bipolar disorders, McElroy 2008, and any Axis I disorder (Fong 2008), thereby limiting the generalisability of the results (Dowling 2015a; Lorains 2011). Finally, there were no studies on atypical antipsychotics other than olanzapine, and studies were insufficient to explore dosage effects or longer‐term outcomes.

Comparisons between categories of pharmacotherapy

The current review extended prior systematic reviews of pharmacotherapy by systematically exploring the comparative efficacy of agents across different pharmacotherapy categories. The search identified very small numbers of studies directly comparing antidepressants with opioid antagonists (k = 2, n = 62), antidepressants with mood stabilisers (k = 2, n = 58), and opioid antagonists with mood stabilisers (k = 1, n = 24). None of the primary studies compared atypical antipsychotics with any other major drug category, indicating that this is an area for future research. The findings from the available studies suggested generally negligible differences across these pharmacotherapy categories. Whilst this may appear counterintuitive in light of the previous findings that antidepressants may be less effective than other pharmacological agents, there is currently little that can be inferred about the relative efficacy of these interventions in the absence of relevant evidence.

Clinical heterogeneity of studies

There were various sources of clinical heterogeneity across studies. Given the small number of trials addressing specific medication types, broad categories of pharmacotherapy were evaluated that combined these types within categories. Studies also differed on the duration of treatment (ranging from 7 to 96 weeks) and the size of their samples (ranging from 13 to 233 participants). Studies differed in the degree to which they included participants with psychiatric comorbidities, and whether they were conducted across multiple or single sites. Studies employed both self‐report measures and structured clinical interviews to establish participant eligibility and outcomes. There were too few studies to conduct subgroup analyses to systematically examine whether such factors were associated with the magnitude of treatment effects.

Tolerability and adverse events

Common adverse effects of participants receiving antidepressants were headaches, nausea, diarrhoea/gastrointestinal issues, insomnia, dizziness/lightheadedness, and dry mouth. Participants receiving opioid antagonists similarly cited common adverse effects of nausea, dry mouth, constipation, insomnia, dizziness, headache, and diarrhoea. However, there was no consistency in the types of adverse effects experienced by participants receiving mood stabilisers, with studies citing various adverse effects including tiredness, headaches, concentration, dizziness, vertigo, dry mouth, and sedation. Similarly, one study reported that the atypical antipsychotic olanzapine resulted in adverse effects such as pneumonia, sedation, and increased hypomania, whilst the other study evaluating this same treatment reported no serious adverse effects. Several participants discontinued treatment due to these adverse effects, with the highest proportion of participants dropping out from the opioid antagonist treatment groups (10.4% to 31.6%), followed by the antidepressant (4.3% to 31.3%), atypical antipsychotic (14.3%), and mood stabiliser (13.3%) treatment groups.

Overall completeness and applicability of evidence

We included 17 randomised trials in this review, but smaller numbers of studies addressed specific pharmacotherapy categories. These included antidepressants (nine studies), opioid antagonists (six studies), mood stabilisers (including anticonvulsants) (four studies), and atypical antipsychotics (two studies). The medications employed within some categories in this review could be considered heterogeneous; for example, the antidepressant category included various types of antidepressants (e.g. SSRIs, NDRIs, SNDRIs), and the mood stabiliser category included mood stabilisers (lithium) and anticonvulsants (topiramate). Moreover, only a small number of specific pharmacological agents were evaluated across multiple studies (fluvoxamine, paroxetine, sertraline, escitalopram, bupropion, naltrexone, nalmefene, lithium, topiramate, olanzapine). 

All studies considered efficacy immediately following the end of treatment, with only one study following participants across a longer follow‐up period (Rosenberg 2013). Treatment effects were defined by outcomes which typically indicate improvement in gambling problems, including gambling symptom severity and responder status. In contrast to the psychological intervention literature (Cowlishaw 2012), the studies in this review were more likely to employ outcomes on psychological symptoms (such as depression and anxiety) and less likely to provide data on gambling behaviour (such as expenditure, frequency, and time spent). The measurement of functional impairment in life domains is also unique to this pharmacological intervention literature. It is also worth noting that the included studies only included patient‐reported outcome measures (PROMs), with none including any patient‐reported experience measures (PREMs), which are questionnaires that collect data on objective participant views of their experience during treatment in terms of the impact of the process of treatment (e.g. communication and timeliness of assistance) (Kingsley 2017).

The current evidence provides moderate levels of external validity overall. The available studies considered both males and females, as well as participants with varying preferences in gambling activity. However, they typically employed criteria excluding individuals with mental health comorbidities, and findings do not necessarily generalise to those with co‐occurring conditions. Arguably, these exclusions lead to particular limitations regarding the generalisability of studies of antidepressants, mood stabilisers, and atypical antipsychotics (which do not generalise to participants with elevated depression, bipolar symptoms, or psychotic symptoms, respectively).

Most studies administered the intervention in outpatient settings, and none were conducted in inpatient or residential contexts. Whilst the restriction to randomised controlled trials in this review allowed for enhanced control over non‐specific improvements with time (Bartley 2013), the findings may not generalise to their effectiveness under real‐world conditions in which clients are currently being referred for pharmacological treatments (Ward 2018). The evidence in this review also had low international relevance, with studies available from only three countries: the United States (12 studies), Israel (3 studies), and Spain (2 studies). The regulation and availability of pharmacological agents, and the nature of gambling problems, may vary across jurisdictions; hence, it remains unclear whether findings will generalise to other countries. Finally, because all studies employed adult samples, the findings of this review have limited applicability to adolescents.

Quality of the evidence

The articles included in this review varied in sample size and methodological quality. The primary studies generally comprised relatively small samples, averaging only 68 participants. GRADE criteria indicated that the certainty of the evidence for all outcomes across all comparisons evaluated in this review was low or very low, indicating limited confidence in the effect sizes presented in the review. The true effect of these pharmacological agents may therefore be substantially different from these estimates of effects.

Most of the studies included in this review provided insufficient information to judge the method of participant randomisation and allocation concealment. Most studies did not explicitly describe the blinding of participants, personnel, and outcome assessors. Around half of the included studies employed a missing data strategy that facilitated ITT analysis, with the remaining studies employing a ‘completers‐only’ analysis. Finally, few studies reported identical outcomes to those reported in trial registries or reported that financial support or study medications were not provided by pharmaceutical companies. Sensitivity analyses conducted to evaluate the impacts of risk of bias were generally consistent with the comparisons using all available studies.

Potential biases in the review process

The current review was predominately restricted to published research. Despite efforts to identify unpublished material (e.g. via trial register searches), the efficacy of these strategies are uncertain, and the exclusion of unpublished trials may have influenced the conclusions of this review. The small number of eligible studies informed several analyses yielding imprecise findings with low statistical power and precluded subgroup analyses to systematically explore sources of heterogeneity across studies. Despite making attempts to recover missing data, this was not possible in all instances. The findings of some studies therefore could not be included, which may have biased the results.

Agreements and disagreements with other studies or reviews

The current findings support previous meta‐analytic findings that opioid antagonists are potentially beneficial pharmacological agents in the treatment of disordered or problem gambling (Bartley 2013; Goslar 2019). In this review, these medications had a consistent medium beneficial effect on gambling symptom severity, albeit with low certainty evidence. Whilst this is consistent with the findings of the Goslar 2019 meta‐analysis, the magnitude of effect and absence of statistical heterogeneity in gambling symptom severity across studies is less consistent with the findings of other meta‐analyses, which identified smaller effect sizes and substantial heterogeneity across studies related to non‐adherence to ITT analyses and an earlier year of publication (Bartley 2013; Mouaffak 2017). It is likely that the pooling of effect sizes across multiple indices of gambling behaviour (gambling symptoms, expenditure, and frequency) in these reviews resulted in smaller effect sizes and heterogeneity for this outcome measure. It seems that separating distinct indices of gambling behaviour produces consistent findings of a beneficial effect of opioid antagonists on gambling symptom severity (Goslar 2019). An alternative explanation for these inconsistent findings is that these previous reviews included studies that investigated combined pharmacological and psychological treatments (Kovanen 2016b; Toneatto 2009b), in which there were estimates or CIs including values which could favour placebo. Although this review attempted to expand on gambling outcomes, the very low to low certainty of the limited amount of available evidence precluded definitive conclusions about the degree to which opioid antagonists can improve depressive symptoms, anxiety symptoms, functional impairment, and responder status. 

With regard to atypical antipsychotics, the findings of the current review revealed that olanzapine may have a beneficial effect on gambling symptom severity, albeit with very low certainty evidence. However, the magnitude of effect was much higher than that identified in the only other meta‐analysis to analyse atypical antipsychotics as a separate drug category (Bartley 2013). Although both reviews included the same two studies in their analyses, Bartley 2013 employed a composite measure of gambling symptom severity and gambling behaviours (frequency, expenditure). Again, this suggests that separating distinct indices of gambling behaviour produces consistent findings of a beneficial effect on gambling symptom severity. We were unable to explore the effects of olanzapine on indices of gambling behaviour or psychological functioning given the very low certainty evidence provided by single studies on most secondary outcomes.

Our findings relating to mood stabilisers (including anticonvulsants) were inconclusive, given the very low certainty of the evidence from a small number of studies and a comparison characterised by high statistical heterogeneity and wide CIs, indicating the possibility that these interventions are no more effective than placebo. Whilst the medium‐to‐large weighted mean effect on gambling symptom severity relative to placebo is consistent with the findings in previous meta‐analyses for mood‐stabilising medications (Bartley 2013; Goslar 2019), the findings of the current review temper any conclusions that can be drawn from the available evidence base. Moreover, the current review is the first to suggest that mood stabilisers do not appear to address issues that extend beyond gambling symptom severity, such as depressive and anxiety symptoms. 

The findings of the current review suggested antidepressants were no more effective than placebo, with small‐to‐medium beneficial effects on gambling symptom severity; however, this effect was based on very low certainty evidence, and was reduced when restricted to studies assessed as at low risk of bias. This is consistent with previous meta‐analyses, which have found small‐to‐medium, but non‐significant, effect sizes (Bartley 2013; Goslar 2019). This study was the first to indicate that antidepressants were also no more effective than placebo on some secondary outcomes, with small beneficial effects on functional impairment and responder status, as well as issues that extend beyond gambling, such as gambling expenditure and depressive symptoms. 

Finally, the systematic comparison of the efficacy of medications across different categories of pharmacotherapy suggested generally negligible effects. This is consistent with the findings from previous meta‐analyses that conducted moderator analyses relating to type of pharmacological treatment (Goslar 2019; Pallesen 2007). However, there were very few available comparisons directly comparing one pharmacological agent with another pharmacological agent, suggesting this is an avenue for further research.

Study flow diagram.

Figures and Tables -
Figure 1

Study flow diagram.

Navigate to figure in ReviewOpen in new tab

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figures and Tables -
Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Navigate to figure in ReviewOpen in new tab

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figures and Tables -
Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 1.1

Comparison 1: Antidepressants versus placebo, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 2: Gambling expenditure

Figures and Tables -
Analysis 1.2

Comparison 1: Antidepressants versus placebo, Outcome 2: Gambling expenditure

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 3: Gambling frequency

Figures and Tables -
Analysis 1.3

Comparison 1: Antidepressants versus placebo, Outcome 3: Gambling frequency

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 4: Time spent gambling

Figures and Tables -
Analysis 1.4

Comparison 1: Antidepressants versus placebo, Outcome 4: Time spent gambling

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 5: Depressive symptoms

Figures and Tables -
Analysis 1.5

Comparison 1: Antidepressants versus placebo, Outcome 5: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 6: Anxiety symptoms

Figures and Tables -
Analysis 1.6

Comparison 1: Antidepressants versus placebo, Outcome 6: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 7: Functional impairment

Figures and Tables -
Analysis 1.7

Comparison 1: Antidepressants versus placebo, Outcome 7: Functional impairment

Navigate to figure in ReviewOpen in new tab

Comparison 1: Antidepressants versus placebo, Outcome 8: Responder status

Figures and Tables -
Analysis 1.8

Comparison 1: Antidepressants versus placebo, Outcome 8: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 2: Opioid antagonists versus placebo, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 2.1

Comparison 2: Opioid antagonists versus placebo, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 2: Opioid antagonists versus placebo, Outcome 2: Depressive symptoms

Figures and Tables -
Analysis 2.2

Comparison 2: Opioid antagonists versus placebo, Outcome 2: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 2: Opioid antagonists versus placebo, Outcome 3: Anxiety symptoms

Figures and Tables -
Analysis 2.3

Comparison 2: Opioid antagonists versus placebo, Outcome 3: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 2: Opioid antagonists versus placebo, Outcome 4: Functional impairment

Figures and Tables -
Analysis 2.4

Comparison 2: Opioid antagonists versus placebo, Outcome 4: Functional impairment

Navigate to figure in ReviewOpen in new tab

Comparison 2: Opioid antagonists versus placebo, Outcome 5: Responder status

Figures and Tables -
Analysis 2.5

Comparison 2: Opioid antagonists versus placebo, Outcome 5: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 3.1

Comparison 3: Mood stabilisers versus placebo, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 2: Gambling expenditure

Figures and Tables -
Analysis 3.2

Comparison 3: Mood stabilisers versus placebo, Outcome 2: Gambling expenditure

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 3: Gambling frequency

Figures and Tables -
Analysis 3.3

Comparison 3: Mood stabilisers versus placebo, Outcome 3: Gambling frequency

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 4: Time spent gambling

Figures and Tables -
Analysis 3.4

Comparison 3: Mood stabilisers versus placebo, Outcome 4: Time spent gambling

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 5: Depressive symptoms

Figures and Tables -
Analysis 3.5

Comparison 3: Mood stabilisers versus placebo, Outcome 5: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 6: Anxiety symptoms

Figures and Tables -
Analysis 3.6

Comparison 3: Mood stabilisers versus placebo, Outcome 6: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 7: Functional impairment

Figures and Tables -
Analysis 3.7

Comparison 3: Mood stabilisers versus placebo, Outcome 7: Functional impairment

Navigate to figure in ReviewOpen in new tab

Comparison 3: Mood stabilisers versus placebo, Outcome 8: Responder status

Figures and Tables -
Analysis 3.8

Comparison 3: Mood stabilisers versus placebo, Outcome 8: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 4.1

Comparison 4: Atypical antipsychotics versus placebo, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 2: Gambling expenditure

Figures and Tables -
Analysis 4.2

Comparison 4: Atypical antipsychotics versus placebo, Outcome 2: Gambling expenditure

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 3: Gambling frequency

Figures and Tables -
Analysis 4.3

Comparison 4: Atypical antipsychotics versus placebo, Outcome 3: Gambling frequency

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 4: Time spent gambling

Figures and Tables -
Analysis 4.4

Comparison 4: Atypical antipsychotics versus placebo, Outcome 4: Time spent gambling

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 5: Depressive symptoms

Figures and Tables -
Analysis 4.5

Comparison 4: Atypical antipsychotics versus placebo, Outcome 5: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 6: Anxiety symptoms

Figures and Tables -
Analysis 4.6

Comparison 4: Atypical antipsychotics versus placebo, Outcome 6: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 4: Atypical antipsychotics versus placebo, Outcome 7: Responder status

Figures and Tables -
Analysis 4.7

Comparison 4: Atypical antipsychotics versus placebo, Outcome 7: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 5: Antidepressants versus opioid antagonists, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 5.1

Comparison 5: Antidepressants versus opioid antagonists, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 5: Antidepressants versus opioid antagonists, Outcome 2: Depressive symptoms

Figures and Tables -
Analysis 5.2

Comparison 5: Antidepressants versus opioid antagonists, Outcome 2: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 5: Antidepressants versus opioid antagonists, Outcome 3: Anxiety symptoms

Figures and Tables -
Analysis 5.3

Comparison 5: Antidepressants versus opioid antagonists, Outcome 3: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 5: Antidepressants versus opioid antagonists, Outcome 4: Responder status

Figures and Tables -
Analysis 5.4

Comparison 5: Antidepressants versus opioid antagonists, Outcome 4: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 6: Antidepressants versus mood stabilisers, Outcome 1: Gambling symptom severity

Figures and Tables -
Analysis 6.1

Comparison 6: Antidepressants versus mood stabilisers, Outcome 1: Gambling symptom severity

Navigate to figure in ReviewOpen in new tab

Comparison 6: Antidepressants versus mood stabilisers, Outcome 2: Depressive symptoms

Figures and Tables -
Analysis 6.2

Comparison 6: Antidepressants versus mood stabilisers, Outcome 2: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 6: Antidepressants versus mood stabilisers, Outcome 3: Anxiety symptoms

Figures and Tables -
Analysis 6.3

Comparison 6: Antidepressants versus mood stabilisers, Outcome 3: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 6: Antidepressants versus mood stabilisers, Outcome 4: Responder status

Figures and Tables -
Analysis 6.4

Comparison 6: Antidepressants versus mood stabilisers, Outcome 4: Responder status

Navigate to figure in ReviewOpen in new tab

Comparison 8: Opioid antagonists versus mood stabilisers, Outcome 1: Depressive symptoms

Figures and Tables -
Analysis 8.1

Comparison 8: Opioid antagonists versus mood stabilisers, Outcome 1: Depressive symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 8: Opioid antagonists versus mood stabilisers, Outcome 2: Anxiety symptoms

Figures and Tables -
Analysis 8.2

Comparison 8: Opioid antagonists versus mood stabilisers, Outcome 2: Anxiety symptoms

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 1: Random sequence generation

Figures and Tables -
Analysis 11.1

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 1: Random sequence generation

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 2: Allocation concealment

Figures and Tables -
Analysis 11.2

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 2: Allocation concealment

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 3: Blinding of participants/personnel

Figures and Tables -
Analysis 11.3

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 3: Blinding of participants/personnel

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 4: Blinding of outcome assessors

Figures and Tables -
Analysis 11.4

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 4: Blinding of outcome assessors

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 5: Incomplete outcome data

Figures and Tables -
Analysis 11.5

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 5: Incomplete outcome data

Navigate to figure in ReviewOpen in new tab

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 6: Other bias ‐ industry funding

Figures and Tables -
Analysis 11.6

Comparison 11: Sensitivity analyses: antidepressants versus placebo, Outcome 6: Other bias ‐ industry funding

Navigate to figure in ReviewOpen in new tab

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 1: Random sequence generation

Figures and Tables -
Analysis 12.1

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 1: Random sequence generation

Navigate to figure in ReviewOpen in new tab

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 2: Blinding of participants/personnel

Figures and Tables -
Analysis 12.2

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 2: Blinding of participants/personnel

Navigate to figure in ReviewOpen in new tab

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 3: Blinding of outcome assessors

Figures and Tables -
Analysis 12.3

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 3: Blinding of outcome assessors

Navigate to figure in ReviewOpen in new tab

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 4: Incomplete outcome data

Figures and Tables -
Analysis 12.4

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 4: Incomplete outcome data

Navigate to figure in ReviewOpen in new tab

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 5: Other bias ‐ industry funding

Figures and Tables -
Analysis 12.5

Comparison 12: Sensitivity analyses: opioid antagonists versus placebo, Outcome 5: Other bias ‐ industry funding

Navigate to figure in ReviewOpen in new tab

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 1: Blinding of participants/personnel

Figures and Tables -
Analysis 13.1

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 1: Blinding of participants/personnel

Navigate to figure in ReviewOpen in new tab

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 2: Incomplete outcome data

Figures and Tables -
Analysis 13.2

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 2: Incomplete outcome data

Navigate to figure in ReviewOpen in new tab

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 3: Selective reporting

Figures and Tables -
Analysis 13.3

Comparison 13: Sensitivity analyses: mood stabilisers versus placebo, Outcome 3: Selective reporting

Navigate to figure in ReviewOpen in new tab

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 1: Allocation concealment

Figures and Tables -
Analysis 14.1

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 1: Allocation concealment

Navigate to figure in ReviewOpen in new tab

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 2: Blinding of participants/personnel

Figures and Tables -
Analysis 14.2

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 2: Blinding of participants/personnel

Navigate to figure in ReviewOpen in new tab

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 3: Blinding of outcome assessors

Figures and Tables -
Analysis 14.3

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 3: Blinding of outcome assessors

Navigate to figure in ReviewOpen in new tab

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 4: Incomplete outcome data

Figures and Tables -
Analysis 14.4

Comparison 14: Sensitivity analyses: atypical antipsychotics versus placebo, Outcome 4: Incomplete outcome data

Navigate to figure in ReviewOpen in new tab
Summary of findings 1. Antidepressants compared to placebo for the treatment of disordered and problem gambling

Antidepressants compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with antidepressants

Gambling symptom severity
assessed with: various clinician‐administered or self‐report measures
follow‐up: range 8 weeks to 24 weeks

The mean gambling symptom severity was 18.2 for the placebo group at post‐treatment.

SMD 0.32 lower
(0.74 lower to 0.09 higher)

‐

225
(5 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure
assessed with: various self‐report or author‐derived measures
follow‐up: range 8 weeks to 24 weeks

The mean gambling expenditure was 47.3 for the placebo group at post‐treatment.

SMD 0.27 lower
(0.6 lower to 0.06 higher)

‐

144
(3 RCTs)

⊕⊕⊝⊝
Low 1 3

Gambling frequency
assessed with: visual analogue scale
follow‐up: 24 weeks

The mean gambling frequency was 15.2 for the placebo group at post‐treatment.

SMD 0.08 lower
(0.59 lower to 0.42 higher)

‐

60
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Time spent gambling
assessed with: Timeline Follow Back
follow‐up: 10 weeks

The mean time spent gambling was 42.8 for the placebo group at post‐treatment.

SMD 0.17 lower
(0.8 lower to 0.46 higher)

‐

39
(1 RCT)

⊕⊕⊝⊝
Low 4

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 8 weeks to 16 weeks

The mean depressive symptoms was 3.6 for the placebo group at post‐treatment.

SMD 0.19 lower
(0.6 lower to 0.23 higher)

‐

90
(3 RCTs)

⊕⊕⊝⊝
Low 1 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 9 weeks

The mean anxiety symptoms was 3.8 for the placebo group at post‐treatment.

SMD 0.23 higher
(0.38 lower to 0.85 higher)

‐

41
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: range 10 weeks to 16 weeks

The mean functional impairment was 7.2 for the placebo group at post‐treatment.

SMD 0.15 lower
(0.53 lower to 0.22 higher)

‐

110
(2 RCTs)

⊕⊕⊝⊝
Low 1 3

Responder status
assessed with: improvement based on the Clinical Global Impression ‐ Improvement scale or abstinence from gambling
follow‐up: range 8 weeks to 24 weeks

Study population

RR 1.24
(0.93 to 1.66)

268
(6 RCTs)

⊕⊝⊝⊝
Very low 1 4 5

396 per 1000

491 per 1000
(368 to 657)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). 

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence 

High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. 

Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. 

Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.

Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded two levels for inconsistency: high heterogeneity as indicated by I2 and point estimates including both beneficial and harmful effects.
3Downgraded one level for imprecision: small sample size.
4Downgraded two levels for imprecision: small sample size and very wide CIs.
5Downgraded one level for inconsistency: minor heterogeneity, but point estimates include both beneficial and harmful effects.

Figures and Tables -
Summary of findings 1. Antidepressants compared to placebo for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 2. Opioid antagonists compared to placebo for the treatment of disordered and problem gambling

Opioid antagonists compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: opioid antagonists
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with opioid antagonists

Gambling symptom severity
assessed with: various clinician‐administered or self‐report measures
follow‐up: range 10 weeks to 16 weeks

The mean gambling symptom severity was 21.3 for the placebo group at post‐treatment.

SMD 0.46 lower
(0.74 lower to 0.19 lower)

‐

259
(3 RCTs)

⊕⊕⊝⊝
Low 1 2

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 18 weeks

The mean depressive symptoms was 9.1 for the placebo group at post‐treatment.

SMD 0.76 lower
(1.29 lower to 0.23 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 18 weeks

The mean anxiety symptoms was 9.6 for the placebo group at post‐treatment.

SMD 1.39 lower
(1.96 lower to 0.83 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: 18 weeks

The mean functional impairment was 8.4 for the placebo group at post‐treatment.

SMD 0.53 lower
(1.06 lower to 0.01 lower)

‐

77
(1 RCT)

⊕⊕⊝⊝
Low 3

Responder status
assessed with: improvement on various measures or gambling abstinence
follow‐up: range 10 weeks to 16 weeks

Study population

RR 1.65
(0.86 to 3.14)

562
(4 RCTs)

⊕⊝⊝⊝
Very low 1 2 4

402 per 1000

664 per 1000
(346 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for imprecision: small sample size.
3Downgraded two levels for imprecision: small sample size and very wide CIs.
4Downgraded two levels for inconsistency: substantial to considerable heterogeneity, and study‐specific estimates include both beneficial and harmful effects.

Figures and Tables -
Summary of findings 2. Opioid antagonists compared to placebo for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 3. Mood stabilisers compared to placebo for the treatment of disordered and problem gambling

Mood stabilisers compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: mood stabilisers
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with mood stabilisers

Gambling symptom severity
assessed with: Pathological Gambling ‐ Yale Brown Obsessive Compulsive Scale
follow‐up: range 10 weeks to 14 weeks

The mean gambling symptom severity was 20 for the placebo group at post‐treatment.

SMD 0.92 lower
(2.24 lower to 0.39 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3 4

Gambling expenditure
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean gambling expenditure was 317.9 for the placebo group at post‐treatment.

SMD 0.33 lower
(1.07 lower to 0.41 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Gambling frequency
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean gambling frequency was 3.4 for the placebo group at post‐treatment.

SMD 0.49 higher
(0.26 lower to 1.24 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Time spent gambling
assessed with: Pathological Gambling Behavioral Self‐Report Scale
follow‐up: 10 weeks

The mean time spent gambling was 149.3 for the placebo group at post‐treatment.

SMD 0.33 lower
(1.07 lower to 0.41 higher)

‐

29
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

Depressive symptoms
assessed with: various self‐report measures
follow‐up: range 10 weeks to 14 weeks

The mean depressive symptoms was 5.6 for the placebo group at post‐treatment.

SMD 0.15 lower
(1.14 lower to 0.83 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 3 4 5

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 10 weeks to 14 weeks

The mean anxiety symptoms was 6.1 for the placebo group at post‐treatment.

SMD 0.17 lower
(0.64 lower to 0.3 higher)

‐

71
(2 RCTs)

⊕⊝⊝⊝
Very low 1 3 6

Functional impairment
assessed with: Sheehan Disability Scale
follow‐up: 14 weeks

The mean functional impairment was 10.5 for the placebo group at post‐treatment.

SMD 0.34 lower
(0.95 lower to 0.27 higher)

‐

42
(1 RCT)

⊕⊝⊝⊝
Very low 1 4

Responder status
assessed with: improvement based on the Clinical Global Impression ‐ Improvement scale
follow‐up: range 10 weeks to 14 weeks

Study population

RR 2.69
(1.14 to 6.32)

40
(1 RCT)

⊕⊝⊝⊝
Very low 1 3 4

227 per 1000

611 per 1000
(259 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: mostly insufficient reporting with some trials at high risk of bias.
2Downgraded one level for inconsistency: moderate to substantial heterogeneity.
3Downgraded one level for indirectness: the population within this comparison had strict inclusion criteria relating to comorbid bipolar disorder.
4Downgraded two levels for imprecision: small sample size and very wide CIs.
5Downgraded two levels for inconsistency: moderate to substantial heterogeneity, and point estimates include both beneficial and harmful effects.
6Downgraded one level for imprecision: small sample size.

Figures and Tables -
Summary of findings 3. Mood stabilisers compared to placebo for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 4. Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling

Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: atypical antipsychotics
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with placebo

Risk with atypical antipsychotics

Gambling symptom severity
assessed with: various clinician‐administered measures
follow‐up: range 7 weeks to 13 weeks

The mean gambling symptom severity was 6.6 for the placebo group at post‐treatment.

SMD 0.59 lower
(1.1 lower to 0.08 lower)

‐

63
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean gambling expenditure was 42.0 for the placebo group at post‐treatment.

SMD 0.16 lower
(1.03 lower to 0.7 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling frequency
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean gambling frequency was 2.5 for the placebo group at post‐treatment.

SMD 0.04 lower
(0.9 lower to 0.83 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Time spent gambling
assessed with: gambling behaviour diary
follow‐up: 7 weeks

The mean time spent gambling was 1.5 for the placebo group at post‐treatment.

SMD 0.26 lower
(1.13 lower to 0.6 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 7 weeks

The mean depressive symptoms was 3.5 for the placebo group at post‐treatment.

SMD 0.12 higher
(0.74 lower to 0.99 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 7 weeks

The mean anxiety symptoms was 2.8 for the placebo group at post‐treatment.

SMD 0.84 higher
(0.07 lower to 1.75 higher)

‐

21
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: improvement based on the Clinical Global Impression scale
follow‐up: 13 weeks

Study population

RR 0.93
(0.62 to 1.40)

42
(1 RCT)

⊕⊕⊝⊝
Low 1 4

714 per 1000

664 per 1000
(443 to 1000)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: mostly insufficient reporting with some trials at high risk of bias. 
2Downgraded one level for indirectness: the population within this comparison had strict inclusion criteria relating to gambling modality.
3Downgraded two levels for imprecision: small sample size and very wide CIs. 
4Downgraded one level for imprecision: small sample size.

Figures and Tables -
Summary of findings 4. Atypical antipsychotics compared to placebo for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 5. Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling

Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: opioid antagonists

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with opioid antagonists

Risk with antidepressants

Gambling symptom severity
assessed with: Clinical Global Impression ‐ Severity Scale
follow‐up: 12 weeks

The mean gambling symptom severity was 12.2 for the opioid antagonist group at post‐treatment.

SMD 0.08 lower
(0.86 lower to 0.71 higher)

‐

25
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 12 weeks to 2 years

The mean depressive symptoms was 7.8 for the opioid antagonist group at post‐treatment.

SMD 0.22 higher
(0.29 lower to 0.72 higher)

‐

62
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 12 weeks to 2 years

The mean anxiety symptoms was 8.2 for the opioid antagonist group at post‐treatment.

SMD 0.21 higher
(0.29 lower to 0.72 higher)

‐

62
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: gambling abstinence
follow‐up: 12 weeks

Study population

RR 1.01
(0.54 to 1.87)

36
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 4

526 per 1000

532 per 1000
(284 to 984)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.
4Downgraded one level for imprecision: small sample size.

Figures and Tables -
Summary of findings 5. Antidepressants compared to opioid antagonists for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 6. Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling

Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: antidepressants
Comparison: mood stabilisers

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with mood stabilisers

Risk with antidepressants

Gambling symptom severity
assessed with: Pathological Gambling ‐ Yale Brown Obsessive Compulsive Scale
follow‐up: 12 weeks

The mean gambling symptom severity was 12.5 for the mood stabiliser group at post‐treatment.

SMD 0.07 higher
(0.64 lower to 0.77 higher)

‐

31
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: range 12 weeks to 2 years

The mean depressive symptoms was 8.6 for the mood stabiliser group at post‐treatment.

SMD 0.02 higher
(0.53 lower to 0.56 higher)

‐

58
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: range 12 weeks to 2 years

The mean anxiety symptoms was 10.4 for the mood stabiliser group at post‐treatment.

SMD 0.16 higher
(0.39 lower to 0.7 higher)

‐

58
(2 RCTs)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status
assessed with: gambling abstinence
follow‐up: 12 weeks

Study population

RR 0.63
(0.29 to 1.33)

31
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 4

600 per 1000

378 per 1000
(174 to 798)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting with some trials at high risk of bias.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.
4Downgraded one level for imprecision: small sample size.

Figures and Tables -
Summary of findings 6. Antidepressants compared to mood stabilisers for the treatment of disordered and problem gambling
Navigate to table in Review
Summary of findings 7. Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling

Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling

Patient or population: treatment of disordered and problem gambling
Intervention: opioid antagonists
Comparison: mood stabilisers

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Risk with mood stabilisers

Risk with opioid antagonists

Gambling symptom severity ‐ not measured

‐

‐

‐

‐

‐

Gambling expenditure ‐ not measured

‐

‐

‐

‐

‐

Gambling frequency ‐ not measured

‐

‐

‐

‐

‐

Time spent gambling ‐ not measured

‐

‐

‐

‐

‐

Depressive symptoms
assessed with: Hamilton Depression Rating Scale
follow‐up: 2 years

The mean depressive symptoms was 10.3 for the mood stabiliser group at post‐treatment.

SMD 0.71 lower
(1.61 lower to 0.2 higher)

‐

24
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Anxiety symptoms
assessed with: Hamilton Anxiety Rating Scale
follow‐up: 2 years

The mean anxiety symptoms was 4.8 for the mood stabiliser group at post‐treatment.

SMD 0.26 lower
(1.15 lower to 0.62 higher)

‐

24
(1 RCT)

⊕⊝⊝⊝
Very low 1 2 3

Functional impairment ‐ not measured

‐

‐

‐

‐

‐

Responder status ‐ not measured

‐

‐

‐

‐

‐

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; RCT: randomised controlled trial; SMD: standardised mean difference.

GRADE Working Group grades of evidence
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.
Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

1Downgraded one level for risk of bias: insufficient reporting.
2Downgraded one level for indirectness: all‐male sample.
3Downgraded two levels for imprecision: small sample size and wide CIs.

Figures and Tables -
Summary of findings 7. Opioid antagonists compared to mood stabilisers for the treatment of disordered and problem gambling
Navigate to table in Review
Comparison 1. Antidepressants versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1.1 Gambling symptom severity Show forest plot

5

225

Std. Mean Difference (IV, Random, 95% CI)

‐0.32 [‐0.74, 0.09]

1.2 Gambling expenditure Show forest plot

3

144

Std. Mean Difference (IV, Random, 95% CI)

‐0.27 [‐0.60, 0.06]

1.3 Gambling frequency Show forest plot

1

60

Std. Mean Difference (IV, Random, 95% CI)

‐0.08 [‐0.59, 0.42]

1.4 Time spent gambling Show forest plot

1

39

Std. Mean Difference (IV, Random, 95% CI)

‐0.17 [‐0.80, 0.46]

1.5 Depressive symptoms Show forest plot

3

90

Std. Mean Difference (IV, Random, 95% CI)

‐0.19 [‐0.60, 0.23]

1.6 Anxiety symptoms Show forest plot

1

41

Std. Mean Difference (IV, Random, 95% CI)

0.23 [‐0.38, 0.85]

1.7 Functional impairment Show forest plot

2

110

Std. Mean Difference (IV, Random, 95% CI)

‐0.15 [‐0.53, 0.22]

1.8 Responder status Show forest plot

6

268

Risk Ratio (M‐H, Random, 95% CI)

1.24 [0.93, 1.66]
Figures and Tables -
Comparison 1. Antidepressants versus placebo
Navigate to table in Review
Comparison 2. Opioid antagonists versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

2.1 Gambling symptom severity Show forest plot

3

259

Std. Mean Difference (IV, Random, 95% CI)

‐0.46 [‐0.74, ‐0.19]

2.2 Depressive symptoms Show forest plot

1

77

Std. Mean Difference (IV, Random, 95% CI)

‐0.76 [‐1.29, ‐0.23]

2.3 Anxiety symptoms Show forest plot

1

77

Std. Mean Difference (IV, Random, 95% CI)

‐1.39 [‐1.96, ‐0.83]

2.4 Functional impairment Show forest plot

1

77

Std. Mean Difference (IV, Random, 95% CI)

‐0.53 [‐1.06, ‐0.01]

2.5 Responder status Show forest plot

4

562

Risk Ratio (M‐H, Random, 95% CI)

1.65 [0.86, 3.14]
Figures and Tables -
Comparison 2. Opioid antagonists versus placebo
Navigate to table in Review
Comparison 3. Mood stabilisers versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

3.1 Gambling symptom severity Show forest plot

2

71

Std. Mean Difference (IV, Random, 95% CI)

‐0.92 [‐2.24, 0.39]

3.2 Gambling expenditure Show forest plot

1

29

Std. Mean Difference (IV, Random, 95% CI)

‐0.33 [‐1.07, 0.41]

3.3 Gambling frequency Show forest plot

1

29

Std. Mean Difference (IV, Random, 95% CI)

0.49 [‐0.26, 1.24]

3.4 Time spent gambling Show forest plot

1

29

Std. Mean Difference (IV, Random, 95% CI)

‐0.33 [‐1.07, 0.41]

3.5 Depressive symptoms Show forest plot

2

71

Std. Mean Difference (IV, Random, 95% CI)

‐0.15 [‐1.14, 0.83]

3.6 Anxiety symptoms Show forest plot

2

71

Std. Mean Difference (IV, Random, 95% CI)

‐0.17 [‐0.64, 0.30]

3.7 Functional impairment Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.34 [‐0.95, 0.27]

3.8 Responder status Show forest plot

1

40

Risk Ratio (M‐H, Random, 95% CI)

2.69 [1.14, 6.32]
Figures and Tables -
Comparison 3. Mood stabilisers versus placebo
Navigate to table in Review
Comparison 4. Atypical antipsychotics versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

4.1 Gambling symptom severity Show forest plot

2

63

Std. Mean Difference (IV, Random, 95% CI)

‐0.59 [‐1.10, ‐0.08]

4.2 Gambling expenditure Show forest plot

1

21

Std. Mean Difference (IV, Random, 95% CI)

‐0.16 [‐1.03, 0.70]

4.3 Gambling frequency Show forest plot

1

21

Std. Mean Difference (IV, Random, 95% CI)

‐0.04 [‐0.90, 0.83]

4.4 Time spent gambling Show forest plot

1

21

Std. Mean Difference (IV, Random, 95% CI)

‐0.26 [‐1.13, 0.60]

4.5 Depressive symptoms Show forest plot

1

21

Std. Mean Difference (IV, Random, 95% CI)

0.12 [‐0.74, 0.99]

4.6 Anxiety symptoms Show forest plot

1

21

Std. Mean Difference (IV, Random, 95% CI)

0.84 [‐0.07, 1.75]

4.7 Responder status Show forest plot

1

42

Risk Ratio (M‐H, Random, 95% CI)

0.93 [0.62, 1.40]
Figures and Tables -
Comparison 4. Atypical antipsychotics versus placebo
Navigate to table in Review
Comparison 5. Antidepressants versus opioid antagonists

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

5.1 Gambling symptom severity Show forest plot

1

25

Std. Mean Difference (IV, Random, 95% CI)

‐0.08 [‐0.86, 0.71]

5.2 Depressive symptoms Show forest plot

2

62

Std. Mean Difference (IV, Random, 95% CI)

0.22 [‐0.29, 0.72]

5.3 Anxiety symptoms Show forest plot

2

62

Std. Mean Difference (IV, Random, 95% CI)

0.21 [‐0.29, 0.72]

5.4 Responder status Show forest plot

1

36

Risk Ratio (M‐H, Random, 95% CI)

1.01 [0.54, 1.87]
Figures and Tables -
Comparison 5. Antidepressants versus opioid antagonists
Navigate to table in Review
Comparison 6. Antidepressants versus mood stabilisers

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

6.1 Gambling symptom severity Show forest plot

1

31

Std. Mean Difference (IV, Random, 95% CI)

0.07 [‐0.64, 0.77]

6.2 Depressive symptoms Show forest plot

2

58

Std. Mean Difference (IV, Random, 95% CI)

0.02 [‐0.53, 0.56]

6.3 Anxiety symptoms Show forest plot

2

58

Std. Mean Difference (IV, Random, 95% CI)

0.16 [‐0.39, 0.70]

6.4 Responder status Show forest plot

1

31

Risk Ratio (M‐H, Random, 95% CI)

0.62 [0.29, 1.33]
Figures and Tables -
Comparison 6. Antidepressants versus mood stabilisers
Navigate to table in Review
Comparison 8. Opioid antagonists versus mood stabilisers

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

8.1 Depressive symptoms Show forest plot

1

24

Std. Mean Difference (IV, Random, 95% CI)

‐0.71 [‐1.61, 0.20]

8.2 Anxiety symptoms Show forest plot

1

24

Std. Mean Difference (IV, Random, 95% CI)

‐0.26 [‐1.15, 0.62]
Figures and Tables -
Comparison 8. Opioid antagonists versus mood stabilisers
Navigate to table in Review
Comparison 11. Sensitivity analyses: antidepressants versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

11.1 Random sequence generation Show forest plot

1

71

Std. Mean Difference (IV, Random, 95% CI)

‐0.10 [‐0.56, 0.37]

11.2 Allocation concealment Show forest plot

2

110

Std. Mean Difference (IV, Random, 95% CI)

‐0.07 [‐0.45, 0.30]

11.3 Blinding of participants/personnel Show forest plot

1

39

Std. Mean Difference (IV, Random, 95% CI)

‐0.02 [‐0.65, 0.61]

11.4 Blinding of outcome assessors Show forest plot

1

39

Std. Mean Difference (IV, Random, 95% CI)

‐0.02 [‐0.65, 0.61]

11.5 Incomplete outcome data Show forest plot

3

144

Std. Mean Difference (IV, Random, 95% CI)

‐0.52 [‐1.04, ‐0.01]

11.6 Other bias ‐ industry funding Show forest plot

1

39

Std. Mean Difference (IV, Random, 95% CI)

‐0.02 [‐0.65, 0.61]
Figures and Tables -
Comparison 11. Sensitivity analyses: antidepressants versus placebo
Navigate to table in Review
Comparison 12. Sensitivity analyses: opioid antagonists versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

12.1 Random sequence generation Show forest plot

2

223

Std. Mean Difference (IV, Random, 95% CI)

‐0.49 [‐0.79, ‐0.20]

12.2 Blinding of participants/personnel Show forest plot

2

182

Std. Mean Difference (IV, Random, 95% CI)

‐0.43 [‐0.74, ‐0.11]

12.3 Blinding of outcome assessors Show forest plot

2

223

Std. Mean Difference (IV, Random, 95% CI)

‐0.49 [‐0.79, ‐0.20]

12.4 Incomplete outcome data Show forest plot

1

77

Std. Mean Difference (IV, Random, 95% CI)

‐0.57 [‐1.10, ‐0.04]

12.5 Other bias ‐ industry funding Show forest plot

2

113

Std. Mean Difference (IV, Random, 95% CI)

‐0.47 [‐0.89, ‐0.06]
Figures and Tables -
Comparison 12. Sensitivity analyses: opioid antagonists versus placebo
Navigate to table in Review
Comparison 13. Sensitivity analyses: mood stabilisers versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

13.1 Blinding of participants/personnel Show forest plot

1

29

Std. Mean Difference (IV, Random, 95% CI)

‐1.63 [‐2.50, ‐0.77]

13.2 Incomplete outcome data Show forest plot

2

71

Std. Mean Difference (IV, Random, 95% CI)

‐0.92 [‐2.24, 0.39]

13.3 Selective reporting Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.29 [‐0.90, 0.32]
Figures and Tables -
Comparison 13. Sensitivity analyses: mood stabilisers versus placebo
Navigate to table in Review
Comparison 14. Sensitivity analyses: atypical antipsychotics versus placebo

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

14.1 Allocation concealment Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.64 [‐1.26, ‐0.02]

14.2 Blinding of participants/personnel Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.64 [‐1.26, ‐0.02]

14.3 Blinding of outcome assessors Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.64 [‐1.26, ‐0.02]

14.4 Incomplete outcome data Show forest plot

1

42

Std. Mean Difference (IV, Random, 95% CI)

‐0.64 [‐1.26, ‐0.02]
Figures and Tables -
Comparison 14. Sensitivity analyses: atypical antipsychotics versus placebo
Navigate to table in Review