Effectiveness of Non-Pharmacological Interventions on Gaming Disorder: A Systematic Review and Meta-Analysis

Literature search

The literature search was conducted in five electronic databases: MEDLINE (via PubMed), Embase, CENTRAL (via Cochrane Library), PsycINFO, and CINAHL (on Ebsco Host). Additionally, the reference lists were manually searched. The search keywords were categorized into two groups: 1) “gaming disorder” (MeSH Terms) OR “internet gaming disorder” (MeSH Terms) AND 2) “intervention” (Title/Abstract) OR “effect” (Title/Abstract) OR “treatment” OR “therapy” (Title/Abstract). Detailed search strategies for all databases are provided in Supplementary Table 3.

Inclusion and exclusion criteria

Studies in English published up to May 12, 2024, were screened according to the participants, intervention, comparison, outcomes, time, setting, and study design (PICOTS-SD) criteria. Two independent reviewers (Ock and Lee) screened the titles and abstracts. The full texts of potentially eligible studies were obtained for further evaluation. Any disagreement between the two reviewers was resolved through discussion with a third researcher (Kim).

Participants (P): The participants included individuals diagnosed with GD (including IGD), based on diagnostic criteria derived from the Internet Addiction Test (IAT) and the DSM-5, with no age restrictions. This category also encompassed normal gamers as well as those with co-occurring mental disorders alongside GD. In contrast, studies focusing primarily on addiction symptoms or disorders related to online SNS), pornography, shopping, or gambling were excluded. Studies with a minimum sample size of 10 participants per group were included.

Intervention (I): All types of interventions for GD (including IGD), except pharmacological, were included. This encompassed indirect interventions, such as parental strategies to manage their children’s gaming behavior.

Comparison (C): Both inactive controls (no intervention, waitlist, sham, or pseudo-training) and active controls (other types of interventions) were allowed as comparators.

Outcomes (O): The primary outcomes included only GDrelated variables measured using objective or subjective diagnostic criteria, such as the IAT and the DSM-5.

Time (T): No restrictions were applied on the study duration or the follow-up period for trials.

Setting (S): No restrictions were applied regarding the study setting.

Study design (SD): Only randomized controlled trials (RCTs) were included. There were no restrictions on the follow-up period of the trial.

Quality assessment

The risk of bias in the included studies was independently assessed by two reviewers using the Cochrane Risk of Bias Tool for RCTs (RoB2) [31]. The assessment followed the guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions [32]. The primary outcomes were analyzed using the intention-to-treat (ITT) approach.

The RoB2 tool addresses five domains of bias: (D1) arising from the randomization process, (D2) due to deviations from intended interventions, (D3) due to missing outcome data, (D4) in the measurement of the outcome, and (D5) in the selection of the reported result. The risk of bias in each domain was rated as “high,” “some concerns,” or “low” and visualized using a traffic light plot. The overall risk of bias was summarized using a bar chart to provide a comprehensive overview of potential biases in the included studies.

Data extraction

Multiple publications using the same dataset were considered as a single study; however, data were extracted for all related publications. To record the basic characteristics of the included studies, a standardized data extraction form was utilized. In cases where a study contained two experimental groups which were independently compared to a control group, each comparison was treated as a separate study [33].

Statistical analysis

A meta-analysis was conducted using Comprehensive Meta-Analysis software (version 3.0; https://www.meta-analysis.com/) to assess the effectiveness of NPIs in reducing the severity of GD. Effect sizes and 95% confidence intervals (CIs) were derived from the pooled data, with a 5% significance level. Effect sizes were calculated using means and standard deviations, with Hedges’s g applied under a random-effects model [34,35].

Heterogeneity was evaluated using the I² statistic, with thresholds of 25%, 50%, and 75% indicating low, moderate, and high heterogeneity, respectively [36]. In addition, Cochran’s Q test was performed, with a p-value of <0.1 interpreted as evidence of significant heterogeneity [37]. In cases where the correlation coefficient between pre- and post-treatment measurements within the same group was not reported, a conservative estimate of r=0.7 was applied [38]. Publication bias was initially assessed through visual inspection of a comparison-adjusted funnel plot for asymmetry. Further analysis was conducted using the Egger test, where a p-value of <0.05 was considered indicative of potential publication bias [39]. To further evaluate the risk of publication bias, the trim-and-fill method was employed under both fixed-effect and random-effect models [40].

Sensitivity analysis was conducted by sequentially excluding each study to evaluate its influence on the pooled estimate of GD severity. This approach was used to assess the robustness and reliability of the overall findings. To explore heterogeneity across studies and examine variations in effect sizes based on specific factors, subgroup analyses were performed. The following categories were considered: intervention type (psychotherapy, behavioral, or other), intervention goal (prevention or treatment), age group (adolescent or adult), control type (active or inactive), outcome measure (DSM-5 or IAT), geographical region (Asia or Europe), and RoB2 ratings (low, some concerns, or high).

Search results

The preliminary search identified 3,508 potentially relevant articles. After removing 898 duplicates, 2,610 records remained for initial screening. Following a review of titles and abstracts, 65 studies were selected for full-text assessment. Studies were excluded for the following reasons: lack of a relevant population [41-43], absence of reported outcomes (GD severity) [20,44-57], differing study designs [50,58-72], inability to access the full text despite thorough searches [73-82], or being ongoing studies without conclusive results (Supplementary Table 4) [83-86]. Two additional studies were identified through reference checking of related meta-analyses and reviews. Finally, 18 studies were analyzed (Figure 1 and Supplementary Table 5).

Figure 1.

PRISMA flow of study selection process. N indicates the number of studies at each stage of the review.

Characteristics of the included studies

The characteristics of the included studies are presented in Table 1. All studies were RCTs published in English between 2013 and 2023. A total of 10 studies were conducted in Asia (Korea, China, and Hongkong) [25,87-95] with 778 participants (39.90%); 6, in Europe (Germany, Switzerland, Turkey) [96-101] with 1,102 participants (56.51%); and 2, in other regions (Nigeria, USA) [101,102] with 70 participants (3.59%). None reported conflicts of interest, and all were peer reviewed.

Table 1.

Characteristics of studies and interventions of randomized controlled trials, included in a systematic review and meta-analysis to reduce gaming disorder

Participants

Of the total participants (n=1,950), 908 were allocated to the NPI group and 1,042 to the control group. The mean number of participants per study was 50.44 (range: 12–167) in the intervention group and 57.89 (range: 12–255) in the control group. The average age of participants ranged from 10.22 to 32.10 years in the NPI group and from 9.97 to 31.50 years in the control group. The proportion of male participants varied widely across studies, ranging from 18.8% to 100%, with 6 studies (k=6, where k denotes the number of studies) exclusively including male participants.

Interventions

The most commonly employed intervention type was psychotherapy (k=10), with most studies utilizing CBT (k=8) (Tables 1 and 2) [87,89,92,93,97,98,100-103]. Four studies implemented behavioral interventions, such as abstinence or response inhibition training using computerized tasks [25,94,95,101]. For prevention, a parental program (k=1) [91] and physical exercise (k=1) [99] were included. Brain stimulation interventions involved tran-scranial direct current stimulation (tDCS, k=2) [88,90]. The intervention durations ranged from 4 hours to 6 months, with a mean duration of 6.37 weeks (Table 1). The number of sessions varied between 1 and 26, with an average of 10.88. The length of each session ranged from 20 minutes to 4 hours, with an average of 78.58 minutes per session.

Table 2.

Number of studies for each type of treatment within each of the groups

Comparison group

The comparison groups (control group) were primarily composed of no-treatment (no intervention) and waiting list controls (Table 1). For brain stimulation and certain behavioral interventions, sham interventions or pseudo-training were employed. Four studies utilized treatment-as-usual (TAU) as the control group, which included active comparators such as CBT [87], support groups [102], educational programs [91], family therapy as usual (FTAU) [100], virtual reality therapy [93], and basic counseling [92].

Outcome

All studies utilized self-report questionnaires to assess GD outcomes. The most frequently used tool was IAT, employed in 11 studies to categorize GD as a subdomain of internet-related activities (Table 3). This included various instruments such as the IAT [26,94], Young’s Internet Addiction Test (YIAT) [90], Korean Internet Addiction Test (K-Scale) [91], German version of the Compulsive Internet Use Scale (CIUS) [98], Generalized Problematic Internet Use Scale 2 (GPIUS2) [103], Young’s Internet Addiction Scale (YIAS) [87,93], Assessment of Internet and Computer Game Addiction–Self-report (AICA-S) [97,101], and Online Game Cognitive Addiction Scale (OGCAS) [92], which provided more detailed assessments of GD. Additionally, outcome assessment tools based on the DSM-5 diagnostic criteria were utilized in seven studies for diagnosing GD, employing tools such as the DSM-5 [95,100,102], Internet Gaming Disorder Scale (IGD-scale) [25,96], Internet Gaming Disorder Scale–short form (IGDS9-SF) [99], and Chen Internet Addiction Scale-gaming version (CIAS-G) [89].

Table 3.

Number of studies for each type of outcome measurement within each of the groups

Quality of the included studies

The risk of bias in the included studies was assessed with equal weights for all domains (Figure 2A). Four studies had a “low risk” across all domains [97,101-103]. In contrast, three had overall “high risk,” particularly in D1 and D3 [91,93,100].

Figure 2.

Evaluation of the risk of bias in the 18 studies: (A) individual assessments and (B) overall percentages. Green indicates low risk of bias, yellow indicates some concerns, and red indicates high risk.

A domain-specific evaluation showed that, in D1, 10 studies employed sequence generation software or random number generators for random allocation, leading to a rating of “low risk [25,89,90,92,94,97,99,101-103].” In contrast, the studies by Li et al. [91] and Park et al. [93] lacked specific information on the randomization methods and allocation concealment, with statistically significant baseline differences between intervention groups, and were classified as “high risk.” The remaining six studies lacked clarity on randomization and allocation concealment, leading to an assessment of “some concern” [87,88,95,96,98,100].

In D2, nine studies indicated that either participants or intervention providers were aware of the intervention group; however, the resulting attrition had a limited impact on the study outcomes; thus, these studies were rated as having “some concern [87,88,90-92,95,96,98,99].” Conversely, nine studies were assessed as “low risk” because neither participants nor providers were aware of the intervention group or no attrition affected the results [26,89,93,94,97,100-103].

In D3, 15 studies provided complete outcome data, resulting in a “low risk” rating [26,87-89,92-97,99-103]. In contrast, the studies by Li et al. [91] and Lindenberg et al. [98] had missing data, with insufficient evidence to confirm the absence of bias, leading to a rating of “some concern.” Notably, Nielsen et al. [100] reported IAT data only at baseline and at 5-week follow-up, with missing data for planned assessments at weeks 1 and 25, resulting in a classification of “high risk.”

In D4, 12 studies had appropriate outcome measurements with consistent methods across intervention groups, leading to a rating of “low risk [26,88,90,93,94,96-99,101-103].” However, six studies had evaluators who were aware of the intervention group and were rated as “some concern [87,89,91,92,95,100].”

In D5, 14 studies utilized data according to pre-specified analytical plans, which resulted in a “low risk” assessment [87-92,94,96-99,101-103]. Conversely, four studies lacked detailed information on multilevel analytical procedures, leading to a rating of “some concern [25,93,95,100].”

Distribution of risk of bias by domain (Figure 2B) revealed some “high risk” studies in D1 and D3. Notably, approximately 10% of studies in D1 were rated as “high risk,” while in D3, only one was classified as “high risk,” with a few others as “some concern.” In D2 and D5, although some studies had “some concern,” the majority had “low risk.”

Results of individual studies

The analysis of the effect of NPIs on GD included a total of 19 comparison-control studies. The NPI group (n=908) demonstrated a statistically significant reduction in GD symptoms compared to the control group (n=1,057) (Figure 3). The pooled effect size (Hedges’s g) was -0.82, indicating a significant effect of NPIs in reducing GD (95% CI, -1.23 to -0.52; p<0.001).

Figure 3.

Forest plot showing effect sizes (Hedges’s g) for non-pharmacological interventions on gaming disorder in 19 comparisons from 18 studies. Squares represent effect sizes with 95% CIs for individual studies, while diamonds indicate the pooled effect size. CI, confidence interval.

The Forest plot analysis revealed that the study by Ede et al. [103] had a notably larger effect size compared to other studies, with the remaining studies displaying a wide range of effect sizes. The heterogeneity analysis showed a high level of variability, with an I² value of 90.36% and a Tau value of 0.653, indicating substantial heterogeneity (p<0.0001).

Synthesis of meta-analysis results

Sensitivity analysis confirmed that the pooled effect size (Hedges’s g) was stable even after sequentially removing individual studies (Figure 4). The estimated pooled effect size for GD severity ranged -0.72 (95% CI, -1.05 to -0.39) to -0.96 (95% CI, -1.37 to -0.55) after excluding each study individually, with no significant changes in the overall effect size. Notably, even when the study by Ede et al. [103], which had an outlier effect size, was excluded, the pooled effect size remained stable.

Figure 4.

Sensitivity analysis assessing the effect on pooled effect size (Hedges’s g) by removing each study sequentially. Squares show recalculated effect sizes with 95% CIs after excluding one study at a time, while the diamond represents the overall pooled effect size, indicating stability across analyses. CI, confidence interval.

Publication bias

The funnel plot analysis revealed some asymmetry, suggesting potential publication bias (Figure 5). Most studies were clustered toward the top, indicating larger sample sizes; however, the study by Ede et al. [103] was outside the CIs, showing an outlier effect. Both Begg and Egger tests confirmed statistically significant publication bias (Begg, p=0.043; Egger, p=0.002; SE=1.189; intercept=-4.418).

Figure 5.

Funnel plot assessing publication bias in studies on the effect of non-pharmacological interventions on gaming disorder.

The trim-and-fill method to adjust for publication bias showed a pooled effect size (Hedges’s g) of -0.46 (95% CI, -0.55 to -0.37) before adjustment and -1.25 (95% CI, -1.17 to -0.90) post-adjustment, indicating a pronounced effect.

Subgroup analysis

Subgroup analyses evaluated the variation in the effectiveness of NPIs (Table 4). The mixed-effects model analysis revealed that psychotherapy had the most significant effect, with a Hedges’s g of -1.34 (95% CI, -1.97 to -0.70), indicating a greater reduction in GD symptoms compared to behavioral interventions. In intervention goals, treatments were more effective than preventions, with a Hedges’s g of -1.13 (95% CI, -1.67 to -0.59). Additionally, the adult group showed a stronger effect compared to the adolescent group (Hedges’s g=-1.14 vs. -0.45). Notably, studies with a low risk of bias exhibited the most substantial effects (Hedges’s g=-4.08).

Table 4.

Subgroup analyses of non-pharmacological interventions for the prevention and reduction of gaming disorder across various demographic and methodological factors (based on 19 comparative studies)