Search and Study selection
We searched four databases (PubMed, Cochrane Library, SCOPUS, and Airiti Library) for articles published between 2000 and 2017. Two independent scientists (Chang and Tzang) screen the literature and identified meta-analysis of randomized control trial studies of the efficacies for IA/IGD in children and young adults.
The following full electronic search strategies were used: (“Internet Addiction” OR “Internet Gaming Disorder” OR “Problematic Internet Use” OR “Internet Usage” OR “Internet Overuse” OR “Problematic Internet” OR “Online Addiction” OR “Cybersex” OR “Computer Game Addiction”) AND (treatment OR therapy OR program OR training OR psychotherapy OR curriculum OR psychosocial OR Escitalopram OR Methylphenidate OR Bupropion OR Atomoxetine OR Fluoxetine OR pharmacological OR non-pharmacological OR behavior OR Treat* OR Therapy* OR Intervention OR Psychotherapy) AND (child OR “elementary students” OR adolescent OR “Young Adult”). To avoid overlooking any crucial studies, we also searched reference lists of the retrieved articles to identify other eligible studies.
Inclusion and Exclusion criteria
Studies were included if they met the following inclusion criteria: (1) the study addressed the “internet addiction” topic (or related topic, e.g. internet gaming disorder); (2) the age of all participants was less than 25 years (intended to include college students); (3) the study included a pharmacological or non-pharmacological treatment (mainly psychological without excluding others, e.g. alternative medicines or therapies) aimed to decrease IA related problems (mainly, severity of IA, online time spent, depression and anxiety).
Studies meeting the following criteria were excluded: (1) the study did not provide sufficient information to evaluate the (Hedges’ g) effect sizes, and, the authors did not response to our mail or willing to provide additional information; (2) the study did not report results for at least one of the aforementioned four outcome variables; (3) the study was a case study; (4) unavailability of the full text of the study; (5) repeated study or the participants overlapped either partially or completely with the subjects of another study or any one included in a meta-analysis. There were 124 eligible studies constituted this study after application of the inclusion and exclusion criteria.
Data Extraction
The following is all variable for efficacy of gaming disorder: (1) the last name of the first author, (2) publication year, (3) type of treatment interventions, (4) the treatment durations (in weeks), (5) number of subjects in the analysis, (6) mean age of subjects, (7) type of comorbid diagnosis (IA, IA+ADHD, and IA+ Depression), (8) type of scales used to evaluate the treatment effects, (9) group therapy (Yes vs. No), and (10) parents involved in the therapy program (Yes vs. No).
Type of treatment interventions
To explore potential influential factors and reduce heterogeneity, we had recoded the type of treatment intervention, type of scales used to evaluate the treatment effects, and the types of subjects as following. Treatment interventions were classified as following five type of treatment: (1) Treat 1: Drugs (Methylphenidate, Escitalopram, Bupropion, Atomoxetine, and Fluoxetine); (2) Treat 2: CBT (Cognitive Behavioral Therapy), Positive Psychology Interventions, Psychological Intervention, Traditional Psychology Intervention, and Integrated Internet Addiction Prevention Program); (3) Treat 3: MLC (Multi-level Counseling Program), So1ution- Focused Brief Therapy (SFBT) with Family Therapy, Adlerian Group Counseling, Multi-family Group Therapy, Solution-Focused Group Counseling, Interpersonal Group Program); (4) Treat 4: Drug + CBT or MLC; (5) Treat 5: Others + CBT or MLC; (6) Treat 6: Others (Electro-acupuncture (EA), Psychology Nursing Intervention, Sports, Negative Health Effect Intervention, Adlerian Group Counseling, Electroencephalographic (EEG) Biofeedback Treatment, Han’s acupoint nerve stimulation (HANS), and Knowledge, Altitude, Belief, and Practice (KABP)).
Type of symptom measurement scales
The type of scales used to evaluate the treatment effects were classified as follows: (1) Type 1: Anxiety (Hamilton Anxiety Scale (HAMA), Screening for Child Anxiety Related Emotional Disorders (SCARED), Self-Rating Anxiety Scale (SAS), SCL-90_Anxiety, State-Trait Anxiety Inventory (STAI)); (2) Type 2: Depression (Hamilton Depression Scale (HAMD), Hamilton Depression Rating Scales (HDRS), Self-Rating Depression Scale (SDS), SCL-90_Depression; Children Depression Inventory (CDI)); (3) Type 3: Severity of internet addiction (Chen/Chinese Internet Addiction Scale (CIAS), Chinese Internet Addiction Scale Revised (CIAS_R), Computer Gaming Addiction Intervention (CGAI), Checklist for the Assessment of Internet and Computer Game Addiction (AICA-C), Impulsive-Compulsive Internet Use Disorder version of the Yate-Brown Obsessive Compulsive Scale (IC-IUD-YBOCS), IAD (Internet Addiction Disorder), self-rating scale (ISS), Internet Addiction Disorder Diagnosis Questionnaire (IAD_DQ), Young's Diagnostic Questionnaire for Internet addiction (YDQ), Young Internet Addiction Scale (YIAS), YIAS-Korean version (YIAS-K), Internet Overuse Self-Rating Scale (IOSRS), Callan Internet Use Scale (CIUS), Self-rating Internet Using Scale (SIUS), Internet Addiction Test (IAT), Adolescent Pathological Internet Use Scale (APIUS), IA Knowledge, Altitude, and Practice (KAP); (4) Type 4: Online time spent scale (Time Management Disposition Scale (TMDS), the severity of Internet use (Hours of Internet Use, SI); (5) Type 5: Others (SCL-90 (Total), Strength and Difficulties Questionnaire (SDQ), Negative health effect (NHE)).
Type of comorbid diagnosis
The types of comorbid diagnosis were classified as follows: (1) PType 1: IA; (2) PType 2: IA comorbid with depression (IA + Depression); (3) PType 3: IA comorbid with ADHD (IA+ ADHD).
Types of study design
The term ‘effect size’ is frequently used in the social sciences, particularly in the context of meta-analysis. Effect sizes typically refer to versions of the standardized mean difference (SMD) as Hedges’ (adjusted) g. However, the value of SMD will be influenced significantly by the choice of control group, mainly, depends on whether the study used an active control (e.g., Drug (Methylphenidate) vs. Drug (Atomoxetine)) or a placebo control (e.g. Drug (Methylphenidate) vs. Placebo). The value of SMD of later should be significantly higher than the former. To be comparable and consistent, in this study, we classified types of study design as follows: (1) StudType 1: For placebo control studies with available mean pre-post change, the SMD of Hedges’ g was calculated using the mean pre-post change in both treatment and control groups. For comparing with active control studies, we treated this type of study as there were two independent groups with mean pre-post change available and calculated each group's unbiased SMDs of Hedges’ g according to the following StudType 2; (2) StudType 2: One group with mean pre-post change available, for convenience, we first used 0.7 as the correlation coefficient between pre- and post-treatment scores to estimate the standard errors of the changes. And then, we used the mean pre-post change and the standard errors of the changes to calculate the unbiased SMD of Hedges’ g; (3) StudType 3: Two groups with mean post-treatment scores available only, that is, the pre-test information of both groups were not presented in the study. The unbiased SMD of Hedges’ g was calculated using the post-test means and standard deviations from both treatment and control groups; (4) StudType 4: A meta-analysis paper, sometimes, only presented unbiased SMD of Hedge's g and its standard errors. In that case, we used both information directly to analyze.
Statistical analysis
Computer application is using STATA/SE version V13.0 for WINDOWS software (Stata Corporation, College Station, TX) to analyze all data. We first used the fixed effects meta-analysis to evaluate the pooled effects size. The random effects meta-analysis will be followed if the testing results of between studies’ heterogeneity is highly significant.
However, the SMDs of Hedges’ g were expected to be comparable but not identical across studies due to highly significant heterogeneity among them (e.g., 5 types of treatment intervention and 6 types of IA scales). For example, there were five types of scales used to evaluate the effect size and as shown in previous section, each scale type included many different formats' scales under same evaluation purpose. It should be noted that the SMD method does not correct for differences in the direction of the scale. Among all evaluation scales, there were mainly two different kinds of trends to represent the treatment effects, that is, either the higher value represents the better treatment effect or, from the reverse direction, the worse effect. If some scales it is essential to multiply the mean values from one set of studies by –1 (or alternatively to subtract the mean from the maximum possible value for the scale) to ensure that all the scales point in the same direction. To be consistent and to facilitate the interpretation of the results, we rearranged all the values of SMDs of Hedges’ (adjusted) g be in one direction according to the characteristics of their original scales, that is, a larger effect size indicates a better effect estimate.
Additional analysis
Meta-regression is appropriate for testing whether certain confounders on the outcome, selected variables (the independent variables) on the effect size (the dependent variable). The influence of the meta-regression analyses was used to compare the potential influential factors on treatment effect. The dependent variable was the SMD of Hedges’ (adjusted) g and assessed by the aforementioned four types of study designed data. The independent variables were publication year, treatment duration, mean age, and aforementioned recoded type of treatment, scales for treatment effect, types of comorbid diagnosis, types of study designed data, group therapy (Yes vs. No), and parents involved in the therapy program (Yes vs. No). The risk of bias was evaluated by funnel plots and Egger’s regression intercept tests for bias. The level of significance was set at p < 0.05.