32 Chapter 2 categories (i.e., general functioning, mental health treatment adherence, psychiatric functioning, psychological functioning, substance use, and criminal behavior) with corresponding subcategories. The complete overview of outcome categories and subcategories can be found in Appendix A, Table 2. Furthermore, outcome characteristics included time of assessment (i.e., intermediate, post, and follow-up assessments). Sample characteristics included age, sex (i.e., percentage of males in the sample), and main psychiatric disorder (i.e., depression and anxiety disorders, schizophrenia and other psychotic spectrum disorders, substance use disorders, and other disorders). Intervention characteristics included type of social network intervention (i.e., group, individual, and combined), type of caregiver (i.e., volunteer, peer supporter, professional, and other), whether or not a personal network member was involved, and duration of the intervention. Study characteristics included type of intervention group (i.e., social network intervention added to treatment as usual and social network intervention alone), type of control group (i.e., treatment as usual with or without an additional intervention and a passive control group), and risk of bias (i.e., high, some concerns, low). The risk of bias was assessed using the revised Cochrane risk of bias tool for randomized trials (Sterne et al., 2019). The overall quality of the scientific evidence was rated using the GRADEpro Guideline Development Tool (GRADEpro GDT, 2021; Schünemann, Brożek, Guyatt, & Oxman, 2013). Corresponding authors of studies were contacted to obtain relevant missing data. Uncertainties or discrepancies in data selection and extraction were resolved through discussion at each stage, until consensus was reached. Meta-analysis One or multiple effect sizes were calculated for each study. Cohen’s d effect sizes reflecting standardized mean differences were computed using means (M) and standard deviations (SD) from intention-to-treat samples with formulas from Lipsey and Wilson (2001). If not available, other values such as frequencies or proportions, M and standard errors (SE), t values or p values were used to compute Cohen’s d. Formulas were used to convert M and SD frommedians, interquartile ranges (IQR), minimum andmaximum values, if appropriate raw statistics were missing (Luo, Wan, Liu, & Tong, 2018; Wan, Wang, Liu, & Tong, 2014). In case of quasi-experimental designs or significant baseline differences between conditions we used a formula for pre-post control designs to calculate Cohen’s d (Morris, 2008). Outliers were checked by searching for effect sizes with standardized scores larger than 3.29 or smaller than -3.29 (Tabachnick, Fidell, & Ullman, 2007). We corrected five outliers to 3.29 above or below average. Continuous moderator variables were centered around its mean and categorical variables with k categories were converted to k-1 dummy variables through binary coding.
RkJQdWJsaXNoZXIy MjY0ODMw