60 Chapter 2 An important strength of this meta-analysis was the extensive search strategy and data extraction that resulted in a comprehensive overview of the effectiveness of social network intervention on outcomes that have not been quantified by previous meta-analyses. Examination of multiple effect sizes allowed us to examine all data while achieving maximum statistical power and accurately dealing with dependency in effect sizes within studies (Assink & Wibbelink, 2016). Noteworthy, researchers have made suggestions to improvemultilevel approaches to sufficiently account for the different forms of dependency in effect sizes within studies (e.g., dependency in effect sizes of outcomes measures at multiple timepoints and dependency in effect sizes of multiple outcomes; see e.g., Assink & Wibbelink, 2023; Fernández-Castilla et al., 2021a). Additionally, this study attempted to contribute to personalized care by examining a wide range of potential moderators. Lastly, we found no indication of publication bias on the majority of treatment outcomes after careful examination of multiple methods. This study has several methodological limitations that might have affected our estimation of effectiveness. First, despite our intention to provide a comprehensive overview of the effectiveness of social network interventions for psychiatric patients, we were unable to include the results of three eligible studies in this meta-analyses due to missing data (i.e., unadjusted means and SDs were unavailable) (O'Connell et al., 2018; O’Connell, Flanagan, Delphin-Rittmon, & Davidson, 2020; UKATT Research Team, 2005). However, we expect that inclusion of these studies would not have impeded our findings, as significant improvements on treatment outcomes (e.g., substance use, psychiatric symptoms, and mental healthcare use) were reported in two studies comparing TAU with the addition of a social network intervention to TAU alone in psychiatric patients (O'Connell et al., 2018, 2020). Second, data regarding interventions and sample characteristics (i.e., intensity of interventions and psychiatric disorders) were not reported fully. Therefore, it was not possible to differ between subgroups of participants and interventions. Although one subgroup effect for groups with more females was found in our study, these results should be considered as a first exploration. Third, different experimental and control groups were used. However, differences in groups did not seem to influence the results. Fourth, we did not account for intervention adherence levels, which might have impeded our findings, as adherence may vary in social network interventions (Siette et al., 2017). Lastly, the overall risk of bias was moderate-to-high, and the quality of the evidence as well as power, based on the GRADE approach and funnel plots, was moderate-to-low (Appendix A), thus results need to be interpreted with caution (Button et al., 2013). Nonetheless, it should be noted that a low quality of evidence and statistical power is often found in psychosocial intervention studies, and the reliability of quality assessments can be low (Cuijpers, 2003;
RkJQdWJsaXNoZXIy MjY0ODMw