24 Chapter 2 used in multiple reviews of observational studies (27-30). We used the version of the tool adapted by Taylor et al, which judges parameters such as sample selection, description of the cohort, and adequate reporting (31). We excluded 4 out of 10 items because these were only applicable for studies with comparison groups. We added 2 items, to assess whether the association between clinical and personal recovery was a primary or secondary outcome of the study, and to assess the method by which a diagnosis was made. This resulted in 8 items. For every study, we calculated a total quality score, see table 1. Data analysis Effect size calculation To assess the strength of the relationship between measures of clinical and personal recovery, Pearson’s correlation coefficient r was used as the measure of effect size. According to Cohen, r = .10 is considered a small effect, explaining 1% of the total variance); r = .30 a medium effect, accounting for 9% of the total variance) and r = .50 a large effect, accounting for 25% of the variance (32). When correlation coefficients were not given, they were either calculated from reported data according to Lipsey and Wilson, or estimated from reported standardized regression coefficients (33). Regarding the personal recovery category of hope, the direction of the reported correlation coefficient was reversed wherever necessary, such that all included effect sizes represented the association between clinical recovery and hope, instead of hopelessness. Integration of dependent effect sizes To ensure that every study only contributed one effect size to the analysis, we calculated average effect sizes within one study if multiple effect sizes were reported on similar outcomes (34-36). This was the case if 2 or more measures were used to assess overall symptom severity, or (an aspect of) personal recovery. When studies only reported the outcome of assessed subscales and not of the total score, we averaged effect sizes of subscales to estimate an overall effect size. For example, reported correlation coefficients between personal recovery and positive symptoms, negative symptoms and general psychopathology of the PANSS were averaged, to calculate an overall correlation between personal recovery and symptom severity. Meta-analysis procedure To account for expected heterogeneity between studies, random-effects models according to Hedges and Vevea (37) were calculated to obtain a combined effect weighted for sample size. The relationships between symptom severity on the one
RkJQdWJsaXNoZXIy MTk4NDMw