Changes in social functioning over the course of psychotic disorders–A meta-analysis

In this meta-analysis we investigated changes in social functioning and its moderators in patients with a psychotic disorder but different durations of illness at baseline. We included longitudinal studies assessing the course of five domains of social functioning in patients with a psychotic disorder. Effect sizes of change between baseline and follow-up within these domains were analyzed in four subgroups based on durations of psychotic disorder at baseline: less than 2 years, between 2 and 5 years, between 5 and 10 years, and more than 10 years. The influence of baseline confounders was analyzed using meta-regression and sensitivity analysis. We included 84 studies analyzing 33,456 participants. We found a medium improvement (d = 0.60) in overall social functioning over time, with a greater improvement for studies investigating patients with a duration of illness of less than 5 years. We found minor improvement in specific domains of social functioning, such as vocational functioning (d = 0.31), prosocial behavior (d = 0.36), activities (d = 0.15), and independence (d = 0.25). Improvement in social functioning was associated with lower baseline levels of negative symptoms, higher baseline levels of quality of life, and, specifically, improved vocational functioning, with rehabilitation and combined treatment. Social functioning in patients with psychotic disorders improves over time, especially for patients with shorter illness durations. Reduction of negative symptoms and improving quality of life might reinforce improvement of social functioning.


Introduction
Psychotic disorders often lead to functional limitations and substantially impact individuals, their loved ones and society (Linscott and Van Os, 2013;Van Os and Reininghaus, 2016;Sullivan et al., 2020). The majority of patients with psychotic disorders have difficulties maintaining their societal roles, such as being employed or maintaining relationships, also after symptomatic remission (Bellack et al., 2016;Madeira et al., 2016). This often leads to a more chronic course of psychotic disorders (Linscott and Van Os, 2013;Santesteban-Echarri et al., 2017). Therefore, improving social functioning, which is defined as regaining societal roles (Mueser and Tarrier, 1998), is a major aim in recovery-oriented treatment and research.
Changes in social functioning are associated with a wide variety of factors, such as symptomatic remission, duration of untreated psychosis, neurocognition and social cognition, hope, optimism, and quality of life (Coşkun and Altun, 2018;Górna et al., 2014;Hasson-Ohayon et al., 2009;Heeramun-Aubeeluck et al., 2015;Javed and Charles, 2018). These changes could also be associated with duration of illness and follow-up durations. Previous reports indicated that levels of social functioning are lower for patients with longer illness durations (Frascarelli et al., 2015;Preston, 2000) and that social impairments persist over time (Wiersma et al., 2000). However, it is unknown to what extent long-term changes in social functioning depend on baseline illness duration, duration of follow-up or other factors. This information is crucial to identify optimal windows of opportunity to enhance improvement in social functioning.
We investigated changes in social functioning between baseline and follow-up assessments in longitudinal studies investigating patients with different durations after onset of a psychotic disorder at baseline (duration of illness) and follow-up periods (duration of follow-up). Furthermore, we investigated which factors are associated with any changes. We included studies that investigated patients with any type of psychotic disorder (including other psychiatric disorders with psychotic features), because patients with different types of psychotic disorders show comparable changes in social functioning over time (Bottlender et al., 2010;Rymaszewska et al., 2007;Möller et al., 2000). A previous meta-analysis (Santesteban-Echarri et al., 2017) has already explored the impact of factors influencing social recovery. However, this is the first meta-analysis evaluating both changes in social functioning over time in patients with different durations of illness at baseline and different follow-up periods and which factors contribute to these changes. We aimed to answer the following questions: 1) To what extent do different domains of social functioning change over the course of psychotic disorders? 2) Which factors at baseline are associated with changes in social functioning over time?

Search strategy
Records were identified through searches in PubMed, PsycInfo, CINAHL, and Cochrane of peer-reviewed journals until July 2018. The search was based on terms related to psychotic disorders, chronicity, course, recovery, and remission (see Supplementary Material A). Additional references were traced through reference lists of identified studies and systematic reviews.

Eligibility criteria
Two authors (LdW & KK) independently executed study selection. Disagreements were resolved by consensus. The included studies meet the following criteria: 1) Patient population: Adults (age ≥ 18) meeting a DSM or ICD diagnosis of what is currently indicated as schizophrenia spectrum and other psychotic disorders (American Psychiatric Association, 1980;American Psychiatric Association, 2000;American Psychiatric Association, 2013), or other psychiatric disorders with psychotic features (i.e., the presence of delusions, hallucinations, and/or thought interferences without insight [Linscott and Van Os, 2013]). 2) Study design: Longitudinal cohort study or randomized controlled trial, with at least 1 year follow-up, assuring long-term follow-up evaluations.
3) Outcomes: Studies reporting uncorrected quantitative measurements of social functioning for at least two time points. In case of multiple follow-up assessments, all measurements were analyzed. 4) Publication: Published in English in peer-reviewed journals.

Outcome domains
After study selection, we categorized the study outcomes into five separate outcome domains: 1) overall social functioning: overall functioning in any domain; 2) prosocial behavior: level of social skills, relationships, and social adaptive behavior; 3) independence: level of independent functioning; 4) activities: level of engagement in prosocial and leisure activities; 5) vocational functioning: involvement in employment or education.
The selection of outcome domains was based on primary or secondary outcomes used in the included studies, frequently used functional outcome assessment instruments (e.g. Birchwood et al., 1990;Morosini et al., 2000;De Wolf et al., 2012), and categorizations used in previous studies (Lloyd et al., 2008). Supplementary Material B provides a complete overview of assessments of each outcome domain.

Assessment of duration of illness and follow-up subgroups
The included studies investigated patients with different durations of illness at baseline, and assessed outcomes over different follow-up periods. Previous studies attempted to stratify patients with psychotic disorders into different stages of illness (Lieberman et al., 2001;McGorry et al., 2010;Tandon et al., 2009). However, included studies lacked detailed information to follow this stratification process. Therefore, we categorized studies according to the patients' duration of illness at baseline, partly based on categorizations of early and chronic stages of psycho in previous studies (e.g. Breitborde et al., 2009;Frascarelli et al., 2015;Preston, 2000), as follows: 1) duration of illness <2 years; 2) duration of illness between 2 and 5 years; 3) duration of illness between 5 and 10 years; 4) duration of illness >10 years. Studies of which duration of illness was unknown, were analyzed separately. Within each baseline duration of illness subgroup we also divided the included studies into separate subgroups based on their follow-up periods: 1) follow-up duration <2 years; 2) follow-up duration between 2 and 5 years; 3) follow-up duration between 5 and 8 years; 4) follow-up duration >8 years.
All subgroups are presented in Text Box 1. This overview shows that combinations of illness duration and duration of follow-up do not lead to mutually exclusive categories of study data. However, given the nature of the studies we selected, we considered clustering studies in these separate duration of illness and follow-up subgroups the optimal classification for this study.

Selection and assessment of moderators of outcome
First, we investigated the influence of study design (RCT's versus cohort studies) and diagnosis (studies only including schizophrenia patients versus studies also including other psychotic disorders) on the outcomes.
Other potential moderators of social functioning were selected following a two-step procedure. First we identified 52 significant moderators in included studies and comparable meta-analyses (Santesteban-Echarri et al., 2017;Fusar-Poli et al., 2015;Ś witaj et al., 2012). Second, we applied the following three criteria reported in the Cochrane Handbook 5.1 (Higgins and Green, 2011): 1) reported by at least 10 of the selected studies; 2) ability to be clustered in separate multivariate models; 3) Not closely related to each other to prevent multicollinearity. This resulted in 19 moderators, which we clustered in seven multivariate regression models: 1) treatment variables: implementation of rehabilitation, psychotherapy, antipsychotic use, and combined treatment to (a subsample of) the participants; 2) symptoms: positive symptoms, negative symptoms, depression, and substance use at baseline; 3) demographic variables: years of education and gender; 4) study characteristics: publication year, and attrition rate; 5) overall neurocognition at baseline; 6) illness related variables: clinical stabilization at baseline, age at onset, DUP, and setting in which the study is executed (i.e., naturalistic or healthcare); 7) subjective quality of life and level of social functioning at baseline.
From continuous moderators that were evaluated by different assessment instruments (i.e., assessments of symptoms, neurocognition, subjective quality of life, and social functioning) we calculated percentile scores based on normative data to ensure that each moderator was assessed in the same scale range. Operationalizations of each moderator are reported in Supplementary Material E.

Quality assessment
Quality assessment was conducted using the Quality in Prognostic Studies (QUIPS) tool (Hayden et al., 2013) and was based on six criteria: participation, attrition, prognostic factor measurement, handling confounders, outcome measurement, and analysis and reporting. Based on these criteria a high, moderate or low risk of bias score was assigned for each study.
Two authors (LdW & MO) independently conducted quality assessment of 10% of the studies. The level of agreement was fair to good (κ = 0.56). Disagreements were resolved by consensus. We investigated the influence of study quality on outcomes by sensitivity analysis.

Meta-analytic procedure
Meta-analyses were conducted using RevMan 5.3 (The Nordic Cochrane Centre, 2014). Effect sizes were calculated by comparing study outcomes between baseline and follow-up assessment. For clinical trials the total study sample, clustering both treatment and control groups, was analyzed. Overall effect sizes of categorical outcomes were converted into Cohen's d (Chinn, 2000) to show homogeneous and consistent patterns for both continuous and categorical outcomes. Magnitude of effect was considered marginal and clinically not relevant when d < 0.2, small when d ≥ 0.2 and <0.5, medium when d ≥ 0.5 and <0.8, and large when d ≥ 0.8 (Chinn, 2000). All outcomes were reported with 95% confidence intervals (CIs). We used random effects models, weighted by the method of inverse variance (Higgins, 2008). Statistical heterogeneity was assessed by calculating the I 2 statistic (including 95% CI), describing the percentage of observed heterogeneity not expected by chance (Higgins and Thompson, 2002).

Subgroup analyses
We analyzed differences in effect sizes of change between subgroups regarding the baseline duration of illness and duration of follow-up (Borenstein and Higgins, 2013). Because of the large number of subgroups, there is a high chance of finding type-I errors in one of our subgroup analyses. Therefore, we controlled for multiple testing effects through a Benjamini-Hochberg correction, with the false discovery rate set on 0.3 (Benjamini and Hochberg, 1995).

Calculation of moderating effects
We investigated the influence of potential moderators on the five outcome domains through a meta-regression analysis using Stata version 12 (StataCorp, 2011). We conducted meta-regression analyses for all study outcomes and further investigated the influence of significant moderators within different duration of illness subgroups using a sensitivity analysis, comparing study outcomes of studies with high levels or presence versus low levels or absence of the respective moderator. Because of the large number of moderators and subgroups, we controlled for multiple testing effects in all analyses through a Benjamini-Hochberg correction (Benjamini and Hochberg, 1995).

Handling outliers and publication bias
Potential influence of outliers (i.e. confidence interval [CI] of study outcomes exceeded overall CI) was handled by re-analyzing the metaanalysis after removing the outliers. Potential publication bias was detected by visual inspection of funnel plots.

Study characteristics
We selected 84 studies describing the course of social functioning of 33,456 participants. The mean age of participants was 33.4 years (SD = 11.5), and 39.3% were female (see Table 1).
Thirty-seven studies (44.0%) also included participants with other psychiatric conditions with psychotic features (see Table 1). Thirty-four studies (40.5%) were clinical trials, and 50 (59.6%) were cohort studies. In 38 studies (53.5%) at least 80% of the participants received antipsychotic medication, in 22 studies (31.0%) participants received any kind of rehabilitation intervention, in 33 studies (46.5%) participants received psychotherapy, and in 25 studies (35.2%) participants received combined treatment with at least two of the aforementioned treatment components.
Differences in baseline study and patient characteristics between the baseline duration of illness subgroups are presented in Supplementary material C. The subgroups did not significantly differ in most characteristics. However, we found that study samples with a longer duration of illness were older, had more severe substance abuse, used antipsychotics more often and were more frequently diagnosed with schizophrenia than subgroups with a shorter duration of illness.

Meta-analysis of study outcomes with different durations of psychosis
A general overview of the outcomes, within each duration of illness and follow-up subgroup is presented in Fig. 2 and Table 2. We also added forest plots of study outcomes in Supplementary materials D.

Overall social functioning
In general, we found a medium improvement in overall social functioning (d = 0.60). For the studies with a shorter baseline duration of illness (i.e., < 2 years and 2-5 years). Specifically for the subgroup with a baseline duration of illness <2 years, we found a large improvement in overall social functioning after a longer follow-up duration (χ 2 = 50.83; df = 2; p < 0.01). For the subgroup with a baseline duration of illness between 5 and 10 years we found no improvement and for the subgroup with a baseline duration of illness >10 years we found a small improvement in overall social functioning. Both subgroups with <2 years and 2-5 years of illness at baseline showed larger improvement over time than the subgroup with a baseline duration of illness > 10 years (χ 2 = 15.30; df = 1; p < 0.01 and χ 2 = 7.71; df = 1; p < 0.01).

Prosocial behavior
Overall, we found a small improvement in prosocial behavior (d = 0.36). We observed a large improvement in prosocial behavior for the subgroup with a baseline duration of illness 5-10 years after a short follow-up duration. For the subgroup with a duration of illness >10 years at baseline we found small improvement in prosocial behavior, with no differences between short and long follow-up outcomes. The subgroup with a baseline duration of illness between 5 and 10 years showed a greater improvement after short follow-up than the other subgroups (χ 2 = 13.28; df = 1; p < 0.01; χ 2 = 11.61; df = 1; p < 0.01; χ 2 = 13.28; df = 1; p < 0.01).

Independence
Overall, we found a small improvement in independence (d = 0.25). We found a large improvement of independence after a short follow-up duration in the subgroup with a baseline duration of illness 5-10 years and a small improvement of independence with greater improvement for study outcomes with shorter follow-up durations in the subgroup with a baseline duration of illness >10 years (χ 2 = 21.29; df = 3; p < 0.01).

Activities
Overall, we found no improvement in activities (d = 0.15). We found a small improvement over time for studies with a baseline duration of illness of less than 2 years. We found no improvement over time for subgroups with a longer baseline duration of illness.

Vocational functioning
Overall, we found a small improvement in vocational functioning (d = 0.31). We found a medium improvement after a short follow-up and a large improvement after long follow-up for the subgroup with a baseline duration of illness >10 years. Differences in improvement between short and long follow-up were significant (χ 2 = 27.92; df = 3; p < 0.01). We found no improvement in vocational functioning for the subgroup with a shorter baseline duration of illness (i.e. <2; 2-5 or 5-10 years).

Outliers and publication bias
We found 13 positive and 7 negative outliers for overall social functioning outcomes, 16 positive and 4 negative outliers for prosocial behavior, 1 positive outlier for independence, and 3 negative outliers for vocational functioning. Excluding outliers did not significantly influence any study outcome.
We found asymmetrical funnel plots, indicating publication bias, for overall social functioning and prosocial behavior (see Supplementary Material H). For overall social functioning mainly study outcomes with a baseline duration of illness <2 years and 2-5 years and for prosocial behavior larger studies with a duration of illness between 5 and 10 years at baseline positively influenced the outcomes.

Analysis of potential moderators of outcome at baseline
Meta-regression outcomes and sensitivity analyses are presented in Supplementary Material E and Table 3. For some outcome domains moderators were excluded, because data were available for less than 10 studies.

Overall social functioning
Meta-regression showed that baseline levels of depression, positive symptoms, negative symptoms, subjective quality of life, and overall social functioning were significant moderators for changes in overall social functioning. Subsequently, sensitivity analyses indicated that higher baseline levels of positive symptoms, subjective quality of life, and overall social functioning, and lower baseline levels of negative symptoms was associated with greater improvement in overall social functioning (χ 2 = 16.24; df = 1; p < 0.01; χ 2 = 8.64; df = 1; p < 0.01; χ 2 = 24.76; df = 1; p < 0.01; χ 2 = 8.48; df = 1; p < 0.01). The influence of baseline positive and negative symptoms and baseline subjective quality of life applied to the subgroup with a duration of illness <2 years. For both baseline negative symptoms and overall social functioning the influence also applied to the subgroup with a duration of illness between 5 and 10 years.

Prosocial behavior
Meta-regression outcomes showed that baseline levels of positive symptoms and substance use, and a health care setting were moderators for changes in prosocial behavior. Sensitivity analyses indicated that higher baseline levels of positive symptoms, and studies executed in a  health care setting were associated with greater improvement in prosocial behavior (χ 2 = 9.71; df = 1; p < 0.01; χ 2 = 4.31; df = 1; p < 0.05).

Independence
Meta-regression outcomes showed that study samples with a schizophrenia diagnosis, and baseline levels of independence were significant moderators for changes in independence. Studies evaluating patients with high levels of baseline independence (χ 2 = 9.72; df = 1; p < 0.01) and studies in which not the whole sample had schizophrenia (χ 2 = 13.03; df = 1; p < 0.01) reported greater improvement in independence. The influence of baseline independence also applied to the subgroup with a duration of illness >10 years at baseline (χ 2 = 13.79; df = 1; p < 0.01).

Activities
Meta-regression outcomes showed that publication year was a moderator for changes in activities. Sensitivity analyses indicated that studies that were published less than 10 years ago reported stronger improvement in activities than older studies (χ 2 = 16.24; df = 1; p < 0.01), especially in the subgroup with a duration of illness between 5 and 10 years after baseline (χ 2 = 64.24; df = 1; p < 0.01).

Vocational functioning
Meta-regression showed that rehabilitation, combined treatment, psychotherapy, depression, negative symptoms, positive symptoms, health care setting, publication year, and baseline vocational functioning are significant moderators for changes in vocational functioning.
Finally, studies conducted in a health care setting (χ 2 = 54.29; df = 1; p < 0.01), published less than 10 years ago (χ 2 = 4.04; df = 1; p < 0.05) and studies evaluating patients with high baseline vocational functioning (χ 2 = 31.64; df = 1; p < 0.01) show greater improvement in vocational functioning than studies without these features. These differences applied to subgroups with both a baseline duration of illness <2 years and a baseline duration of illness 5-10 years.

Quality assessment
The quality assessment and its sensitivity analysis are presented in Supplementary Material F and G. High risk of bias, and lower study quality, was specifically indicated on a substantial number of studies for study attrition (26.2%) and prognostic factor measurement (36.9%).
Although the QUIPS items study attrition and prognostic factor measurement significantly influenced all outcome domains, the direction of the influence of these QUIPS items varied. Therefore, we did not find a consistent trend of influence of study quality of any of the QUIPS items.

Discussion
This meta-analysis investigated changes in social functioning and moderators of change in patients with psychotic disorders, with different durations of illness and duration of follow-up.
We observed medium improvement in overall social functioning, with greater improvement in those within the first 5 years of illness after a longer duration of follow-up. We found small improvement in vocational functioning, prosocial behavior and independence, specifically in subgroups with a baseline duration of illness of more than 5 years. We found no overall improvement of activities.
The results we found are in line with previous landmark longitudinal cohort studies, such as IPSS (Leff et al., 1992) that also found long-term improvement of social functioning for patients with psychotic disorders. Results are also in line with earlier studies indicating that patients with shorter illness duration at baseline showed more substantial improvement in social functioning than patients with longstanding psychosis (Frascarelli et al., 2015;Preston, 2000). Our findings also support the idea that the first 5 years after onset of a psychotic disorder could be labeled as a "critical period of recovery" (Birchwood et al., 1998), in which patients can achieve more improvement in social functioning (Luther et al., 2020). However, we observed small or no improvement in the other outcome domains of social functioning during the first five years of illness, though these results were based on a limited number of study outcomes. This emphasizes the need for more studies investigating specific domains of social functioning during early psychosis. The improvement in vocational functioning, prosocial behavior and independence in patients with a longer baseline duration of illness shows hopeful patterns of improvement for chronic patient populations, but also stresses the need for a focus on improvement in these domains for patients with early psychosis.
After controlling for multiple testing effects, we found indications Duration of illness >10 years + = 3 (15.79%)/-= 0 (0.00%)   Outcomes in bold are significant (p < 0.05) after Benjamini-Hochberg correction; Outcomes underlined are no longer significant after Benjamini-Hochberg correction for multiple testing. 1 Significant subgroup differences with the duration of illness <2 years subgroup outcome within the same follow-up cohort.
that high levels of baseline positive symptoms and social functioning, low levels of baseline negative symptoms and studies published in more recent publications were associated with more improvement in multiple domains of social functioning. Furthermore, we found that a high level of baseline subjective quality of life was associated with improvement in overall social functioning and that the presence of specific rehabilitation, or combined treatment, and the absence of psychotherapy were associated with improvement in vocational functioning. The positive influence of high baseline levels of positive symptoms on improvement in social functioning contradicts previous studies indicating that lower severity of psychotic symptoms is an important predictor for social recovery (Alvarez-Jimenez et al., 2012;Bottlender et al., 2010). The results might be explained by the fact that patients with more severe symptoms have a higher level of functional impairment (Rymaszewska et al., 2007) and thereby greater potential for improvement in social functioning. The negative association between baseline levels of positive symptoms and functioning at baseline (r = − 0.48; p < 0.01) in our included studies corroborates this explanation.
Furthermore, the positive association between low levels of baseline negative symptoms and improvement in social functioning is in accordance with previous findings (Albert et al., 2011;Bottlender et al., 2010;Gee et al., 2016;Möller et al., 2000). This might be explained by the conceptual overlap between features of negative symptoms (e.g. apathy and speech problems) and social functioning and the negative association between negative symptoms and neurocognition, social cognition and adherence to treatment (Bliksted et al., 2017;Ventura et al., 2015), which may hamper social recovery. In our report we could not replicate these negative associations, due to lack of study outcomes and lack of heterogeneity of neurocognition assessments. Therefore, we recommend further investigation of the etiology and pathobiology of negative symptoms and possibilities for integrating interventions targeting negative symptoms within functional rehabilitation (Gee et al., 2016;Stiekema et al., 2018;Fervaha et al., 2014).
The positive influence of baseline subjective quality of life on the improvement in overall social functioning confirms previous findings (Burns-Lynch and Musa, 2016;Lambert et al., 2009). This might be explained by the fact that better subjective quality of life might lead to increased engagement in social roles due to increased hope and optimism and a reduced "why try?" effect (Corrigan et al., 2009).
The positive association between recent publications and improvement in activities and vocational functioning might give some first indications for a shift towards greater emphasis on social functioning in standard care for psychosis. We recommend further elaboration of this trend in future research.
Furthermore, studies delivering rehabilitation and combined treatment to the study (sub)sample are associated with improvement in vocational functioning especially for patients with a short illness duration at baseline. This is in line with previous studies indicating beneficial vocational outcomes for vocational rehabilitation programs, such as individual placement and support (IPS), in early intervention services (Bond et al., 2015;Rinaldi et al., 2004). The negative influence of psychotherapy on vocational outcomes might be explained by the fact that most of the psychotherapy studies were not focused on rehabilitation or combined treatment and thereby less focused on vocational rehabilitation.
It is important to consider that we analyzed the whole study sample of each study, so we analyzed both the intervention and the control condition. Therefore, intervention effects do not exclusively explain changes in vocational functioning. The results could be explained by the 'Hawthorne effect' which indicates that being a subject of social investigation might explain the behavior-modifying effect (Wickström and Bendix, 2000). We recommend future research investigating long-term effects of different types of treatment and treatment adherence on different levels of social functioning to put current results into perspective.
Finally, we found a negative association between a diagnosis of schizophrenia and improvement in independence. This indicates that a more severe and chronic pattern of psychotic disorders might affect improvement in this outcome domain. However, both study design and study sample did not influence the other outcome domains in this metaanalysis. Therefore, the broad inclusion norms increase the generalizability of our findings with limited influence on the heterogeneity of study outcomes.
A possible important explanation for the results we found might be explained by the fact that the duration of illness subgroups might be biased and censored because sample characteristics between these subgroups differed at baseline. However, in our meta-analysis we found no indications of such a sampling effect, except for the fact that studies with a longer duration of illness were more often diagnosed with schizophrenia than studies with a shorter duration of illness. This might have influenced outcomes as a schizophrenia diagnosis is negatively associated with improvement in independence. Nevertheless, the influence of this moderator is very limited, so the results could not be explained by sampling effects.
There are several limitations to address. First, the subgroup and sensitivity analyses were often based on a limited number of studies with heterogeneous outcomes, making the outcomes less reliable (Böhning et al., 2017). The high heterogeneity might be explained by the fact that social functioning remains a complex and disputed construct with low psychometric quality (Bellack et al., 2006). Although heterogeneity of study outcomes in complex meta-analyses are often inevitable and could not directly translated to clinical implications of study outcomes (Ioannidis, 2008), we partly explained heterogeneity by executing metaregression analyses on potential moderators of outcomes. Quality assessment also revealed lower quality of a few included studies. However, the sensitivity analysis did not indicate a significant influence of study quality on outcomes. Furthermore, although subgroup and sensitivity analyses were necessary to answer our research questions, the relatively high number of analyses might have caused alpha inflation. Therefore, we executed a Benjamini-Hochberg correction on all significant outcomes to test for potential type-I errors. Furthermore, we could not analyze the influence of potentially relevant moderators, such as stigma, social cognition, premorbid functioning, regional differences or ethnic groups due to limited studies reporting on these factors. These moderators would be valuable to investigate in future research. Finally, indications of publication bias and high numbers of positive outliers might have inflated study outcomes, though analyses of positive outliers does not support this possibility.
Our findings show hopeful patterns of improvement in social functioning in the first 5 years of illness. However, even patients with a longer duration of illness improve in distinct outcome domains of social functioning. This stresses the needs for extensive intervention services. Reduction of negative symptoms and improvement in subjective quality of life might amplify improvement in social functioning. Further research into specific interventions might help to further unlock the social potential of patients with psychotic disorders. 2 Significant subgroup differences with the duration of illness 2-5 years subgroup outcome within the same follow-up cohort. 3 Significant subgroup differences with the duration of illness 5-10 years subgroup outcome within the same follow-up cohort. 4 Significant subgroup differences with the duration of illness >10 years subgroup outcome within the same follow-up cohort. * significant (p < 0.05) * Subgroup differences between follow-up cohorts   Prof. Jaap van Weeghel, PhD., Jaap van Weeghel is a social scientist working at Phrenos Center of Expertise for Mental Illnesses. He is also emeritus professor at Tranzo Scientific Center for Care and Welfare, Department of Social and Behavioral Sciences, Tilburg University, The Netherlands. He studies various aspects of the recovery, rehabilitation and social inclusion of people with psychotic disorders or other severe mental illnesses.
Prof. Ilanit Hasson-Ohayon, PhD. Ilanit Hasson-Ohayon is a rehabilitation psychologist and full professor at the department of psychology in Bar-Ilan University, Israel. She studies different psychological aspects of coping with illnesses and disabilities.
Dr. Jentien Vermeulen. Jentien Vermeulen is a psychiatrist in training and post-doctoral researcher in the field of psychosis and addiction at the Amsterdam University Medical Centers, location AMC. She has a specific track record in the etiology, interventions and prevention of tobacco smoking and severe mental illness, both in research  as in clinical practice. Prof. Dr. Cornelis Mulder. Cornelis Mulder is a psychiatrist and professor of public mental health. He is program leader of the Epidemiological and Social Research institute at Erasmus University Medical Centre, department of psychiatry, psychiatrist and teacher at Antes/ Parnassia Psychiatric Institute. He is involved in research projects concerning help seeking behavior, motivation and compliance, dual diagnosis, victimisation, emergency psychiatry, assertive outreach, and coercion and has published over 200 (inter)national scientific articles (see pubmed), chapters and several books on these matters.
Prof. Dr. Nynke Boonstra Nynke Boonstra is a mental health nurse practicioner at the early intervention service of KieN VIP and professor Healthcare and Innovation in Psychiatry at NHL Stenden University of Applied Science, the Netherlands. She studies various aspects of societal and personal recovery in patients with a psychotic disorder.
Kete Klaver, MSc. Kete Klaver is neuropsychologist and PhD candidate at the Netherlands Cancer Institute in Amsterdam. Currently, she is conducting a randomized controlled trial into the effectiveness of a cognitive rehabilitation program for working cancer survivors.
Matthijs Oud, MSc. Matthijs Oud is a research officer and is working at Trimbos Institute, the Netherlands. His main topic of research is conducting systematic reviews and meta-analyses on different topics of mental health care.
Prof. Dr. Lieuwe de Haan. Lieuwe de Haan, is a psychiatrist and Professor of Psychiatry in Amsterdam Medical Center in the Netherlands. Prof. Dr. Lieuwe de Haan has extensive experience in different research topics, especially focused on early psychosis and schizophrena.
Prof. Dr. Wim Veling. Wim Veling, MD, PhD, is a psychiatrist and adjunct Professor of Psychiatry in the Department of Psychiatry at University Medical Center Groningen, University of Groningen, the Netherlands. His research focuses on the social context of psychosis and other psychiatric disorders. He uses epidemiology and virtual reality as a tool for understanding of psychosocial mechanisms and treatment of psychiatric disorders.