Introduction
Method
Eligibility criteria
Search strategy
Screening and study selection
Narrative synthesis
Quality assessment
Meta-analysis
Eligibility criteria
Data extraction and effect size computation
Statistical analysis
Results
Study selection
Study characteristics
Author, year, country | Sample N(M/F) % SS | Social network measure | Outcome measure | Outcome category | Design (timepoints) | Key findings |
---|---|---|---|---|---|---|
Allison et al. (2013)
UK
| 24 (18/6) 100% | Modified PRQ | PANSS (P, N) HADS | Symptom severity (positive, negative & affective) | Cross-sectional correlational | 1. No sig correlations between size of network and positive (rho = − 0.15), negative (rho = − 0.27), anxiety (rho = − 0.09), or depression (rho = 0.01) symptoms |
Angell & Test (1992)
USA
| 87 (62/25) 98% | CAF—study specific | BPRS-18 (P) | Symptom severity (positive) | Cross-sectional correlational + longitudinal (18 + 24 months) | 1. N = 122 in RCT, but complete data for 87 (71.3%) due to attrition 2. Cross-sectional analyses at T1 showed no sig correlation between positive symptoms and network size (rho = − 0.07) 3. Cross-sectional analyses at T2 showed no sig correlation between positive symptoms and network size (rho = − 0.18) 4. OLS regression showed no sig association between T1 positive symptoms and T2 network size (β = − 0.11) over six month period, controlling for education, gender, age and treatment allocation 5. OLS regression found increase in positive symptoms over a six month period (T1 to T2) was sig associated with declines in network reciprocity (β = − 0.23) but not size (β = − 0.17) |
Becker et al. (1998)
UK
| 143 (80/63) 83% | SNS | LQoLP | QOL | Cross-sectional correlational (baseline) | 1. Sample grouped according to quintiles of network size, from 1 (small network, 1–6 contacts) to 5 (large network, 19–42 contacts) 2. ANOVA showed sig quadric contrast between social network quintiles in terms of average QOL. Average QOL increased up to quintiles 4 and 5; sig contrasts between these two and the lower three quintiles 3. In multiple linear regression analyses, average QOL was sig positively associated with higher network quintiles, with levelling off at around 20 contacts (β = 0.61, CI = 0.25, 0.96). This remained after adjusting for age, BPRS anxiety/depression, service satisfaction, and number of unmet needs |
Cechnicki & Wojciechowska (2008)
1
Poland
| 64 (28/36) 100% | BQ | BPRS-24 (G, N, P) DSM-III: social functioning | Symptom severity (overall, positive, negative) Social functioning | Cross-sectional correlational (7 years after first admission) | 1. Larger social networks sig associated with less severe BPRS overall (rho = − 0.32), positive (rho = − 0.42) and negative symptoms (rho = − 0.38) and improved social functioning (rho = 0.38) 2. Large extra-familial networks sig related to less severe BPRS overall (rho = − 0.29), negative (rho = − 0.30) and positive (rho = − 0.38) symptoms and improved social functioning (rho = 0.44) |
Cechnicki et al. (2008)1
Poland
| 64 (28/36) 100% | BQ | LQoLQ | QOL | Cross-sectional correlational (7 years after first admission) | 1. Large social network sig correlated with higher general subjective satisfaction with QOL (rho = 0.35) 2. Larger extra-familial network did not correlate with general QOL (rho = 0.12) |
Cohen et al. (1997)
USA
| 117 (29/88) 100% | Modified NAP | TSC | QOL | Cross-sectional correlational | 1. 47% (n = 54) satisfied and 53% (n = 63) not satisfied with their lives and QOL over the past 30 years 2. Satisfied group sig more likely to have more network members who could be ‘counted on’ (t = − 2.52, d = − 0.47, CI = − 0.84, 0.10) and greater network density (t = − 2.73, d = − 0.51, CI = − 0.88, − 0.14) 3. No sig group differences in total network size (t = 0.90, d = 0.17, CI = − 0.20, 0.53) 4. Multivariate logistic regression model showed more reliable social contacts (could be ‘counted on’) sig predicted greater satisfaction with QOL, adjusting for socio-demographics (AOR = 4.68, CI = 1.36, 16.08). Network density was not a sig predictor of QOL (AOR = 2.94, CI = 0.43, 20.18) |
Cresswell et al. (1992)
UK
| 40 (31/9) 100% | SNIS | BPRS-18 (G) SANS | Symptom severity (overall and negative) | Cross-sectional correlational | 1. Higher severity of overall symptoms sig related to smaller primary group (family/friends) (r = − 0.36), secondary group (outside family/friends) (r = − 0.47) and primary group seen weekly (r = − 0.32) 2. Higher severity of negative symptoms sig related to smaller network size in primary group (r = − 0.42) and primary group seen weekly (r = − 0.41), but not secondary group (r = − 0.21) |
Dixon et al. (2001)
2
USA
| 218(123/95) 72% | Study specific | PANSS (T) | Symptom severity (overall) | Cross-sectional, correlational | 1. Hierarchical OLS regression showed that total PANSS symptom severity was inversely associated with the size of social support network (r = − 0.21), which remained when controlling for demographic covariates (age, education, gender, ethnicity) (sr2 = 0.041) |
Goldberg et al. (2003)
2
USA
| 218(123/95) 72% | Modified SSSNI | PANSS (GP, P, N) BQOL | Symptom severity (affective, positive, negative) QOL | Cross-sectional, correlational | 1. Smaller network size was sig correlated with more severe negative symptoms (r = − 0.29) and general symptoms (r = − 0.19). Positive symptoms and general satisfaction with QOL were not sig associated with network size (na) 2. Results from ANOVA showed sig differences across five network density categories for PANSS general (F = 3.30), positive (F = 3.14) and negative (F = 3.00) symptoms, with trend for less severe symptoms for low to moderate density and more severe symptoms for no connections. But post-hoc tests only sig for general symptoms. QOL was not sig associated with density (na) |
Hamilton et al. (1989)
USA
| 39 (39/0) 100% | Modified PPKI | NSRS and SANS SAPS | Symptom severity (positive and negative) | Cross-sectional, correlational | 1. NSRS total scores sig negative correlated with social network size (rho = − 0.64) 2. Positive symptoms on SANS did not correlate significantly with network size (na) |
Horan et al. (2006)
USA
| T1: 89 (75/14) T2: 34(na) 100% | Study specific | BPRS-24 (G, P, N, A) SAS | Symptom severity (overall, positive, negative, affective) Social functioning | Cross-sectional correlational (baseline, 15 months) | 1. At baseline (T1), total network size did not sig correlate with BPRS overall (r = 0.01), BPRS positive (r = 0.11), BPRS negative (r = − 0.16), BPRS anxiety/depression (r = 0.07) or current overall social functioning (total SAS) (rho = 0.03). Higher percent kin was sig associated with lower total SAS scores (r = − 0.28). Network density and degree did not correlate with BPRS symptoms or functioning 2. 34/87 participants (38%) completed T2 assessments. At 15 months, smaller total network size correlated with more sig severe BPRS positive symptoms (r = − 0.36), but did not sig correlate with BPRS overall (r = − 0.19), BPRS anxiety/depression (r = 0.05) or BPRS negative (r = − 0.16). Higher scores on BPRS positive correlated with greater density (r = 0.42) and lower per cent kin (r = − 0.37), but not degree. BPRS negative and overall symptoms did not correlate with density, percent kin or degree at T2 3. Current social functioning did not correlate with size (r = 0.19), density (r = 0.05), degree (r = 0.05) or percent kin (r = − 0.05) at T2 |
Howard et al. (2000)
UK
| 302 (143/159) 74% | SNS | GAF (D) | Global functioning | Cross-sectional correlational + longitudinal (baseline, 2 years) | 1. Complete data n = 135 Time 1 and n = 130 Time 2. 188 (62%) completed SNS T1 and 151 (50%) T2; 230 (76%) completed GAF T1 and 215 (71%) T2. Reasons not reported 2. Cross-sectional regressions found positive associations between social network size and GAF disability at T1 (n = 135, β = 0.22, CI = 3.46. 11.62) and T2 (n = 130, β = 0.19, CI = 0.73, 12.89) 4. No sig temporal association between social networks T1 and GAF disability T2 controlling for GAF at T1 (n = 107, β = 0.14, CI = − 0.35, 12.36) 5. The best fitting model from SEM (n = 107) suggested that total social network size at T1 explains some variance in GAF disability at T1 (coefficient = 0.19), and GAF T1 explains some of the variance in GAF over a 2 years period (T2) (coefficient = 0.34), but insufficient power to detect effects |
Macdonald et al. (1998)
Australia
| 46 (34/12) 100% | SRS—two subscales | SAPS SANS BDI | Symptom severity (positive, negative, affective) | Cross-sectional correlational | 1. No sig correlations between network variables and depressive (na) or positive symptoms (r = 0.07). Size of support network sig correlated with increased negative symptoms (r = 0.34) 2. SEM supported a tentative model that negative symptoms has a direct effect (coefficient = − 0.31) on social skill, with an indirect effect (coefficient = 0.34) on the size of social networks via social skill. This model accounted for 15% of the variance in the size of networks |
Sibitz et al. (2011)
Austria
| 157 (85/72) 100% | Study specific | WHOQOL-BREF ADS | QOL Symptom severity (affective) | Cross-sectional, correlational | 1. Increased number of friends sig correlated with lower depressive symptoms (r = − 0.32) and improved subjective QOL (r = 0.27) 2. SEM showed that poorer social network (insufficient number friends) negatively influences subjective QOL only if leading to stigma and low empowerment, which resulted in depression and, in turn, impaired QOL (indirect effect = 0.16). There was no direct effect of QOL (direct effect = − 0.07) on social network |
Thorup et al. (2006)
Denmark
| 547(323/224) 100% | SNS | SAPS SANS GAF (S, F) | Symptom severity (positive, negative) Global functioning | Cross-sectional correlational and longitudinal (baseline, 2 after treatment onset) | 1. 369/547 (67%) completed entire 2 year follow-up. After missing data and attrition, Time 1 N = 505, Time 2 N = 347 2. Reduction in negative symptoms and global functioning sig associated with larger network size (family and friends) at baseline and 2 years 3. Reduction in disorganised symptoms associated with larger network size at 2 years but not baseline 4. Positive symptoms was not sig related to network size at baseline or 2 years 5. Multivariate regression models at 2 years (N = 332) included site, treatment, age and number of contacts at entry as sig covariates. Male gender (− 0.86), older age (− 0.07) and more severe disorganised symptoms (− 0.29) were sig predictors of reduced family network size. More severe negative symptoms (− 0.26), older age (− 0.04) and completed A-level status (0.74) were sig predictors of reduced friendship network size |
Wojciechow et al. (2002)
Poland
| 56 (32/24) 100% | BQ | BPRS-24 (G, N, P) DSM-III: social functioning | Symptom severity (overall, positive, negative) Social functioning | Cross-sectional correlational (3 years after first admission) | 1. Smaller total network size sig correlated with greater intensity of BPRS overall (rho= -0.49), negative (rho= -0.48) and positive (rho= -0.29) symptoms. No sig correlations between size of network and social functioning (rho = 0.14) 2. Larger extra-familial network sig correlated with fewer overall (rho = − 0.42) and positive symptoms (rho = − 0.30) and improved social functioning (rho = 0.55). No sig correlations between extra-familial network and negative symptoms (rho = − 0.25) |
Social network characteristics
Study quality assessment
Study reference | Selection bias | Confounders | Data collection—outcome | Data collection—size | Withdrawals and drop-outs | Analyses |
---|---|---|---|---|---|---|
Allison et al. (2013) | WEAK | WEAK | STRONG | MOD | N/A | MOD |
Angell and Test (1992) | WEAK | STRONG | STRONG | WEAK | MOD | MOD |
Becker et al. (1998) | MOD | STRONG | STRONG | STRONG | N/A | STRONG |
Cechnicki & Wojciechowska (2008)1 | WEAK | WEAK | STRONG | STRONG | N/A | MOD |
Cechnicki et al. (2008)1 | WEAK | WEAK | WEAK | STRONG | N/A | MOD |
Cohen et al (1997) | MOD | STRONG | STRONG | STRONG | N/A | STRONG |
Cresswell et al. (1992) | WEAK | WEAK | STRONG | STRONG | N/A | WEAK |
Dixon et al. (2001)2 | WEAK | STRONG | STRONG | WEAK | N/A | STRONG |
Goldberg et al. (2003)2 | MOD | WEAK | STRONG | WEAK | N/A | STRONG |
Hamilton, et al. (1989) | WEAK | WEAK | STRONG | STRONG | N/A | MOD |
Horan et al. (2006) | WEAK | WEAK | STRONG | STRONG | WEAK | MOD |
Howard, Leese & Thornicroft (2000) | MOD | STRONG | STRONG | STRONG | WEAK | STRONG |
Macdonald et al. (1998) | WEAK | STRONG | STRONG | MOD | N/A | STRONG |
Sibitz et al. (2011) | MOD | STRONG | STRONG | WEAK | N/A | STRONG |
Thorup et al. (2006) | WEAK | MOD | STRONG | MOD | MOD | STRONG |
Wojciechow et al. (2002) | WEAK | WEAK | STRONG | STRONG | N/A | MOD |
Association between social networks and outcomes
Outcome | Studies | Total N | Random effects meta-analysis | Heterogeneity | |||||
---|---|---|---|---|---|---|---|---|---|
Hedge’s g | 95% CI | p value | Q value (df) | p value | I2 (%) | ||||
Overall symptoms | |||||||||
Total | 5 | 467 | − 0.530 | − 0.875 | − 0.184 |
0.003
| 10.822 (4) | 0.029 | 63.037 |
100% SS + high-quality network | 4 | 249 | − 0.595 | − 1.111 | − 0.079 |
0.024
| 10.683 (3) | 0.014 | 71.919 |
Positive symptoms | |||||||||
Total | 7 | 405 | − 0.192 | 0.494 | 0.110 | 0.213 | 12.709 (6) | 0.048 | 52.788 |
100% SS | 6 | 318 | − 0.206 | − 0.581 | 0.169 | 0.281 | 12.683 (5) | 0.027 | 60.578 |
High-quality network | 4 | 248 | − 0.276 | − 0.793 | 0.241 | 0.296 | 11.357 (3) | 0.010 | 73.584 |
Negative symptoms | |||||||||
Total | 8 | 577 | − 0.754 | − 0.997 | − 0.512 |
0.000
| 10.895 (7) | 0.143 | 35.748 |
100% SS | 7 | 358 | − 0.818 | − 1.126 | − 0.509 |
0.000
| 10.128 (6) | 0.119 | 40.757 |
High-quality network | 5 | 288 | − 0.899 | − 1.319 | − 0.480 |
0.000
| 9.789 (4) | 0.044 | 59.138 |
Social functioning | |||||||||
Total | 3 | 209 | 0.361 | − 0.078 | 0.801 | 0.107 | 4.737 (2) | 0.094 | 52.766 |