Eighteen RCTs were included (2,707 participants). Eight trials described appropriate sequence generation and allocation concealment. Six trials used blinded outcomes assessors. Ten trials addressed incomplete data adequately. Seven trials measured relapse prospectively. Eleven trials predefined relapse.
Psychosocial interventions (nine RCTs): Specialist first-episode psychosis programmes were significantly more effective than treatment as usual at preventing relapse (OR 1.80, 95% CI 1.31 to 2.48, NNT=8, I2=0%; three RCTs, 679 participants) and reducing hospital stay (WMD -26.2 days, 95% CI -7.35 to -45.06, I2=0%; three RCTs, 679 participants).
There were no significant differences in relapse rates in comparisons of CBT versus supportive counselling or treatment as usual (one RCT), specialist first-episode psychosis programmes with CBT versus without (two RCTs), specialist first-episode psychosis programmes with relapse prevention therapy versus without (one RCT) and family therapy versus treatment as usual (two RCTs, some heterogeneity I2=76%).
Pharmacological interventions (nine RCTs): There were no significant differences in relapse rates between antipsychotics and placebo (three RCTs, some heterogeneity I2=50%) and between different types of first-generation antipsychotics (one RCT). However the relapse rate was significantly lower with second-generation antipsychotics than with first-generation antipsychotics (OR 1.47, 95% CI 1.07 to 2.01, NNT=10, I2=0%; four RCTs). Treatment maintenance was significantly more effective for relapse prevention than guided discontinuation (OR 2.91, 95% CI 1.33 to 6.37, NNT=5; one RCT), but mean bed days did not differ significantly.
There was no clear evidence of any publication bias. Sensitivity analyses and other findings were reported.