Twenty-five RCTs were included (1,744 patients).
Effect on relapse rate was relatively well-assessed though differences emerged, particularly in comparison with poor control treatments in older studies, and these differences disappeared in recent studies. Finding on patients' symptoms, social functioning or family variables were limited and affected by methodological pitfalls. No difference between intervention models emerged, though behavioural and psychoeducational approaches were better investigated. Patients' inclusion, greater frequency and length of treatment led to better results.
Patient's relapse or readmission (18 RCTs, including 15 RCTs with a control group and 3 RCTs comparing different treatments): results were inconsistent across studies. There were wide variations in relapse rate across studies in both treatment and control groups. At 1 year, median relapse rate in intervention group was 18% (range: 6 - 52) versus 44% (range: 15 - 67) in control. At 2 years, median relapse rate in intervention group was 33% (range: 15 - 50) versus 64% (range: 46 - 83) in control.
Concerns about the use of 'relapse' as an outcome included variable definitions of 'relapse', with criteria for definition of relapse often unclear inconsistent and poorly standardised across studies; definitions based on a threshold of symptoms may be misleading or useless for patients with high levels of psychopathology at baseline; schizophrenia often shows a fluctuating course and any division of patients into categories tend to be artificial; and relapsed patients usually left the trial, and data on the period following relapse were not collected.
Patient's mental state (8 RCTs, including 6 RCTs with a control group and 2 RCTs comparing different treatments): problems with the primary studies included high drop-out rates, and difficulties interpreting results because full data were provided on selected symptoms only or from non-relapsed patients only.
Patient's social and work functioning (11 RCTs, including 9 RCTs with a control group and 2 RCTs comparing different treatments): clearly positive results were seldom found, there was little consistency in the use of measurement tools, and a lack of accuracy in outcome criteria assessment.
Family's well-being and other family-related variables (14 RCTs): results were inconsistent. Problems included the use of a variety of outcome measures of uncertain validity.
Comparison of different intervention models.
Technical model (psychoeducational, behavioural, psychodynamic, systematic).
Interventions within these categories were not homogeneous. It was not possible to identify the effective components of family intervention.
Subject of intervention (single family with or without the patient, multiple family group): there was consistency of negative results from 4 RCTs with groups of relatives without the patient.
Setting (hospital, out-patient clinic, home): results were inconsistent.
Duration: 7 RCTs, in which the family intervention was provided on a short-term basis (no more than 10 sessions over less than 6 months), failed to show convincing and long-lasting results.
Other methodological issues.
Methodological problems included: small sample size without consideration of study power; few authors described the standard treatment fully or checked its application to the control group; inadequate investigator blinding; failure to consider reliability of outcome measures; and failure to perform analyses on an intention to treat basis.