Twenty-one studies (959 families, children and parents) were included.
Twelve studies (57.1%) reported randomisation and 5 studies (23.8%) presented follow-up data.
Inter-rater agreement ranged from kappa 0.62 for outcome variables to kappa 0.77 for participant characteristics.
Overall, CM treatment improved outcomes compared with control (d=0.54, 95% CI: 0.39, 0.69). Statistically significant heterogeneity was found (P<0.01). The fail-safe N was 36, i.e. 36 studies would be required to reduce the overall ES to d=0.2.
For outcome constructs, the subgroup analysis showed that CM treatment improved outcomes compared with the control (P<0.05). The ESs were statistically homogeneous for child cognitive performance (d=0.28), child self-report (d=0.44), behaviour observation of child (d=0.30), parent self-report (d=0.53) and behavioural observation of the family (d=0.21). The smallest ES was shown for objective behavioural observations of the family (d=0.21, 95% CI: 0.05, 0.37; fail-safe N=0; based on 2 studies), whilst the largest ES was shown for parent self-reporting parenting attitudes (d=0.53, 95% CI: 0.43, 0.63; fail-safe N=12; based on 7 studies).
The subgroup analysis showed that treatment effects were influenced by:
the type of control (larger effects for no-treatment controls compared with placebo or case-management control groups; d=0.99 versus d=0.38 and d=0.35, respectively),
theoretical orientation (non-behavioural treatments d=0.87, behavioural treatments d=0.40, and combination d=0.59, although significant statistical heterogeneity was found in combination treatments), and
the type of CM (ES for child sexual abuse interventions were larger than those for general CM, d=0.69 versus d=0.40).
Significant statistical heterogeneity was found for studies of sexual abuse interventions (P<0.05). The post hoc analysis, found that duration of treatment was significantly longer for non-behavioural treatments (13.5 months versus 2.9 months, P=0.006).
No effect was shown for treatment modality or study quality, or for mandated versus volunteer participants.