Ten studies (n=776,909) were included. Of these, three were RCTs (n=3,057), two were cluster-RCTs (n=683,277) and the remaining five were interrupted time series (n=90,575).
All of the RCTs were considered good quality; one cluster-RCTs and one interrupted time series study were also considered good quality. All of the other included studies were considered fair quality. The cluster-RCTs, which were considered only to be fair quality, generally failed to report baseline measurements clearly. The interrupted time series studies generally failed to use an appropriate time trend analysis.
Overall, there was a significant reduction in Caesarean section rates in the intervention groups (RR 0.81, 95% CI: 0.75, 0.87; 10 studies). However, significant statistical heterogeneity was found between the studies (I-squared 87.6%, p<0.00001).
Meta-regression analysis showed that the strategy used, the study design and the identification of barriers all contributed significantly to the overall variability between studies. Subgroup analysis was performed for the following groups of studies: audit and feedback strategy (4 studies); quality improvement strategy (4 studies); multifaceted strategy (2 studies); controlled design (5 studies); non-controlled design (5 studies); no identification of barriers (6 studies) and identification of barriers (3 studies). All the results remained significant, as did the significant heterogeneity between many of the subgroups.
There was a significant reduction in Caesarean section rates between groups for the following indicators: dystocia (6 studies), repeat Caesarean section (4 studies), foetal distress (3 studies) and maternal indications(1 study).
There was no significant difference between study groups for stillbirth rate (1 study), perinatal and neonatal mortality (3 studies), admission to intensive care (5 studies) or perinatal and maternal morbidity (3 studies).
No evidence of publication bias was found.