Twenty-three RCTs (n=2,326) were included.
The authors stated that lack of reporting of allocation concealment was the main methodological problem.
Withdrawal for any reason (23 RCTs, n=2,326): statistical heterogeneity was found (chi-squared p=0.007; I-squared 47.4%). There was a reduction in the risk of withdrawal for any reason in patients receiving benzodiazepines compared with placebo (random-effects RR 0.78, 95% CI: 0.62, 1.0, p=0.05).
Withdrawal due to lack of efficacy (20 RCTs, n=2,061): no statistical heterogeneity was found (I-squared 0%; p=0.15). There was a large reduction in the risk of withdrawal due to lack of efficacy in patients receiving benzodiazepines compared with placebo (fixed-effect RR 0.29, 95% CI: 0.18, 0.45, p<0.00001).
Withdrawal due to adverse events (19 RCTs, n=1,950): very little heterogeneity was found (I-squared 26.8%; p=0.15). There was an increase in the risk of withdrawal due to adverse events in patients receiving benzodiazepines compared with placebo (fixed-effect RR 1.54, 95% CI: 1.17, 2.03, p=0.002).
Exploration of heterogeneity.
The risk of withdrawal for any reason associated with benzodiazepines compared with placebo was significantly reduced in studies using DSM-III criteria (11 RCTs) and significantly increased in studies using DSM-IV criteria (3 RCTs); in studies using DSM-III-R criteria there was no significant difference between treatment groups (8 RCTs). No heterogeneity was found for any of these subgroup analyses.
The risk of withdrawal for any reason was significantly reduced in patients receiving diazepam (12 statistically homogeneous RCTs) and alprazolam (4 statistically heterogeneous RCTs) compared with placebo; there was no significant difference between lorazepam and placebo (7 statistically heterogeneous RCTs).
Meta-regression showed that 74% of the variation between studies was accounted for by year of publication, with smaller treatment effects being found in more recent publications.
There was some evidence of publication bias (Begg’s test p<0.012).