Twenty-seven RCTs (1,624 patients) were included. One RCT used a crossover design, whilst all the others were parallel group studies.
The methodological design of the trials was often poor, as was the quality of the data reporting. The median Oxford validity score was 2 (range: 1 to 4). The primary studies exhibited several methodological problems: only one trial attempted to blind the study drugs; 40% of the studies did not mention the number of, and reasons for withdrawals; most trials had fewer than 50 patients; and few trials reported a baseline measurement of sedation.
Ten RCTs were sponsored by the manufacturers of the two drugs being studied.
Efficacy of sedation (19 RCTs).
The event scatter suggested that propofol was more efficacious than midazolam. The efficacy ranged from 54 to 97% with propofol, and from 26 to 95% with midazolam.
Duration of adequate sedation (9 RCTs, 615 patients).
The duration of adequate sedation was significantly greater with propofol than with midazolam; the WMD was 2.9 hours (95% CI: 0.2, 5.6, p=0.04).
Weaning time (12 RCTs, 633 patients).
Short-term sedation (9 RCTs): the weaning time was less with propofol than with midazolam. The average weaning time ranged from 0.8 to 4.3 hours after sedation with propofol, and from 1.5 to 7.2 hours after sedation with midazolam; the WMD was 2.2 hours (95% CI: 0.8, 3.7).
Long-term sedation (3 RCTs, 156 patients): there was a lack of evidence of difference in weaning times. The 3 RCTs favoured propofol with WMDs of 1, 13.1 and 35.8 hours, respectively.
Length of ICU stay (4 RCTs): no meaningful conclusions could be made.
Adverse drug reactions.
The specific adverse reactions, i.e. arterial hypotension and hypertriglyceridaemia, were reported significantly more often with propofol than with midazolam.
Arterial hypotension (7 RCTs): the pooled RR was 2.5 (95% CI: 1.3, 4.5) and the NNT was 12 (95% CI: 8, 33).
Hypertriglyceridaemia (4 RCTs): the pooled RR was 12.1 (95% CI: 2.9, 49.7) and the NNT was 6 (95% CI: 4, 10).
Mortality (8 RCTs): there was no significant difference between the intervention groups; the RR was 0.8 (95% CI: 0.5, 1.3).