Eighteen RCTs (n=1,785, range 22 to 421) were included in the review. Seventeen studies reported baseline comparability of groups, four reported blinding, 12 described methods of randomisation, five described allocation concealment and five used intention-to-treat analysis. Overall study quality was considered inadequate.
Using a fixed-effect model, the intervention group had significantly less pain (SMD -1.84, 95% CI -2.45 to -1.22; two RCTs) and fewer wound infections (SMD 0.32, 95% CI 0.18 to 0.56; three RCTs). Using a random-effects model, the intervention group had a significantly lower rate of post-anaesthetic shivering (effect measure unclear 0.01, 95% CI 0.0001 to 0.08; five RCTs). Using either model, the intervention group had significantly lower hypothermia (random effects SMD -4.44, 95% CI -5.92 to -2.95; 14 RCTs) and blood loss (random effects SMD -1.60, 95% CI -1.92 to -1.29; five RCTs). There was significant statistical heterogeneity for the analyses of pain (Χ2 test p=0.001) and blood loss (Χ2 test p=<0.0001). Directions of effect differed for hypothermia (Χ2 test p not reported). For some outcomes the statistical significance of the findings differed according to the choice of analytical model.
The findings of the sensitivity analysis were reported. There were insufficient data to report other outcomes.