Sixteen RCTs were included in the review (n=3,887 participants). Six trials reported that the person interpreting the deep vein thrombosis screening imaging was blinded to treatment; it was not possible to blind patients or care givers. Randomisation resulted in equal distribution of confounding factors in all trials. Duration of follow-up ranged from five days to three months.
There was no significant difference in the risk of deep vein thrombosis (16 RCTs) or pulmonary embolism (15 studies) between patients who received mechanical compression compared with those who received subcutaneous heparin. There was substantial heterogeneity for deep vein thrombosis (I2=62%), but not for pulmonary embolism (I2=0%).
The type of heparin (unfractionated or low molecular weight) was the only factor significantly associated with risk of deep vein thrombosis in the meta-regression analysis (p=0.03). The risk of deep vein thrombosis was significantly increased with mechanical compression compared to low molecular weight heparin (RR 1.80, 95% CI 1.16 to 2.79; eight RCTs), but not compared with unfractionated heparin (eight RCTs). None of the other variables investigated in the meta-regression analyses modified the effect on deep vein thrombosis risk. As trials were homogeneous for pulmonary embolism, subgroup analysis was not conducted.
The risk of bleeding was significantly reduced in patients who received mechanical compression compared with those who received subcutaneous heparin (RR 0.46, 95% CI 0.31 to 0.70; 10 RCTs). There was some evidence of heterogeneity (I2=42%). When analysed by type of bleeding, results were significantly lower for mechanical compression for major bleeding, minor bleeding, and the risk of undergoing transfusion. When results were stratified by heparin type, the risk of bleeding was significantly increased for both unfractionated and low molecular weight heparin; none of them modified the effect on bleeding risk in the meta-regression analysis. None of the individual trials were found to substantially alter the pooled estimates.
There was no evidence of publication bias (p>0.2).