Five RCTs were included in the review (n=3,103, range 104 to 1,400). The main limitation was failure to report adequate allocation concealment. The mean Jadad score was 2 out of 5. Three RCTs described their method of randomisation, two reported withdrawals and dropouts and none were double blinded. Participant characteristics appeared to be well-balanced between the groups in four RCTs.
The intervention was associated with a significantly lower risk of progression (HR 0.66, 95% CI 0.56 to 0.77, p<0.01; five RCTs), a significantly lower risk of death (HR 0.77, 95% CI 0.67 to 0.89, p<0.01; four RCTs) and significantly higher likelihood of response (RR 1.50, 95% CI 1.06 to 2.10, p=0.02; five RCTs). However, for all these analyses there was statistically significant heterogeneity, which in each case was attributed mainly to the same RCT. There was no statistically significant difference between the groups in all-cause mortality at 60 days (four RCTs).
In the intervention group, there were significantly more grade 3 or 4 adverse events (OR 1.79, 95% CI 1.52 to 2.11, p<0.01; five RCTs) and a significantly higher risk of discontinuing treatment due to adverse effects (OR 1.38, 96% CI 1.14 to 1.66, p<0.01; four RCTs), with no significant heterogeneity. The intervention was associated with significantly higher rates of grade 3 or 4 hypertension (five RCTs), thromboembolic/thrombotic events (three RCTs), bleeding (four RCTs), and gastrointestinal perforation (four RCTs). There was no statistically significant difference between the groups in rates of proteinuria, diarrhoea or leukopenia.
No obvious publication bias was detected.