Fourteen RCTs (n=6,785) were included. Seven assessed protease-inhibitor based regimens and seven assessed non-NRTI-based regimens.
Nine RCTs reported adequate allocation concealment. Twelve RCTs used placebo controls.
Triple therapy significantly reduced the risk of clinical progression compared with dual therapy (OR 0.65, 95% CI: 0.52, 0.81). Evidence of statistical heterogeneity was detected (P=0.09).
The meta-regression found that larger treatment effects were associated with triple regimens based on protease inhibitors (P<0.0001), smaller treatment effects were associated with triple regimens using didanosine (P<0.0001), and larger differences in treatment effects were associated with studies with a higher proportion of patients with AIDS. These variables were found to account for the between-trial heterogeneity. Length of follow-up, year of publication, publication type, median age, NRTI naive, CD4 cell count and viral load, censoring effect and trial quality were not associated with treatment effect.
There was little evidence of publication bias, as demonstrated by funnel plot asymmetry (P=0.27).
Direct comparisons found that triple regimens based on protease inhibitors showed larger treatment effects than triple regimens based on non-NRTIs: the OR was 0.49 (95% CI: 0.41, 0.58) for protease inhibitor regimens and 0.90 (95% CI: 0.71, 1.15) for non-NRTI regimens.
An indirect comparison of the two triple regimens showed that protease inhibitor-based regimens had larger treatment effects than non-NRTI-based regimens (OR 0.54, 95% CI: 0.49, 0.73). The results were similar after adjusting for the use of didanosine and the proportion of patients with AIDS.
Protease inhibitor-based regimens increased CD4 cell counts compared with non-NRTI-based regimens; an indirect comparison showed an additional increase of 25 cells per microL (95% CI: -17, 68). Protease inhibitor-based regimens also suppressed viral replication better than non-NRTI-based regimens; an indirect comparison showed that the odds of reaching an undetectable viral load was 6.0 (95% CI: 2.2, 16.6).