Generalizing the per-protocol treatment effect: The case of ACTG A5095

Area of Science:

• Clinical Trials
• Epidemiology
• Biostatistics

Background:

• Intention-to-treat (ITT) analyses in randomized trials estimate treatment assignment effects, irrespective of adherence.
• Decision-makers often require per-protocol (PP) effect estimates and generalizability to specific populations.
• This study aimed to generalize findings from the ACTG A5095 trial to US individuals diagnosed with HIV between 2008-2014.

Purpose of the Study:

• To replicate conventional ITT estimates from the ACTG A5095 trial.
• To estimate ITT effects accounting for informative dropout and PP effects accounting for protocol deviations.
• To generalize these effects to a target population of US individuals diagnosed with HIV during 2008-2014.

Main Methods:

• Replication of conventional intention-to-treat (ITT) estimates.
• Estimation of dropout-weighted ITT and per-protocol (PP) effects using inverse probability weighting.
• Generalization of ITT and PP effects to a target US HIV-diagnosed population using inverse odds of sampling weights.

Main Results:

• The 2-year risks of virologic failure or death were similar across ITT and PP analyses in both the trial and target populations.
• In the US target population, the 2-year risk difference for virologic failure or death comparing four-drug to three-drug regimens was -0.4% (ITT), -0.9% (dropout-weighted ITT), and -0.7% (PP).
• Hazard ratios were consistently close to 1 across all comparisons and methods, indicating no substantial difference in effectiveness.

Conclusions:

• No significant benefit of a four-drug antiretroviral regimen over a three-drug regimen was observed.
• Findings were consistent across conventional ITT, dropout-weighted ITT, and PP analyses.
• The effectiveness of the four-drug regimen did not demonstrate superiority in the trial sample or the generalized US HIV-diagnosed target population.