Analyzing plasma HIV-1 RNA measurements as multiple recurrent events in clinical trials

Primary endpoints based on levels of HIV-1 RNA are currently used to evaluate efficacy of antiretroviral therapy in AIDS/HIV clinical trials. These endpoints, however, are generally based on few HIV-1 RNA measurements. More specifically, when the endpoint is the time to virologic failure (usually defined as two consecutive HIV-1 RNA levels above a threshold value), viral load measurements recorded after failure are simply ignored in the analysis. Instead of one virologic failure used in a classical survival analysis, we suggest that a failure be defined as any plasma HIV-1 RNA level above a predefined threshold value. This then becomes an analysis of multiple events of the same type, that is, recurrent virologic failures. Many survival models based on the Cox proportional hazards model have been proposed to handle multiple event data. The two distinct classes of models (marginal and conditional) answer two distinct types of clinical question and should be considered as complementary rather than competing models. We applied the different models to data from a clinical trial in HIV-1-infected patients.