Viral load as a primary outcome in human immunodeficiency virus trials: a review of statistical analysis methods

In Human Immunodeficiency Virus infection, several statistical methods are available to analyze viral load (HIV-1 RNA) used as a surrogate outcome in trials of antiretroviral treatments. We compared the most frequently used methods and applied them to one of these trials, where HIV-1 RNA was measured using two lower limits of detection. Methods were reviewed for different properties dealing with validity, interpretation, and handling. Compared to change of HIV-1 RNA at the end of follow-up or HIV-1 RNA area-under-the-curve during follow-up minus baseline, the most attractive methods appeared to be HIV-1 RNA undetectability, HIV-1 RNA reduction at the end of follow-up with censoring adjustment, and mixed linear model on HIV-1 RNA.