Analysis of repeated virological measurements based on cell dilution assays

Virological measurement has become increasingly popular as an endpoint in the evaluation of new drugs to treat human immunodeficiency virus (HIV) diseases. A commonly used quantity to measure the viral load is the proportion of infected cells in a patient's blood. Although the literature suggests various patient-specific estimation procedures for such proportions based on a series of cell dilutions, the resulting estimates are highly unstable. Moreover, it seems inappropriate to use those estimates as raw data points to make inferences about the group differences in comparative studies. In this article, under a two-arm clinical trial setting, we propose semi-parametric methods to estimate the treatment difference with this particular quantity evaluated repeatedly over time. Our proposals are conceptually simple and implemented easily in practice. We also propose model checking techniques to examine the adequacy of the fitted model. Data from a recent trial conducted by the AIDS Clinical Trials Group to evaluate the relative merit of zidovudine (ZDV) and dideoxyinosine (ddI) illustrate the methods. Our analysis indicates that patients treated with high dose ddI tended to have significantly lower viral load than those treated with a low dose combination of ZDV and ddI.