Estimates of the virological benefit of antiretroviral therapy are both assay- and analysis-dependent

Abstract

To assess the potential discrepancies in reported changes in plasma viral load (PVL) depending on how values below the detection limit of the assay are handled in the data analysis phase of a randomized controlled clinical trial. Data from a recently completed clinical trial comparing combinations of zidovudine, didanosine and nevirapine were analysed. In this trial, PVL was measured using an assay with a lower quantification limit of 400 HIV-1 RNA copies/ml initially. All PVL values less than 500 copies/ml were retested with a more sensitive assay with a lower quantification limit of 20 copies/ml. Several summary measures for assessing change in PVL were calculated using three different methods to adjust for PVL values less than the quantification limit of the assay. The differences between these measures were evaluated. We found that the magnitude of the discrepancy between summary measures used to report changes in PVL depended on the proportion of subjects with PVL less than the quantification limit of the assay, how those observations were handled in the data analysis, and the relative difference between the quantification limits of the conventional and more sensitive assay. The lack of consensus in reporting of PVL data in the literature makes the interpretation of published trial results difficult. In the absence of agreement on the most appropriate summary measure of PVL data, we recommend that all summaries include information on the quantification limit of the assay used, the proportion of observations at or below the quantification limit and how these observations were handled in the data analysis.