Relative contributions of baseline patient characteristics and the choice of statistical methods to the variability of genotypic resistance scores: the example of didanosine

Abstract

Our aim was to investigate the respective role of statistical methodology and patients' baseline characteristics in the selection of mutations included in genotypic resistance scores. As an example, the FORUM database on didanosine including 1453 patients was used. We split this population into four samples based on countries of enrolment (France n = 474, Italy n = 440, USA/Canada n = 219, others n = 320). We used both a continuous outcome measure (the viral load reduction at week 8) and a binary outcome measure (viral load decline at week 8 <0.6 log(10) or > or =0.6 log(10)) and both parametric and non-parametric methods for each outcome. Overall, 61 distinct mutations were selected by at least one method in at least one data set. The variability due to baseline characteristics varies from 79% to 88%, i.e. for a given method applied to the four data sets >80% of the mutations are selected only once. The variability induced by the methodology varies from 49% to 56%, i.e. for a given data set approximately 50% of the mutations are selected by at least two methods. Baseline patient characteristics contribute more than the choice of statistical method to the variability of the mutations to be included in the genotypic resistance scores.