524   The Virtual Phenotype Is an Independent Predictor of Clinical Response.

N. Graham*¹, M. Peeters², W. Verbiest², R. Harrigan³, and B. Larder³.
¹Vircolab, Baltimore, MD and Virco,²Mechelen, Belgium and³Cambridge, UK.

Background:Retrospective studies have correlated baseline drug resistance (phenotype and genotype) with clinical response. We wished to assess the ability of the virtual phenotype, an averaged phenotype based on matching of genotypes with phenotypes, to predict response.

Methods:Baseline samples from 191 patients enrolled in VIRA3001 (prospective phenotypic resistance study) were analysed in relation to viral load response (<400 RNA copies/ml) at week 16. Virtual phenotypes were derived from the genotype (ABI sequencing) using a large relational database. Uni- and multivariate analyses were performed on data sets of genotype, phenotype (Antivirogram) and virtual phenotype, in accordance with the resistance collaborative group data analysis plan (DAP).

Results:The genotype analysis (dropouts as censored, DAC) was a significant predictor of response in the univariate model with an odds ratio (OR) of 0.69 (CI = 0.51—0.93), p = 0.015, but not in the multivariate model, OR = 0.81 (CI = 0.57—1.14), p = 0.22. In contrast, the virtual phenotype was highly significant in both models,

also using the DAC analysis. With a 4-fold susceptibility cutoff for all drugs in the univariate model, the OR = 0.38 (CI = 0.25—0.6), p < 0.0001, and in the multivariate model the OR = 0.52 (CI = 0.31—0.87), p = 0.013. Using drug-specific, biologically defined cutoffs, in the univariate model the OR= 0.39 (CI = 0.26—0.58), p < 0.0001, and in the multivariate model the OR = 0.49 (CI = 0.31—0.76), p = 0.0014. The DAF (dropouts as failures) analyses showed consistent superiority for virtual phenotype over genotype, although the significance was correspondingly lower for all of the categories.

Conclusions:In this study, the virtual phenotype, but not the genotype, was a significant, independent predictor of clinical response. The use of biologically relevant cutoff values appeared to enhance the predictive value of the virtual phenotype.