Association of antiretroviral resistance genotypes with response to therapy--comparison of three models

Genotype-based resistance assays are commonly used to aid treatment in HIV-infected individuals failing antiretroviral therapy. The relationship between genotype and antiretroviral therapy comes mostly from in vitro assays of the response to a single drugs although there is a need for a prediction of clinical response to combination therapy. We have compared three different methods of analysing genotype data as a predictor of clinical response in a small clinical cohort of highly antiretroviral-experienced individuals failing therapy. No method performed well beyond 8 weeks into a new therapeutic regimen. A model based on the number of 'primary' mutations was statistically significant, but a multiple regression model, which identified specific mutations explained threefold more variation in response. Optimal prediction in this dataset was given by a model obtained from a classification tree analysis, in which genotype at amino acid sites 135 and 202 were combined with amino acid site 184, which explained over 50% of the deviance in the data and had a classification success of 86%.