Bayesian Evaluation of CD Count as a Surrogate Endpoint in Patients with Advanced HIV Infection

We examine CD lymphocyte responses and survival patterns in persons with ad vanced HIV infection who were randomly assigned to treatment with one of two antiretro viral drugs didanosine ddI and zalcitabine ddC Analyses of CD lymphocyte count over time are performed using longitudinal random e ects models which allow for the possibil ity of a drug induced CD increase Probit regression is also used to directly estimate the probability of such an increase in each treatment group Finally we assessed the prognostic value of CD response and drug group as predictors of survival time Throughout Markov chain Monte Carlo methods enable computation of the relevant posterior summaries We nd that while patients given ddI had a substantially greater chance of a CD response after two months than those given ddC only CD responding ddC patients experienced improved survival Our results thus provide compelling evidence against the common practice of using CD count as a surrogate marker for survival in persons with advanced HIV disease