CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Prediction of the time to treatment success from longitudinal left-censored viral load measurements Authors:  Geert Molenberghs - UHasselt (Belgium)
Tarylee Reddy - South African Medical Research Council (South Africa) [presenting]
Marc Aerts - UHasselt (Belgium)
Abstract: Viral load measurements play a key role in monitoring antiretroviral treatment success in HIV positive patients. In this setting, treatment success is formally defined as two consecutive viral load measurements less than 1000 copies/ml. Statistical challenges that arise in the analysis of longitudinal viral load measurements include: the handling of left-censored measurements and the non-linear evolution of viral load over time. We present a novel approach to estimate the time to treatment success, taking into account left-censoring and the biphasic evolution of viral load. In the first stage of the approach a mixed model, with random intercept and two random slopes is fitted to the data, where the partial information provided by the censored values is incorporated into the likelihood function. In the second stage, using the estimates from the mixed model, the probability of treatment success and the estimated time to treatment success is computed using a previous methodology. We apply the proposed methodology to an HIV/AIDS clinical trial, presenting the estimated time to treatment success and predicted viral load trajectory for selected patients.