Two-component generalized bent-cable models

This paper presents an innovative Bayesian method for assessing the status and progression of HIV infection using biomarkers such as the CD4 count and viral load variables. A distribution of viral loads in subjects starting antiretroviral treatment and followed over time may show a mixture of two subgroups: one subgroup representing “non-progressor” subjects and another subgroup of “progressor” subjects. In modeling these groups, we use a mass point and a right-skewed continuous distribution in which trajectories of repeated observations of viral load exhibit multiphasic features along with a gradual transition period. The commonly used method for describing these phasic patterns is a bent-cable model with a quadratic function for modeling the gradual transition phase. A quadratic function may be too restrictive for adequately modeling the transition period. Thus, we we extend the bent-cable model within the context of a two-component Tobit growth model by relaxing the assumption of a quadratic function for a gradual transition period and accounting for measurement errors in CD4 count. The proposed methods are demonstrated using real data from an AIDS clinical study.