The Influence of the Time Course of CD4 and Viral Load on Clinical Outcome Events Before and During Antiretroviral Therapy
Nick Holford (1), Rory Leisegang (2), Gary Maartens (2)
1. Department of Pharmacology & Clinical Pharmacology, University of Auckland, New Zealand. 2. Division of Clinical Pharmacology, University of Cape Town, Cape Town, South Africa.
Objectives: Mechanistic mixed effect models for HIV infection using CD4 cell counts and viral load (VL) are only practical for small data sets and short term follow up because of computational problems related to solving stiff differential equations. Existing models are from industrialized countries and even those purporting to show African HIV outcomes have been developed with data from the USA. Our objective has been to develop methods for describing the time course of CD4 and VL that can be used as predictors of clinical outcomes in HIV-infected patients in long term follow up settings in southern Africa.
Methods: CD4 and VL in HIV-infected patients can be described with empirical continuous functions of time before and after the initiation of antiretroviral therapy (ART). The parameters of a hazard function for clinical outcome events can be estimated using a combined Weibull and Gompertz baseline distribution with CD4, VL and ART as time varying explanatory factors.
Results: The simplest model describing CD4 used an exponential drop from baseline prior to ART. When ART is started a shallow sigmoid Emax model predicted the initial rapid rise and subsequent slow rise towards an asymptote. VL was essentially constant prior to ART and then fell exponentially. More complex models can be used for example to describe an initial slow decline before rapid loss of CD4 cells leading to ART. The hazard of death could be explained more by the time course of CD4 than by VL.
Conclusions: Prediction of survival in HIV-infected patients depends on time varying factors which are modified by ART. Survival analyses using only baseline characteristics are likely to be misleading about future events. Empirical mixed effect models for CD4 and VL offer a practical method for understanding time to event outcomes such as death, hospitalization, and treatment failure.
Acknowledgment: We are grateful to the patients who provided the data used to develop these methods and to our clinical collaborators at Madwaleni Hospital (Lynne Wilkinson,Tom Boyles, Richard Cooke) and Aid for AIDS (Michael Hislop, Leon Regensberg).