Markers and residual time to AIDS

The value of immunological and virological markers as predictors of progression to AIDS, or death by AIDS, is a topic of much current interest. Mostly, the influence of markers is investigated in a time-dependent or a baseline proportional hazard model, relating time-varying or baseline marker values to the instantaneous AIDS risk. Low CD4 numbers have been shown to increase the AIDS risk. Based on this finding, the decision to start treatment has mainly been based on a person's CD4 count. Over the last couple of years, high viral load has been added as a criterion to start treatment. Relative hazards do not directly reveal information on time to AIDS. A better approach is to base the decision to start treatment on an estimate of a person's residual time to AIDS or death by AIDS. Discrimination and calibration are introduced as criteria to compare prediction models. Two models to predict residual time to AIDS are discussed. One is a proportional hazards model based on baseline marker values; the other one is a combined model of a time-dependent proportional hazards model and a longitudinal model for marker development.