Marker values at the time of an AIDS diagnosis.

In this paper statistical methods are proposed to estimate the distribution of a CD4 T-cell number at the time of a clinical AIDS endpoint from serial measurements of CD4 T-cell values in a cohort study. The statistical formulation of the problem is that of survival analysis with interval censored data, but in which the endpoints are obtained with measurement error. A measurement error likelihood is developed, assuming normality of the CD4 distribution at AIDS. A maximum likelihood estimation procedure and a Gibbs sampling approach are implemented.