ESTIMATING RISKS OF PROGRESSING TO AIDS WHEN COVARIATES ARE MEASURED WITH ERROR

For some time, there has been interest in quantifying the risks of pro gressing to Aids from infection with the human immunodeficiency virus (HIV) associated with changes in the immunological markers. Until rece ntly, little attention has been paid to the considerable variability o f these markers and the corresponding effect on estimates of relative risk. We estimate the risks of progression to Aids associated with imm unological markers among a group of homosexual men infected with HIV w ho have been followed at quarterly intervals for 5 years. Errors in me asurement in the markers are shown to bias these estimates. Smoothing techniques are compared for their effectiveness in reducing the amount of variation in the markers and the subsequent bias in the coefficien ts. A simulation study compares the effectiveness of various methods o f smoothing in reducing the bias of the estimated coefficient.