New methods are being developed to incorporate treatment effects into models for determining HIV seroprevalance and projecting AIDS incidence by backcalculation, and publications have appeared on rapid flexible methods of backcalculation and 9 on the reliability of estimates of seroprevalence and AIDS incidence projections obtained by backcalculation. Work on the use of surrogate markers in clinical trials to treat HIV infection and work on other aspects of such clinical trials have been published. Work on the effects of errors in measurements of exposure indicates that unexpected biases away from the null hypothesis can arise, even when errors are nondifferential, if exposures are polychotomous. Extensions show that combining exposure categories in this setting can lead to biases away from the null hypothesis or even reverse the direction of an association, and similar results were obtained for clinical trials in which the endpoint is a composite of several component clinical categories. Manuscripts have been prepared describing new methods of analysis of case-control data that incorporate validation sampling to relate surrogate exposures to """"""""gold standard"""""""" exposures, and that use an initial validation sample to design the size of a subsequent larger sample using only surrogate exposure data. Work on the reliability of next-of-kin interview data and comparisons with death certificate information are in press. Methods for selecting controls for case-control studies have been reviewed, and papers have appeared on the weighted selection of controls and on optimal allocation of controls when the costs of obtaining controls vary across strata. Appropriate logistic analyses have been developed for controls obtained by cluster sampling. Two papers describing practical and theoretical issues in selecting a design for studying a large assembled cohort appeared. Papers appeared on the analysis of attributable risk based on logistic regression and on estimates of absolute risk from cohort studies. Related work for population-based case-control studies is in progress.
