A Comparison of Alternative Approaches to Analyzing Subgroup Differences in Survival After AIDS Diagnosis When the Proportional Hazards Assumption Does Not Hold (303205)

Keywords: Acquired Immune Deficiency Syndrome, HIV/AIDS Surveillance, AIDS Survival, Survival Analysis, Kaplan-Meier, proportional hazards, Cox regression, Propensity Scores

We examine three distinct classes of statistical methods for comparing survival after AIDS diagnosis across population subgroups. We describe and compare the traditional approaches of unstandardized Kaplan-Meier estimation and the Cox proportional hazards model with weighting procedures (including inverse probability of treatment weighting and propensity scores) that adjust for covariate effects without relying on the proportional hazards assumption. Rather than actual case counts, the weighting procedures use weighted sums of events and persons at risk to estimate survival. The adjustment weights represent the size of each covariate stratum within a subpopulation relative to its size in a standard population. This work is motivated by a study of racial disparities in survival after AIDS diagnosis among heterosexually acquired cases of HIV disease reported to the Centers for Disease Control and Prevention. Covariates included gender, age group, race/ethnicity, year of diagnosis, and CD4 count at diagnosis. We include a description of the advantages and disadvantages of each class of methods and present the results of an analysis of racial differences in survival.