Objectives: To assess the practical significance of the following sources of bias for estimates of the AIDS incubation period in a large seroconverter cohort: estimation of the time of seroconversion; presentation with an HIV-related illness; preferential inclusion of survivors; loss to follow-up and analysis cut-off date; the inclusion of Kaposi's sarcoma as an AIDS event; death without an AIDS diagnosis; and representativeness of the HIV population.
Methods: Standard non-parametric survival methods were used to estimate the AIDS incubation period distribution. The practical importance of each type of bias was assessed using various sensitivity analyses.
Results: The potential sources of bias of most practical importance in this study were the right-censoring strategy and that due to lack of documentation of a negative HIV antibody test. Five different right-censoring strategies gave estimates of the median time to AIDS ranging from 8.1 to 10.8 years for the 1202 individuals enrolled in the UK Register of HIV Seroconverters. HIV-infected persons with a history of a previous antibody negative test which could not be verified appeared to progress to AIDS more rapidly than persons with such verification (Relative risk = 1.8, 95% confidence intervals = 1.3-2.3).
Conclusions: As a number of possible causes of bias can impact on results, care must be taken to document them and control for them wherever possible. In our study, this was particularly relevant in relation to the documentation of a previous HIV antibody negative test and the choice of analysis cut-off dates. As methods may differ between cohorts, comparison of the published results from one cohort with those of another may be misleading.