Estimating human immunodeficiency virus (HIV) prevalence from sentinel seroprevalence studies is difficult. We characterize these studies and show that most are investigations of incompletely defined (hypothetical) cohorts and are usually based on nonprobability samples. Prevalence in HIV sentinel serosurveys is also time-averaged and vulnerable to several time-dependent sources of bias (e.g., migration, deaths, and changes in incidence). Assumptions must be made that these time-dependent biases did not meaningfully affect the data, and this can be helped by reducing the period of investigation. Furthermore, we show that "reliability" can not be adequately measured by standard error, that "internal validity" is vulnerable to self-selection bias and laboratory problems, and that "generalizability" is limited. We propose that what is needed is a procedure (like formal metaanalysis methods) incorporating information from several separate HIV sentinel seroprevalence studies, in a manner that is reproducible and can take into consideration the differences between studies.
- Acquired immunodeficiency syndrome
- human immunodeficiency virus type 1
- survey methods
ASJC Scopus subject areas