A Bayesian Model for Estimation of HIV Incidence Using Serial Prevalence Data

• Activity Number: 479
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Nicole Bohme Carnegie*+
• Companies: New York University
• Keywords: HIV incidence ; serial prevalence data ; Bayesian modelling

Abstract

One of the most common HIV monitoring systems in resource-limited settings is antenatal or neo-natal seroprevalence surveillance. Under this system, we have prevalence estimates in women of child-bearing age repeated, usually annually, over an extended period of time. We can then attempt to extract incidence rates over time, based on changes in population prevalence and known factors influencing the likelihood of an infected person being tested, for example excess mortality or disease-related infertility.

We build on the flexible method for estimating incidence from serial prevalence data, using a relative inclusion ratio (RIR) to adjust for differential inclusion of infected and uninfected persons in the sample developed by Ades and Medley (1994). We reformulate the model in a Bayesian framework and extend it to include adjustment for the sensitivity and specificity of tests. The resulting model is applied to data from King County, Washington, with a new specification of the RIR for application to surveillance data in public health clinics.