Abstract
Incidence of an infectious disease such as HIV is traditionally estimated by following a cohort of susceptible individuals and registering new cases that occur during the follow-up period. However, in certain situations cohort follow-up may be unfeasible but a cross-sectional study can be done easily.
Biological markers correlated with time since infection have been proposed for estimating HIV incidence in cross-sectional studies. The incidence is estimated from the number of "recent infections", i.e. the samples whose biomarker is below some prespecified threshold.
However, it has been observed that the cross-sectional incidence estimates obtained in this way can be severely biased. We investigate the theoretical sources of the bias in cross-sectional incidence estimation and identify conditions on the biomarker that make the bias tolerable.
We show how to evaluate a collection of several biomarkers measured in conjunction on a set of infected blood samples with "known" times since infection and explain how to use such validation data to develop cross-sectional incidence estimates with optimized performance across a range of different populations.