Testing bias in calculating HIV incidence from the Serologic Testing Algorithm for Recent HIV Seroconversion

Abstract

Incidence is critical in monitoring HIV infection in populations but often difficult to measure. The Serologic Testing Algorithm for Recent HIV Seroconversion (STARHS) can estimate HIV incidence from a single specimen at low cost. Nevertheless, HIV testing patterns may introduce bias, rendering interpretation of the STARHS result problematic. We found empirical evidence of such bias in Ontario using the STARHS formula with varied window periods In a hypothetical population of homosexual men, we calculated HIV incidence from the STARHS assay on the basis of incidence density, study duration, STARHS window period and intertest interval. We also incorporated the increased likelihood of a newly infected person having an HIV test due to seroconversion illness or high-risk behaviours ('seroconversion effect' or SCE). We also varied the intertest interval inversely as a function of incidence density. To adjust incidence estimates for bias, we fit empirical STARHS data to an algebraic formula expressing measured HIV incidence as a function of SCE and incidence. Incidence density estimates were unbiased when SCE or incidence density-interval interactions were absent. However, estimated incidence density was higher than true incidence density in the presence of SCE, as much as seven-fold higher under certain conditions. The goodness-of-fit provided estimates with an excellent fit, yielding plausible results. HIV incidence from STARHS may be strongly biased because of early testing in recently infected persons, resulting in substantial overestimation, at least amongst men who have sex with men. Thus, incidence estimates from STARHS must be interpreted with considerable caution. Nevertheless, incidence estimates may be adjusted to yield unbiased results.