Research question
Estimates of HIV incidence, the rate of new infections in a population, are essential for monitoring the progress of HIV epidemics and for targeting and evaluating interventions that prevent HIV acquisition and transmission.  Incidence estimates can be obtained through repeated testing of individuals in longitudinal surveillance. Such surveillances are, however, difficult to establish and costly to maintain, and will thus not be feasible sources of incidence information in most settings.

An operationally less demanding approach to estimating incidence uses tests that distinguish recent from non-recent infection in cross-sectional data.  While several tests for recent HIV infection (TRI) have been developed, the BED IgG-Capture Enzyme Immunoassay (BED assay), has been frequently applied, especially in developing countries. However, from the time of the development of the BED assay in 2002, there has been debate over how to correctly analyze data generated through its use.  In particular, the imperfect nature of the BED test raised questions about correct test calibration and optimal mathematical approaches to estimate incidence point and uncertainty estimates.

We pursued the following research questions:

·         What are the correct calibration parameters for use of the BED assay in rural South Africa?

·         Are these calibration parameters significantly different from those obtained in two previous studies in other settings?

·         What are the optimal mathematical approaches to estimate incidence point estimates using data from TRI?

·         What are the optimal mathematical approaches to estimate incidence uncertainty estimates using data from TRI?

·         Are incidence estimates in past published applications of the BED assay likely to have been accurate?

·         Can information from BED assay testing be used to prioritize communities in sub-Saharan Africa to HIV prevention interventions?

Fit of the research question within the AC portfolio/strategic plan
The research on TRI fits in closely with a number of central research themes at Africa Centre.   Obviously, TRI are a source of information on HIV incidence and can be used to identify communities and sub-populations at high risk of HIV acquisition or HIV transmission.  In addition, TRI can be used to improve the accuracy of information obtained in longitudinal HIV surveillance by, firstly, reducing the length of the intervals within which HIV seroconversion is known to have occurred and, secondly, by providing information on time since seroconversion in those individuals who participate only once in the longitudinal surveillance.

Data sources and methods
Only a few centres worldwide have the data to calibrate TRI, because longitudinal information on time since HIV seroconversion is necessary for such calibration.  We used the following data sources in our research

·         Longitudinal data on HIV status from the Africa Centre HIV surveillance

·         Cross-sectional HIV status data from the Africa Centre HIV surveillance and data on HIV infection recency.  The latter data was obtained in a BED assay study, co-funded by the Centers for Disease Control, in collaboration with Adrian Puren at the
South African National Institute for Infectious Disease in Johannesburg.

·         Theoretical mathematical models and sensitivity analyses

·         Systematic review of the literature 

 Findings

·         In a rural community in South Africa with high HIV prevalence, the key BED assay calibration parameter (the long-term false recent ratio or FRR) is substantially lower than previous estimates.

·         The BED assay performs well in HIV incidence estimation if the locally measured FRR is used, but significantly underestimates incidence when a FPR estimate based on previous studies elsewhere is used. Thus, locally valid estimates of the calibration parameters should be used in TRI-based incidence estimation.

·         The vast majority of the more than 50 published past applications of the BED assay are likely to have substantially mis-estimated HIV incidence, because they did not use correct mathematical approaches or locally valid calibration parameters.

·         In future TRI applications, it is particularly important to use an accurate estimate for the FRR, even when further clinical or biomarker information is used (since this additional information may not eliminate all false-recent individuals, such as those who never progress above the assay threshold).  As far as possible, further clinical or biomarker information should be used to reduce the FRR.

·         It is important to ensure that additional information used to exclude or reclassify individuals is applied consistently in both the study estimating the FRR and the subsequent incidence estimation study (e.g. it is incorrect to calculate incidence in a sample that includes individuals on ART using an FRR estimated in a sample that excluded individuals on ART).

·         Methods that account for false-recent results should be used to estimate incidence. In particular, the third generation estimators of McWalter/Welte/Bärnighausen or McWalter/ Welte should be used, because they are both parsimonious (as opposed to second generation estimators which are over-parameterized) and the least biased of all currently available estimators.

·         CIs for incidence estimates should be computed using approaches that take into account parameter uncertainty.

·         TRI data can be used to prioritize communities for HIV prevention interventions according to HIV acquisition risk and, in conjunction with data on late-stage disease, according to HIV transmission risk.

Policy implications
We have demonstrated that TRI can be valid sources of HIV incidence point and uncertainty estimates, but that past TRI applications have mostly mis-estimated incidence.  Our research has led to concrete recommendations on how to ensure that future TRI applications yield valid incidence point and uncertainty estimates.  We have further demonstrated how information obtained from TRI can be used to prioritize communities for HIV prevention interventions.