Estimates of HIV incidence may be unreliable: techniques need refinement

The BED method for estimating HIV incidence should be standardised so that it can be reliably used in an African setting, and the technique needs to be improved to avoid counting people with longstanding HIV infection as recent infections, according to two South African research groups. Their findings were presented at the South African AIDS Conference in Durban earlier this month.

The BED assay is an enzyme-based method that allows the time of HIV infection to be estimated for research purposes. It is not routinely used for diagnosis of HIV infection in the clinic. The procedure is based on measuring the proportion of anti-HIV antibody immunoglobulinG (IgG), and comparing it to the total blood IgG using an ELISA method that permits the concentration of blood IgG to be measured. The time of HIV infection can then be estimated because, in principle, the ratio of IgG (normal vs HIV) varies in accordance with disease progression.

The BED technique is often used in conjunction with mathematical techniques to estimate the incidence of HIV and has been widely used in South Africa, Uganda and the USA. However, despite its popularity, there have been complaints that the method can be inaccurate.

Thomas McWalter, of the University of the Witswatersrand in Johannesburg, South Africa, hoped to shed light on the accuracy of the BED assay by testing whether it had been uniformly applied in a cross-section of published BED incidence estimates.

His research team searched journal databases and discovered a total of 1136 publications that included review articles, full texts and abstracts. Of this article group, a total of 27 studies were included in a final review that took location, period, population, reported HIV incidence, the incidence estimation formula, and adjustment for false-recent results into account.

McWalter’s team discovered substantial divergence in methods in the studies. Differences were found in laboratory parameters, optical density cut-offs (the measure of the ratio of HIV-IgG to total-IgG seen as sufficient for people to be considered seroconverted), window periods, incidence estimate methods and sensitivity analysis methods (mathematical methods for testing the consistency of results).

McWalter's team then applied a universal, standardised set of procedures and parameters (such as optical density cut-offs and incidence estimate formulae) to each of the published data sets. This resulted in dramatic changes to many of the calculated incidence values.

The research indicates that using different methods gives rise to different incidence values and so clear and standardised procedures are needed if incidence values are to be comparable.

Further research presented at the conference by Professor John Hargrove of the South African Centre for Epidemiological Modeling and Analysis (SACEMA) highlighted additional problems with the technique.

It is well known that a subset of HIV-positive people tested with the BED assay always appear to have been recently infected even when they have been infected for many months or years. The reason for this slow (or non-existent) progression to a full antibody response is not well understood, although its occurrence is well documented. To complicate matters, another subset of people who initially seroconvert later revert to a non-seroconverted state according to the BED assay, an event, which if frequent enough, also produces incidence inaccuracies. (It should be noted that this phenomenon does not apply to standard HIV testing using the ELISA-based test.)

The Hargrove research team tested the extent to which these inaccuracies existed within a Zimbabwean dataset of 14,000 mothers, each of whom had been followed after giving birth. Of those who were HIV-positive, 5.2% were found to have BED absorbances less than 0.8 after their twelve-month retest and had, therefore, been falsely identified in the initial study as recent seroconverters. Consequently, the estimated BED annual incidence at twelve months for this group (7.6%) was 2.2 times higher than it should have been.

The Hargrove team stress that, unless the proportion of long-term false-recent cases is adjusted for, the BED method cannot be reliably used, even when estimating relative changes in HIV incidence. He further emphasised that the proportion of people who test false-recent may be location-specific and that estimates of the proportion of people in this group should be obtained in any incident estimate study.

Both bodies of research indicate that the BED technique needs to be improved and standardised before reliable HIV incidences can be estimated from cross-sectional studies. These findings have strong implications for public policy because, without an accurate estimate of HIV incidence, it is impossible to measure whether or not prevention and treatment programmes are having an effect on HIV incidence.