Incidence Inference Based on Cross-Sectionally Obtained Biomarkers for ‘Recent Infection’

Biomarkers for recent infection (as measured by tests often referred to as incidence assays) allow for incidence estimation from cross-sectional surveys, potentially overcoming biases and shortcomings associated with longitudinal (cohort) studies. Host or viral biomarkers are typically used to classify HIV-positive subjects as putatively ‘recently’ or ‘non-recently’ infected. For example:

  • a low proportion of HIV-specific Immunoglobin G (IgG) in total IgG, as measured by the widely used BED assay, or
  • a low antibody avidity (strength of binding to antigen) as measured by one of a number of modified ELISA assays

can be used as evidence of an immature antibody response to the virus, and thus to identify recent infection. A timeline (with links to references), capturing key developments of tests for recent infection, is provided. Cross-sectional survey counts of ‘recently infected’, ‘non-recently infected’ and ‘uninfected’ individuals can then be related to incidence, together with (estimates of) parameters capturing the characteristics of the biomarker in the population of interest.

Formal analysis can refine the intuition that greater ‘prevalence’ of recent infection implies greater incidence. However, the performance of currently available, and perhaps all conceivable, incidence assays is problematic, with some HIV-positive subjects classified as recently infected at times long after infection. Accounting for these ‘false recent’ results in a formal incidence estimator has led to some disagreement and controversy. Consensus may be emerging that recent infection tests are appropriately characterised through two parameters, namely describing:

  • the average time that individuals spend in the biomarker-defined state of recent infection, and
  • the proportion of ‘false-recent’ results in individuals with long-standing infection.

For reliable population-level surveillance using cross-sectional methodologies, there is a need for inexpensive, easy-to-use HIV biomarkers. To maximise statistical power, test developers face a difficult trade off:

  • The ‘recent infection’ state should not be too short, so that there will be sufficient numbers of individuals detected to be in this state at any point in time. A mean duration of recent infection of the order of a year is perhaps ideal from a statistical and temporal resolution point of view.
  • Any biological process which plays out, on average, over more than a few weeks, will inevitably exhibit substantial inter-subject variability, leading to a non-zero (or not even close to zero) false-recent rate. The formal accounting for the false-recent rate requires at least an unbiased estimate of this parameter, and brings intrinsic loss of precision, which becomes rapidly egregious as the false-recent rate rises above a few percent.

Consensus is needed within the field to help standardize methodology and terminology, develop and optimise HIV incidence assays, and provide guidelines for the use of this approach for incidence estimation. A brief review (with references) of recent progress in cross sectional incidence estimation methodology is provided.

Development of better technology for recent infection testing would be a more attractive investment if there were uses other than epidemiological surveillance. The idea of using the same, or slightly modified, technology to provide feedback to individual patients on likely status of infection as ‘recent’ of ‘non-recent’ has begun to attract considerable interest, and is being implemented in the UK, Italy and France. Wider uptake of this application, perhaps requiring some advances in methodological consensus, could substantially increase the incentive for investment in the relevant research and development.