HIV incidence

Methods for monitoring incidence can vary depending on the epidemic setting and are typically categorized either as direct or indirect measures. Direct measurement at a population level is preferred but can often be difficult to obtain. As a result, most if not all countries rely on indirect measures or triangulate direct and indirect methods.

Strategies for directly measuring HIV incidence include longitudinal follow-up and repeat testing among individuals who do not have HIV infection and estimation using a laboratory test for recent HIV infection and clinical data in the population. Longitudinal monitoring is often costly and difficult to perform at a population level. Laboratory testing of individuals to determine the recency of infection also raises cost and complexity challenges since a nationally representative population-based survey is typically required to obtain estimates.

Instead of direct measurement, the recommended source is the Spectrum AIDS impact model. Depending on the country’s epidemic setting and structure of surveillance data, Spectrum fits the national incidence trend from epidemic start to current year, considering population-based household HIV serosurveys, seroprevalence from sentinel surveillance and/or routine HIV testing in antenatal care, HIV case and AIDS-related death surveillance programme and clinical data, and in some instances data about risk behaviour and prevention coverage.

Note that case surveillance data counting new diagnoses of HIV should not be interpreted as a direct measure of people newly infected with HIV in the reporting year. As a result of reporting delays and underdiagnosis, the number of newly reported cases may not accurately reflect the actual number of people who were newly infected during the same period. Case reports may be useful, however, for triangulating or validating a Spectrum estimate—and in the case surveillance and vital registration (CSAVR) option within Spectrum, used by countries with strong case surveillance systems, it could be a model input