Using tests for recent infection to estimate incidence: problems and prospects for HIV

Abstract

Tests for recent infection (TRIs), such as the BED assay, provide a convenient way to estimate HIV incidence rates from cross-sectional survey data. Controversy has arisen over how the imperfect performance of a TRI should be characterised and taken into account. Recent theoretical work is providing a unified framework within which to work with a variety of TRI- and epidemic-specific assumptions in order to estimate incidence using imperfect TRIs, but suggests that larger survey sample sizes will be required than previously thought. This paper reviews the framework qualitatively and provides examples of estimator performance, identifying the characteristics required by a TRI to estimate incidence reliably that should guide the future development of TRIs.

FIGURE 1

Uncertainty of incidence point estimates as a result of sample size and background incidence CV: coefficient of variation; pyar: person years at risk; TRI: test for recent infection. The coefficient of variation of estimates of incidence using a TRI depends on the sample size of the survey and the true incidence rate. Note that a sample size of 10,000 approximates to the typical size of household-based surveys in Sub-Saharan Africa, and that incidence in South Africa (where there is one of the largest epidemics) is estimated to be about two per 100 pyar). Assumptions: ω= 155 days; ε=0.05; no TRI parameter uncertainty; steady-state epidemic conditions; mean survival with HIV: 11 years [–33].

FIGURE 2

Probability of correctly inferring a reduction in incidence pyar: person years at risk; TRI: test for recent infection. The probability of detecting a reduction in incidence between two surveys, when incidence has actually been reduced by half, as a function of the sample size of the surveys (both assumed to be the same) and the baseline incidence rate. Assumptions: ω=155 days; ε=0.05; no TRI parameter uncertainty; significance α=5%; steady-state epidemic conditions at first survey, with equal prevalence at second survey; mean survival with HIV: 11 years [31,33].

Figure 3

Uncertainty of incidence point estimates as a result of TRI parameter uncertainty CV: coefficient of variation; pyar: person years at risk; TRI: test for recent infection. Coefficient of variation of incidence estimator, using a BED-like assay on a sample size of 5,000, in a population exposed to an incidence of two per 100 pyar, as a function of the uncertainty in the TRI parameters, assumed to be normally distributed. Assumptions: ω= 155 days; ε=0.05; steady-state epidemic conditions; mean survival with HIV: 11 years [31,33].

FIGURE 4

Systematic error in incidence point estimates as a result of systematic error in TRI parameters pyar: person years at risk; TRI: test for recent infection. Systematic error expressed as a percentage of the correct estimate, excluding counting error, observed in the incidence estimator, using a BED-like assay, as a function of a precisely known systematic error in the TRI parameters. Assumptions: ω=155 days; ε=0.05; steady-state epidemic conditions; mean survival with HIV: 11 years [31,33].

FIGURE 5

Uncertainty of incidence point estimates as a result of TRI performance CV: coefficient of variation; pyar: person years at risk; TRI: test for recent infection. Coefficient of variation of incidence estimator, on a sample size of 5,000, in a population exposed to an incidence of two per 100 pyar, as a function of the TRI parameters. Assumptions: no TRI parameter uncertainty; steady-state epidemic conditions; mean survival with HIV: 11 years [31,33].