On the statistical accuracy of biomarker assays for HIV incidence

Objective: To evaluate the statistical accuracy of estimates of current HIV incidence rates from cross-sectional surveys, and to identify characteristics of assays that improve accuracy. Methods: Performed mathematical and statistical analysis of the cross-sectional estimator of HIV incidence to evaluate bias and variance. Developed probability models to evaluate impact of long tails of the window period distribution on accuracy. Results: The standard cross-sectional estimate of HIV incidence rate is estimating a time-lagged incidence where the lag time, called the shadow, depends on the mean and the coefficient of variation of window periods. Equations show how the shadow increases with the mean and the coefficient of variation. We find with an assay such as BED capture enzyme immunoassay, if only 0.5% are elite controllers who remain in the window until death, then the shadow is over 2.3 years, implying that estimates reflect HIV incidence more than 2 years in the past rather than current levels. If even 5% of AIDS cases are unrecognized and not excluded from the numbers in the window, then the shadow is more than 2.2 years. Conclusions: Small perturbations to the tail of the window period distribution can have large effects on the accuracy of current HIV incidence estimates. The shadow and mean window period are useful for comparing the accuracy of assays. The results help explain differences reported between cohort and cross-sectional HIV incidence estimates. Screening out elite or viremic controllers by RNA polymerase chain reaction testing, and persons with advanced HIV disease (with AIDS or on antiretrovirals) may considerably improve the accuracy of HIV incidence estimates based on BED or similar assays.