Modeling the Risk of HIV Transfusion Transmission

Blood donations are routinely screened for HIV to prevent an infectious unit from being released to the blood supply. Despite improvements to blood screening assays, donations from infected donors remain undetectable during the window period (WP), when the virus has not yet replicated above the lower limit of detection (LOD) of a screening assay. To aid in the quantitative risk assessments of WP donations, a dose-response model describing the probability of transfusion-transmission of HIV over a range of viral RNA copies was developed.

An exponential model was chosen based on data fit and parsimony. A data set from a HIV challenge study using a nonhuman primate model and another data set from reported human blood transfusions associated with HIV infected donors were separately fit to the model to generate parameter estimates. A Bayesian framework using No-U-Turn Sampling (NUTS) and Monte Carlo simulations was performed to generate posterior distributions quantifying uncertainty in parameter estimation and model predictions.

The parameters of the exponential model for both nonhuman primate and human data were estimated with a mean (95% credible intervals) of 2.70 × 10 -2 (7.74 × 10 -3 , 6.06 × 10 -2 ) and 7.56 × 10 -4 (3.68 × 10 -4 , 1.31 × 10 -3 ), respectively. The predicted ID 50 for the animal and human models was 26 (12, 90) and 918 (529, 1886) RNA copies transfused, respectively.

This dose-response model can be used in a quantitative framework to estimate the probability of transfusion-transmission of HIV through WP donations. These models can be especially informative when assessing risk from blood components with low viral load.