Timing HIV infection with nonlinear viral dynamics

Abstract

AbstractIn HIV prevention trials, precise identification of infection time is critical to quantify drug efficacy but difficult to estimate as trials may have relatively sparse visit schedules. The last negative visit does not guarantee a boundary on infection time because viral nucleic acid is not present in the blood during early infection. Here, we developed a framework that combines stochastic and deterministic within-host mathematical modeling of viral dynamics accounting for the early unobservable viral load phase until it reaches a high chronic set point. The infection time estimation is based on a population non-linear mixed effects (pNLME) framework that includes the with-in host modeling. We applied this framework to viral load data from the RV217 trial and found a parsimonious model capable of recapitulating the viral loads. When adding the stochastic and deterministic portion of the best model, the estimated infection time for the RV217 data had an average of 2 weeks between infecting exposure and first positive. We assessed the sensitivity of the infection time estimation by conducting in silico studies with varying viral load sampling schemes before and after infection. pNLME accurately estimates infection times for a daily sampling scheme and is fairly robust to sparser schemes. For a monthly sampling scheme before and after first positive bias increases to -7 days. For pragmatic trial design, we found sampling weekly before and monthly after first positive allows accurate pNLME estimation. Our estimates can be used in parallel with other approaches that rely on viral sequencing, and because the model is mechanistic, it is primed for future application to infection timing for specific interventions.