Using Network-based Modeling to Implement Strategies for Reducing HIV Drug Resistance

Abstract

Mathematical models have evolved to capture epidemiological and infectious disease data and complex relationships between the disease, host, and environment. Infectious disease and epidemic models serve as ethical approaches for introducing strategies intended to prevent, control, and reduce disease progression, transmission, and exposure, especially when the risk of harm and costs are high. These models also allow researchers, clinicians, and other health professionals to explore strategies and interventions that have yet to be widely explored, either due to limited information or resources. This thesis uses a network-based, stochastic modeling system called EvoNet to capture the spread of the human immunodeficiency virus (HIV) and explore strategies that may reduce HIV drug resistance (HIV-DR), within-host and between-host. The first aim focuses on predicting the frequency and emergence of drug-resistant mutations, given the use of first-line antiretroviral treatment, patient-specific information such as drug adherence levels, and the incorporation of pharmacogenomics (PGx) and pharmacokinetic (PK) data. The second aim investigates the effect of host genetic variation and drug resistance mutations in the population, using PGx and PK study data from two sub-Saharan African populations. Lastly, in the third aim, we build a treatment-switching optimization routine and develop a method similar to the grid search optimization method, and use an R package that performs simulated annealing. The objective of using these two optimization methods is to find the best parameter values that reduce the levels of drug resistance in the simulated population and to also inform how health professionals may prioritize individuals for second-line treatment. The results of these aims include the following: (1) the inclusion of host genetic or PGx data influences the frequency of drug resistance mutations, within-host and how rapidly the drug resistance mutations emerge, (2) the presence of individuals with drug-resistant strains in the population at the start of the model simulation yields higher levels of predicted drug resistance and in particular, transmitted drug resistance, and (3) the adapted grid-search optimization approach had a higher computational time burden than simulated annealing but provided a wider range of options for group prioritization for second-line treatment conditions that dramatically reduced HIVDR in the simulated population. In all, these methodologies and results may extend to future investigations of new drugs and treatment regimens.