Comparative Analysis of HIV Estimation Methods and Results in Locations with Case Reports and High-Quality Vital Registration Data

Abstract

This thesis provides an overview of HIV burden estimation methods and results in high-income locations with high-quality vital registration data and case reports. In addition, it presents an investigation of the sensitivity of estimates to variation in model specification and input parameters. The study compared estimates of HIV burden from bespoke models in Australia, the Netherlands, and the United States to estimates produced using models developed by the Joint United Nations Programme on HIV/AIDS (UNAIDS) and the Institute for Health Metrics and Evaluation (IHME). The sensitivity of the results produced with the IHME model to differences in input parameter estimates from UNAIDS and IHME was investigated by conducting a Shapley decomposition. The decomposition breaks down the individual contributions of differences in a set of input parameters to variation in modeled estimates by averaging across multiple model runs with permutations of the input parameters. Upon comparison, the overall estimates from each model exhibited substantial heterogeneity, revealing the strengths and weaknesses of each modeling strategy in fitting to particular types of input data. From the decomposition analysis, antiretroviral therapy and disease progression parameters stood out as the largest contributors to variation between estimates from the IHME model, with one of the these two parameters explaining the majority of variation eight of the nine location-statistic combinations evaluated. This study indicates the importance of pursuing harmony in modelling strategies in order to provide a coherent understanding of trends in HIV burden in high-income countries. In addition, future investigation of available data for better estimation of ART coverage may reduce some of the observed variation in modeled estimates.