PURPOSE: We calibrated two versions of a dynamic compartmental model of human immunodeficiency virus (HIV) disease progression and transmission in the United States to evaluate different assumptions regarding the effect of antiretroviral therapy (ART) on disease progression of people living with HIV (PLWH).
METHODS: The two model versions had similar structures but different assumptions regarding the effects of ART on disease progression. In Model 1, we assumed that PLWH who were treated with ART experienced no disease progression while their HIV RNA viral load was suppressed (VLS). When not VLS, their disease progression was equivalent to that among PLWH who were never treated with ART. In Model 2, PLWH exposed to ART experienced slower (but non-zero) disease progression before and after VLS. Both models were calibrated to approximate the percentage of PLWH diagnosed, percentage of PLWH prescribed ART (Model 1) or VLS (Model 2), and incidence reported in surveillance data, with Model 2 additionally calibrated to fit prevalence. Model 2 was later validated against other model outputs, including AIDS deaths among the diagnosed PLWH.
RESULTS: Model 1 and Model 2 were able to meet all their calibration targets, but Model 1's structure consistently underestimated HIV prevalence and death. Only Model 2 could be parameterized to simultaneously fit all targets while performing well in model validation.
CONCLUSIONS: To closely replicate the relatively low number of deaths among HIV-infected persons in the United States and a prevalence of approximately 1 million infected persons, our model calibrations showed that more realistic assumptions, such as extending survival benefits of ART to those whose viral load is not suppressed must be included, given the relatively small proportion of persons who achieve VLS at any one point in time. We also demonstrated that multiple-target calibration is necessary to match complex epidemiologic data.
