Estimation of HIV treatment-efficacy by combining structural nested mean models with pharmacokinetic models of antiretroviral drug exposure

Abstract

The aim of treating HIV-1-infected patients is to achieve and maintain suppression of viral load (VL). Achievement of this aim is thwarted by variable adherence to prescribed anti-retroviral drugs. Variable adherence to an antiretroviral regimen creates variability in the patient's internal exposure to the drugs. Structural nested mean models (SNMMs) enabled us to estimate, during the initial phase of treatment, the relationship between variable internal exposure and VL, accounting for measured time-varying confounders and feedback relations using an antiretroviral regimen containing lopinavir/ritonavir (LPV/RTV, LPV/r). Our final SNMM predicts that the short term effect of treatment is modified by the most recent past VL, with higher initial VL's being associated with larger treatment-induced reductions in VL for a given internal exposure to the drugs. Variation in internal exposure to LPV/r in the interquartile interval (P25%-P75%) only slightly affects the overall reduction in VL, supporting the conclusion that the relatively long duration of action of LPV/r lessens the impact on VL of the most frequently recurring intermittent lapses in dosing.