Abstract
BACKGROUND: Determining a therapeutic window for maintaining antiretroviral drug concentrations within an appropriate range is required for identifying effective dosing regimens. The limits of this window are typically calculated using predictive models. We propose that target concentrations should instead be calculated based on counterfactual probabilities of relevant outcomes and describe a counterfactual framework for this. METHODS: The proposed framework is applied in an analysis including longitudinal observational data from 125 HIV-positive children treated with efavirenz-based regimens within the CHAPAS-3 trial, which enrolled children < 13 years in Zambia/Uganda. A directed acyclic graph was developed to visualize the mechanisms affecting antiretroviral concentrations. Causal concentration-response curves, adjusted for measured time-varying confounding of weight and adherence, are calculated using g-computation. RESULTS: The estimated curves show that higher concentrations during follow-up, 12/24 h after dose, lead to lower probabilities of viral failure (> 100 c/mL) at 96 weeks of follow-up. Estimated counterfactual failure probabilities under the current target range of 1-4 mg/L range from 24% to about 2%. The curves are almost identical for slow, intermediate and extensive metabolizers and show that a mid-dose concentration level of ≥ 3.5 mg/L would be required to achieve a failure probability of < 5%. CONCLUSIONS: Our analyses demonstrate that a causal approach may lead to different minimum concentration limits than analyses that are based on purely predictive models. Moreover, the approach highlights that indirect causes of failure, such as patients' metabolizing status, may predict patients' failure risk, but do not alter the threshold at which antiviral activity of efavirenz is severely reduced.