Assessing the Effectiveness of Antiretroviral Adherence Interventions: Using Marginal Structural Models to Replicate the Findings of Randomized Controlled Trials

• Published in: Journal of acquired immune deficiency syndromes (1999) Vol. 43 Suppl 1; no. Supplement 1; pp. S96 - S103
• Main Authors: Petersen, Maya L, Wang, Yue, van der Laan, Mark J, Bangsberg, David R
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Hagerstown, MD Lippincott Williams & Wilkins, Inc 01-12-2006; Lippincott Williams & Wilkins
• Subjects: AIDS/HIV; Anti-HIV Agents - administration & dosage; Anti-HIV Agents - therapeutic use; Biological and medical sciences; Fundamental and applied biological sciences. Psychology; HIV Infections - drug therapy; Human viral diseases; Humans; Infectious diseases; Medical sciences; Microbiology; Miscellaneous; Patient Compliance; Patient Education as Topic - organization & administration; Pharmacists; Randomized Controlled Trials as Topic - methods; Treatment Outcome; Viral diseases; Virology; Antiretroviral agent; Microbiology; Retroviridae; Lentivirus; Adhesion; Virology; randomized controlled trial; Virus; Treatment; HIV; G-computation; Antiviral; Models; counterfactual; inverse probability of treatment; Human immunodeficiency virus; causal inference
• ISSN: 1525-4135; 1944-7884
• DOI: 10.1097/01.qai.0000248344.95135.8d

Randomized controlled trials of interventions to improve adherence to antiretroviral medications are not always feasible. Marginal structural models (MSM) are a statistical methodology that aims to replicate the findings of randomized controlled trials using observational data. Under the assumption of no unmeasured confounders, 3 MSM estimators are available to estimate the causal effect of an intervention. Two of these estimators, G-computation and inverse probability of treatment weighted (IPTW), can be implemented using standard software. G-computation relies on fitting a multivariable regression of adherence on the intervention and confounders. Thus, it is related to the standard multivariable regression approach to estimating causal effects. In contrast, IPTW relies on fitting a multivariable logistic regression of the intervention on confounders. This article reviews the implementation of these methods, the assumptions underlying them, and interpretation of results. Findings are illustrated with a theoretic data example in which MSM are used to estimate the effect of a behavioral intervention on adherence to antiretroviral therapy.