A simulation study on implementing marginal structural models in an observational study with switching medication based on a biomarker

• Published in: Journal of biopharmaceutical statistics Vol. 28; no. 2; pp. 350 - 361
• Main Authors: Kim, Hyang, Cable, Greg
• Format: Journal Article
• Language: English
• Published: England Taylor & Francis 04.03.2018; Taylor & Francis Ltd
• Subjects: Biomarkers; Causal inference; confounding; directed acyclic graph; marginal structural model; measurement errors; observational study; selection bias; switching medication; Causal inference; selection bias; confounding; measurement errors; directed acyclic graph; switching medication; marginal structural model; observational study
• ISSN: 1054-3406; 1520-5711; 1520-5711
• DOI: 10.1080/10543406.2017.1402783

Assessing treatment effectiveness in longitudinal data can be complex when treatments are not randomly assigned and patients are allowed to switch treatment to other or no treatment, often in a manner that is driven by changes in one or more variables associated with patient or clinical characteristics. There can be confounding of the treatment effect from a time-varying variable, i.e., one which is affected by previous exposure and can in turn also influence subsequent treatment changes. Precision medicine relies on validated biomarkers to better classify patients by their probable response to treatment. However, biomarkers may be a source of time-varying confounding, which are affected by prior treatment in the evaluation and are also subject to measurement errors. The impact of switching medications based on a biomarker has received less attention. We conducted simulation studies to explore biased estimation under various scenarios when marginal structural model estimations are employed. Holding model misspecification issues constant, bias is severe in the presence of multiple switching, along with measurement error and missing data in the covariates.