Keywords: unmeasured confounding, sensitivity analyses, systematic review, best practice proposal
Background: Unmeasured confounding is a fundamental analytical challenge impacting the validity of comparative observational research. While methods have been proposed as solutions, implementation remains scarce due to the variety of research scenarios and the complexity and varying information required to implement each approach. Thus, guidance is needed to assist researchers toward analytical solutions applicable for specific research settings. Objective: To provide an overview of methodological approaches and best practice recommendations for addressing unmeasured confounding. Methods/Results: A literature review identified 12 methods for estimating causal effects adjusted for unmeasured confounders, including falsification outcome, instrumental variable, Bayesian modeling, etc. A best practice proposal was developed using a “decision tree” approach that leads to methodological choices based on research needs and the available information on the unmeasured confounders. Conclusion: This systematic review and the proposed best practice guidance equips researchers to address concerns with unmeasured confounding, thus improving the validity of comparative observational research