Doubly Robust Causal Modeling to Evaluate Device Implantation

Evidence from randomized clinical trials (RCTs) is the standard for evaluating treatment safety and efficacy. However, observational studies using large databases are increasingly being used due to their greater generalizability and feasibility. These studies are threatened by confounders, which are factors that affect the treatment group and outcome. Statistical approaches can control for confounders by modeling their associations with the treatment, or both. Correctly specifying these models is challenging, and model misspecification can lead to residual confounding. Doubly robust approaches model associations of confounders with both the treatment and outcome, providing two opportunities to correctly model confounders.