All Times EDT
Keywords: Causal estimand, statistical estimand, Randomized clinical trials, Observational studies, counterfactual outcomes, identification
Scientific research is generally motivated by a causal question, e.g., does anti-diabetic drug X reduce the blood glucose concentration by yy%, as compared with the standard of care, among patients with type I diabetes? The causal question reflects the research objective and is often expressed as a causal estimand that is rigorously defined as what is to be estimated before a study starts. A causal estimand summarizes the distribution of counterfactuals or potential outcomes that cannot be directly estimated and therefore must be mapped from the counterfactual world to the real-world data via a statistical estimand. In this talk, we will first briefly review the current literature in estimands (e.g., ICH E9 [R1]) and then describe the similarities and differences of estimands that are commonly used in randomized clinical trials and real-world studies. After giving some considerations in defining an estimand in the real-world setting, we will outline the design and analytical methods to estimate causal estimands and then provide some examples in randomized and non-randomized studies. A roadmap o choose an appropriate estimand in clinical research will be discussed.