(C-168) The Big Four: An Enhanced Framework for Drawing Causal Inferences from Observational Data

Background: Observational studies play an increasingly important role in causal inference. However, the literature addressing potential sources of bias that threaten internal or external validity remains surprisingly disjointed.

Objectives: To introduce a structured framework aimed at generating high-quality, causally interpretable real-world evidence that informs policy and practice.

Methods: Drawing on a wealth of insights from the field of epidemiology, we propose the “four targets” framework — comprising target estimand, target population, target trial, target validity — to guide researchers in producing credible and actionable causal inference from observational data.

Results: First, “target trial” emulation is a useful approach for design and analyses in observational research and mitigating biases related to internal validity (e.g., immortal time bias and selection bias). However, target trial emulation alone may be insufficient to establish more generalizable causal knowledge, as external validity is still a concern in observational studies due to many reasons (e.g., stringent eligibility criteria, and the automatic exclusion of patients without treatment access when using the active comparator, new user design). To enhance external validity, one must also specify the “target population” (the population about which inference is to be made) and the “target estimand” (the treatment effect to be estimated in a target population in response to a well-defined research question). By employing this paradigm, one can establish and maximize a joint measure of both internal and external validity, called “target validity,” enhancing the value of effect estimates for decision-making and planning. We will demonstrate the utility of the "four targets" framework using a case study from the electronic health records from the Veterans Health Administration, in which we compare the effectiveness of high-dose (24-32 mg/day) versus standard-dose (16-24 mg/day) buprenorphine for treating opioid use disorder.

Conclusions: Our “four targets” framework offers a cohesive approach to unify the scattered literature, facilitating policy- or practice-relevant causal inference from observational data. When using observational studies to infer causality, it is recommended to clearly define the ‘target estimand’ and ‘target population’, then outline and emulate a ‘target trial’ as closely as possible for study design and analysis, and finally discuss biases in terms of possible violations to ‘target validity’.