Causal inference from observational data in neurosurgical studies: a mini-review and tutorial.

Abstract

Establishing a causation relationship between treatments and patient outcomes is of essential importance for researchers to guide clinical decision-making with rigorous scientific evidence. Despite the fact that randomized controlled trials are widely regarded as the gold standard for identifying causal relationships, they are not without its generalizability and ethical constraints. Observational studies employing causal inference methods have emerged as a valuable alternative to exploring causal relationships. In this tutorial, we provide a succinct yet insightful guide about identifying causal relationships using observational studies, with a specific emphasis on research in the field of neurosurgery. We first emphasize the importance of clearly defining causal questions and conceptualizing target trial emulation. The limitations of the classic causation framework proposed by Bradford Hill are then discussed. Following this, we introduce one of the modern frameworks of causal inference, which centers around the potential outcome framework and directed acyclic graphs. We present the obstacles presented by confounding and selection bias when attempting to establish causal relationships with observational data within this framework. To provide a comprehensive overview, we present a summary of efficient causal inference methods that can address these challenges, along with a simulation example to illustrate these techniques.