Pragmatic Clinical Trials in the Rubric of Structural Causal Models

TL;DR

This paper introduces a generalized structural causal model framework for pragmatic clinical trials, enabling better causal inference and analysis techniques tailored to their unique design characteristics.

Contribution

It proposes a novel SCM representation for PCTs, bridging the gap between explanatory and observational studies, and demonstrates its application with real clinical data.

Findings

A new graphical model for PCTs is established.

Analysis techniques like intention-to-treat are adapted within the SCM framework.

Application to real data shows the model's practical utility.

Abstract

Explanatory studies, such as randomized controlled trials, are targeted to extract the true causal effect of interventions on outcomes and are by design adjusted for covariates through randomization. On the contrary, observational studies are a representation of events that occurred without intervention. Both can be illustrated using the Structural Causal Model (SCM), and do-calculus can be employed to estimate the causal effects. Pragmatic clinical trials (PCT) fall between these two ends of the trial design spectra and are thus hard to define. Due to its pragmatic nature, no standardized representation of PCT through SCM has been yet established. In this paper, we approach this problem by proposing a generalized representation of PCT under the rubric of structural causal models (SCM). We discuss different analysis techniques commonly employed in PCT using the proposed graphical model, such as intention-to-treat, as-treated, and per-protocol analysis. To show the application of our proposed approach, we leverage an experimental dataset from a pragmatic clinical trial. Our proposition of SCM through PCT creates a pathway to leveraging do-calculus and related mathematical operations on clinical datasets.