Causal Inference
Standardized Tools for Constructing DAGs: Advancing Causal Inference and Risk Assessment in Pharmaceutical Studies Sherrine Eid* Laura Watson Sherrine Eid
Causal inference and directed acyclic graphs (DAGs) are powerful tools in pharmaceutical research, facilitating analysis of critical questions and decision-making. By visualizing causal relationships, DAGs help identify confounders, mediators, and colliders, guiding the selection of covariates to control confounding and improve causal estimates. They play a role in detecting and mitigating biases, thus improving study design and analysis robustness. For comparative effectiveness research, DAGs clarify causal pathways, supporting the evaluation of treatment effects and drug efficacy and safety. They inform and optimize study designs for randomized controlled trials and observational studies. DAGs support risk assessments by analyzing real-world data for adverse drug reactions and long-term safety. They also support data integration by identifying compatible datasets and combining findings while preserving causal interpretability. Beyond analysis, DAGs are educational tools for communicating causal relationships and informing regulatory decisions and healthcare policies.
Specifically in bias detection, DAGs provide a framework for visualizing relationships among variables to assess potential biases in causal inference. Bias typically arises from three sources: data source (e.g., systematic inclusion/exclusion of subjects or stakeholder influence), study design, and analysis methods. While SAS procedures like PROC ASSESSBIAS address biases from analysis, they overlook biases from data sources and study design. DAGs bridge this gap by identifying biases from all sources, offering a broader understanding of causal inference.
Standardized tools for constructing DAGs would ensure consistency in analysis, enabling reproducible and comparable results across studies. SAS Viya, SAS 9, R, and Python offer strengths and limitations for conducting causal inference and leveraging DAGs in pharmaceutical research. We explore these comparative strength and limitations in this work.