Presentation (CTI)

OBJECTIVES: Causal directed acyclic graphs (DAGs) are central to transparent causal inference in real-world evidence and health outcomes research. However, developing DAGs is often a time-intensive process requiring manual extraction and documentation of causal information from existing literature and domain expertise. Commonly used tools, such as DAGitty support robust DAG specification and construction but are not primarily designed for AI-assisted evidence extraction, versioned collaborative workflows, or living evidence annotation. Evolving causal inference workflows increasingly require more interactive, evidence-linked, and updateable approaches. We developed DAGLive, an AI-augmented, human-in-the-loop, evidence linked, version-controlled DAG development platform to support these emerging needs.
METHODS: An interactive cloud-based web application was created by integrating a large language model to support AI-assisted extraction and structuring of causal information from research questions and study documents, with human review, editable outputs, traceability, and version-controlled refinement of DAGs.
RESULTS: The resulting application, DAGLive, supports multiple DAG development pathways: (1) AI-assisted text-to-DAG generation from a natural-language research question, (2) automated extraction of causal variables and pathways from uploaded study PDFs, (3) spreadsheet import via Excel or CSV using defined variable and pathway schemas, and (4) fully manual in-app construction. The application supports login-based authentication, project-level organization, version history, and comparison of DAGs or project versions. DAG elements are visually distinguished by node and pathway types, and scenario-mode filters allow users to isolate specific pathway subsets. Users can annotate pathways with evidence sources, and use grouped variable views to navigate between high-level DAG structures and individual node-level pathways.
CONCLUSIONS: DAGLive integrates AI-assisted generation, document-driven extraction, evidence annotation, version control, and interactive visualization into a unified DAG development workflow. By making causal assumptions explicit, traceable, and updateable, the application may support more transparent and efficient causal inference workflows. The application is intended for open-source release to support transparency, reproducibility, and community-led refinement.