Causal Inference
Practical hands-on guidance for creating directed acyclic graphs in epidemiologic sub-fields where they are under-utilized Rachel R Yorlets* Rachel Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Yorlets Department of Biostatistics, Boston University School of Public Health, Boston, MA, USA
Background
Directed acyclic graphs (DAGs) are essential for communicating epidemiologists’ assumptions about context-specific data-generating processes. DAGs are underutilised in tuberculosis (TB) research and global health more broadly because implementation is challenging: TB research involves complex social factors that are difficult to represent on a DAG. We aimed to develop and document a reproducible process for creating DAGs for TB.
Methods
In our underlying study, we aimed to understand how care-seeking factors are associated with time to TB treatment initiation in Lima, Peru; we surveyed participants on exposures representing pre-TB care access. We began with an initial DAG drafted by a TB expert who works in Lima. Next, we undertook an iterative decision-making process on how to 1) operationalize the exposures and outcome, 2) depict study selection, 3) represent symptomaticity as an effect measure modifier, and whether to include 4) mediators, 5) measurement error, or 6) multiple time points. Finally, we created a structured process for identifying, defining, and assessing the inclusion of measured and unmeasured covariates. We illustrated each step with annotated interim DAGs.
Results
We provide documentation and templates for collaborative decision-making steps that multidisciplinary teams can use to develop DAGs. Our documented process includes a scoping literature review, consultations with context experts, systematic revision of covariate definitions, and consensus-building on DAG structure with a focus on omitted edges and nodes. At each step, we focus on DAG-building aspects common to implementation-heavy fields where causal inference methods are uncommon. We share the final DAG.
Discussion
With increased transparency on how to apply DAGs to complex research questions, we help make foundational methods translational to fields like TB. We emphasize processes for DAG development that allow analyses to be grounded in evidence and subject-matter expertise.
