[Directed acyclic graphs in statistical modelling of epidemiological studies]

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Research

Background:

• Observational epidemiological studies can suffer from distorted exposure-event relationships due to confounding.
• Traditional "step-by-step" regression models lack a strong causal basis for confounder selection.
• Directed Acyclic Graphs (DAGs) provide a visual and qualitative method to assess confounding and guide statistical modeling.

Purpose of the Study:

• To compare "step-by-step" regression and DAGs for identifying the minimum set of confounders.
• To evaluate causal relationships in arsenic exposure studies.
• To determine optimal confounder control strategies in statistical modeling.

Main Methods:

• Utilized data from 66 individuals in northern Chile (Cáceres et al., 2010).
• Constructed a causal DAG using DAGitty v2.3 to identify potential confounders.
• Applied a step-by-step backwards multiple linear regression methodology for comparison.

Main Results:

• The causal diagram identified 12 non-causal open pathways.
• The minimum adjustment set for confounders included 'sex', 'body mass index', and 'fish and seafood ingest'.
• Multivariate model confusion retention involved overweight status, gender, and 'water intake' interaction with GSTT1.

Conclusions:

• Employing DAGs before statistical modeling enhances the comprehensiveness and coherence of causal analyses.
• DAGs facilitate biologically plausible interpretations of causal relationships in public health research.
• This approach improves the reliability of findings in observational studies.