Using compartmental models to simulate directed acyclic graphs to explore competing causal mechanisms underlying...

Area of Science:

• Epidemiology and Biostatistics
• Causal Inference and Modeling

Background:

• Accurate estimation of exposure effects requires understanding causal relationships between variables.
• Directed acyclic graphs (DAGs) are crucial in epidemiology for causal process understanding and bias mitigation.
• Compartmental models (CMs) represent population-level causal mechanisms through disease state flows.

Purpose of the Study:

• To extend the mapping between DAGs and CMs for comparing competing causal mechanisms.
• To demonstrate the use of DAG-derived CMs for simulating epidemiological studies and analyzing bias.
• To evaluate the robustness of study results against design biases and underlying causal factors.

Main Methods:

• Developed a framework linking DAGs to CMs for simulation purposes.
• Simulated a longitudinal cohort study to investigate the obesity paradox in diabetic populations.
• Conducted statistical analyses on simulated data to assess bias and causal mechanisms.

Main Results:

• The DAG-derived CM framework successfully simulated epidemiological studies.
• Simulations illustrated how study design biases, such as reverse causation, can induce the obesity paradox.
• The approach effectively evaluated the impact of different biases and causal structures on study outcomes.

Conclusions:

• DAGs can be transformed into computational ('in silico') laboratories for systematic bias evaluation.
• This framework aids researchers in informing epidemiological analyses and study designs.
• The methodology provides a robust tool for understanding complex causal relationships and potential biases.