Exploratory causal modeling in epidemiology: are all factors created equal?

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Epidemiological research often explores etiological factors using statistical models.
• Stepwise multiple regression is a common method, but it assumes equal proximity of all factors to the outcome.
• This assumption is often unrealistic, especially for factors of different types (e.g., psychological, biological).

Purpose of the Study:

• To demonstrate the consequences of analyzing sequentially caused relationships using models that assume equally proximate causation.
• To compare the performance of logistic modeling versus simple pathway analysis in identifying true causation in simulated data.

Main Methods:

• Conducted Monte Carlo simulations with data exhibiting well-defined sequential causal relationships.
• Applied logistic modeling and simple pathway analysis to the simulated data.
• Evaluated the accuracy of each method in identifying distant versus proximal causal factors.

Main Results:

• Logistic modeling assuming equal proximity was misleading when analyzing distant causal factors.
• Simple pathway analysis successfully identified true causation in simulated causal pathways.
• The relative risk of an intermediate cause must be sufficiently large for a distant variable's effect to propagate; the distant variable's prevalence is less critical than the intermediate variable's low prevalence.

Conclusions:

• Models assuming equal proximity of causal factors to an outcome can obscure or dismiss more distant etiological influences.
• Simple pathway analysis is a more appropriate method for exploring complex causal chains with varying factor proximities.
• Current reliance on stepwise multiple regression hinders the upstream exploration of distant causal factors in epidemiology.