Inference for causal interactions for continuous exposures under dichotomization.

Area of Science:

• Epidemiology
• Biostatistics
• Medical Research

Background:

• Dichotomization of continuous exposure variables is prevalent in medical and epidemiological studies.
• This practice has faced criticism due to potential inefficiencies and biases.
• The implications of dichotomization for studying interactions between variables require careful consideration.

Purpose of the Study:

• To investigate the consequences of dichotomizing continuous covariates on the analysis of interactions.
• To demonstrate how inferences about causal interactions can be made on the original continuous exposure scale, even after dichotomization.
• To provide a method to prevent erroneous conclusions about interactions caused by improper exposure variable modeling.

Main Methods:

• The study theoretically analyzes the impact of dichotomizing a continuous exposure variable within interaction analyses.
• It proposes methods to draw valid inferences about causal interactions on the original continuous scale.
• The approach involves examining different dichotomization points to understand interaction effects at various exposure levels.

Main Results:

• Inferences regarding causal interactions can be validly drawn concerning the original continuous exposure scale, despite dichotomization.
• The proposed methods help mitigate incorrect conclusions about interaction presence arising from modeling errors.
• Analyzing various dichotomization points provides deeper insights into the nature of causal interactions between exposures.

Conclusions:

• Dichotomization of continuous exposures, when analyzed appropriately, can yield valid insights into causal interactions.
• This approach offers a way to avoid misinterpretations of interaction effects stemming from modeling inaccuracies.
• The study's findings are applicable to real-world scenarios, such as investigating combined effects of smoking and arsenic exposure on skin lesions.