Variable selection: current practice in epidemiological studies.

Area of Science:

• Epidemiology
• Biostatistics
• Medical Research Methodology

Background:

• Variable selection is crucial for accurate epidemiologic analysis, impacting confounding control and prognostic accuracy.
• Traditional methods like stepwise selection are widely used but often criticized.
• Modern techniques offer potential improvements but their adoption is unclear.

Purpose of the Study:

• To evaluate the prevalence of different variable selection techniques in contemporary epidemiologic studies.
• To assess the adoption rate of modern methods, such as shrinkage and penalized regression.
• To identify barriers to the implementation of advanced variable selection strategies.

Main Methods:

• Review of publications in leading epidemiological journals from 2008.
• Categorization and quantification of variable selection methods employed.
• Analysis of the reported use of stepwise, shrinkage, and penalized regression techniques.

Main Results:

• Stepwise selection methods were the predominant approach used in the reviewed literature.
• No instances of shrinkage methods were identified in the selected publications.
• A significant gap exists between the availability of modern methods and their application.

Conclusions:

• The epidemiological literature shows a strong reliance on traditional, often controversial, variable selection methods.
• Editors, reviewers, and authors have not adequately promoted or adopted newer, less controversial techniques.
• Further efforts are needed to integrate and validate the feasibility of advanced statistical methods in practice.