Outcome modelling strategies in epidemiology: traditional methods and basic alternatives

Area of Science:

• Epidemiology
• Biostatistics
• Statistical modeling

Background:

• Excessive covariate control in regression models can lead to data sparsity and multicollinearity, especially with large covariate numbers relative to study size.
• Traditional methods for selecting covariates, such as stepwise regression and the 'change-in-estimate' (CIE) approach, have limitations.
• Accurate estimation of exposure effects is crucial in epidemiological and clinical research.

Purpose of the Study:

• To review traditional covariate selection strategies for outcome-regression models.
• To discuss the shortcomings of common methods like stepwise regression and CIE.
• To propose basic, accessible alternatives for minimizing mean squared error (MSE) and improving effect estimates within standard software.

Main Methods:

• Review of existing literature on covariate selection in regression modeling.
• Analysis of traditional methods including stepwise regression and the 'change-in-estimate' (CIE) approach.
• Development and discussion of alternative, simpler methods for covariate selection.

Main Results:

• Traditional covariate selection methods can exacerbate data problems.
• Stepwise regression and CIE have notable limitations that can compromise effect estimates.
• Basic, programming-free methods can approximate optimal covariate selection by minimizing MSE.

Conclusions:

• Careful consideration of covariate selection is essential to avoid model issues.
• Simpler, accessible methods can effectively improve the accuracy of effect estimates.
• Minimizing mean squared error (MSE) is a practical approach for robust statistical modeling in standard software.