eComment. The importance of choosing a proper predictor variable selection method in logistic regression analyses

In their recently published manuscript Vogt et al. investigated factors associated with permanent pacemaker implantation after sutureless bioprosthetic aortic valve replacement [1]. In the statistical analysis section of the manuscript, authors reported that they included only the variables, which are significantly different among groups. We think that a delicately chosen predictor variable selection method will make this study more precise. In logistic regression analysis, selection of predictor variables in a regression model can influence the outcome. To overcome the problems in selecting predictor variables, there are some methods available in statistical software programmes. The purpose of multiple logistic regression is to define the functional relationship between predictor variables and outcome.

During statistical model building variables are minimized as much as possible so that the most parsimonious model that describes the data is found. Commonly used variable selection methods are hierarchic selection, forced entry, and stepwise methods. In hierarchic selection researcher determines the possible variables entering into the model based on previous studies. Variables which have already proven to be a predictor enter the model first; other variables are incorporated subsequently. Forced entry is a method in which all predictors forced into the model. This method is not suitable for high number of variables like Vogt et al.'s research (they described 61 variables). Stepwise regression predictor variable selection is based on mathematical criteria. There are two different stepwise selection methods: forward and backward. In forward selection, which involves starting with no variables in the model, chi-square statistic is computed for each effect and the largest of these statistics is determined. The computer adds the variable if it improves the model. In backward elimination, which involves starting with all candidate variables, testing the deletion of each variable using the results of the Wald test for individual variables are examined. The variable that does not improve the model is removed. As a result, the validity and quality of research rely heavily on statistical methodology. In logistic regression analyses, researchers must select a suitable predictor entry method for their studies.

Conflict of interest: none declared.