Dichotomization of continuous variables in logistic regression models.

It is highly to observe in the biomedical literature, that the continuous variables such as systolic blood pressure and cholesterol, are dichotomized and used in such manner in a given statistical analysis. The consequences of such transformation can be very varied. In this article, how the dichotomization of a continuous exposure variable affects the quality of the prediction of a response using the logistic regression is examined. One can conclude that, in almost all the studied stimulation, the percentages of misclassification can increase drastically, reaching over there times the probability of misclassification of the ones using the original variable. Therefore, we recommend to avoid dichotomization.