Classification trees and logistic regression applied to prognostic studies: A comparison using meningococcal disease as an example

Abstract

The authors used logistic regression and classification trees to develop prediction models for fatal outcomes in meningococcal disease in a cohort of 829 children hospitalized for meningococcal disease during 1989-1990 in Rio de Janeiro. The area under the receiver operator characteristic (ROC) curve was 92 per cent for logistic regression and 88 per cent for classification trees. Logistic regression may be preferred when the main objective is to obtain explicit measures for statistical inference and measures of the force of the association between each variable and the outcome. However, estimation of the probability of dying for each patient involves manipulation of the logistic regression formula, which would not easily be done in an emergency room. Classification trees provided comparable discrimination between fatal and non-fatal outcomes, and yielded a graphical display of the results that is easier to understand and is straightforward to apply in clinical settings.