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

The authors used logistic regression, and classification trees to develop p rediction models for fatal outcomes in meningococcal disease in a cohort of 829 children hospitalized for meningococcal disease during 1989-1990 in Ri o de Janeiro. The area under the receiver operator characteristic (ROC) cur ve was 92 per cent for logistic regression and 88 per cent for classificati on trees. Logistic regression may be preferred when the main objective is t o obtain explicit measures for statistical inference and measures of the fo rce of the association between each variable and the outcome. However, esti mation of the probability of dying for each patient involves manipulation o f the logistic regression formula, which would not easily be done in an eme rgency room. Classification trees provided comparable discrimination betwee n fatal and non-fatal outcomes, and yielded a graphical display of the resu lts that is easier to understand and is straightforward to apply in clinica l settings.