[Classification of dengue hemorrhagic fever using decision trees in the early phase of the disease].

Abstract

Introduction: dengue is a viral disease with endemic behavior. At the beginning of the illness it is not possible to know which patients will have an unfavorable evolution and develop a severe form of dengue. However, some warning symptoms and signs may be present.

Objective: to apply decision tree techniques to the exploration of signs of severity in the early phase of the illness.

Methods: the study sample was made up of 230 patients admitted with dengue to "Pedro Kouri" Institute of Tropical Medicine in 2001. The variables considered for the classification were the signs, symptoms and laboratory exams on the third day of evolution of the illness. The algorithm of classification and regression trees using the Gini's index was applied. Different loss matrices to improve the sensitivity were considered.

Results: the algorithm CART, corresponding to the best loss, had a sensitivity of 98,68% and global error of 0,36. Without considering loss, it obtained its sensitivity reached 74% with an error of 0,25. In both cases, the most important variables were platelets and hemoglobin.

Conclusions: the study submitted rules of decision with high sensitivity and negative predictive value of utility in the clinical practice. The laboratory variables resulted more important from the informational viewpoint than the clinical ones to discriminate clinical forms of dengue.