Prediction of pneumonia in the postoperative period of cardiac surgery using the classification tree

Background

The classification and regression tree (CART) is an easier alternative to logistic regression as a predictor of the chance of bedside event occurrence, not requiring complex calculations.

Objective

To assess the chance of predicting pneumonia using the CART model.

Patients

A total of 1158 patients undergoing cardiac surgery in the Instituto Nacional de Cardiologia Laranjeiras and at the Hospital Pró-Cardíaco, in the city of Rio de Janeiro, from January 2000 to September 2002.

Study design

A classical cohort.

Methods

Data collected in the databank of the surgical intensive care unit, using the Gini index for selection of the variables with a stop rule based on misclassification and equal prioris. The tree was constructed using Statistica 6.0.

Results

The following variables were selected with cutoffs discriminated by the program with sensitivity and specificity of 90.63% and 63.23%, respectively.

Conclusions

The model provides promising results, and should be validated and reassessed in subsamples or in a new sample.