The identification of a subgroup with greater mortality among patients undergoing surgical cardiac valve replacement (SCVR) may prevent inadequate management and also identify subgroups requiring review of the therapeutic strategies in surgical intensive care units (SICU).

To define the inhospital mortality (HM) on the first postoperative day (FPOD) using preoperative (PREOP), perioperative (PEROP), and FPOD variables.

A classical cohort with data consecutively collected at a public SICU (A, 326 patients) from January 2001 to February 2003, and at a private SICU (B, 121 patients) from June 2000 to February 2003. All 46 variables were previously defined according to the major prognostic indices in the literature, which were correlated with HM. A classification and regression tree (CART; using the Gini index with a FACT stop rule of 0.10 and equal priori) was created and followed by pruning based on misclassification and crossvalidation.

Based on CART, eight relevant variables were selected. The model had an accuracy of 81.33, sensitivity of 95%, and specificity of 80% for HM prediction.

CART may provide interesting solutions regarding the management of patients in the postoperative period of SCVR. Variables: FPOD SOFA score, PEROP fluid balance, FPOD epinephrine >0.1 or norepinephrine >0.1, patient's sex, left atrial length on ECHO, alveoloarterial O₂ tension gradient >250, PREOP creatinine, body mass index <20.