Improvement of SOFA's predictive power for death when associated with central venous oxygen saturation intermittently obtained in the first 24 hours of cardiac surgery postoperation
© BioMed Central Ltd 2005
Published: 7 March 2005
Central venous oxygen saturation (ScVO2) as well as SOFA, have been considered important parameters for follow-up, prognostic estimate, and therapeutic target in the management of critically ill patients.
To analyze the impact of ScVO2 on the postoperative (PO) period of cardiac surgery (CS), for the inhospital mortality predictive power of SOFA.
A cohort of 132 consecutive patients selected from January 2004 to August 2004 and divided into the following two groups: GI, death (n = 11, 8.3%); and GII, survivors. Blood samples were collected through a central venous catheter properly positioned in the right atrium. The ScVO2 measurements were taken in the postoperative period as follows: immediately (SV0), after 6 hours (SV1), after 24 hours (SV2), and identified the lower ScVO2 in each patient at the first 24 hours PO (SVL). SOFA was also registered on the first day PO. Inhospital mortality was considered when death occurred at any time during hospitalization. The t test was used for statistical analysis, followed by six logistic regression (LR), classification tables and ROC curves.
Considering the total of the sample amount (132 patients), the mean SOFA was 4.03 ± 2.35 considering a GI value of 5.72 ± 3 and a GII value of 3.8 ± 2.2 (P = 0.012). ScVO2 mean values and the t test results of GI compared with those of GII were as follows, respectively: SV0 54.8 ± 12.6% vs 65.4 ± 8.9% (P < 0.0001), SV1 56.6 ± 7.3% vs 68.5 ± 5.9% (P < 0.001), SV2 61.1 ± 7% vs 69.3 ± 5.3% (P < 0.001) and SVL 50 ± 10% vs 62.7 ± 7.6% (P < 0.001). After the LR and the classification table have predicted a 50% mortality, the isolated SOFA score obtained a 91.7% accuracy (AUCROC 0.683, P = 0.045, confidence interval 0.499–0.867). From all the tested variables in LR with the SOFA score, the one that obtained a greater accuracy was SV2 (93.9%, AUCROC 0.846, P < 0.001, confidence interval 0.737–0.954) with sensibility of 73% and specificity of 79% with a likelihood ratio (+) 3.52 and (-) 0.34.