- Meeting abstract
- Open Access
Comparison between prognostic scores of patients undergoing cardiac surgery: which is the best score to predict mortality and length of stay in a surgical intensive care unit?
© BioMed Central Ltd 2003
- Published: 25 June 2003
- Public Health
- Logistic Regression
- Intensive Care Unit
- Univariate Analysis
- Cardiac Surgery
Several studies on prognostic scores (PS) in cardiac surgery (CS) and their comparisons have been published in the literature. However, these studies have been carried out with populations whose demographic characteristics and prevalence of pathologies differ from those found among ours. We know no study aiming at predicting length of stay (LOS) in surgical intensive care units (SICU).
A classical cohort with data of 1458 patients consecutively collected, and the three automatically calculated PS, from June 2000 (group B, 594 patients) and January 2001 (group A, 865 patients) to February 2003. The statistical analysis comprised univariate analysis with the Student t test, analysis of variance, and Mann–Whitney and Pearson tests, followed by logistic and multinominal regression, and the receiver operating characteristic curve.
The three PS had significantly different prediction of mortality and of LOS in the SICU (P < 0.0001). In predicting mortality and LOS in the SICU longer than 7 days, the three PS analyzed did not provide a good correlation (Nagelkerke's R2 of 0.134 = 13.4% and of 0.226 = 22.6%, respectively). The less significant PS for prediction of mortality is the preCleveland (P = 0.054) as compared with the EuroSCORE and the post-Cleveland (P < 0.0001). Comparing the receiver operating characteristic curves of the three PS for LOS in the SICU longer than 7 days and prediction of mortality, the following was observed, respectively: EuroSCORE, 0.575–0.745; preCleveland, 0.550–0.769; and post-Cleveland, 0.769–0.769.
Prognostic scores are not intended to predict LOS in the SICU. With regard to prediction of mortality, the three receiver operating characteristic curves are similar, and the logistic regression is worse for preCleveland. Although none of the PS analyzed seemed adequate to be used in this group of patients undergoing CS, post-Cleveland was the best among the three.