Prediction for major adverse outcomes in cardiac surgery: comparison of three prediction models

Recent advances in medical treatment have altered the profile of patients referred for cardiac surgery. The proportion of high risk patients has increased dramatically. Numerous multifactorial risk scores have been developed to predict outcomes after cardiac surgery. However, these additive risk models were all developed outside of Asia and have never been validated in Taiwan. We applied the Parsonnet score, Tu score and logistic regression to a population in Taiwan who received cardiac surgery to predict the mortality, morbidity and likelihood of prolonged stay in the intensive care unit (ICU).

This retrospective study included 622 adult patients who received cardiac surgery during a 2-year period at Taichung Veterans General Hospital. The patients were randomly divided into a reference set (n = 423) and a validation set (n = 199). The Parsonnet score and Tu score were calibrated separately with the reference set to determine mortality, morbidity and likelihood of prolonged ICU stay. We developed a separate logistic regression model for each of the three outcomes by using the reference set. The validation set was used to test these models.

The area under the receiver operating characteristic (ROC) curve (AUC) of the Parsonnet score, Tu score and logistic regression for predicting in-hospital mortality were 0.843, 0.714 and 0.867, respectively. The AUC of the Parsonnet score, Tu score and logistic regression for predicting major morbidity were 0.784, 0.736 and 0.808, respectively. The AUC of the Parsonnet score, Tu score and logistic regression for predicting likelihood of prolonged ICU stay were 0.701, 0.689 and 0.764, respectively.

The Parsonnet score performed as well as the logistic regression models in predicting major adverse outcomes. The Parsonnet score appears to be a very suitable model for clinicians to use in risk stratification of cardiac surgery.