Comparison of Three Preoperative Multifactorial Risk Indices to Predict Mortality and Prolonged Length of Stay in Hospital After Cardiac Surgery

Abstract

Introduction Many multifactorial risk indices developed after multivariate analysis have been proposed to predict mortality and other outcomes after cardiac surgery. [1-3] However few risk indices have been validated in centers other than where they were developed. Furthermore, comparisons of their predictive value within a single institution have not been made. This study compares 3 existing multifactorial risk indices [1-3] as predictors of mortality and prolonged length of stay (LOS) in our institution after cardiac surgery. The objective is to determine the most appropriate risk index to conduct risk-adjusted evaluation of quality of care in our institution. Methods This is a prospective observational study of all patients undergoing a cardiac procedure in our institution between November 12, 1996 and September 29, 1997. Pediatric cardiac procedures, heart transplants and other thoracic procedures without a concomitant cardiac surgery were exclusion criteria. Preoperative and intraoperative data were collected by the attending anesthetists who completed the database forms at the time of surgery. The accuracy and completeness of the data were verified by a research assistant. The database contains 130 preoperative variables pertaining to the severity of the patient's disease and comorbid factors and 80 variables related to intraoperative procedures and events. Using this information, each patient was given a risk score according to 3 risk classifications: 1) the Parsonnet's classification, which uses 15 risk factors with predetermined scores and a possible maximum of 98 points, and 2 factors with scores given on the basis of physician's judgement, accounting for another possible maximum of 60 points; 2) the Higgins' classification which has 13 risk factors with predetermined scores and a possible maximum of 31 points; 3) the Tuman's classification which includes 12 risk factors with predetermined scores and a possible maximum of 23 points. Each risk classification was tested to see how it predicts mortality and postoperative LOS in hospital >or=to 7, 14, 21 and 28 days. The accuracy of each classification in the prediction of those variables was analyzed by building receiver operating characteristic (ROC) curves for each outcome. Classifications providing an area greater than 0.70 under the ROC curve for all variables were considered clinically useful outcome predictors. The risk indices were compared by Chi-square test and analysis of variance. P < 0.05 was considered significant. Results A total of 1322 cardiac surgical patients were studied. There were 39 deaths (2.95 %). The mean +/- SD postoperative LOS in hospital was 8.4 +/- 9.6 days with a median of 6 days. The areas under the ROC curves for prediction of mortality and LOS in hospital were: (Table 1)Table 1Discussion The three tested multifactorial risk classifications have acceptable predictive accuracy and thus, are potentially useful to conduct risk-adjusted evaluation of mortality and prolonged LOS in hospital after cardiac surgery. The Parsonnet's classification is slightly but significantly more accurate, possibly because its scoring system also involves some clinical judgement. Clinical judgement was based on a single investigator's opinion (JYD) in this study. Whether inter-individual variability, as expected when using the classification as designed, may affect the predictive accuracy of the Parsonnet's classification remains to be determined.