DEVELOPMENT AND VALIDATION OF A BAYESIAN MODEL FOR PERIOPERATIVE CARDIAC RISK ASSESSMENT IN A COHORT OF 1,081 VASCULAR SURGICAL CANDIDATES

Objectives. This study sought to develop and validate a Bayesian risk prediction model for vascular surgery candidates. Background. Patients who require surgical treatment of peripheral vascular disease are at increased risk of perioperative cardiac morbidity and mortality. Exist ing prediction models tend to underestimate risk in vascular surgery c andidates. Methods. The cohort comprised 1,081 consecutive vascular su rgery candidates at five medical centers. Of these, 567 patients from two centers (''training'' set) were used to develop the model, and 514 patients from three centers were used to validate it (''validation'' set). Risk scores were developed using logistic regression for clinica l variables: advanced age (>70 years), angina, history of myocardial i nfarction, diabetes mellitus, history of congestive heart failure and prior coronary revascularization. A second model was developed from di pyridamole-thallium predictors of myocardial infarction (i.e., fixed a nd reversible myocardial defects and ST changes). Model performance wa s assessed by comparing observed event rates with risk estimates and b y performing receiver-operating characteristic curve (ROC) analysis. R esults. The postoperative cardiac event rate was 8% for both sets. Pro gnostic accuracy (i.e., ROC area) was 74 +/- 3% (mean +/- SD) for the clinical and 81 +/- 3% for the clinical and dipyridamole-thallium mode ls. Among the validation sets, areas were 74 +/- 9%, 72 +/- 7% and 76 +/- 5% for each center. Observed and estimated rates were comparable f or both sets. By the clinical model, the observed rates were 3%, 8% an d 18% for patients classified as low, moderate and high risk by clinic al factors (p < 0.0001). The addition of dipyridamole-thallium data re classified >80% of the moderate risk patients into low (3%) and high ( 19%) risk categories (p < 0.0001) but provided no stratification for p atients classified as low or high risk according to the clinical model . Conclusions. Simple clinical markers, weighted according to prognost ic impact, will reliably stratify risk in vascular surgery candidates referred for dipyridamole-thallium testing, thus obviating the need fo r the more expensive testing. Our prediction model retains its prognos tic accuracy when applied to the validation sets and can reliably esti mate risk in this group.