Cardiovascular Risk Scores to Predict Perioperative Stroke in Noncardiac Surgery

Background and Purpose— Perioperative stroke is associated with significant morbidity and mortality. Conventional cardiovascular risk scores have not been compared to predict acute stroke after noncardiac surgery. Methods— Patients undergoing noncardiac surgery between 2009 and 2010 were identified from the US National Surgical Quality Improvement Program (n=540 717). Patients were prospectively followed for 30 days postoperatively for the primary outcome of stroke. Established cardiovascular and perioperative risk scores (CHADS 2 , CHA 2 DS 2 -VASc, Revised Cardiac Risk Index, Mashour et al risk score, Myocardial Infarction or Cardiac Arrest risk score, and National Quality Improvement Project American College of Surgeons surgical risk calculator) were assessed to predict perioperative stroke. Results— Stroke occurred in the perioperative period of 1474 noncardiac surgeries (0.27%). Patients with perioperative stroke were older, more frequently male, had lower body mass index, and were more likely to have undergone vascular surgery or neurosurgery than patients without stroke ( P <0.001 for each comparison). All risk prediction models were associated with increased risk of perioperative stroke (C statistic [AUC] range, 0.743–0.836). The Myocardial Infarction or Cardiac Arrest risk score (AUC, 0.833) and American College of Surgeons surgical risk calculator (AUC, 0.836) had the most favorable test characteristics and a greater ability to discriminate perioperative stroke when compared with Revised Cardiac Risk Index, CHADS 2 , CHA 2 DS 2 -VASc, and Mashour risk scores ( P for comparison, <0.001; Delong). Risk scores did not provide consistent discriminative ability across surgery types and were least predictive in vascular surgery (AUC range, 0.588–0.672). Conclusions— The Myocardial Infarction or Cardiac Arrest risk score and American College of Surgeons surgical risk calculator surgical risk scores provide excellent risk discrimination for perioperative stroke in most patients undergoing noncardiac surgery. Stroke prediction was less optimal in patients undergoing vascular surgery.