The prediction of postoperative stroke or death in patients with preoperative atrial fibrillation undergoing non-cardiac surgery: a VISION sub-study.

Summary Background The optimal means of pre-operative risk stratification in patients with atrial fibrillation (AF) is uncertain. Objective To examine the accuracy of AF thromboembolic risk models (the CHADS2, CHA2DS2-VASc, and R2CHADS2 scores) for predicting 30-day stroke and/or all-cause mortality after non-cardiac surgery in patients with preoperative AF, and to compare these risk scores with the Revised Cardiac Risk Index (RCRI). Patients/Methods A multicentre (8 countries, 2007–2011) prospective cohort study of patients ≥ 45 years of age undergoing inpatient non-cardiac surgery, who were followed until 30 days after surgery. We calculated c-statistics for each risk prediction model and net reclassification improvements (NRIs) compared with the RCRI. Results The 961 patients with preoperative AF were at higher risk of any cardiovascular event in the 30 days postoperatively compared with the 13 001 patients without AF: 26.6% vs. 9.0%; adjusted odds ratio, 1.58; 95% confidence interval [CI], 1.33–1.88. All thromboembolic risk scores predicted postoperative death just as well as the RCRI (with c-indices between 0.67 and 0.72). Compared with the RCRI (which had a c-index of 0.64 for 30-day stroke/death), the CHADS2 (c-index, 0.67; NRI, 0.31; 95% CI, 0.02–0.61) significantly improved postoperative stroke/mortality risk prediction, largely due to improved discrimination of patients who did not subsequently have an event. Conclusions In AF patients, the three thromboembolic risk scores performed similarly to the RCRI in predicting death within 30 days and the CHADS2 score was the best predictor of postoperative stroke/death regardless of type of surgery.