Predicting New-Onset Post-Coronary Artery Bypass Graft Atrial Fibrillation With Existing Risk Scores.

Abstract

New-onset atrial fibrillation (AF) after coronary artery bypass graft (CABG) operation is associated with poorer survival. Blanket prophylaxis efforts have not appreciably decreased incidence, making targeted prevention for high-risk patients desirable. We compared predictive abilities of existing scores developed/used to predict adverse CABG outcomes (Society of Thoracic Surgeons' [STS] risk of mortality) or AF not associated with cardiac operation (the Cohorts for Heart and Aging Research in Genomic Epidemiology [CHARGE]-AF score, the CHA₂DS₂-VASc score), and a risk model for predicting postoperative AF following cardiac operations (POAF score), with age (the most consistently identified post-CABG AF risk factor). Data submitted to the STS Adult Cardiac Surgery Database were used to assess new-onset AF in 8,976 consecutive patients without preoperative AF undergoing isolated CABG from 2004 to 2010 at five participating centers. Five logistic regression models (for CHA₂DS₂-VASc score, CHARGE-AF score, POAF score, STS risk score, and age, respectively, all modeled with restricted cubic splines) with a random effect for site were fitted to predict post-CABG AF. Estimates were used to compute and compare receiver operating characteristic (ROC) areas. New-onset AF occurred in 2,141 patients (23.9%). The ROC area was greatest for CHARGE-AF (0.68, 95% confidence interval [CI]: 0.67-0.69), followed by age (0.66, 95% CI: 0.65-0.68), POAF score (0.65, 95% CI: 0.64-0.66), CHA₂DS₂-VASc (0.59, 95% CI: 0.58 to 0.60), and STS risk of mortality (0.58, 95% CI: 0.56-0.59). CHARGE-AF was significantly more predictive than age (p < 0.0001); the other scores were significantly less predictive (p < 0.0001). Only CHARGE-AF performed better than age alone. Its performance was moderate and comparable with published risk models specifically targeted at new-onset post-isolated CABG AF. Future research should continue to focus on developing better predictive models.