Validation of the Risk Estimator Decision Aid for Atrial Fibrillation (RED-AF) for predicting 30-day adverse events in emergency department patients with atrial fibrillation

Abstract

Study objective: In the United States, nearly 70% of emergency department (ED) visits for atrial fibrillation result in hospitalization. The incidence of serious 30-day adverse events after an ED evaluation for atrial fibrillation remains low. This study's goal was to prospectively validate our previously reported Risk Estimator Decision Aid for Atrial Fibrillation (RED-AF) model for estimating a patient's risk of experiencing a 30-day adverse event.

Methods: This was a prospective cohort study, which enrolled a convenience sample of ED patients presenting with atrial fibrillation. RED-AF, previously derived from a retrospective cohort of 832 patients, assigns points according to age, sex, coexisting disease (eg, heart failure, hypertension, chronic obstructive pulmonary disease), smoking, home medications (eg, β-blocker, diuretic), physical examination findings (eg, dyspnea, palpitations, peripheral edema), and adequacy of ED ventricular rate control. Primary outcome was occurrence of greater than or equal to 1 atrial fibrillation-related adverse outcome (ED visits, rehospitalization, cardiovascular complications, death) within 30 days. We identified a clinically relevant threshold and measured RED-AF's performance in this prospective cohort, assessing its calibration, discrimination, and diagnostic accuracy.

Results: The study enrolled 497 patients between June 2010 and February 2013. Of these, 120 (24%) had greater than or equal to 1 adverse event within 30 days. A RED-AF score of 87 was identified as an optimal threshold, resulting in sensitivity and specificity of 96% (95% confidence interval [CI] 91% to 98%) and 19% (95% CI 15% to 23%), respectively. Positive and negative predictive values were 27% (95% CI 23% to 32%) and 93% (95% CI 85% to 97%), respectively. The c statistic for RED-AF was 0.65 (95% CI 0.59 to 0.71).

Conclusion: In this separate validation cohort, RED-AF performed moderately well and similar to the original derivation cohort for identifying the risk of short-term atrial fibrillation-related adverse events in ED patients receiving a diagnosis of atrial fibrillation.

Conflict of interest statement

Conflicts of Interest/Disclosures: There are no conflicts of interest in connection with this submission or are there any copyright constraints. No industry financial support or compensation has been or will be received for conducting this study. Dr. Barrett serves as a scientific consultant for Red Bull GmbH, Fuschl am See, Salzburg and Boehringer Ingelheim Pharmaceuticals, Inc. Ridgefield, Connecticut. Dr. Self is a paid member of the scientific advisory board for BioFire Diagnostics, Inc.

Copyright © 2014 American College of Emergency Physicians. Published by Elsevier Inc. All rights reserved.

Figures

Figure 1

RED-AF model based on data from the original derivation cohort. Points are assigned for each of the 12 predictors. The total points correspond to an absolute predicted risk for 30-day adverse events.[8]

Figure 2

Calibration plot for the RED-AF clinical prediction model in the validation cohort. This plot illustrates the calibration accuracy of the original model (“Apparent”) and the bootstrap model (“Bias-corrected”) for 30-day adverse events with locally weighted scatterplot smoothing used to model the relationship between actual and predicted probabilities. As can be seen, the model’s calibration function estimate is moderately nonlinear, with the corrected calibration showing good agreement with the apparent calibration.

Figure 3

Receiver Operating Characteristic (ROC) Curves for RED-AF. The Figure presents the ROC curves for RED-AF in both the original derivation cohort (n=832) and the validation cohort (n=497).

Appendix Figure 1