How well does physician risk assessment predict stroke and bleeding in atrial fibrillation? Results from the Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT-AF)

Abstract

Assessments of stroke and bleeding risks are essential to selecting oral anticoagulation in patients with atrial fibrillation (AF). We aimed to assess outcomes according to physician assessed risk, with comparison to empirical risk scores. This was a prospective, observational study of 9,715 outpatients with AF enrolled in ORBIT-AF, a US national registry. Stroke and bleeding risks were quantified by physician assignment, CHADS2 and CHA2DS2-VASc stroke scores, and ATRIA and HAS-BLED bleeding scores. Outcomes were stroke or systemic embolism and major bleeding during a median follow-up of 28 months. Physician-assigned risk was associated with thromboembolic events: low risk (0.71 per 100 patient-years [95% CI 0.56-0.91], n=3,991), intermediate risk (0.98 [95% CI 0.79-1.20], n=4,148), and high risk (1.84 [95% CI 1.43-2.37], n=1,576, P<.0001), and major bleeding: low (3.43 [95% CI 3.07-3.82], n=4,250), intermediate (4.55 [95% CI 4.03-5.15], n=2,702), and high (5.76 [95% CI 4.42-7.50], n=468; P<.0001). Discrimination of stroke risk was similar with CHADS2 (c=0.59, 95% CI 0.57-0.61) vs physician assessment (c=0.58, 95% CI 0.55-0.62). Among patients on oral anticoagulation, bleeding risk discrimination was higher with ATRIA (c=0.63, 95% CI 0.61-0.65) and HAS-BLED (c=0.60, 95% CI 0.59-0.62) than with physician assessment (0.55, 95% CI 0.53-0.57). Physician-assessed risk categories did not add significantly to empirical risk scores, in Cox models for outcomes (Padjusted>.05 for all physician assessments vs Padjusted<.05 for empirical scores). Physician-assigned risk showed a graded relationship with outcomes, and both physician-based and empirical scores yielded only moderate discrimination. Although empirical scores provided valuable risk stratification information (with or without physician judgment), physician assessment added little to existing scores. These data support the use of empirical scores for stroke and bleeding risk stratification, and the need for novel approaches to risk stratification in this population.