Comparison of the CHADS2, CHA2DS2 -VASc and HAS-BLED scores for the prediction of clinically relevant bleeding in anticoagulated patients with atrial fibrillation: The AMADEUS trial

Many of the risk factors for stroke in atrial fibrillation (AF) are also important risk factors for bleeding. We tested the hypothesis that the CHADS(2) and CHA(2)DS(2)-VASc scores (used for stroke risk assessment) could be used to predict serious bleeding, and that these scores would compare well against the HAS-BLED score, which is a specific risk score designed for bleeding risk assessment. From the AMADEUS trial, we focused on the trial's primary safety outcome for serious bleeding, which was "any clinically relevant bleeding". The predictive value of HAS-BLED/CHADS(2)/CHA(2)DS(2)-VASc were compared by area under the curve (AUC, a measure of the c-index) and the Net Reclassification Improvement (NRI). Of 2,293 patients on VKA, 251 (11%) experienced at least one episode of "any clinically relevant bleeding" during an average 429 days follow up period. Incidence of "any clinically relevant bleeding" rose with increasing HAS-BLED/CHADS(2)/CHA(2)DS(2)-VASc scores, but was statistically significant only for HAS-BLED (p<0.0001). Only HAS-BLED demonstrated significant discriminatory performance for "any clinically relevant bleeding" (AUC 0.60, p<0.0001). There were significant AUC-differences between HAS-BLED (which had the highest AUC) and both CHADS(2) (p<0.001) and CHA(2)DS(2)VASc (p=0.001). The HAS-BLED score also demonstrated significant NRI for the outcome of "any clinically relevant bleeding" when compared with CHADS(2) (p=0.001) and CHA(2)DS(2)-VASc (p=0.04). In conclusion, the HAS-BLED score demonstrated significant discriminatory performance for "any clinically relevant bleeding" in anticoagulated patients with AF, whilst the CHADS(2) and CHA(2)DS(2)-VASc scores did not. Bleeding risk assessment should be made using a specific bleeding risk score such as HAS-BLED, and the stroke risk scores such as CHADS(2) or CHA(2)DS(2)-VASc scores should not be used.