The Hosmer-Lemeshow test for calibration was also performed by Lip et al2 in conjunction with all C statistics, and none of the P values was ≤ .05 for any of the risk scores (ie, lack of goodness of fit was not indicated). For HAS-BLED in particular, the P values were .24 for all patients and .13 for the warfarin patient cohort. Furthermore, using multivariate Cox regression models, Lip et al2 tested whether the HAS-BLED score added significantly to models already incorporating the four older scores, one at a time. In all four instances, HAS-BLED was associated with predictive improvement when inserted into models already incorporating the older scores. In contrast, none of the other four older scores added significantly when inserted one at a time into a model already including HAS-BLED. Thus, we hope we have clarified the interpretation of the predictive accuracy of the HAS-BLED model, as well as its comparison with the other bleeding risk models, in the second validation study, which had a much larger sample size and used other statistical methods beyond the C statistic,2 as suggested by Dr Corbanese.