Predictive ability of bleeding risk scores in the routine clinical practice

Previous predictive models of bleeding in acute coronary syndromes (ACSs) used different definitions of bleeding and some of them come from populations lacking important predictors of haemorrhagic complications. Our group previously developed a predictive model of bleeding (PMB), including clinically meaningful variables, providing an optimal predictive ability. We aimed to compare the ability of this PMB with the main available bleeding risk scores for predicting major bleeding according to different definitions in non-selected ACS patients from daily clinical practice.

All ACS patients admitted to the Coronary Care Unit were prospectively included. CRUSADE, Mehran and ACTION bleeding risk scores were calculated for each patient. In-hospital bleeding was recorded using the CRUSADE, TIMI, Mehran, ACTION and BARC definitions. For reasons of clinical relevance, BARC 3 and 5 categories were considered severe BARC bleeding for this study. The predictive ability of the PMB and other bleeding risk scores was assessed by binary logistic regression, ROC curves and areas under the curves (AUCs).

We included 1976 patients. Mean age was 62.1 years. Almost all patients underwent angiography, 65% of them by the radial approach. The incidence of major bleeding was: CRUSADE bleeding 3.9% (77/1976); Mehran bleeding 4.8% (94/1976); ACTION bleeding 3.9% (78/1976); and BARC 3/5 bleeding 2.4% (48/1976). The PMB showed the best ability for predicting major bleeding regardless of the definition used. The differences were specially significant for predicting BARC 3/5 bleeding (AUC: PMB 0.87, Mehran score 0.68, CRUSADE score 0.70 and ACTION score 0.70). The predictive ability of CRUSADE, ACTION and Mehran scores was similar for all the definitions analysed.

Current bleeding risk scores showed a similar predictive ability for major bleeding regardless of the definitions used. Including other clinically meaningful predictors of bleeding into the new PMB significantly improved its predictive ability in the clinical scenario of ACS.