Prognostic Value of BARC Bleeding Complications in STEMI
The data analyzed in this study were obtained from STEMI patients accepted for PPCI at the Academic Medical Center, University of Amsterdam, between January 1, 2003, and July 31, 2008. The study complied with the principles of the Declaration of Helsinki, and the local ethics committee approved the study protocol. In general, patients qualified for PPCI if they had typical ischemic chest pain and at least a 1-mm ST-segment elevation in 2 or more contiguous leads, a new left bundle branch block, or a true posterior myocardial infarction. Patients received a standard 300- to 600-mg loading dose of clopidogrel. If a coronary stent was implanted, clopidogrel was prescribed for at least 1 month to patients with a bare metal stent and for 6 to 12 months to patients with a drug-eluting stent. Patients were routinely pretreated with 300 mg of aspirin and 5,000 IU of unfractionated heparin. An additional heparin bolus was administered at the catheterization laboratory if necessary to achieve a targeted activated clotting time of 300 s followed by an infusion of 12 U/kg/h with titration to achieve a target activated partial thromboplastin time (aPTT) of 1.5 to 2.0 times the control. Glycoprotein IIb/IIIa inhibitors were used at the discretion of the operator.
Procedural and angiographic data were prospectively collected and entered into a dedicated database by interventional cardiologists and specialized nurses. Chart review for consecutive STEMI patients with available aPTT measurements was performed in the context of a study designed to investigate the relationship between aPTT and clinical outcome in STEMI patients treated with PPCI. A detailed description of the study protocol has been published previously. Source documentation (laboratory results, discharge letters, case history, and nurse reports) was collected for every hospital admission for each patient in both the PCI center and in referring hospitals and was assessed for the occurrence of clinical events, including hemorrhage. Follow-up information regarding vital status was obtained from computerized, long-term mortality records from the Dutch National Death Index between January 1, 2012, and April 30, 2012.
The study cohort consisted of all STEMI patients included in our study database who were alive at the end of the procedure. The primary outcome of this study was 1-year all-cause mortality. All hospitalizations were reviewed for the presence of bleeding by 1 author (W.J.K.) who had full access to the patient's clinical and laboratory records. Complicated cases were discussed and adjudicated with 2 other authors (J.G.P.T. and J.P.S.H.). For each bleeding event, the following items were recorded in the study database: the date and source of the bleeding, the hemoglobin decrease associated with the bleeding event (adjusted for the amount of transfusions), the amount of blood transfusions attributable to the bleeding event, the discontinuation of antithrombotic therapy associated with bleeding (as well as the use of aspirin, thienopyridine, vitamin K antagonists, heparin, and glycoprotein IIb/IIIa inhibitors before and after the bleeding was recorded), use of diagnostic procedures (including imaging techniques), surgery to control bleeding, the use of vasoactive agents for bleeding, and other medical and nonmedical interventions. In addition, for all patients who underwent coronary artery bypass graft (CABG), we recorded the need for reoperation because of bleeding after closure of the sternum, the amount of blood transfusions within a 48-h window post-CABG, and the chest tube output in the first 24 h after CABG.
Bleeding events occurring during the initial hospitalization were assessed and classified according to TIMI (Thrombolysis In Myocardial Infarction), GUSTO (Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries), ISTH (International Society on Thrombosis and Haemostasis), and BARC criteria. A detailed description of these bleeding classifications is given in Online Table 1 . If a patient experienced multiple in-hospital bleeding events, the most severe was classified rather than the first event. The occurrence of BARC type 1 bleeding was not considered in the present paper.
For each bleeding classification, 1-year mortality rates for the various bleeding categories were estimated using Kaplan-Meier analyses. A landmark was set at 30 days. BARC type 5 bleeding (fatal bleeding) was not assessed as a separate bleeding category because it is not possible to associate fatality itself with mortality. To ensure a fair comparison with the other bleeding classifications, patients with fatal bleedings were excluded from the analyses. We calculated hazard ratios for 1-year mortality for the different bleeding categories of the TIMI, ISTH, GUSTO, and BARC bleeding classifications by using univariable and multivariable Cox regression analyses adjusting for baseline predictors of mortality. To identify baseline variables with a significant unadjusted association with 1-year mortality, univariable Cox regression analyses were first performed for the variables listed in Table 1 and Table 2 (with the exception of in-hospital hemoglobin decrease and procedural data specific to CABG). To identify independent baseline predictors of 1-year mortality, we next performed a stepwise backward selection Cox regression analysis including variables with a significant association according to univariable analysis (p < 0.10). The entry criterion was set at p < 0.05, and the exit criterion was set at p = 0.10.
The increased discriminative value of the models after addition of in-hospital bleeding (according to the different bleeding definitions) to the baseline predictors of 1-year mortality was estimated by using 3 measures: change in Harrell's c-index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Bootstrapping (400 runs) was used to compare these measures among the different multivariable models. To investigate the prognostic value of the individual data elements of the different bleeding classifications for 1-year mortality, we developed additional Cox models. Details regarding these models are described in the Online Appendix.
Data were complete for mortality and for 19 of 40 variables. Missing patient-level covariates were assumed to be missing at random and were imputed with the use of multiple imputations. The imputation procedure and subsequent Cox proportional hazards regression estimation, estimation of Harrell's c-index, NRI, and IDI were performed according to Rubin's protocol. Analyses were performed with SPSS version 19.0 (IBM SPSS Statistics, IBM Corporation, Armonk, New York) and R (R Foundation for Statistical Computing, Vienna, Austria).
Methods
Source Population and Procedures
The data analyzed in this study were obtained from STEMI patients accepted for PPCI at the Academic Medical Center, University of Amsterdam, between January 1, 2003, and July 31, 2008. The study complied with the principles of the Declaration of Helsinki, and the local ethics committee approved the study protocol. In general, patients qualified for PPCI if they had typical ischemic chest pain and at least a 1-mm ST-segment elevation in 2 or more contiguous leads, a new left bundle branch block, or a true posterior myocardial infarction. Patients received a standard 300- to 600-mg loading dose of clopidogrel. If a coronary stent was implanted, clopidogrel was prescribed for at least 1 month to patients with a bare metal stent and for 6 to 12 months to patients with a drug-eluting stent. Patients were routinely pretreated with 300 mg of aspirin and 5,000 IU of unfractionated heparin. An additional heparin bolus was administered at the catheterization laboratory if necessary to achieve a targeted activated clotting time of 300 s followed by an infusion of 12 U/kg/h with titration to achieve a target activated partial thromboplastin time (aPTT) of 1.5 to 2.0 times the control. Glycoprotein IIb/IIIa inhibitors were used at the discretion of the operator.
Procedural and angiographic data were prospectively collected and entered into a dedicated database by interventional cardiologists and specialized nurses. Chart review for consecutive STEMI patients with available aPTT measurements was performed in the context of a study designed to investigate the relationship between aPTT and clinical outcome in STEMI patients treated with PPCI. A detailed description of the study protocol has been published previously. Source documentation (laboratory results, discharge letters, case history, and nurse reports) was collected for every hospital admission for each patient in both the PCI center and in referring hospitals and was assessed for the occurrence of clinical events, including hemorrhage. Follow-up information regarding vital status was obtained from computerized, long-term mortality records from the Dutch National Death Index between January 1, 2012, and April 30, 2012.
Study Design, Bleeding Definitions, and Adjudication
The study cohort consisted of all STEMI patients included in our study database who were alive at the end of the procedure. The primary outcome of this study was 1-year all-cause mortality. All hospitalizations were reviewed for the presence of bleeding by 1 author (W.J.K.) who had full access to the patient's clinical and laboratory records. Complicated cases were discussed and adjudicated with 2 other authors (J.G.P.T. and J.P.S.H.). For each bleeding event, the following items were recorded in the study database: the date and source of the bleeding, the hemoglobin decrease associated with the bleeding event (adjusted for the amount of transfusions), the amount of blood transfusions attributable to the bleeding event, the discontinuation of antithrombotic therapy associated with bleeding (as well as the use of aspirin, thienopyridine, vitamin K antagonists, heparin, and glycoprotein IIb/IIIa inhibitors before and after the bleeding was recorded), use of diagnostic procedures (including imaging techniques), surgery to control bleeding, the use of vasoactive agents for bleeding, and other medical and nonmedical interventions. In addition, for all patients who underwent coronary artery bypass graft (CABG), we recorded the need for reoperation because of bleeding after closure of the sternum, the amount of blood transfusions within a 48-h window post-CABG, and the chest tube output in the first 24 h after CABG.
Bleeding events occurring during the initial hospitalization were assessed and classified according to TIMI (Thrombolysis In Myocardial Infarction), GUSTO (Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries), ISTH (International Society on Thrombosis and Haemostasis), and BARC criteria. A detailed description of these bleeding classifications is given in Online Table 1 . If a patient experienced multiple in-hospital bleeding events, the most severe was classified rather than the first event. The occurrence of BARC type 1 bleeding was not considered in the present paper.
Statistical Analysis
For each bleeding classification, 1-year mortality rates for the various bleeding categories were estimated using Kaplan-Meier analyses. A landmark was set at 30 days. BARC type 5 bleeding (fatal bleeding) was not assessed as a separate bleeding category because it is not possible to associate fatality itself with mortality. To ensure a fair comparison with the other bleeding classifications, patients with fatal bleedings were excluded from the analyses. We calculated hazard ratios for 1-year mortality for the different bleeding categories of the TIMI, ISTH, GUSTO, and BARC bleeding classifications by using univariable and multivariable Cox regression analyses adjusting for baseline predictors of mortality. To identify baseline variables with a significant unadjusted association with 1-year mortality, univariable Cox regression analyses were first performed for the variables listed in Table 1 and Table 2 (with the exception of in-hospital hemoglobin decrease and procedural data specific to CABG). To identify independent baseline predictors of 1-year mortality, we next performed a stepwise backward selection Cox regression analysis including variables with a significant association according to univariable analysis (p < 0.10). The entry criterion was set at p < 0.05, and the exit criterion was set at p = 0.10.
The increased discriminative value of the models after addition of in-hospital bleeding (according to the different bleeding definitions) to the baseline predictors of 1-year mortality was estimated by using 3 measures: change in Harrell's c-index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Bootstrapping (400 runs) was used to compare these measures among the different multivariable models. To investigate the prognostic value of the individual data elements of the different bleeding classifications for 1-year mortality, we developed additional Cox models. Details regarding these models are described in the Online Appendix.
Data were complete for mortality and for 19 of 40 variables. Missing patient-level covariates were assumed to be missing at random and were imputed with the use of multiple imputations. The imputation procedure and subsequent Cox proportional hazards regression estimation, estimation of Harrell's c-index, NRI, and IDI were performed according to Rubin's protocol. Analyses were performed with SPSS version 19.0 (IBM SPSS Statistics, IBM Corporation, Armonk, New York) and R (R Foundation for Statistical Computing, Vienna, Austria).