CRUSADE (30) | 71 277 community-treated patients with NSTEMI | Hematocrit, creatinine clearance, baseline heart rate, baseline systolic blood pressure, female sex, signs of CHF on presentation, previous vascular disease and diabetes mellitus | Each independent variable was assigned weighted integers according to its coefficient value in the regression model. The sum of the weighted integers (range 1 to 100 points) estimates the risk of in-hospital major bleeding, with a curvilinear relation between CRUSADE bleeding score and predicted probabilities of major bleeding | Patients who died within 48 hours were excluded and early bleeding events may be underestimated. Patients on oral anticoagulation were excluded; similarly those with previous bleeding events or bleeding disorders were not included. CRUSADE is designed to predict in-hospital bleeding events |
ACUITY (31) | 17 421 patients with ACS (UA, NSTEMI and STEMI) | Age, female sex, serum creatinine, white blood cell count, anemia, NSTEMI, STEMI and the use of heparin plus glycoprotein IIb/IIIa inhibitor (rather than bivalirudin alone) | Each independent variable was assigned weighted integers according to its coefficient value in the regression model. The sum of the weighted integers (range 1 to 52 points) estimates the risk of 30-day non-CABG major bleeding, with curvilinear relation between ACUITY bleeding score and predicted probabilities of bleeding | Posthoc analysis of patients included in 2 RCTs. Potential variables of interest were not available to be incorporated in the model. Potent P2Y12 inhibitors were not studied |
REACH (32) | 64 589 at risk of CAD or with stable CAD | Age, peripheral arterial disease, CHF, diabetes, hypertension, smoking, antiplatelets, oral anticoagulants, hypercholesterolemia | Each factor was assigned a single point, except for CHF, hypertension, smoking and non-ASA antiplatelet therapy, which were assigned 2 points. Oral anticoagulation or DAPT were assigned 4 points. A score > 10 was associated with 6-fold increase in risk of serious bleeding over 2 years | The definition of serious bleeding used for the analyses was either a hemorrhagic stroke or bleeding leading to both hospitalization and transfusion. This may underestimate the rate of major bleeding events. Data regarding potent P2Y12 inhibitors were limited. The exposure to oral anticoagulation was extrapolated and did not account for potential changes over study follow-up |
DAPT (33)* | 11 648 patients who tolerated DAPT for 1 year without ischemic or bleeding events | Age, cigarette smoking, diabetes mellitus, MI at presentation, previous PCI or previous MI, paclitaxel-eluting stent, stent diameter < 3 mm, CHF or LVEF < 30%, and vein graft stent | Each variable was assigned a single point except for age (65 to < 75 yr and ≥ 75 yr, for which patients were assigned −1 or −2, respectively). Those with CHF, LVEF or vein graft stent were assigned 2 points. Total scores ranged from −2 to 10, and those with scores ≥ 2 were considered high risk and extended DAPT was recommended. Patients with low scores (< 2) were considered low risk and extended DAPT was not recommended | DAPT score showed moderate accuracy in the derivation and validation cohort. It is designed to inform the duration of DAPT rather than predicting future bleeding events |
PARIS (34) | 4190 patients treated with DES (almost 60% had stable presentation) | Age, body mass index, triple therapy at discharge, anemia, current smoking and renal dysfunction | An integer-based risk score was developed for major bleeding (and ischemic events) at 2 years by assigning each variable a score of 2, except for anemia (score of 3) and age (higher score proportional to older patients). The score ranges from 0 to14 and ≥ 8 is considered high bleeding risk | Most patients were treated with clopidogrel, which limits generalizability to potent P2Y12 inhibitors. Duration of DAPT was not randomized and decision to stop antiplatelet was according to the clinician’s discretion |
PRECISE-DAPT (35)* | 14 963 patients treated with DAPT after PCI were pooled from 8 RCTs with independent adjudication of events | Age, creatinine clearance, hemoglobin, white blood cell count and previous spontaneous bleeding. | Independent predictors of bleeding events that were identified in the multivariate regression model were assigned points based on the magnitude of association of each predictor with bleeding. A score ≥ 25 is considered HBR, and extended DAPT has been associated with increased bleeding in this group, unlike patients with low scores. PRECISE-DAPT showed improved integrated discrimination and reclassification performance compared with the PARIS score | The accuracy of the model in the validation cohort ranged from moderate to good. Frailty was not included as part of the risk model. Most patients were treated with clopidogrel and those on oral anticoagulation were excluded. Prediction of bleeding events in patients on prasugrel was poor |
ARC-HBR trade-off model (36), (37) | 6641 patients who underwent PCI and were identified as HBR were pooled from 6 studies | Age, hemoglobin, renal dysfunction, liver disease, cancer, planned major surgery, COPD, current smoker, complex PCI procedure, oral anticoagulation at discharge | Variables were classified as a major or minor criterion for HBR. Major criterion is considered to confer risk of major bleeding of ≥ 4% or ≥ 1% intracranial hemorrhage at 1 year. This score outperformed PARIS and PRECISE-DAPT (alternative model without white blood cell count) | The criteria used were modified from those proposed in the ARC consensus to allow the use of available data. Infrequent, but wellrecognized, predictors were pooled as a single variable in the model (i.e., liver disease, cancer and planned surgery) |
PRAISE (38)* | 19 826 patients presenting with ACS from 2 registries | Age, sex, diabetes, hypertension, hyperlipidemia, PAD, eGFR, previous MI, previous PCI, previous CABG, previous stroke, previous bleeding, malignancy, STEMI, LVEF, multivessel disease, complete revascularization, vascular access, DES and treatment with BBs, ACE inhibitors, ARBs, statins, oral anticoagulation and PPIs | Four machine learning models that used different classifiers were developed to predict occurrence of all-cause death, recurrent MI and major bleeding 1 year after discharge. Each model’s performance was assessed using a range of learning metrics and the best performing model was selected. When calculated, PRAISE provides 3 outcomes of the calculated score for death, MI, and major bleeding. | It was not possible to fully compare PRAISE with PARIS or PRECISE-DAPT given the insufficient clinical data. Although PRAISE has been prospectively validated in external cohorts, it has not been used in RCTs to aid decision-making of DAPT duration |