Abstract
Frequently used models, such as the HAS-BLED, ATRIA, ORBIT, and GARFIELD-AF evaluate the risk of bleeding when using an anticoagulant, for example warfarin, in patients with non-valvular atrial fibrillation. Limited studies are available reporting a model with a good discriminative ability to predict the bleeding risk score when using direct oral anticoagulants.
Patient data were collected from King Abdulaziz Medical City, King Fahad Cardiac Center, and Prince Sultan Cardiac Center in Riyadh, from outpatients, inpatients, or primary care clinics. In total, 1722 patients with a prescription for a new oral anticoagulant, Dabigatran, Rivaroxaban, or Apixaban, were enrolled. A resampling approach for variable selection was used and a five-fold cross-validation to assess the model fit and misclassification probabilities. The analysis used the receiver operating characteristics curve (ROC) and the concordance (c) statistic to assess the validation models' discriminative power. The final penalized likelihood parameters were used for the development of the risk prediction tool. The accuracy of a classification and the prediction are reported with the sensitivity, specificity, and Brier score.
Bleeding occurred in 11.15% of cases, of which 23.08% required a blood transfusion and 51.65% had a reduction in haemoglobin of more than 2 gm. The variable selection model identified 15 predictors associated with major bleeding. The discriminative ability of the model was good (c-statistic 0.75, p = 0.035). The Brier score of the model was 0.095. With a fixed cut-off probability value of 0.12 for the logistic regression equation, the sensitivity was 72.7%, and the specificity 66.3%.
This model demonstrated a good performance in predicting the bleeding risk in Arab patients treated with novel oral anticoagulants. This easy to use bleeding risk score will allow the clinician to quickly classify patients according to their risk category, supporting close monitoring and follow-up for high-risk patients, without laboratory and radiological monitoring.