External Validation of HASBLED and ORBIT bleeding risk scores in Wales-AF population

Abstract

Abstract Background Anticoagulant therapy (AC) is recommended for Atrial Fibrillation (AF) patients when formal assessment of thromboembolic and bleeding risk suggests a net benefit of AC treatment. Recent UK National Institute of Health and Care Excellence guidelines consider that the ORBIT bleeding risk score(1) provides a more accurate assessment of bleeding risk with AC than the HASBLED score(2) and is recommended for routine clinical use in the UK. Purpose To compare the performance of HASBLED and ORBIT scores to predict hospitalisations for bleeding (HB) in a large population of UK patients with AF receiving AC. Methods We conducted a retrospective longitudinal analysis using linked primary and secondary care health records in the All-Wales SAIL databank. Patients with a diagnosis of AF treated with AC were identified between 2012 and 2018. Both HASBLED and ORBIT scores were calculated for each patient annually based on identified comorbidities, demographics and prescription data. All HB were evaluated. Logistic regression models were used to compare sensitivity and specificity of each score for HB prediction. The Area Under Curve for the Receiver operating Characteristic plot (AUC) was generated for each score to illustrate the risk discrimination ability of each prediction schemes. Results A total of 107,137 (45% female, mean age=74) AF patients were evaluated over the study period. The number of anticoagulated AF patients increased from 27,959 in 2012 (49.3% of cohort) to 48,595 in 2018 (66.8%), providing a total of 265,410 patient years of AC therapy for analysis. There were 710 HB (2.5% of AC patients) in 2012 increasing to 1,146 (2.4%) in 2018. The predictive power of HASBLED and ORBIT increased slightly over the period of study: with observed HASBLED AUC of 60.8 and ORBIT AUC of 64.8 in 2018 (Figure 1). Over the period of study, the observed HB rates for AC AF patients with HAS BLED scores of 0-3 and ORBIT scores of 0-5 were similar to those observed in the original studies, but observed HB for patients with higher HASBLED and ORBIT scores were less consistent (Fig. 2) Conclusion Our findings demonstrated that HASBLED and ORBIT were relatively limited in their predictive performance for HB in a large, real-world AC AF population, with ORBIT providing more accurate prediction across the overall risk range. This highlights the need to develop and validate new bleeding risk scores from the wide range of clinical and demographic factors in AF patients to improve effectiveness of risk communication and AC prescribing in AF.Figure 1Figure 2