DOAC Score for bleeding risk in atrial fibrillation

Clinical background

Assessment of bleeding risk is a necessary complement to stroke risk assessment in atrial fibrillation, but has long been the weaker side of the available tools. HAS-BLED, the most widely used bleeding score, was derived in the warfarin era, in which a labile INR is a central variable, which makes it unsuitable for patients treated with direct oral anticoagulants (DOACs). Its discrimination has moreover been limited in several validation studies. The DOAC score was developed specifically to estimate bleeding risk in patients with atrial fibrillation on a DOAC, without dependence on INR measurements, and with variables reflecting the contemporary DOAC-treated population.

The score must not be used to decide whether anticoagulation should be started or stopped. Its role is to identify patients at increased bleeding risk in whom modifiable factors can be addressed, and to support shared decision-making when stroke risk and bleeding risk must be weighed against one another.

Calculating the DOAC score

The score sums ten variables as follows:

$$\text{DOAC score} = P_{\text{age}} + P_{\text{CrCl/eGFR}} + P_{\text{underweight}} + P_{\text{stroke/TIA}} + P_{\text{diabetes}} + P_{\text{hypertension}} + P_{\text{NSAID}} + P_{\text{liver}} + P_{\text{bleeding}} + P_{\text{antiplatelet}}$$

where the point values are:

The maximum score is 23. An age below 65 years scores 0, and a creatinine clearance/eGFR above 60 mL/min scores 0.

The derivation cohort consisted of 5,684 patients with atrial fibrillation treated with dabigatran 150 mg twice daily in the RE-LY trial, a multinational trial across 951 centres in 44 countries, with recruitment between 2005 and 2007 [1]. The primary outcome was major bleeding by the ISTH criteria at one year. Over a median follow-up of 1.74 years, 386 patients (6.8%) had a major bleeding event. The model was internally validated by bootstrapping (200 iterations) and gave an optimism-corrected C-statistic of 0.73 with good calibration (goodness-of-fit P = 0.57). Each additional point was associated with a 48.7% increase in major bleeding (95% CI: 38.9 to 59.3%, P < 0.001) [1].

The model was then refined in the GARFIELD-AF registry (N = 12,296), which included patients on dabigatran, edoxaban, rivaroxaban and apixaban, in order to broaden its application to all DOACs [1].

Interpretation in practice

The score translates into five risk bands based on the one-year incidence of major bleeding in RE-LY [1]:

The score boundaries for the risk bands are not fixed thresholds in the calculator, but rest on the observed bleeding rates in the derivation cohort. In clinical work it is more relevant to regard the score as a spectrum in which a higher score warrants more aggressive modification of risk factors, rather than as a single cut-off triggering a binary decision.

A patient with a high score because of non-modifiable factors (advanced age, reduced creatinine clearance) should not forgo anticoagulation if the stroke risk warrants it. The score points instead to where extra vigilance and possible dose adjustment are needed.

Validation and performance

In RE-LY the DOAC score significantly outperformed HAS-BLED (C-statistic 0.73 versus 0.60, P < 0.001) [1]. In the GARFIELD-AF registry the corresponding figures were 0.71 versus 0.66 (P = 0.025) [1].

External validation was carried out in two cohorts. In COMBINE-AF, a pooled cohort of five randomised trials (N = 25,586, excluding RE-LY), the DOAC score performed better than HAS-BLED (C-statistic 0.67 versus 0.63, P < 0.001) [1]. In RAMQ, an administrative database from Quebec (N = 11,945), the difference was similar (0.65 versus 0.58, P < 0.001) [1]. Discrimination thus fell from 0.73 in the derivation cohort to 0.65 to 0.67 in external cohorts, a typical but appreciable deterioration.

A meta-analysis from 2026 including nine studies with a total of 89,688 patients found a pooled C-statistic of 0.68 for the DOAC score versus 0.63 for HAS-BLED for major bleeding, a statistically significant but moderate difference [2]. Against ORBIT the performance was comparable (0.67 versus 0.66) [2]. For specific outcomes such as intracranial bleeding and gastrointestinal bleeding, no significant difference was seen between the scores [2].

In a Spanish observational study (N = 14,672, of whom 1,484 were on a DOAC), the superiority of the DOAC score was confirmed, with a C-statistic of 0.71 versus 0.64 for HAS-BLED (P = 0.03) and a greater net benefit in decision curve analysis [3]. In a Japanese multicentre study of patients who had undergone TAVR (N = 1,230), the DOAC score was associated with bleeding events (HR 1.22 per point increase), but discrimination was low (AUC 0.59) and not significantly better than HAS-BLED in the DOAC subgroup [4].

In a Chinese cohort (N = 2,532) the DOAC score did not perform better than HAS-BLED (AUC 0.615 versus 0.674), and HAS-BLED had superior calibration [5]. This underlines that the performance of the tool varies between populations and ethnicities.

Calibration has generally been acceptable in the low to intermediate risk groups, but several studies have shown that the DOAC score tends to overestimate bleeding risk in the highest risk bands [2].

Limitations

The DOAC score applies to patients with atrial fibrillation treated with a DOAC. It has not been validated for patients on warfarin, and the variables were chosen on the basis of DOAC-specific determinants. With a mechanical valve prosthesis or significant mitral stenosis a DOAC is not indicated, and the score should not be used in those settings.

The score rests on baseline variables and does not capture dynamic changes in renal function, haemoglobin, medication use or frailty. A patient whose creatinine clearance deteriorates during follow-up will have an underestimated risk if the score is not recalculated. Repeated reassessment is recommended, particularly in older patients, in whom renal function and comorbidity change quickly.

The observed bleeding rates in the risk bands are taken from the RE-LY cohort and are cohort-dependent. In contemporary real-world populations with a lower bleeding incidence the score may overestimate risk, particularly in the high-risk bands [2]. In Asian populations, where the bleeding pattern differs with a higher proportion of intracranial bleeds, the results have been mixed [4, 5].

A common mistake is to use the score as a reason to withhold anticoagulation. The 2024 ESC guidelines emphasise that bleeding risk scores should identify modifiable risk factors and support shared decision-making, not provide grounds for avoiding anticoagulation when the stroke risk warrants it [6].

References

1. Aggarwal R et al. Development and Validation of the DOAC Score: A Novel Bleeding Risk Prediction Tool for Patients With Atrial Fibrillation on Direct-Acting Oral Anticoagulants. Circulation 2023. PMID: 37621213
2. Guo Y et al. DOAC Score Versus HAS-BLED and ORBIT for Predicting Bleeding Events in Atrial Fibrillation on Direct Oral Anticoagulants. Clin Cardiol 2026. PMID: 42041083
3. Abu-Assi E et al. Relative performance evaluation of four bleeding risk scores in atrial fibrillation patients. What does the new DOAC score provide? Int J Cardiol 2024. PMID: 38579940
4. Harano Y et al. Association Between Direct Oral Anticoagulant Score and Bleeding Events in Patients With Atrial Fibrillation Following Transcatheter Aortic Valve Replacement. J Am Heart Assoc 2024. PMID: 39508173
5. Fan M et al. Assessing the applicability of the DOAC, HAS-BLED and ORBIT risk scores in Chinese patients on non-vitamin K antagonist oral anticoagulants. Br J Clin Pharmacol 2025. PMID: 39844428
6. Van Gelder IC et al. 2024 ESC Guidelines for the management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS). Eur Heart J 2024. PMID: 39210723