Clinical background
The decision on anticoagulation in atrial fibrillation is a trade-off between the benefit of stroke prevention and the risk of bleeding. HEMORR₂HAGES was developed to quantify that bleeding risk in older patients with atrial fibrillation taking warfarin, and thereby to complement stroke risk scores such as CHA₂DS₂-VASc. The purpose is not to decide whether anticoagulation should be started, but to identify patients in whom the bleeding risk is high enough to warrant intensified preventive measures, dose adjustment or closer follow-up [1]. The tool dates from the warfarin era and has in later literature largely been displaced by the simpler and equally well validated HAS-BLED, but HEMORR₂HAGES is still used and has in some populations shown comparable or slightly better discrimination [2,3].
Calculating HEMORR₂HAGES
The score is the sum of eleven factors:
where each variable means:
| Variable | Meaning | Points |
|---|---|---|
| Hepatic or renal disease (chronic liver disease or dialysis-dependent/severe renal impairment) | 1 | |
| Ethanol (alcohol) misuse | 1 | |
| Malignancy (previous or newly diagnosed) | 1 | |
| Older (age >75 years) | 1 | |
| Reduced platelet count or function (e.g. thrombocytopenia or concomitant antiplatelet treatment) | 1 | |
| Rebleeding risk (a history of major or clinically relevant bleeding) | 2 | |
| Uncontrolled hypertension | 1 | |
| Anaemia | 1 | |
| Genetic factors linked to bleeding (e.g. CYP2C9 polymorphism; include only if the genotype is known) | 1 | |
| Excessive fall risk (e.g. gait disturbance, neurological disease, marked frailty) | 1 | |
| Previous stroke | 1 |
The maximum score is 12 (ten factors give 1 point each and prior bleeding gives 2). The derivation cohort consisted of 3,791 Medicare beneficiaries with atrial fibrillation from seven American states, collected through quality registry organisations. The outcome was hospital admission with an ICD-9 code for bleeding, and all patients were prescribed warfarin [1]. The bleeding rate per 100 patient-years rose with increasing score: 1.9 at 0 points, 2.5 at 1, 5.3 at 2, 8.4 at 3, 10.4 at 4 and 12.3 at 5 or more points [1].
Interpretation in practice
HEMORR₂HAGES has no universally accepted threshold for low, intermediate or high risk, but in the original cohort the bleeding rate approximately doubled from 1 to 2 points (2.5 to 5.3 per 100 patient-years) and more than tripled from 0 to 3 points (1.9 to 8.4) [1]. In external validation studies the following bands have been used:
| Score | Risk category | Clinical action |
|---|---|---|
| 0–1 | Low | Continue anticoagulation as indicated; no further risk-modifying measure required |
| 2–3 | Intermediate | Identify and correct modifiable factors: blood pressure control, falls prevention, avoid concomitant antiplatelet treatment unless indicated, review renal and liver function |
| ≥4 | High | Careful consideration of the net benefit of anticoagulation; intensified INR monitoring (with warfarin) or dose adjustment; consideration of alternative stroke prevention such as percutaneous left atrial appendage occlusion if the bleeding risk is judged overwhelming |
A higher score should lead to more thorough measures to reduce the bleeding risk rather than to automatic withholding of the anticoagulant [1].
Validation and performance
In the derivation cohort, HEMORR₂HAGES achieved a c-statistic of 0.67 for predicting bleeding in patients prescribed warfarin, significantly better than other bleeding scores available at the time (P < 0.001) [1].
In a network meta-analysis of 18 studies with a total of 321,888 patients, HEMORR₂HAGES was judged to have balanced sensitivity and specificity, comparable to HAS-BLED [4]. A systematic review and meta-analysis comparing HAS-BLED with HEMORR₂HAGES, ATRIA and ORBIT found, however, that HEMORR₂HAGES placed a larger proportion of patients who actually bled in the low- and intermediate-risk categories than HAS-BLED did. In the HAS-BLED low-risk group, the bleeding rate was 87% lower than in the corresponding HEMORR₂HAGES group, and in the intermediate group 39% lower [2]. This indicates that HAS-BLED calibrates better in its risk categorisation, even though both tools have similar discrimination.
In an external validation in 1,080 patients undergoing PCI, of whom 104 were on oral anticoagulants, HEMORR₂HAGES showed a c-statistic of 0.767 for major bleeding (BARC type 3–5) at two years of follow-up, compared with 0.727 for HAS-BLED (P = 0.07 for the difference, not significant) and 0.716 for PARIS [3]. In the subgroup on anticoagulants the c-statistic was 0.724 for HEMORR₂HAGES [3]. This was a population that did not primarily have atrial fibrillation, and the results support the view that HEMORR₂HAGES may have broader applicability but not superior performance compared with HAS-BLED.
Limitations
HEMORR₂HAGES has several important limitations:
Derived in the warfarin era. All the patients in the derivation cohort were treated with warfarin [1]. The score is not validated for patients on direct oral anticoagulants (DOACs), and the bleeding risk with DOACs is generally lower and the pattern of risk factors partly different. No external validation specifically for DOACs has been identified.
The genetic factor. CYP2C9 polymorphism is included as a variable, but genotyping is not routinely available in clinical practice [1]. In practice this factor is usually omitted, which reduces the maximum score to 11 and makes the tool easier to use but at the same time underuses one of its original strengths.
Competing tools. HAS-BLED, which is simpler (seven variables) and does not require genotyping, is recommended in European and American guidelines as the first choice for bleeding risk assessment in atrial fibrillation. HEMORR₂HAGES has fewer variables in common with routinely recorded data and requires more clinical assessment, which reduces its practical usefulness [2,5].
Insufficient discrimination. A c-statistic of 0.67 in the derivation cohort means moderate, not strong, discrimination [1]. Even in external validations the c-statistic rarely exceeds 0.75 [3]. None of the available bleeding scores reaches the discrimination that the stroke scores achieve, and clinical judgement remains central.
Not for a mechanical valve or mitral stenosis. With a mechanical valve prosthesis or significant mitral stenosis, anticoagulation is mandatory irrespective of the bleeding score, and HEMORR₂HAGES must not be used to justify withholding it.
References
- Gage BF et al. Clinical classification schemes for predicting hemorrhage: results from the National Registry of Atrial Fibrillation (NRAF). Am Heart J 2006. PMID: 16504638
- Zeng J et al. Comparison of HAS-BLED with other risk models for predicting the bleeding risk in anticoagulated patients with atrial fibrillation: A PRISMA-compliant article. Medicine 2020. PMID: 32569222
- Doomun I et al. Predictive Value of HAS-BLED and HEMORR2HAGES Bleeding Risk Scores After Percutaneous Coronary Intervention. Tex Heart Inst J 2024. PMID: 38982874
- Chang G et al. Accuracy of HAS-BLED and other bleeding risk assessment tools in predicting major bleeding events in atrial fibrillation: A network meta-analysis. J Thromb Haemost 2020. PMID: 31782613
- Wang Y, Bajorek B. Safe use of antithrombotics for stroke prevention in atrial fibrillation: consideration of risk assessment tools to support decision-making. Ther Adv Drug Saf 2014. PMID: 25083260
