ATRIA Bleeding Risk Score

Clinical background

The decision on anticoagulation in atrial fibrillation requires the benefit of stroke prevention to be weighed against the risk of bleeding. In patients at high embolic risk the benefit usually predominates, but in the intermediate categories, where the advantage of stroke prevention is less self-evident, the bleeding risk becomes decisive for whether treatment is started. The ATRIA bleeding risk score was developed precisely to quantify this risk using variables that are readily available in clinical practice, and thereby to provide a better basis for decision than unsystematic clinical judgement.

The score is above all a tool for identifying patients at increased bleeding risk, in whom extra caution and risk-reducing measures are warranted. It must not be used as the sole reason to withhold anticoagulation in patients at high stroke risk.

Calculating the ATRIA bleeding risk score

The score rests on five clinical variables, weighted according to the size of the regression coefficients in the derivation model:

$$\text{Score} = 3 \cdot I_{\text{anaemia}} + 3 \cdot I_{\text{severe renal disease}} + 2 \cdot I_{\text{age} \geq 75} + 1 \cdot I_{\text{previous bleeding}} + 1 \cdot I_{\text{hypertension}}$$

where $I$ is an indicator variable taking the value 1 if the risk factor is present and 0 if it is absent. Anaemia is defined as a haemoglobin <13 g/dL in men and <12 g/dL in women. Severe renal disease is defined as an eGFR <30 mL/min or dialysis dependence. Previous bleeding means any history of bleeding, whatever the site. The maximum score is 10.

The derivation cohort consisted of 9,186 adult patients with non-valvular atrial fibrillation in Kaiser Permanente Northern California, identified between 1996 and 1997 and followed to 2003 [1]. In total the cohort contributed 32,888 person-years on warfarin. During follow-up, 461 validated incident major bleeds were observed, corresponding to 1.4% per year. The model was developed with Cox regression using time-dependent covariates, and the variables were selected by bootstrap methodology over 1,000 samples. Internal validation was carried out by split-sample testing (a 2:1 split between derivation and validation cohorts).

Interpretation in practice

The score is categorised into three risk bands according to the observed bleeding rate in the derivation cohort:

In the derivation cohort the bleeding events were heavily concentrated in the high-risk group: 42% of all bleeds occurred in only 10.2% of person-years, while the low-risk group accounted for 83% of the follow-up time [1].

An important point of application is that the score must not be read as a binary decision aid. A high score does not mean that anticoagulation is contraindicated, but that risk-reducing measures should be taken and that the patient should be informed of the increased risk. At an intermediate score in particular, clinical assessment of the overall stroke risk becomes decisive.

Validation and performance

In the internal split-sample validation, the continuous score showed a c-index of 0.74 (95% CI 0.70–0.74) and the three-level categorical version 0.69 [1]. Calibration was acceptable on goodness-of-fit testing (p = 0.29). ATRIA was compared with six previously published bleeding risk scores and showed the highest c-index together with a net reclassification improvement of 27.7% to 56.6% against all of them.

External validation has produced more mixed results. In a meta-analysis of nine studies comparing HAS-BLED with ATRIA, ORBIT and HEMORR₂HAGES, ATRIA placed more bleeding events in its low-risk category than HAS-BLED did [2]. In the ATRIA low-risk group, 1.82% bleeding events were observed versus 0.50% for HAS-BLED, suggesting that ATRIA is less successful at identifying patients who genuinely do not bleed. In the intermediate- and high-risk categories the differences between the scores were less marked.

In the RE-LY trial, which comprised 18,113 patients randomised to dabigatran or warfarin, ATRIA, like the other bleeding scores, showed only moderate discrimination [3]. ORBIT performed best there. An American study of 39,539 patients starting a direct oral anticoagulant found that ATRIA, HAS-BLED and ORBIT all performed similarly and modestly, with no statistically significant differences in net reclassification improvement [4]. In a French primary care cohort (the CACAO study, 3,082 patients) all 13 bleeding scores tested, including ATRIA, showed only poor to moderate discrimination in patients treated with both direct oral anticoagulants and vitamin K antagonists [5].

Limitations

The ATRIA bleeding risk score was derived in a warfarin-treated cohort and models warfarin-associated bleeding. Its applicability to direct oral anticoagulants is less well validated, and the studies that have tested the score in DOAC cohorts consistently show lower discrimination than in the derivation cohort [3, 4, 5].

The score lacks several variables regarded clinically as important for bleeding risk: labile INR, liver function, concomitant antiplatelet or NSAID treatment, and a tendency to fall. In the derivation cohort there was no access to over-the-counter drugs such as aspirin and NSAIDs, which may have underestimated their contribution [1]. The absence of INR stability as a variable is particularly relevant for warfarin-treated patients, in whom a labile INR is one of the strongest single predictors of bleeding.

All five variables are binary, which means that the score does not capture the degree of renal impairment or the severity of anaemia. A patient with an eGFR of 29 receives the same score as a dialysis-dependent patient. Age is treated as a threshold variable at 75 years, without modelling the continuous effect of age.

The score applies to non-valvular atrial fibrillation. With a mechanical valve prosthesis or significant mitral stenosis, anticoagulation is indicated whatever the bleeding score, and ATRIA must not be used to justify withholding it.

The risk is not static. A patient with a low score today may pass into intermediate or high risk by developing anaemia or renal failure, or simply by turning 75. The score should therefore be reassessed when the clinical situation changes and at annual review.

Place in current practice

In contemporary practice, warfarin has largely been replaced by direct oral anticoagulants as first-line treatment in non-valvular atrial fibrillation, in line with the ESC guidelines. Since ATRIA is validated primarily for warfarin, and DOAC cohorts show poorer performance, HAS-BLED is often the more widely used alternative, particularly as it includes labile INR as a variable and thereby better reflects warfarin-treated patients, while also having broader external validation for DOACs. The 2024 ESC guidelines emphasise that bleeding risk scores should be used to identify and address modifiable risk factors rather than to exclude patients from anticoagulation [6].

References

1. Fang MC, Go AS, Chang Y, et al. A new risk scheme to predict warfarin-associated hemorrhage: the ATRIA Study. J Am Coll Cardiol. 2011;58(4):395-401. PMID: 21757117
2. Zeng J, Yu P, Cui W, 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 (Baltimore). 2020;99(25):e20782. PMID: 32569222
3. Proietti M, Hijazi Z, Andersson U, et al. Comparison of bleeding risk scores in patients with atrial fibrillation: insights from the RE-LY trial. J Intern Med. 2018;283(3):282-292. PMID: 29044861
4. Yao X, Gersh BJ, Sangaralingham LR, et al. Comparison of the CHA₂DS₂-VASc, CHADS₂, HAS-BLED, ORBIT, and ATRIA Risk Scores in Predicting Non-Vitamin K Antagonist Oral Anticoagulants-Associated Bleeding in Patients With Atrial Fibrillation. Am J Cardiol. 2017;120(9):1549-1556. PMID: 28844514
5. Gaboreau Y, Frappé P, Vermorel C, et al. Oral anticoagulant safety in family practice: prognostic accuracy of Bleeding Risk Scores (from the CACAO study). Fam Pract. 2024;41(1):9-17. PMID: 38281089
6. Boriani G, Mei DA, Vitolo M, et al. The 2024 ESC guidelines on atrial fibrillation: essential updates for everyday clinical practice. Intern Emerg Med. 2025;20(5):1299-1306. PMID: 40514614