Six HAS-BLED details that change the score

HAS-BLED reports how often major bleeding occurred within one year in patients with atrial fibrillation who had a given number of risk factors. Pisters and colleagues derived it in 2010 from the Euro Heart Survey on atrial fibrillation, with patients enrolled in 2003 and 2004, most of whom were taking a vitamin K antagonist.¹ It is an observed group rate, not an individual probability.

The outcome is major bleeding as defined by the ISTH in 2005: intracranial bleeding, bleeding requiring hospitalisation, a haemoglobin fall of more than 2 g/dL, or bleeding requiring transfusion, excluding haemorrhagic stroke.¹² The derivation paper prints the haemoglobin threshold as "2 g/L", a misprint: a 2 g/L fall is clinically trivial, and the ISTH definition it cites specifies 2 g/dL.

It is a points model with nine items worth one point each. The score was derived in 3,456 patients with known one-year follow-up, of whom 53 (1.5%) had a major bleed; follow-up was incomplete for part of the wider survey, which the authors acknowledge.¹

Source: Pisters 2010¹; ISTH 2005².

The two "A" items and the two "D" items each score separately, so the maximum is 9. Pisters 2010 defines no low, moderate or high categories; it reports a rate for each score.¹

Source: Pisters 2010¹. Definitions as signed off at clinical review of the Curie Calc HAS-BLED calculator, August 2026.

Is the alcohol threshold 8 drinks or 8 units?

Eight units, and the difference is large. The derivation paper classifies alcohol use of 8 or more units a week as a risk factor and says "units" throughout, without defining a unit in grams.¹ Curie Calc, on clinical review, reads them as UK units, consistent with the European cohort: a UK unit is about 8 g of ethanol, so 8 units is about 64 g a week. A US standard drink is about 14 g, so the same threshold is roughly 4 to 5 US drinks, not 8.³

Reading the item as "8 drinks a week" raises the bar by about 1.75 times. A patient drinking five US standard drinks a week (about 70 g) meets the original threshold and scores a point; under a "drinks" reading, the same patient scores 0. For reference, one US standard drink is 12 oz of beer at about 5%, 5 oz of wine at about 12%, or 1.5 oz of spirits at about 40%.³

Does labile INR apply to a patient on a DOAC?

No. The item asks about unstable or high INRs, or time in therapeutic range under 60%, which only has meaning for a vitamin K antagonist. HAS-BLED was derived in 2003 and 2004, before direct oral anticoagulants were in use.¹ For a patient on a DOAC, the item scores 0.

That has a consequence for reading the result. The published rates come from a cohort in which the item could be present, so a DOAC patient's total is not directly comparable to them. The score still discriminates about as well in DOAC-treated patients as in VKA-treated ones: a meta-analysis of 39 studies found a C-statistic of 0.63 in both groups.⁴

Scores built for DOAC patients exist. The DOAC Score had a higher C-statistic than HAS-BLED in its derivation and validation cohorts (C-statistic 0.73 vs 0.60 in RE-LY, and 0.67 vs 0.63 in a pooled trial cohort),⁵ and a 2026 meta-analysis of nine studies found a modest advantage (0.68 vs 0.63) with no difference from ORBIT.⁶ A 2025 cohort of Chinese patients on DOACs found HAS-BLED and the DOAC Score comparable.⁷ No replacement has yet shown a consistent advantage across settings.

Does a TIA count as stroke?

No. The S item is prior stroke, "particularly lacunar". TIA and systemic embolism are not counted.¹ In the derivation paper, stroke or TIA appears together only as a baseline characteristic; the bleeding risk factor itself is stroke alone.

This is a different question from CHA₂DS₂-VASc, where stroke, TIA and thromboembolism together score 2 points. A patient with a prior TIA scores in CHA₂DS₂-VASc but not in HAS-BLED, so the answer should not be copied from one score to the other. Read the CHA₂DS₂-VASc guide.

Does treated hypertension count?

Only if it is uncontrolled. The H item is a systolic blood pressure above 160 mmHg.¹ A history of hypertension, treated or not, does not score unless the pressure is above that level. A patient at 150 mmHg scores 0.

Here too the two AF scores ask different questions. CHA₂DS₂-VASc's H is a history of hypertension or treatment for it; HAS-BLED's H is current control. A patient can score the point in one and not the other.

What counts as abnormal kidney or liver function?

Fixed cutoffs, each worth a point. Abnormal renal function is chronic dialysis, a kidney transplant, or a serum creatinine of 200 µmol/L or above (2.26 mg/dL).¹ The cutoff is inclusive: a creatinine of exactly 200 µmol/L scores. An eGFR is not part of the definition.

Abnormal liver function is cirrhosis, or a bilirubin above twice the upper limit of normal together with an AST, ALT or alkaline phosphatase above three times normal.¹ Mildly raised liver enzymes alone do not score. Kidney and liver are separate items, so a patient can score both.

What about age 65 and the two "D" points?

Over 65 scores; exactly 65 does not. The E item is age over 65.¹ And the two D items are separate: concomitant antiplatelet or NSAID use is one point, and alcohol at or above the threshold is another. A patient taking aspirin who also drinks 10 units a week scores 2 for the D items.

Concomitant antiplatelet or NSAID use scores one point whenever it is present, whatever the indication.

The published rates are major bleeds per 100 patient-years in the derivation cohort. They rise with the score but not smoothly, and above a score of 5 there are too few patients to estimate a rate. These per-score rates come from the paper's Table 5, which counts 3,071 patients and 48 major bleeds.

Pisters 2010, Euro Heart Survey, one-year major bleeding (ISTH 2005), mostly vitamin K antagonist-treated patients.¹ Group averages, not individual forecasts. Curie Calc displays these rounded to one decimal place.

Three things are worth knowing about this table:

• Score 0 sits above score 1. That is in the published data and reflects small numbers of events, not an error.
• The higher scores rest on few patients. Only 8 patients scored 5, so the 12.5 figure comes from a single bleed. Two patients scored 6, with no bleeds observed, and none scored higher; that is too few to estimate a rate, so Curie Calc shows none above 5.
• Discrimination is modest. The derivation paper reported a C-statistic of 0.72; across 39 later studies the pooled figure is 0.63.¹⁴ A meta-analysis found the score under-predicts bleeding in the middle of its range.⁸

What the guidelines say. The 2023 ACC/AHA atrial fibrillation guideline states that bleeding risk scores should not be used in isolation to determine eligibility for oral anticoagulation, but to identify and modify bleeding risk factors.⁹ The 2024 ESC guideline recommends assessing and managing modifiable bleeding risk factors and no longer names a specific bleeding score.¹⁰

Following these positions, Curie Calc shows HAS-BLED without risk tiers and places a short note beside every result: a high score identifies modifiable bleeding risk factors for review, and it is not a threshold for withholding anticoagulation. Several of the nine items can change: uncontrolled blood pressure, antiplatelet or NSAID use, alcohol, and labile INR on a vitamin K antagonist.

Synthetic cases. Any criterion not mentioned is absent. Each result was checked on the live Curie Calc HAS-BLED calculator.

Try these cases in the HAS-BLED calculator

HAS-BLED was derived in atrial fibrillation; valvular AF and mitral stenosis were excluded from the derivation cohort.¹ Curie Calc also lists end-stage kidney disease and active cancer as populations where the score should not be used, because both carry bleeding risk the nine items do not capture.

The score was built before DOACs, so for a DOAC-treated patient the labile INR item cannot score and the total is not directly comparable to the published rates. Its rates come from a European cohort enrolled in 2003 and 2004, with incomplete one-year follow-up.¹ It estimates bleeding risk only and says nothing about stroke risk, which CHA₂DS₂-VASc addresses. The two scores count different outcomes on different scales.