VTE-BLEED Score

Clinical background

The decision whether to continue anticoagulation beyond the first three to six months after venous thromboembolism (VTE) requires a balance between the risk of recurrence and the risk of bleeding. For patients with unprovoked or weakly provoked VTE, the risk of recurrence without treatment is about 10 per cent at one year and up to 36 per cent at ten years, while the case fatality of a major bleed on anticoagulation is two to three times that of a recurrence [3]. Once the annual bleeding risk exceeds about 2.5 per cent, no net clinical benefit from extended anticoagulation is expected, even at a high risk of recurrence [3]. The VTE-BLEED score was developed to identify patients at increased bleeding risk during stable anticoagulation, that is, after the initial 30-day phase, and thereby to support decisions about treatment duration [1].

Calculating VTE-BLEED

VTE-BLEED is the sum of six variables according to the following formula:

$$\text{VTE-BLEED} = \text{cancer} \times 2 + \text{male with uncontrolled hypertension} \times 1 + \text{anaemia} \times 1{.}5 + \text{previous bleeding} \times 1{.}5 + \text{age} \geq 60 \times 1{.}5 + \text{renal dysfunction} \times 1{.}5$$

The variables are defined as: active cancer (2 points), male with a systolic blood pressure $\geq 140$ mmHg (1 point, male patients only), anaemia (Hb <13 g/dL in men or <12 g/dL in women), previous major or clinically relevant non-major bleeding, age $\geq 60$ years and renal dysfunction (creatinine clearance/eGFR <60 mL/min). The score ranges from 0 to 9, with a threshold at 2 points separating low risk from elevated risk.

The score was derived in a post hoc analysis of the pooled RE-COVER trials, two double-blind randomised trials comparing dabigatran with warfarin in 5,107 VTE patients [1]. The model was developed from the dabigatran arm using logistic regression over the whole follow-up period. The primary outcome was major bleeding after day 30, that is, during the stable treatment phase. In the derivation cohort the score had a c-statistic of 0.72 (95% CI 0.67 to 0.76) [1]. Patients at low risk (<2 points) made up 74 per cent of the population and had a bleeding incidence of 2.8 per cent, compared with 12.6 per cent in the elevated-risk group ($\geq 2$ points), corresponding to an OR of 5.0 (95% CI 3.5 to 7.1) [1].

A methodological advantage is that none of the VTE-BLEED variables overlaps with predictors of VTE recurrence. This eliminates the problematic situation in which one and the same variable drives up both bleeding risk and recurrence risk, making a meaningful trade-off between the two difficult [2].

Interpretation in practice

In the external validation from the COMMAND VTE registry, the cumulative 5-year incidence of major bleeding was 13.2 per cent in the high-risk group compared with 5.4 per cent in the low-risk group, and the difference persisted consistently over time [2]. In the prospective cohort of patients with unprovoked VTE receiving extended anticoagulation, Wells et al. measured an annual bleeding incidence of 3.1 events per 100 person-years in the high-risk group compared with 1.1 per 100 person-years in the low-risk group using a modified VTE-BLEED [3].

A central clinical question is whether patients whom VTE-BLEED classifies as being at high risk of bleeding also have an increased risk of recurrence, which would limit the score's usefulness. In a post hoc analysis of the PADIS-PE study, in which 308 patients with unprovoked pulmonary embolism were followed after anticoagulation was stopped, the cumulative incidence of recurrence was 16.4 per cent in the VTE-BLEED high-risk group compared with 14.6 per cent in the low-risk group, with an adjusted hazard ratio of 1.16 (95% CI 0.62 to 2.19) [4]. The difference was not statistically significant, supporting the use of VTE-BLEED for decisions about treatment duration without the bleeding prediction simultaneously signalling an increased risk of recurrence.

Validation and performance

For the stable treatment phase, the score's actual field of application, the c-statistic was 0.75 (95% CI 0.61 to 0.89) for dabigatran and 0.78 (95% CI 0.68 to 0.86) for warfarin, with no statistically significant difference between the treatment arms ($p = 0{.}77$) [1]. This indicates that the score performs equivalently regardless of the type of anticoagulation.

The external validation in the COMMAND VTE registry, an unselected Japanese multicentre cohort of 2,124 patients on anticoagulation extended beyond 30 days, confirmed the score's long-term predictive ability, with a significant difference in cumulative bleeding incidence between the groups over up to five years of follow-up [2]. In the prospective North American cohort of patients with unprovoked or weakly provoked VTE on extended anticoagulation, a modified VTE-BLEED performed comparably to both a modified HAS-BLED and a newly derived model, with an annual bleeding rate above 2.5 per 100 person-years in the high-risk group and below this threshold in the low-risk group [3].

In a Chinese multicentre study of 1,121 patients with VTE treated with DOACs, an external validation reported an AUC of 0.746 for VTE-BLEED, compared with 0.773 for the RIETE score and 0.558 for the Hokusai score [5]. A machine learning study based on the RIETE registry found that an XGBoost algorithm outperformed both VTE-BLEED and RIETE only in the prospective validation cohort, but not in the external validation cohort, where VTE-BLEED had an F1 value of 9.75 per cent compared with 17.3 per cent for RIETE [6].

One review emphasises that VTE-BLEED is particularly suited to identifying patients at low bleeding risk, in whom extended anticoagulation can be given with good safety, rather than to identifying with certainty those in whom treatment should be stopped [7].

Limitations

VTE-BLEED was derived and validated for patients on stable anticoagulation after the first 30 days of treatment. It should not be used to assess bleeding risk during the acute phase, where HAS-BLED or the RIETE score may be more appropriate [7]. None of the available bleeding scores, VTE-BLEED included, has been tested in a randomised trial in which the score governs the decision to withhold anticoagulation altogether, and the evidence for basing such a decision on the score alone is therefore limited [7].

In the prospective cohort of patients with unprovoked VTE, a modified version of VTE-BLEED was used, since active cancer was excluded (cancer being one of the score's variables) and uncontrolled hypertension was approximated using treatment for hypertension as a proxy variable [3]. This illustrates that the score's variable definitions may need to be adapted in populations without cancer, and that interpretation of the hypertension variable is sensitive to how it is measured.

The only VTE-BLEED variable that potentially varies over time is male with uncontrolled hypertension, while the other variables are as a rule stable [4]. This is an advantage over other bleeding scores, but it also means that a patient with well-controlled blood pressure at the time of assessment may later develop uncontrolled hypertension, so that reassessment is warranted if the clinical status changes.

References

1. Klok FA, Hösel V, Clemens A et al. Prediction of bleeding events in patients with venous thromboembolism on stable anticoagulation treatment. Eur Respir J. 2016;48(5):1369-1376. PMID: 27471209
2. Nishimoto Y, Yamashita Y, Morimoto T et al. Validation of the VTE-BLEED score's long-term performance for major bleeding in patients with venous thromboembolisms: From the COMMAND VTE registry. J Thromb Haemost. 2020;18(3):624-632. PMID: 31785073
3. Wells PS, Tritschler T, Khan F et al. Predicting major bleeding during extended anticoagulation for unprovoked or weakly provoked venous thromboembolism. Blood Adv. 2022;6(15):4605-4616. PMID: 35679460
4. Klok FA, Presles E, Tromeur C et al. Evaluation of the predictive value of the bleeding prediction score VTE-BLEED for recurrent venous thromboembolism. Res Pract Thromb Haemost. 2019;3(3):364-371. PMID: 31294323
5. Wu G, Chen J, Chen Y et al. Bleeding prediction scores in patients with venous thromboembolism using direct oral anticoagulants. Ann Hematol. 2025;104(6):3477-3486. PMID: 40455258
6. Mora D, Mateo J, Nieto JA et al. Machine learning to predict major bleeding during anticoagulation for venous thromboembolism: possibilities and limitations. Br J Haematol. 2023;201(5):971-981. PMID: 36942630
7. Nopp S, Ay C. Bleeding risk assessment in patients with venous thromboembolism. Hämostaseologie. 2021;41(4):267-274. PMID: 33626580