IMPROVE Score for bleeding risk

Clinical background

In medical inpatients, venous thromboembolism is a leading cause of morbidity and mortality, yet fear of bleeding complications often deters clinicians from prescribing pharmacological prophylaxis. The IMPROVE bleeding risk score aims to quantify the in-hospital bleeding risk at admission, so that the decision about pharmacological VTE prophylaxis can be set against an estimated bleeding risk rather than against undefined anxiety. The score complements thrombosis risk assessments (for example Padua) and is particularly useful when the overall picture is unclear: patients with a moderate thrombotic risk but a high bleeding risk, or the reverse.

Calculating the IMPROVE bleeding risk score

The score is a weighted sum of eleven variables assessed on admission:

$$\text{Score} = R_{\text{renal}} + S_{\text{sex}} + A_{\text{age}} + C_{\text{cancer}} + R_{\text{rheum}} + V_{\text{CVC}} + I_{\text{ICU}} + L_{\text{liver}} + T_{\text{platelets}} + B_{\text{bleeding}} + U_{\text{ulcer}}$$

where each variable is weighted as follows:

• $R_{\text{renal}}$: renal function (GFR) $\geq 60$ mL/min scores 0 points, 30–59 mL/min scores 1, $< 30$ mL/min scores 2.5
• $S_{\text{sex}}$: female scores 0, male scores 1
• $A_{\text{age}}$: $< 40$ years scores 0, 40–84 years scores 1.5, $\geq 85$ years scores 3.5
• $C_{\text{cancer}}$: active cancer scores 2 points, absent scores 0
• $R_{\text{rheum}}$: rheumatological disease scores 2 points, absent scores 0
• $V_{\text{CVC}}$: a central venous catheter scores 2 points, absent scores 0
• $I_{\text{ICU}}$: admission to an ICU or CCU scores 2.5 points, absent scores 0
• $L_{\text{liver}}$: hepatic failure (INR >1.5) scores 2.5 points, absent scores 0
• $T_{\text{platelets}}$: a platelet count $< 50 \times 10^9\text{/L}$ scores 4 points, absent scores 0
• $B_{\text{bleeding}}$: bleeding in the 3 months before admission scores 4 points, absent scores 0
• $U_{\text{ulcer}}$: an active gastroduodenal ulcer scores 4.5 points, absent scores 0

The derivation cohort comprised 15,156 medical inpatients in the multinational observational study IMPROVE (International Medical Prevention Registry on Venous Thromboembolism) [1]. The modelled outcome was major or clinically relevant bleeding during the hospital stay, within 14 days of admission, with a cumulative incidence of 3.2% in the cohort. The strongest independent risk factors were an active gastroduodenal ulcer (OR 4.15), bleeding within 3 months before admission (OR 3.64) and a low platelet count (OR 3.37) [1]. The point weights were derived from the odds ratios in a multivariable regression model, and a total of $\geq 7$ defined a high bleeding risk according to the threshold proposed in the derivation study [1].

Interpretation in practice

The score is an aid, not a rule. A patient with a high bleeding risk and a low thrombotic risk should generally not receive pharmacological prophylaxis, whereas a patient with a high bleeding risk but a very high thrombotic risk may need an individualised judgement in which reduced-dose prophylaxis or mechanical prophylaxis may be appropriate [2,3]. Relying on the bleeding score alone to withhold prophylaxis from a patient at pronounced thrombotic risk is the commonest misuse.

Validation and performance

External validation in Chinese medical inpatients (DissolVE-2): In a retrospective multicentre study at 60 Chinese hospitals, 5,076 medical patients with complete data for calculating the score were included [2]. The cumulative incidence of clinically relevant bleeding within 14 days was 2.6% (95% CI 2.3–3.4). The area under the ROC curve was 0.73 ($p < 0{.}001$) for clinically relevant bleeding. Patients with a score $\geq 7$ had a hazard ratio of 7.17 (95% CI 5.05–10.18) for clinically relevant bleeding and 13.95 (95% CI 7.28–26.73) for major bleeding, compared with patients scoring $< 7$. Sensitivity was 44.1% and the positive predictive value 10.9%, reflecting the fact that bleeding is a relatively rare outcome and that false positives are common [2].

External validation in COVID-19: In a prospective multicentre study comprising 3,886 hospitalised COVID-19 patients at six Chinese university hospitals (December 2022 to January 2023), the AUC for major bleeding at 90 days was 0.84 (95% CI 0.77–0.91), with good calibration on calibration plots [3]. Patients with a score $\geq 7$ had an adjusted hazard ratio of 6.63 (95% CI 3.62–12.15) for major bleeding after adjustment for LMWH prophylaxis. In subgroup analysis, LMWH prophylaxis increased the risk of major bleeding specifically in older patients at high bleeding risk (adjusted HR 5.45 at 14 days, 95% CI 1.15–25.94), underlining that the score can guide the choice of prophylactic strategy in older, severely ill patients [3].

US external validation: In a prospective cohort at the Walter Reed Army Medical Center (1,668 medical inpatients), a score $\geq 7$ was confirmed to be associated with an increased risk of major bleeding after adjustment for pharmacological prophylaxis, but the association for the composite outcome of major and clinically relevant bleeding did not quite reach statistical significance [4].

Systematic assessment of risk factors: An expert panel used GRADE methodology and a Delphi process to assess systematically which variables should be included in a bleeding risk model for medical inpatients [5]. The panel identified age $\geq 65$ years, renal failure, thrombocytopenia, an active gastroduodenal ulcer, liver disease, recent bleeding and critical illness as the variables with the strongest evidence [5]. Rheumatological disease, by contrast, was excluded, and male sex, active cancer and a central venous catheter were considered to require further research to confirm their importance [5]. This means that the IMPROVE bleeding risk score may overstate the weight of these three variables in individual patients.

Limitations

The score applies only to medical inpatients with acute illness. It is not validated for surgical patients, patients already on therapeutic anticoagulation, or patients with traumatic injury. Pregnancy and chronic illness without acute deterioration were excluded from the derivation cohort [1].

Sensitivity at the threshold of $\geq 7$ is low, around 44% in the largest external validation [2], meaning that the majority of patients who do bleed are not identified as high risk. The positive predictive value is likewise low, around 11% [2], so most patients scoring $\geq 7$ will not bleed. The score flags a level of risk, not a prognosis, and should not be used as the sole reason to withhold pharmacological prophylaxis from a patient at high thrombotic risk.

Several variables in the score, particularly rheumatological disease, male sex and a central venous catheter, have been questioned by an expert panel and may contribute to misclassification [5]. The score furthermore measures risk on admission and does not capture complications or treatments arising during the hospital stay.

References

1. Decousus H, Tapson VF, Bergmann JF, et al. Factors at admission associated with bleeding risk in medical patients: findings from the IMPROVE investigators. Chest 2011;139(1):69–79. PMID: 20453069
2. Zhang Z, Zhai Z, Li W, et al. Validation of the IMPROVE bleeding risk score in Chinese medical patients during hospitalization: findings from the DissolVE-2 study. Lancet Reg Health West Pac 2020;4:100054. PMID: 34327391
3. Tao Y, Xu F, Han J, et al. External validation of the IMPROVE risk score for predicting bleeding in hospitalized COVID-19 patients. J Gen Intern Med 2025;40(9):1995–2004. PMID: 39979703
4. Hostler DC, Marx ES, Moores LK, et al. Validation of the International Medical Prevention Registry on Venous Thromboembolism Bleeding Risk Score. Chest 2016;149(2):372–9. PMID: 26867833
5. Darzi AJ, Karam SG, Spencer FA, et al. Risk models for VTE and bleeding in medical inpatients: systematic identification and expert assessment. Blood Adv 2020;4(12):2557–2566. PMID: 32542391