Clinical background
Acutely ill medical inpatients are at increased risk of venous thromboembolism (VTE), but the absolute incidence is low in the group as a whole, and pharmacological prophylaxis carries a small but real bleeding risk. The decision about prophylaxis therefore requires an individual judgement. The IMPROVE risk score was developed to provide a structured estimate of VTE risk and thereby to support the choice between pharmacological prophylaxis, mechanical prophylaxis or none. The score rests on variables that are readily available clinically and is intended for use on admission and during the hospital stay, not in outpatient care.
Calculating the IMPROVE risk score
The IMPROVE risk score is the sum of seven weighted risk factors:
The variables are binary (yes = 1, no = 0). The score ranges from 0 to 12.
The first four variables (previous VTE, known thrombophilia, paresis/paralysis of a lower limb, active cancer) are available on admission. The remaining three (immobilisation ≥7 days, care in an ICU or CCU, age >60 years) accrue during the hospital stay. The calculator uses the full seven-variable model, which in the derivation cohort had somewhat better discrimination than the four-variable model based on admission data alone.
The derivation cohort consisted of 15,156 medical inpatients in the international observational study IMPROVE (International Medical Prevention Registry on Venous Thromboembolism), with data collected across several countries. The outcome measure was symptomatic VTE within 3 months of admission. Of the 184 patients who developed symptomatic VTE, 76 had pulmonary embolism and 67 had lower-limb deep vein thrombosis. The cumulative VTE incidence was 1.0%, and 45% of events occurred after discharge [1]. The c-statistic was 0.65 for the admission model and 0.69 for the in-hospital model [1].
Interpretation in practice
In the derivation cohort, the different score levels corresponded to the following observed VTE risk within 3 months [1]:
| Score | Observed VTE risk | Clinical action |
|---|---|---|
| 0–1 | Low (<1.5%) | Withhold pharmacological prophylaxis if there is a bleeding risk. Consider mechanical prophylaxis if immobility persists. |
| 2–3 | Elevated (~1.5%) | Individual judgement. Consider pharmacological prophylaxis if the bleeding risk is acceptable. |
| ≥4 | High (~5.7%) | Pharmacological prophylaxis recommended unless a bleeding risk contraindicates it. |
A score of 2 or more was, in the derivation cohort, also associated with higher all-cause mortality and VTE-related in-hospital mortality [1]. This reinforces that the threshold of ≥2 marks a patient group in whom the decision about prophylaxis should not be deferred.
Since almost half of VTE events occurred after discharge, patients with a high score should also be considered for extended prophylaxis after discharge, particularly if risk factors such as immobilisation or active cancer persist. A modified version of the IMPROVE risk score, in which a score of 2–3 is combined with a raised D-dimer (>2 times the upper reference limit), has been shown to identify a subgroup with an almost threefold increased VTE risk (7.94% versus 2.83% at day 35) in whom extended prophylaxis with rivaroxaban produced a significant reduction in risk [2]. This modification was used as an inclusion criterion in the MARINER trial but is not identical to the calculator's seven-variable model.
Validation and performance
External validation of the IMPROVE risk score has given mixed results. In a prospective Swiss multicentre study (the RISE cohort) of 1,352 medical inpatients at three university hospitals, IMPROVE was compared head-to-head with the Padua and Geneva scores [3]. A total of 28 VTE events occurred within 90 days (incidence 2.1%). IMPROVE had the highest specificity of the models (70.4%) but low sensitivity. Discrimination was poor for all the models, with an AUC below 60% for each [3]. After adjustment for the use of prophylaxis and study site, the difference in VTE risk between high- and low-risk groups was not statistically significant for any of the models [3].
In a Chinese multicentre study of 595 medical inpatients (310 with VTE, 285 without), IMPROVE performed worst of the four models compared, with an AUC of 0.52 ± 0.03, compared with Padua (0.72), Caprini (0.74) and a locally developed instrument (0.80) [4]. The study had a case-control design with half the patients having VTE, which means that the results do not reflect a realistic incidence but nonetheless illustrate the model's limited discrimination.
In summary, the c-statistic in the derivation was 0.69 [1], but in external cohorts it has fallen to between 0.52 and below 0.60 [3, 4]. This indicates that the model's ability to distinguish patients who will develop VTE from those who will not is limited in clinical practice.
Limitations
The IMPROVE risk score applies to acutely ill medical inpatients. It is not validated for surgical patients, trauma patients, pregnant women or patients already on therapeutic anticoagulation. Where therapeutic anticoagulation is indicated (e.g. atrial fibrillation with a high stroke risk, established VTE, a mechanical valve prosthesis), the score should not be used to justify prophylaxis, since these patients should be anticoagulated in any case.
The model does not capture all relevant risk factors. Acute infection, heart failure, respiratory failure and a central venous catheter are factors shown in other studies to be independent risk markers for VTE in medical patients but are absent from the IMPROVE risk score [4]. The D-dimer, which has independent prognostic value, is likewise not part of the original model.
The low VTE incidence in the target population (about 1–2%) means that even a model with reasonable sensitivity generates a large number of false positives, which may lead to overtreatment with prophylaxis and unnecessary bleeding risk. Conversely, a low score cannot exclude VTE with certainty, particularly in patients with risk factors the model does not cover.
The score is not static. Immobilisation and intensive care may develop during the hospital stay, and a patient initially judged to be at low risk may later cross the threshold for prophylaxis. Reassessment should take place if the clinical condition changes.
Place in current practice
VTE prophylaxis in medical inpatients rests primarily on clinical judgement and local protocols, and the IMPROVE risk score is not established as a standard instrument to the same extent as the Padua score in European practice. The 2018 ASH guidelines recommend pharmacological prophylaxis for acutely ill medical inpatients with an acceptable bleeding risk, and state that either the Padua or the IMPROVE score may be used for risk stratification [5]. There is no international consensus on which of the two instruments should be preferred.
References
- Spyropoulos AC, Anderson FA Jr, FitzGerald G, et al. Predictive and associative models to identify hospitalized medical patients at risk for VTE. Chest. 2011;140(3):706–714. PMID: 21436241
- Spyropoulos AC, Lipardi C, Xu J, et al. Modified IMPROVE VTE Risk Score and Elevated D-Dimer Identify a High Venous Thromboembolism Risk in Acutely Ill Medical Population for Extended Thromboprophylaxis. TH Open. 2020;4(1):e59–e65. PMID: 32190813
- Häfliger E, Kopp B, Darbellay Farhoumand P, et al. Risk Assessment Models for Venous Thromboembolism in Medical Inpatients. JAMA Netw Open. 2024;7(5):e249980. PMID: 38728035
- Chen X, Shi H, Chang J, et al. External Validation of the Risk Assessment Model of Venous Thromboembolism in Multicenter Internal Medicine Inpatients. Clin Appl Thromb Hemost. 2024;30:452621. PMID: 38632943
- Schünemann HJ, Cushman M, Burnett AE, et al. American Society of Hematology 2018 guidelines for management of venous thromboembolism: prophylaxis for hospitalized and nonhospitalized medical patients. Blood Adv. 2018;2(22):3198–3225. PMID: 30482763
