Geneva Risk Score for VTE prophylaxis

Clinical background

Pharmacological thromboprophylaxis reduces the risk of venous thromboembolism (VTE) in selected medical inpatients, but the absolute risk reduction is small in the general population and treatment carries an increased bleeding risk. The decision to give or withhold prophylaxis therefore requires a structured risk assessment. Without an instrument, prophylaxis tends to be given inconsistently: in the ESTIMATE cohort, 48 per cent of low-risk patients received prophylaxis unnecessarily, while 38 per cent of high-risk patients were left untreated [5].

The Geneva risk score was developed to identify medical patients at increased VTE risk in whom pharmacological prophylaxis is justified and, just as importantly, to identify with confidence those whose risk is so low that prophylaxis can be withheld. The score is empirically derived, based on the published literature, consensus guidelines and clinical expertise rather than on data-driven statistical modelling [4]. A simplified version, with only two point levels (2 and 1 point), was later derived from the same cohort to improve usability [3].

Calculating the Geneva risk score

The score is a weighted sum of risk factors assessed at admission:

$$\text{Geneva risk score} = 2 \times \sum_{i} M_i + \sum_{j} m_j$$

where $M_i$ are major factors (each worth 2 points) and $m_j$ are minor factors (each worth 1 point). Major factors comprise decompensated heart failure, respiratory failure, acute stroke (<3 months), recent myocardial infarction (<4 weeks), acute infection or sepsis, acute rheumatological disease, active malignancy, myeloproliferative syndrome, nephrotic syndrome, previous venous thromboembolism and known thrombophilia. Minor factors comprise immobilisation ≥3 days, recent long-distance travel (>6 hours), age >60 years, BMI >30 kg/m², chronic venous insufficiency, pregnancy or the postpartum period, hormonal treatment (oestrogen/HRT) and dehydration.

The threshold for high risk is a score of $\geq 3$.

The derivation cohort for the simplified version consisted of 1,478 acutely ill medical patients from eight Swiss hospitals, recruited prospectively in the ESTIMATE study (NCT01277536) [1]. The median age was about 70 years, and the cumulative 90-day incidence of symptomatic VTE or VTE-related death was 1.6 per cent in the whole cohort [3]. The original score was validated in the same cohort, and the simplified version was derived retrospectively from these data with discrimination and calibration preserved compared with the original version [3].

Interpretation in practice

The most important function of the score is to identify low-risk patients. In the ESTIMATE cohort, the low-risk group had a VTE risk of 0.6 per cent (all patients) and 0.8 per cent (those without prophylaxis), a level at which the risk of bleeding complications from prophylaxis probably outweighs the benefit [1]. The negative likelihood ratio was 0.28 (95 per cent CI 0.10 to 0.83) for the Geneva risk score, compared with 0.51 (95 per cent CI 0.28 to 0.93) for the Padua Prediction Score, indicating that the Geneva score is somewhat better at ruling out high risk [1].

For patients who fall into the high-risk group, the decision about prophylaxis must additionally be weighed against bleeding risk. In the ESTIMATE cohort, active bleeding at admission and thrombocytopenia (with or without cancer) were the strongest factors leading to high-risk patients being left without prophylaxis [5].

Validation and performance

The ESTIMATE study (external validation of the original score). In the prospective multicentre cohort of 1,478 Swiss medical patients, the Geneva risk score classified 65 per cent as high risk and 35 per cent as low risk [1]. The cumulative 90-day incidence of VTE or VTE-related death was 3.2 per cent in the high-risk group compared with 0.6 per cent in the low-risk group (p = 0.002). Among patients who did not receive prophylaxis, the corresponding figures were 3.5 per cent and 0.8 per cent (p = 0.029) [1]. Discrimination, measured as the AUC at 30 days, was 0.81 (95 per cent CI 0.73 to 0.89) in a comparative analysis of the same cohort [3].

Comparison with the Padua Prediction Score. In the ESTIMATE cohort, the Geneva risk score performed comparably to the Padua Prediction Score in terms of discrimination, but with one important difference: among patients without prophylaxis, Padua could not separate low-risk patients significantly (3.2 per cent vs 1.5 per cent, p = 0.130), while the Geneva score maintained a significant separation (3.5 per cent vs 0.8 per cent, p = 0.029) [1]. This suggests that the Geneva score is more reliable for identifying patients in whom prophylaxis can safely be withheld.

The RISE study (prospective validation of the simplified version). In a later prospective multicentre cohort (RISE, 1,352 medical patients at three Swiss university hospitals, 2020 to 2022), the results were considerably less favourable [2]. The AUC for the simplified Geneva score was 58.1 per cent (95 per cent CI 55.4 to 60.7), and for the original version 53.8 per cent (95 per cent CI 51.1 to 56.5). Sensitivity was 78.6 per cent and specificity 37.2 per cent. After adjustment for the use of prophylaxis and study site, the difference in VTE risk between the high- and low-risk groups was no longer significant (subhazard ratio 2.04, 95 per cent CI 0.83 to 5.05, p = 0.12) [2]. The 90-day VTE incidence was 2.6 per cent in the high-risk group and 1.2 per cent in the low-risk group, an absolute difference of only 1.4 percentage points.

Systematic review. A systematic review of 51 validation studies of 24 different risk assessment models found that the c-statistic was generally weak (<0.7) for most models, with considerable heterogeneity between studies [6]. The Geneva risk score was evaluated in four studies. No model performed obviously better than the others, and the review concluded that there was insufficient evidence to recommend any specific model over another [6].

Limitations

The score applies only to non-surgical inpatients with acute illness. Patients with an indication for therapeutic anticoagulation (e.g. atrial fibrillation with a high stroke risk, established VTE, a mechanical heart valve) should not be assessed with the score, since the question of prophylaxis does not arise. Intensive care patients and postoperative patients have different risk profiles and are not covered.

The empirical derivation, without data-driven statistical modelling, means that the point weights rest on expert judgement rather than on measured odds ratios in a cohort [4]. This carries a risk of suboptimal calibration, particularly in populations differing from the Swiss university hospital setting in which the score was validated.

The prospective RISE study calls the score's clinical utility into question: discrimination was poor (AUC below 60 per cent) and the risk separation between groups was not significant after adjustment [2]. This applies to a population that broadly matches the intended target group, which makes the finding hard to explain away as poorer applicability.

A practical pitfall is that the score is often used carelessly as a yes/no instrument, when in fact it only provides a probability. A value of 2 points does not automatically mean that prophylaxis should be given, and a value of 3 does not guarantee that the benefit outweighs the bleeding risk. Bleeding risk must be assessed separately, and no established bleeding score has been validated in combination with the Geneva risk score.

References

1. Nendaz M, Spirk D, Kucher N, et al. Multicentre validation of the Geneva Risk Score for hospitalised medical patients at risk of venous thromboembolism. Explicit ASsessment of Thrombembolic RIsk and Prophylaxis for Medical PATients in SwitzErland (ESTIMATE). Thromb Haemost. 2014;111(3):531-8. PMID: 24226257
2. 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
3. Blondon M, Spirk D, Kucher N, et al. Comparative Performance of Clinical Risk Assessment Models for Hospital-Acquired Venous Thromboembolism in Medical Patients. Thromb Haemost. 2018;118(1):82-89. PMID: 29304528
4. Stuck AK, Spirk D, Schaudt J, et al. Risk assessment models for venous thromboembolism in acutely ill medical patients. A systematic review. Thromb Haemost. 2017;117(4):801-808. PMID: 28150851
5. Spirk D, Nendaz M, Aujesky D, et al. Predictors of thromboprophylaxis in hospitalised medical patients. Explicit ASsessment of Thrombembolic RIsk and Prophylaxis for Medical PATients in SwitzErland (ESTIMATE). Thromb Haemost. 2015;113(5):1127-34. PMID: 25608607
6. Pandor A, Tonkins M, Goodacre S, et al. Risk assessment models for venous thromboembolism in hospitalised adult patients: a systematic review. BMJ Open. 2021;11(7):e045672. PMID: 34326045