Clinical background
VA-ECMO for refractory cardiogenic shock is a resource-intensive treatment with a high rate of complications and substantial mortality. The decision to start support requires a rapid assessment of whether the patient has a realistic chance of surviving to discharge and, beyond that, to cardiac recovery, transplantation or durable mechanical circulatory support. Without a structured tool the assessment becomes intuitive and varies between centres and clinicians. The SAVE score was constructed to standardise this assessment and to provide an estimate of the probability of hospital survival from variables available before or at the time of cannulation.
Calculating the SAVE score
The SAVE score is a weighted sum of 12 variables collected before or at the time of VA-ECMO cannulation, with a constant of 6 subtracted:
where is the weight for variable and is the value of the variable. The variables and their points are:
| Variable | Category | Points |
|---|---|---|
| Age | 18–38 years | +7 |
| 39–52 years | +4 | |
| 53–62 years | +3 | |
| ≥63 years | 0 | |
| Weight | <65 kg | +1 |
| 65–89 kg | +2 | |
| >89 kg | 0 | |
| Cause of cardiogenic shock | Other cause | 0 |
| Myocarditis | +3 | |
| Refractory VT/VF | +2 | |
| After heart or lung transplantation | +3 | |
| Congenital heart disease | −3 | |
| Chronic renal failure | Yes | −6 |
| HCO₃ ≤15 mmol/L before ECMO | Yes | −3 |
| Duration of mechanical ventilation before ECMO | ≤10 hours | 0 |
| 11–29 hours | −2 | |
| ≥30 hours | −4 | |
| Peak inspiratory pressure ≤20 cmH₂O | Yes | +3 |
| Cardiac arrest before ECMO | Yes | −2 |
| Diastolic blood pressure ≥40 mmHg before ECMO | Yes | +3 |
| Pulse pressure ≤20 mmHg before ECMO | Yes | −2 |
| Acute hepatic failure before ECMO | Yes | −3 |
| CNS dysfunction before ECMO | Yes | −3 |
The score ranges theoretically from −35 to +17 and is divided into five risk classes:
| Class | Score range | Expected survival |
|---|---|---|
| I | >5 | 75% |
| II | 1 to 5 | 58% |
| III | −4 to 0 | 42% |
| IV | −9 to −5 | 30% |
| V | ≤−10 | 18% |
The derivation cohort consisted of 3,846 patients with refractory cardiogenic shock treated with VA-ECMO, extracted from the international ELSO registry between January 2003 and December 2013. In all, 1,601 patients (42%) were alive at discharge. Multivariable logistic regression with bootstrapping and internal as well as external validation was used to identify factors independently associated with hospital survival [1].
Interpretation in practice
The SAVE score is an aid to risk stratification, not a decision tool that alone determines whether ECMO should be started or stopped. The score should be interpreted in the context of the patient's overall prognosis, the reversibility of the underlying cause and access to definitive treatment such as transplantation or mechanical circulatory support.
| Class | Score | Survival | Clinical action |
|---|---|---|---|
| I | >5 | ~75% | Favourable prognosis. ECMO is indicated if the other criteria are met. |
| II | 1–5 | ~58% | Moderate prognosis. ECMO is reasonable; plan the bridge to definitive treatment early. |
| III | −4 to 0 | ~42% | Uncertain prognosis. Individual assessment of reversibility and treatment options. |
| IV | −9 to −5 | ~30% | Increased risk. ECMO may be considered if a reversible cause is present, but the expectation of survival is low. |
| V | ≤−10 | ~18% | Very high risk. ECMO is generally not justified unless a specific reversible cause is identified. |
A SAVE score of zero corresponds to roughly 50% survival. Negative scores are driven above all by chronic renal failure, a long duration of ventilation before ECMO, a low bicarbonate, cardiac arrest and organ dysfunction, while young patients with myocarditis or refractory arrhythmia gain positive points.
Validation and performance
In the derivation cohort, discrimination was moderate with an AUROC of 0.68 (95% CI 0.64–0.71). External validation in an Australian cohort of 161 patients, by contrast, showed excellent discrimination with an AUROC of 0.90 (95% CI 0.85–0.95) [1]. This discrepancy has attracted attention in later validation studies.
In a North American validation at Toronto General Hospital comprising 120 patients treated with VA-ECMO between 2011 and 2018, a c-statistic of 0.77 (95% CI 0.69–0.86) was achieved. Survival to discharge was 45%, comparable to the 42% of the derivation cohort. Calibration was limited, however: the SAVE score systematically underestimated survival. For risk class II the observed survival was 67% versus 58% predicted, class III 78% versus 42%, class IV 61% versus 30% and class V 29% versus 18%. No patients in the cohort fell into class I [2].
A systematic review from 2023 identified 58 prognostic models for ECMO patients, of which only 14 (24%) had been externally validated. The SAVE score was the most validated model, with 20 external validations. The pooled c-statistic for SAVE and RESP (a corresponding model for respiratory ECMO) lay between 0.66 and 0.70, described as moderate discrimination and comparable to general intensive care scores such as SAPS II, APACHE II and SOFA. Only 1 of the 58 models met the PROBAST criteria for good methodological quality [3].
In a validation against the MIMIC-IV database (n=101), the SAVE score performed less well than in earlier studies, with an AUC-ROC of 0.578 and significant miscalibration. Decision curve analysis showed minimal net clinical benefit [4].
In summary, discrimination varies between cohorts, from excellent (0.90) in the original Australian validation to close to chance (0.58) in MIMIC-IV. Calibration tends to underestimate survival in contemporary cohorts, which may lead to patients being wrongly judged too high risk for ECMO.
Limitations
The SAVE score has several important limitations that must be considered in clinical use:
Derived from an already treated population. All the patients in the ELSO registry had already started VA-ECMO. The score therefore cannot answer whether a patient would survive if ECMO were started, only estimate survival among those already being treated. This is the single most important limitation and is highlighted in the systematic review as a problem for all existing ECMO models [3].
Time period and practice. The data were collected between 2003 and 2013. ECMO technology, patient selection and postoperative management have developed since then, which may explain why contemporary cohorts perform better than the score predicts.
Poor calibration in contemporary cohorts. In the Toronto cohort, the score underestimated survival in every risk class from II to V [2]. A patient in class V (predicted survival 18%) had an observed survival of 29%. This carries a risk of patients being denied ECMO on the basis of a score that overestimates mortality.
Variables at cardiac arrest. In pre-ECMO cardiac arrest, reliable haemodynamic measurements are often unavailable, and some variables (diastolic blood pressure, pulse pressure) must be estimated retrospectively, which introduces uncertainty into the calculation.
Mix of aetiologies. The score gives different weights to specific aetiologies (myocarditis, refractory VT/VF, transplantation, congenital heart disease), but the category "other cause" is heterogeneous and includes acute myocardial infarction, sepsis, valvular heart disease and other conditions with differing prognoses.
Not applicable to ECPR. Patients cannulated during ongoing cardiopulmonary resuscitation (ECPR) without return of spontaneous circulation constitute a specific subgroup in which the SAVE variables are difficult to establish and its predictive value is even less well validated.
References
- Schmidt M, Burrell A, Roberts L, et al. Predicting survival after ECMO for refractory cardiogenic shock: the survival after veno-arterial-ECMO (SAVE)-score. Eur Heart J 2015;36(33):2246–56. PMID: 26033984.
- Amin F, Lombardi J, Alhussein M, et al. Predicting survival after VA-ECMO for refractory cardiogenic shock: validating the SAVE score. CJC Open 2021;3(1):71–81. PMID: 33458635.
- Pladet LCA, Barten JMM, Vernooij LM, et al. Prognostic models for mortality risk in patients requiring ECMO. Intensive Care Med 2023;49(2):131–141. PMID: 36600027.
- Zhang ZW, Wan QH. Integrating bibliometric analysis and external validation of the SAVE score in VA-ECMO for refractory cardiogenic shock: insights from global trends and the MIMIC-IV database. J Thorac Dis 2026;18(7):783. PMID: 42583220.
