PRESET Score for survival on ECMO

Clinical background

The decision to start veno-venous ECMO in severe ARDS is resource-intensive and carries a substantial risk of complications. Patient selection is decisive for the outcome, but there is no consensus on how to make it. Several prediction scores have been published, including RESP, PRESERVE, ECMOnet and Roch, but they vary in the variables they include and often lack adequate external validation. The PRESET score was developed to offer a simple, categorical instrument based on extrapulmonary predictors available immediately before cannulation, and as a complement to lung-focused scores such as RESP [1].

Calculating the PRESET score

The PRESET score is the sum of five variables measured immediately before ECMO is started:

$$\text{PRESET} = P_{\text{MAP}} + P_{\text{lactate}} + P_{\text{pH}} + P_{\text{platelets}} + P_{\text{hospital days}}$$

where each component scores within the following ranges:

The total score ranges from 0 to 15. The variables are categorised with thresholds taken from the Wald statistics of the multivariable logistic regression, in which each variable's relative contribution to the model determines its point weight [1].

The derivation cohort consisted of 108 consecutive ARDS patients who received veno-venous ECMO at University Hospital Essen, Germany, between December 2009 and February 2015. The median SOFA score was 14 (interquartile range 12–16), the median SAPS II 62.5 (57–72.8), the median ICU stay 17 days (1–124) and ICU mortality was 62%. The leading causes of ARDS were bacterial pneumonia (45%) and H1N1 influenza (19%). Of the patients, 86% were cannulated at the referring hospital and transported on ECMO [1].

The five variables were identified by entering 11 variables with a univariate p ≤ 0.1 into a backward stepwise logistic regression. Lactate, hospital days before ECMO, mean arterial pressure, platelets and arterial pH were independently associated with ICU mortality and formed the basis of the score [1].

Interpretation in practice

PRESET divides patients into three risk classes with clearly differentiated mortality. In the derivation cohort the distribution was as follows:

Class I identifies patients with low mortality in whom ECMO is justified and the expected outcome is favourable. Class III indicates very high mortality (93% in the derivation) and should prompt a critical discussion of whether ECMO is meaningful, particularly if other limiting factors are present such as end-stage lung disease without the possibility of transplantation. Class II is the broad middle group in which the score alone cannot determine the decision and overall clinical assessment must be weighed in [1].

Median survival time in the derivation cohort differed markedly between the classes: class I 58 days (31–undefined), class II 23 days (13–40), class III 12 days (1–16) [1].

Validation and performance

In the derivation cohort, PRESET showed an AUC of 0.823 (95% CI 0.74–0.90, p < 0.001) and outperformed all the scores compared (ECMOnet AUC 0.69, RESP AUC 0.64, PRESERVE and Roch lower) in pairwise ROC comparison [1].

Internal validation was performed prospectively in 82 consecutive patients at the same centre (February 2015 to January 2017, mortality 55%). Discrimination was excellent, with an AUC of 0.845 (95% CI 0.76–0.93, p < 0.001). Mortality by class was 14%, 67% and 91% respectively, confirming the pattern of the derivation cohort [1].

External validation was carried out in 59 patients at Marienhospital Osnabrück, another German centre (February 2013 to December 2015, mortality 47%). The AUC fell to 0.70 (95% CI 0.56–0.84, p = 0.008). Calibration remained good (Hosmer–Lemeshow χ² = 4.2, p = 0.655). Mortality by class was 27%, 50% and 70% respectively, showing that class III still identifies high risk but with a lower absolute mortality than in the derivation [1].

Later external validations have shown mixed results. In a COVID-19 cohort (n = 105, University of Maryland), the mean PRESET score was significantly lower in survivors (6.03 versus 8.11, p < 0.001), and patients with a PRESET ≤6 had 97.7% survival compared with 32.5% for those with a PRESET ≥8. Multivariable logistic regression showed an OR of 2.84 (95% CI 1.75–4.63) for survival per point lower [2]. In another large German cohort (n = 283, 2012–2022), discrimination was weak for all the scores tested, and the authors advise against using them alone for patient selection [3]. A Czech multicentre study in influenza A ARDS found an AUC of 0.57 for PRESET, considerably worse than APACHE II in the same cohort [4]. In a large international COVID-19 cohort (n = 1,147), the AUROC for PRESET lay between 0.58 and 0.62, on a par with the other ECMO scores, with inadequate calibration [5].

In summary, PRESET performs best in the population from which it was derived: bacterial and viral pneumonia with a high burden of organ failure at German ECMO centres. In COVID-19 cohorts and other aetiologies, discrimination falls appreciably.

Limitations

PRESET was derived exclusively in veno-venous ECMO and does not apply to veno-arterial ECMO or cardiogenic shock. Contraindications such as end-stage lung disease without the possibility of transplantation were excluded at the derivation stage and are not part of the score [1].

All five variables are extrapulmonary. The score captures organ failure and circulatory compromise but contains no lung-specific parameters such as the oxygenation index, PaCO₂ or ventilator settings. This is a deliberate design that makes PRESET a complement to, not a replacement for, lung-focused scores such as RESP. A patient with extreme hypoxaemia but a normal lactate, MAP and pH may have a low PRESET score despite life-threatening respiratory failure [1].

The derivation cohort is small (n = 108) and single-centre, with a high proportion of interhospital ECMO transport (86%), which limits generalisability. The external validation was likewise small (n = 59) and confined to German centres. Later validations in COVID-19 and influenza A have shown poorer performance, particularly in discrimination [3, 4, 5].

The score is static and measured at one point in time. The variables can change rapidly in the ICU, and a patient initially in class II may shift to class III within hours through a rising lactate or a falling pH. The score should therefore be seen as a snapshot at the time of the decision, not as a prognostic marker during ongoing ECMO treatment.

References

1. Hilder M, Herbstreit F, Adamzik M et al. Comparison of mortality prediction models in acute respiratory distress syndrome undergoing extracorporeal membrane oxygenation and development of a novel prediction score: the PRESET-Score. Critical Care 2017. PMID: 29233160
2. Powell EK, Lankford AS, Ghneim M et al. Decreased PRESET-Score corresponds with improved survival in COVID-19 veno-venous extracorporeal membrane oxygenation. Perfusion 2023. PMID: 36114156
3. Mazuru V, Mang S, Ajouri J et al. External Validation of the PREdiction of Survival on Extracorporeal Membrane Oxygenation Therapy (PRESET) Score: A Single-Center Cohort Experience. ASAIO Journal 2024. PMID: 38728743
4. Maca J, Matousek V, Bursa F et al. Extracorporeal membrane oxygenation survival: External validation of current predictive scoring systems focusing on influenza A etiology. Artificial Organs 2021. PMID: 33534922
5. Shah N, Xue B, Xu Z et al. Validation of extracorporeal membrane oxygenation mortality prediction and severity of illness scores in an international COVID-19 cohort. Artificial Organs 2023. PMID: 37032544