Comparison of mortality prediction models in acute respiratory distress syndrome undergoing extracorporeal membrane oxygenation and development of a novel prediction score: the PREdiction of Survival on ECMO Therapy-Score (PRESET-Score)

• Main Authors: Michael Hilder, Frank Herbstreit, Michael Adamzik, Martin Beiderlinden, Markus Bürschen, Jürgen Peters, Ulrich H. Frey
• Format: Article
• Language: English
• Published: BMC 2017-12-01
• Series: Critical Care
• Subjects: Acute respiratory distress syndrome; Extracorporeal membrane oxygenation; Mortality; Prediction; Survival
• Online Access: http://link.springer.com/article/10.1186/s13054-017-1888-6

Abstract Background Extracorporeal membrane oxygenation (ECMO) is a life-saving therapy in acute respiratory distress syndrome (ARDS) patients but is associated with complications and costs. Here, we validate various scores supposed to predict mortality and develop an optimized categorical model. Methods In a derivation cohort, 108 ARDS patients (2010–2015) on veno-venous ECMO were retrospectively analysed to assess four established risk scores (ECMOnet-Score, RESP-Score, PRESERVE-Score, Roch-Score) for mortality prediction (receiver operating characteristic analysis) and to identify by multivariable logistic regression analysis independent variables for mortality to yield the new PRESET-Score (PREdiction of Survival on ECMO Therapy-Score). This new score was then validated both in independent internal (n = 82) and external (n = 59) cohorts. Results The median (25%; 75% quartile) Sequential Organ Failure Assessment score was 14 (12; 16), Simplified Acute Physiology Score II was 62.5 (57; 72.8), median intensive care unit stay was 17 days (range 1–124), and mortality was 62%. Only the ECMOnet-Score (area under curve (AUC) 0.69) and the RESP-Score (AUC 0.64) discriminated survivors and non-survivors. Admission pHa, mean arterial pressure, lactate, platelet concentrations, and pre-ECMO hospital stay were independent predictors of death and were used to build the PRESET-Score. The score’s internal (AUC 0.845; 95% CI 0.76–0.93; p < 0.001) and external (AUC 0.70; 95% CI 0.56–0.84; p = 0.008) validation revealed excellent discrimination. Conclusions While our data confirm that both the ECMOnet-Score and the RESP-Score predict mortality in ECMO-treated ARDS patients, we propose a novel model also incorporating extrapulmonary variables, the PRESET-Score. This score predicts mortality much better than previous scores and therefore is a more precise choice for decision support in ARDS patients to be placed on ECMO.