Mortality prediction models in acute respiratory failure treated with extracorporeal membrane oxygenation: it must be firstly designed for clinicians and beside use

Abstract

In a recent issue of Critical Care, we read with great interest the article by Enger and colleagues [1] regarding the development of novel mortality prediction models on extracorporeal membrane oxygenation (ECMO) in acute respiratory failure and the comparison of its performance with the ECMOnet (Extracorporeal Membrane Oxygenation Network) [2], PRESERVE (Predicting Death for Severe Acute Respiratory Distress Syndrome on Veno-venous Extracorporeal Membrane Oxygenation) [3], and Sequential Organ Failure Assessment scores. We commend the efforts made by the authors to externally validate the ECMOnet and the PRESERVE scores, but we recognize that a problem common to all prediction models, including their own, is that inevitably performance is better in the dataset from which they were derived. A potential limitation of the approach taken by Enger and colleagues is that the calculation of predicted mortality requires the use of a computer-generated algorithm which may limit immediate ease of use and applicability by clinicians. Prediction scores (such as the ECMOnet and the PRESERVE) which contain only categorical variables may be better for beside use. Derivation of a score from the pre-ECMO model and the model day 1 (which contained 4 out of 5 and 7 out of 8 continuous predictors) in the study by Enger and colleagues would have further decreased the discrimination of both models [4]. The recently published Respiratory Extracorporeal Membrane Oxygenation Survival Prediction (RESP) score, derived from over 2,000 patients from the Extracorporeal Life Support Organization (ELSO) registry [5,6], may provide another mortality prediction tool which could also be assessed. Lastly, peak inspiratory pressure (PIP) was used as a substitute for plateau pressure in the validation of the PRESERVE score. It would have been helpful of the authors to provide the PIP threshold used. Substitution of the same threshold for PIP as used for the plateau pressure (that is, 30 cmH2O) would have led to an overestimated low pre-ECMO compliance and potentially impaired performance of the PRESERVE score in this cohort.