Extracorporeal survival scores perform poorly in the covid-19 patients

ABSTRACT

Introduction: Over the last two decades several ECMO survival predictions scores have been developed, with varying internal and external validation. We sought to evaluate the performance of six widely available scores on both our local COVID-19 database and a large international multicenter dataset. Methods: Using an institutional dataset encompassing 15 hospitals in a bi-state region and an international dataset of 42 countries, International Severe Acute Respiratory and emerging Infections Consortium (ISARIC), we evaluated the performance of ECMOnet, Respiratory Extracorporeal Membrane Oxygenation Survival Prediction (RESP), PRedicting dEath for SEvere ARDS on VV-ECMO (PRESERVE), Sequential Organ Failure Assessment (SOFA), Roch and PREdiction of Survival on ECMO Therapy-Score (PRESET) scores in identifying ECMO survival for COVID-19 patients. Results: We identified a total of 67 local and 1,014 ISARIC COVID-19 patients supported on ECMO, with a mortality rates of 48% and 51% respectively. In the local cohort all scores demonstrated poor overall performance with area under the receiver operative curve (AUROC) values between 0.53-0.61;ECMOnet 0.54, RESP 0.53, PRESERVE 0.59, Roch 0.53, PRESET 0.61 and SOFA 0.59. The ISARIC database contained fewer variables, allowing 4 scores to be evaluated. Again, all scores demonstrated poor performance in identifying non-survivors with AUROC between 0.55-0.66;ECMOnet 0.59, Roch 0.66, PRESET 0.55 and SOFA 0.59. Conclusions: Current ECMO prediction scores have poor accuracy and limited clinical utility when applied to both local and international databases of COVID-19 patients. Future work should focus on developing clinically applicable models to identify COVID-19 patients most likely to benefit from ECMO.