**Abstract**

Over the five decades since the first successful reports of extracorporeal membrane oxygenation (ECMO) use, ideal patient selection has been an ongoing question. This has led to the development of several prognostication tools aimed at identifying risk factors associated with poor outcomes. These have spanned neonatal, pediatric and adult patients supported on ECMO for cardiac or respiratory failure. The majority of these scores have focused on mortality as an objective poor outcome with only 2 adult scores looking at long-term neuropsychological outcomes in ECMO survivors. In the development of these scores the authors have mainly relied on registry style data with limited granularity and focused on immediate pre-ECMO data points without incorporation of the evolving patient trajectories leading up to ECMO cannulation. While such scores can be useful in both prognostication and as risk stratification and quality assessment tools, they all lack practicality on an individual patient level with regards to decision making, as these scores have all been developed on data from patients already supported on ECMO without a comparable control cohort, to truly mimic decision making at the bedside. In this chapter we review the currently available ECMO prognostication scores, their limitations and potential future directions.

**Keywords:** ECMO, predictive scores, mortality, predictive analytics, machine learning
