Extracorporeal membrane oxygenation (ECMO) is a critical life-support technique for patients with severe medical conditions. Various factors affect the mortality rates of patients in intensive care units, presenting a significant clinical challenge in accurately predicting outcomes based on a limited set of indicators.

Extracorporeal Membrane Oxygenation (ECMO) is used to provide continuous extracorporeal respiratory and circulatory support for patients with severe cardiopulmonary failure. It is the most important life support method in critical care medicine, and also one of the most complex and expensive treatment methods in intensive care unit (ICU). It takes a lot of resources to maintain. Therefore, it is particularly important to strictly grasp the indications of patients and accurately predict the prognosis of patients to assist clinical decision-making.

Several previous published studies have used clinical scores to predict the prognosis of ECMO patients, but most of them focused on ECMO outcomes in specific patient groups, such as adult respiratory distress syndrome(ARDS), respiratory failure, lung transplantation, cardiogenic shock, and so on. In addition, most of these estimates were calculated using traditional statistical methods and have limited fitting power for data sets with more characteristic variables.

Artificial Intelligence (AI) and Machine Learning (ML) provide a more advanced alternative to traditional statistical methods, and have unparalleled advantages in dealing with data sets with high-dimensional characteristic variables and nonlinear data. And it can self-iterate to improve model performance. In addition, ML, which can process information based on causal or statistical data, may reveal hidden dependencies between clinical indicators and disease prognosis and support clinical decision making, has emerged as the best alternative The primary outcome measures discharged alive from the hospital and died during hospitalization. A total of 69 clinical characteristic indicators were identified based on relevant literature and insights from ECMO experts in critical care medicine. These indicators included demographic data such as age, height, weight, and the medical history of ECMO patients. Additionally, infection indicators were assessed within 24 hours prior to the initiation of ECMO support and within 24 hours after its discontinuation. Furthermore, indicators pertaining to cardiac, renal, and hepatic function, as well as the type of shock (distributive shock, hypovolemic shock, cardiogenic shock, obstructive shock), were included. In addition, the average daily liquid volume within three days after the initiation of ECMO support, the duration of ECMO support, and the ICU length of stay were also considered.

Inclusion Criteria:

• All patients who underwent ECMO in our hospital and were registered in the CSECLS registry database (ClinicalTrials.gov Identifier:NCT04158479) from January 1, 2018 to now were retrospectively collected.

Exclusion Criteria:

• ECMO was discontinued for non-medical reasons.
• Under 18 years of age.

The study involved patients who received ECMO treatment at Zhujiang Hospital of Southern Medical University from 2018 to 2024. Exclusion criteria included patients who received ECMO support at outside hospitals and those under the age of 18. Of the patients, 100 were discharged alive, while 190 died during hospitalization.