Association between blood transfusion and early mortality in patient undergoing extracorporeal membrane oxygenation

TL;DR

This study finds that higher blood transfusion volumes in ECMO patients are linked to increased 90-day mortality, suggesting the need for optimized transfusion practices.

Contribution

The study identifies blood transfusion volumes normalized by hospital stay as a critical predictor of mortality in ECMO patients using machine learning and statistical models.

Findings

Higher transfusion volumes of RBC, FFP, and PC were consistently associated with increased 90-day mortality in ECMO patients.

RBC, FFP, and PC transfusions were among the top 10 predictors of mortality in machine learning models.

Multivariable logistic regression confirmed a significant association between transfusion volumes and mortality (p < 0.001).

Abstract

Blood transfusions are frequently administered to extracorporeal membrane oxygenation (ECMO) patients. However, the relationship between blood transfusion and outcomes in ECMO patients remains unclear. This study investigated the association between blood transfusion and 90-day mortality in ECMO patients. Using a large administrative database (National Health Insurance Service, NHIS) from the Republic of Korea during 2014–2020, ECMO cases were identified. Patients younger than 19 years and those with ECMO durations of less than 1 day were excluded. Transfusion variables included the total volumes of red blood cells (RBC), fresh frozen plasma (FFP), and platelet concentrates (PC), normalized by hospital stays. The primary outcome was 90-day mortality. Variable importance calculation using Random Forest and XGBoost models identified RBC, FFP, and PC transfusions among the top 10 predictors of mortality. Variable importance was calculated to identify the most influential variables in predicting the outcome among 51 variables. A total of 11,874 patients were included. The median age was 62 years (IQR 51–71), and 32.7% were female. 90-day mortality occurred in 6980 patients (58.8%). The proportion of patients who received at least one transfusion during hospitalization was 96.2% for RBC, 68.8% for FFP, and 74.4% for PC. Higher transfusion volumes (50–100%) were consistently associated with an increased risk of death compared to lower volumes (0–49%). Odds ratios for mortality were 6.764 (95% confidence interval (CI), 6.227–7.347) for RBC/hospital day, 3.702 (95% CI 3.426–4.000) for FFP/hospital day, and 5.082 (95% CI 4.691–5.506) for PC/hospital day in the higher transfusion group. In a multivariable logistic regression model, RBC, FFP, and PC transfusions normalized by hospital stay were significantly associated with mortality (p < 0.001). Variable importance calculation using Random Forest and XGBoost models identified RBC, FFP, and PC transfusions among the top 10 predictors of mortality. Blood transfusion normalized by hospital stay emerged as a critical predictor of 90-day mortality in ECMO patients. Thus, optimizing blood transfusion practices is essential to improve patient outcomes and safety.

The online version contains supplementary material available at 10.1038/s41598-025-11702-7.

Figures