Open Access

Seyedmostafa Sheikhalishahi, Sebastian Goss, Lea K. Seidlmayer, Sarra Zaghdoudi, Ludwig C. Hinske, Mathias Kaspar

• Background Blood transfusion (BT) is a critical aspect of medical care for surgical patients in the Intensive Care Unit (ICU). Timely and accurate identification of BT needs can enhance patient outcomes and healthcare resource management. Methods This study aims to determine whether a machine learning (ML) model can be trained to predict the need for blood transfusion (BT) in patients on the ICU after a wide range of surgeries, utilizing only data from the ICU. Results This retrospective study analyzed data from 9,118 surgical ICU patients from the Amsterdam University Medical Centers database (UMCdb). The study included a primary analysis using data from 6 h before ICU admission up to 1, 2, 3, and 6 h after admission, and a secondary analysis using only the data from 6 h before ICU admission and only the data from the first hour after admission. The model integrated 32 relevant clinical variables and compared the performance of XGBoost and logistic regression (LR) algorithms.Background Blood transfusion (BT) is a critical aspect of medical care for surgical patients in the Intensive Care Unit (ICU). Timely and accurate identification of BT needs can enhance patient outcomes and healthcare resource management. Methods This study aims to determine whether a machine learning (ML) model can be trained to predict the need for blood transfusion (BT) in patients on the ICU after a wide range of surgeries, utilizing only data from the ICU. Results This retrospective study analyzed data from 9,118 surgical ICU patients from the Amsterdam University Medical Centers database (UMCdb). The study included a primary analysis using data from 6 h before ICU admission up to 1, 2, 3, and 6 h after admission, and a secondary analysis using only the data from 6 h before ICU admission and only the data from the first hour after admission. The model integrated 32 relevant clinical variables and compared the performance of XGBoost and logistic regression (LR) algorithms. Conclusions The model demonstrated an effective BT prediction, with XGBoost outperforming LR, particularly for a 12-hour prediction window. Notable differences in patient characteristics were observed among those who received BT and those who did not receive BT. The study establishes the feasibility of using ML for the prediction of BT in surgical ICU patients. It underlines the potential of ML models as decision support tools in healthcare, enabling early identification of BT needs.…