Predictive model of preoperative blood preparation for transfusion in isolated coronary artery bypass graft surgery

Autor: S. Hazli Ramly, N. Bachok, S. Mohd, N.E. Bakri, I.F. Gaaffar, S. Kadiman, Y. Ayob, N. Jaafar, N. Louis, A.S. Kepli
Rok vydání: 2020
Předmět:
Zdroj: Journal of Cardiothoracic and Vascular Anesthesia. 34:S27-S28
ISSN: 1053-0770
DOI: 10.1053/j.jvca.2020.09.039
Popis: Introduction Patients undergoing isolated coronary artery bypass graft (CABG) at our heart centre require two packed red blood cells (RBC) pre-operatively based on the data from Maximum Surgical Blood Order Schedule (MSBOS). Transfusion-related mortality and morbidity have been shown to be dose dependent, suggesting that the lowest effective number of units should be transfused. Pre-operative identification of patient's characteristic requiring blood transfusion during procedure will allow for better utilization of blood, reduce wastage and increase cost efficiency. We aim to define a set of parameters to indicate necessity for blood transfusion hence, consideration will be given to supply one instead of two packed RBC to patients who do not require transfusion. Methods This cross-sectional study obtained integrated data-set from the hospital information system and clinical database (Cardio thoracic Surgery Registry) from January 2015 to December 2018. A total of 4,646 patients who underwent isolated CABG were included in this study and 28 independent predictors associated with transfusion were identified and analyzed. A predictive model for preoperative requirement of blood transfusion was developed using logistic regression and machine learning approach ie; Decision tree, Logistic Regression, KNN, Naive Bayes, Gradient Boosted Tree, and Random Forest using RapidMiner. Results In 2015-2018, only 25.7% of patients utilized two packed RBC's whereas 20.2% utilized only one for blood transfusion during procedure. The remainder 54.1% of non-utilized blood were returned to the blood bank within 48 hours. Logistic regression and machine learning algorithm unveiled similar significant predictors to predict the requirement of blood transfusion pre-operatively. There are 24 significant predictors identified from the highest accuracy algorithm which is Random Forest (99.78%), as compared to others machine learning approaches particularly the Decision Tree 79.55%, Logistic Regression 85.25, KNN 75.35%, Naive Bayes 79.76%, and Gradient Boosted Tree 83.32%. Discussion A clinical predictive model was developed for preoperative risk stratification of patients requiring blood transfusion and to further assist clinicians for better utilization of blood bank resources. Variables significantly associated with blood transfusion were female gender, EF 40 serum creatinine>100mmol/L, and previous cardiac procedures. This tool may prospectively optimize the ordering of blood with the potential to reduce adverse events and costs related to transfusion in the future. However, for the consideration to supply only one packed RBC in retrospect to patients who do not require transfusion, more data and predictors should be added in this study.
Databáze: OpenAIRE
Externí odkaz: