The Prediction of Complications of Blood Transfusion in Thalassemia Patients Using Deep Learning Method

Abstract

<strong>Background and purpose:</strong> Thalassemia is the acute hereditary anemia and the most common hemoglobin disorder in the world. The main treatment for this disease is the persistent blood injection, but the injection of blood can have different complications. These complications affect the quality of life of patients and increase the risk of mortality. Moreover, it increases the use of healthcare services and hospital costs. Predicting the risk of complications before blood transfusion, more appropriate alternative treatment can be selected to prevent or reduce the complications. Moreover, identifying high-risk patients and following them after transfusion provides the possibility of timely interventions. So far, several studies have analyzed the effects of blood transfusion and the risk factors of these complications by statistical methods. However, few studies have attempted to predict these complications. In this study, the risk of post-transfusion complications in thalassemia patients is predicted using machine learning algorithms. <strong>Method:</strong> The cross-sectional data were collected from 3489 cases in 12 thalassemia centers in Tehran province and 14 thalassemia centers in Mazandaran province in 2018. A set of different classification models including classic and deep learning techniques were trained and studied on this data set. <strong>Results:</strong> The results showed that machine learning methods have good accuracy to predict the risk of post-transfusion complications. According to the results, the deep learning method has improved the results considerably in comparison to other models (precision=0.21, sensitivity=0.77, f1-score=0.33). <strong>Conclusion:</strong> In this study, machine learning methods were used to predict the occurrence of post-transfusion complications in thalassemia patients. Finally, the deep learning method produced the best prediction results. Using this method, of patients who will suffer complications are detected before transfusions. Appropriate alternative methods can be used for treating these patients in order to prevent or reduce transfusion complications.