AI Driven Prognosis in Pediatric Bone Marrow Transplantation Survival

Pediatric Bone Marrow Transplantation (BMT) is widely used as a treatment innovation that can treat certain types of cancers and hematologic disorders in children. Nevertheless, it is quite worrying that despite all the scientific breakthroughs in medical field, the survival rates for post BMT pediatric patients are still very low. This work aims to improve the probabilistic forecast of pediatric BMT survival rates by applying AI and ML approaches. Having a large dataset of demographic and clinical characteristics of pediatric patients, we subjected the data to an elaborate data cleaning process. This included handling of missing records, converting categorical variables into dummy and dealing with an uneven distribution of the survival status using the Borderline SMOTE method. Then we used mutual information for selecting the features, which helped in the elimination of the non-relevant characteristics. The selected features along with the full features dataset were fed to the ML models namely Random Forest, XGBoost, Logistic Regression, Decision Tree and Support Vector Classifier through Hyperparameter optimization. The results of this study showed that the XGBoost model was the most efficient in identifying the survival status of the pediatric patients after BMT. By facilitating more accurate predictions of survival outcomes, we can equip healthcare professionals with the insights necessary to make more informed clinical decisions, thereby potentially enhancing survival rates for children undergoing BMT.