Predicting the required pre-surgery blood volume in surgical patients based on machine learning

Abstract

Precisely predicting the required pre-surgery blood volume (PBV) in surgical patients is a formidable challenge in China. Inaccurate estimation is associate with excessive costs, postponed surgeries and adverse outcome after surgery due to in sufficient supply or inventory. This study aimed to predict required PBV based on machine learning techniques. 181,027 medical documents over 6 years were cleaned and finally obtained 92,057 blood transfusion records. The blood transfusion and surgery related factors of perioperative patients, surgeons experience volumes and the actual volumes of transfused RBCs were extracted. 6 machine learning algorithms were used to build prediction models. The surgery patients received allogenic RBCs or without transfusion, had total volume less than 10 units, or had the latest laboratory examinations of pre-surgery within 7 days were included, providing 118,823 data points. 39 predictive factors related to the RBCs transfusion were identified. Random forest model was selected to predict the required PBV of RBCs with 72.9% accuracy and strikingly improved the accuracy by 30.4% compared with surgeons experience, where 90% of data was used for training. We tested and demonstrated that both the data-driven models and the random forest model achieved higher accuracy than surgeons experience. Furthermore, we developed a computational tool, PTRBC, to precisely estimate the required PBV in surgical patients and we believe this tool will find more applications in assisting clinician decisions, not only confined to making accurate pre-surgery blood requirement predicting.