Postoperative Discharge Destination Classification via Extreme Gradient Boosting

Abstract

Hypothermia is a medical crisis which arises when a patient’s body significantly loses heat rather than producing heat. It is a vital concern after surgery, exerting multiple effects such as impairment of innate immunity leading to an increase complication and mortality risk. Postoperative care received after a surgical procedure includes discharge destination whether the patient is to be delivered to an intensive care unit (ICU), to be delivered to the general hospital floor or the patient is already allowed to be sent home. Since postoperative hypothermia is a serious risk, discharge decision corresponds roughly to patient’s body temperature measurements. In this paper, we discuss the utilization of extreme gradient boosting (XGBoost) algorithm–a variant of gradient boosted algorithm and an ensemble of classification and regression tree, for classifying patient’s postoperative discharge destination. Patients’ data, including core and surface temperatures, blood pressure, blood oxygen level, stability of patient’s internal and surface temperatures, stability of patient’s blood pressure and postoperative perceived comfort are used as input features in formulating the XGBoost model. Experimental results show high performance of the formulated XGBoost model (accuracy =0.S947, kappa coefficient =0.S407, f-score =0.90) in classifying postoperative patients’ discharge destination compared to the methods used as discussed in section II. A ranking of feature importance is presented in the latter part of this paper.